[PATCH] drm/amdgpu/umsch: add suspend and resume callback

[Lang Yu]

Add missing IP callbacks.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c | 16 
 1 file changed, 16 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c
index 4bd076e9e367..f5fdde5181f7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umsch_mm.c
@@ -844,6 +844,20 @@ static int umsch_mm_hw_fini(void *handle)
return 0;
 }
 
+static int umsch_mm_suspend(void *handle)
+{
+   struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+   return umsch_mm_hw_fini(adev);
+}
+
+static int umsch_mm_resume(void *handle)
+{
+   struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+
+   return umsch_mm_hw_init(adev);
+}
+
 static const struct amd_ip_funcs umsch_mm_v4_0_ip_funcs = {
.name = "umsch_mm_v4_0",
.early_init = umsch_mm_early_init,
@@ -852,6 +866,8 @@ static const struct amd_ip_funcs umsch_mm_v4_0_ip_funcs = {
.sw_fini = umsch_mm_sw_fini,
.hw_init = umsch_mm_hw_init,
.hw_fini = umsch_mm_hw_fini,
+   .suspend = umsch_mm_suspend,
+   .resume = umsch_mm_resume,
 };
 
 const struct amdgpu_ip_block_version umsch_mm_v4_0_ip_block = {
-- 
2.25.1

[PATCH v2] drm/amd/swsmu: update smu v14_0_0 header files and metrics table

[Li Ma]

Update driver if, pmfw and ppsmc header files.
Add new gpu_metrics_v3_0 for metrics table updated in driver if
and reserve legacy metrics table to maintain backward compatibility.
---
v1:
Update header files and add gpu_metrics_v3_0.
v2:
Update smu_types.h, smu headers and drop smu_cmn_get_smc_version in smu v14_0_0.

Signed-off-by: Li Ma 
Reviewed-by: Yifan Zhang 
Reviewed-by: Kenneth Feng 
Acked-by: Alex Deucher 
---
 .../gpu/drm/amd/include/kgd_pp_interface.h|  68 
 .../inc/pmfw_if/smu14_driver_if_v14_0_0.h |  69 ++--
 .../pm/swsmu/inc/pmfw_if/smu_v14_0_0_pmfw.h   |  29 +-
 .../pm/swsmu/inc/pmfw_if/smu_v14_0_0_ppsmc.h  |   5 +-
 drivers/gpu/drm/amd/pm/swsmu/inc/smu_types.h  |   3 +-
 .../drm/amd/pm/swsmu/smu14/smu_v14_0_0_ppt.c  | 318 ++
 6 files changed, 371 insertions(+), 121 deletions(-)

diff --git a/drivers/gpu/drm/amd/include/kgd_pp_interface.h 
b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
index 1d4620322a15..c71e0415e82e 100644
--- a/drivers/gpu/drm/amd/include/kgd_pp_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_pp_interface.h
@@ -1057,4 +1057,72 @@ struct gpu_metrics_v2_4 {
uint16_taverage_soc_current;
uint16_taverage_gfx_current;
 };
+
+struct gpu_metrics_v3_0 {
+   struct metrics_table_header common_header;
+
+   /* Temperature */
+   /* gfx temperature on APUs */
+   uint16_ttemperature_gfx;
+   /* soc temperature on APUs */
+   uint16_ttemperature_soc;
+   /* CPU core temperature on APUs */
+   uint16_ttemperature_core[16];
+   /* skin temperature on APUs */
+   uint16_ttemperature_skin;
+
+   /* Utilization */
+   /* time filtered GFX busy % [0-100] */
+   uint16_taverage_gfx_activity;
+   /* time filtered VCN busy % [0-100] */
+   uint16_taverage_vcn_activity;
+   /* time filtered IPU per-column busy % [0-100] */
+   uint16_taverage_ipu_activity[8];
+   /* time filtered per-core C0 residency % [0-100]*/
+   uint16_taverage_core_c0_activity[16];
+   /* time filtered DRAM read bandwidth [GB/sec] */
+   uint16_taverage_dram_reads;
+   /* time filtered DRAM write bandwidth [GB/sec] */
+   uint16_taverage_dram_writes;
+
+   /* Driver attached timestamp (in ns) */
+   uint64_tsystem_clock_counter;
+
+   /* Power/Energy */
+   /* average dGPU + APU power on A + A platform */
+   uint32_taverage_socket_power;
+   /* average IPU power [W] */
+   uint16_taverage_ipu_power;
+   /* average APU power [W] */
+   uint32_taverage_apu_power;
+   /* average dGPU power [W] */
+   uint32_taverage_dgpu_power;
+   /* sum of core power across all cores in the socket [W] */
+   uint32_taverage_core_power;
+   /* calculated core power [W] */
+   uint16_tcore_power[16];
+   /* maximum IRM defined STAPM power limit [W] */
+   uint16_tstapm_power_limit;
+   /* time filtered STAPM power limit [W] */
+   uint16_tcurrent_stapm_power_limit;
+
+   /* Average clocks */
+   uint16_taverage_gfxclk_frequency;
+   uint16_taverage_socclk_frequency;
+   uint16_taverage_vpeclk_frequency;
+   uint16_taverage_ipuclk_frequency;
+   uint16_taverage_fclk_frequency;
+   uint16_taverage_vclk_frequency;
+
+   /* Current clocks */
+   /* target core frequency */
+   uint16_tcurrent_coreclk[16];
+   /* CCLK frequency limit enforced on classic cores [MHz] */
+   uint16_tcurrent_core_maxfreq;
+   /* GFXCLK frequency limit enforced on GFX [MHz] */
+   uint16_tcurrent_gfx_maxfreq;
+
+   /* Metrics table alpha filter time constant [us] */
+   uint32_ttime_filter_alphavalue;
+};
 #endif
diff --git a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu14_driver_if_v14_0_0.h 
b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu14_driver_if_v14_0_0.h
index cb6948a0b262..b483c8e096e7 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu14_driver_if_v14_0_0.h
+++ b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu14_driver_if_v14_0_0.h
@@ -149,23 +149,37 @@ typedef struct {
   uint32_t MaxGfxClk;
 } DpmClocks_t;
 
-
-// Throttler Status Bitmask
-#define THROTTLER_STATUS_BIT_SPL  0
-#define THROTTLER_STATUS_BIT_FPPT 1
-#define 

Re: [PATCH] drm/amdgpu/pm: update SMU 13.0.0 PMFW version check

[Zhang, Hawking]

[AMD Official Use Only - General]

Reviewed-by: Hawking Zhang 

Regards,
Hawking

Get Outlook for iOS

From: amd-gfx  on behalf of Alex Deucher 

Sent: Friday, October 13, 2023 3:12:47 AM
To: amd-gfx@lists.freedesktop.org 
Cc: Deucher, Alexander 
Subject: [PATCH] drm/amdgpu/pm: update SMU 13.0.0 PMFW version check

Update the PMFW version check the the ROCm optimizations.

Signed-off-by: Alex Deucher 
---
 drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c 
b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
index 56b11f6386df..cb7e1a22d7ef 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
@@ -2562,7 +2562,7 @@ static int smu_v13_0_0_set_power_profile_mode(struct 
smu_context *smu,
 if (smu->power_profile_mode == PP_SMC_POWER_PROFILE_COMPUTE &&
 (amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 
0)) &&
 ((smu->adev->pm.fw_version == 0x004e6601) ||
-(smu->adev->pm.fw_version >= 0x004e7300))) {
+(smu->adev->pm.fw_version >= 0x004e7400))) {
 workload_type = smu_cmn_to_asic_specific_index(smu,

CMN2ASIC_MAPPING_WORKLOAD,

PP_SMC_POWER_PROFILE_POWERSAVING);
--
2.41.0

Re: [PATCH] drm/amdgpu: Add EXT_COHERENT override for NUMA local nodes

[Philip Yang]

  


On 2023-10-12 12:39, David Francis
  wrote:


  On NUMA systems, local memory gets the local mtype, set by an
override callback. If EXT_COHERENT is set, memory will be set as
MTYPE_UC by default, with local memory MTYPE_CC.

Add an option in the override function for this case, and
add a check to ensure it is not used on UNCACHED memory.

Signed-off-by: David Francis 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 13 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|  8 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 33 +++
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c  |  6 ++---
 5 files changed, 41 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 46d27c87..189341944bf1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -761,6 +761,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * @immediate: immediate submission in a page fault
  * @unlocked: unlocked invalidation during MM callback
  * @flush_tlb: trigger tlb invalidation after update completed
+ * @allow_override: change MTYPE for local NUMA nodes
  * @resv: fences we need to sync to
  * @start: start of mapped range
  * @last: last mapped entry
@@ -777,7 +778,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * 0 for success, negative erro code for failure.
  */
 int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-			   bool immediate, bool unlocked, bool flush_tlb,
+			   bool immediate, bool unlocked, bool flush_tlb, bool allow_override,
 			   struct dma_resv *resv, uint64_t start, uint64_t last,
 			   uint64_t flags, uint64_t offset, uint64_t vram_base,
 			   struct ttm_resource *res, dma_addr_t *pages_addr,
@@ -815,6 +816,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	params.immediate = immediate;
 	params.pages_addr = pages_addr;
 	params.unlocked = unlocked;
+	params.allow_override = allow_override;
 
 	/* Implicitly sync to command submissions in the same VM before
 	 * unmapping. Sync to moving fences before mapping.
@@ -1062,6 +1064,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va,
 		trace_amdgpu_vm_bo_update(mapping);
 
 		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,
+	   !(bo->flags & AMDGPU_GEM_CREATE_UNCACHED),
 	   resv, mapping->start, mapping->last,
 	   update_flags, mapping->offset,
 	   vram_base, mem, pages_addr,
@@ -1257,8 +1260,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
 		mapping->start < AMDGPU_GMC_HOLE_START)
 			init_pte_value = AMDGPU_PTE_DEFAULT_ATC;
 
-		r = amdgpu_vm_update_range(adev, vm, false, false, true, resv,
-	   mapping->start, mapping->last,
+		r = amdgpu_vm_update_range(adev, vm, false, false, true, false,
+	   resv, mapping->start, mapping->last,
 	   init_pte_value, 0, 0, NULL, NULL,
 	   );
 		amdgpu_vm_free_mapping(adev, vm, mapping, f);
@@ -2546,8 +2549,8 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, u32 pasid,
 		goto error_unlock;
 	}
 
-	r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr,
-   addr, flags, value, 0, NULL, NULL, NULL);
+	r = amdgpu_vm_update_range(adev, vm, true, false, false, false,
+   NULL, addr, addr, flags, value, 0, NULL, NULL, NULL);
 	if (r)
 		goto error_unlock;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 6e71978db13f..9ea8b5b644e3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -244,6 +244,12 @@ struct amdgpu_vm_update_params {
 	 * @table_freed: return true if page table is freed when updating
 	 */
 	bool table_freed;
+
+	/**
+	 * @allow_override: true for memory that is not uncached: allows MTYPE
+	 * to be overridden for NUMA local memory.
+	 */
+	bool allow_override;
 };
 
 struct amdgpu_vm_update_funcs {
@@ -436,7 +442,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
 			struct amdgpu_vm *vm, struct amdgpu_bo *bo);
 int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-			   bool immediate, bool unlocked, bool flush_tlb,
+			   bool immediate, bool unlocked, bool flush_tlb, bool allow_override,
 			   struct dma_resv *resv, uint64_t start, uint64_t last,
 			   uint64_t flags, uint64_t offset, uint64_t vram_base,
 			   struct ttm_resource *res, dma_addr_t *pages_addr,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
index 9b025fd17b84..80273809c93f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
@@ -842,7 +842,7 @@ static void amdgpu_vm_pte_update_flags(struct amdgpu_vm_update_params 

Re: [PATCH 1/2] drm/amdgpu: correct NBIO v7.11 programing

[Deucher, Alexander]

[Public]

Acked-by: Alex Deucher 

From: Yu, Lang 
Sent: Thursday, October 12, 2023 3:31 AM
To: amd-gfx@lists.freedesktop.org 
Cc: Deucher, Alexander ; Zhang, Yifan 
; Yu, Lang 
Subject: [PATCH 1/2] drm/amdgpu: correct NBIO v7.11 programing

Use v7.7 before, switch to v7.11 now.
Fix incorrect programing.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c   | 56 +--
 .../asic_reg/nbio/nbio_7_11_0_offset.h|  9 ++-
 2 files changed, 33 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c 
b/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
index 6873eead1e19..3a94f249929e 100644
--- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
+++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
@@ -66,19 +66,19 @@ static void nbio_v7_11_sdma_doorbell_range(struct 
amdgpu_device *adev, int insta
   bool use_doorbell, int 
doorbell_index,
   int doorbell_size)
 {
-   u32 reg = SOC15_REG_OFFSET(NBIO, 0, regGDC0_BIF_SDMA0_DOORBELL_RANGE);
+   u32 reg = SOC15_REG_OFFSET(NBIO, 0, regGDC0_BIF_CSDMA_DOORBELL_RANGE);
 u32 doorbell_range = RREG32_PCIE_PORT(reg);

 if (use_doorbell) {
 doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
OFFSET, doorbell_index);
 doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
SIZE, doorbell_size);
 } else {
 doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
SIZE, 0);
 }

@@ -145,27 +145,25 @@ static void nbio_v7_11_enable_doorbell_aperture(struct 
amdgpu_device *adev,
 static void nbio_v7_11_enable_doorbell_selfring_aperture(struct amdgpu_device 
*adev,
 bool enable)
 {
-/* u32 tmp = 0;
+   u32 tmp = 0;

 if (enable) {
-   tmp = REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+   tmp = REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
 DOORBELL_SELFRING_GPA_APER_EN, 1) |
-   REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+ REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
 DOORBELL_SELFRING_GPA_APER_MODE, 1) |
-   REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+ REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
 DOORBELL_SELFRING_GPA_APER_SIZE, 0);

 WREG32_SOC15(NBIO, 0,
-   regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_BASE_LOW,
+   regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_BASE_LOW,
 lower_32_bits(adev->doorbell.base));
 WREG32_SOC15(NBIO, 0,
-   regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_BASE_HIGH,
+   regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_BASE_HIGH,
 upper_32_bits(adev->doorbell.base));
 }

-   WREG32_SOC15(NBIO, 0, regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
-   tmp);
-*/
+   WREG32_SOC15(NBIO, 0, regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL, 
tmp);
 }


@@ -216,12 +214,12 @@ static void nbio_v7_11_ih_control(struct amdgpu_device 
*adev)

 static u32 nbio_v7_11_get_hdp_flush_req_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_GPU_HDP_FLUSH_REQ);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_GPU_HDP_FLUSH_REQ);
 }

 static u32 nbio_v7_11_get_hdp_flush_done_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_GPU_HDP_FLUSH_DONE);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_GPU_HDP_FLUSH_DONE);
 }

 static u32 nbio_v7_11_get_pcie_index_offset(struct amdgpu_device *adev)
@@ -236,27 +234,27 @@ static u32 nbio_v7_11_get_pcie_data_offset(struct 
amdgpu_device *adev)

 static u32 nbio_v7_11_get_pcie_port_index_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_RSMU_INDEX);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_RSMU_INDEX);
 }

 static u32 nbio_v7_11_get_pcie_port_data_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, 

Re: [PATCH] drm/edid: add 8 bpc quirk to the BenQ GW2765

[Harry Wentland]

On 2023-10-12 14:49, Hamza Mahfooz wrote:
> The BenQ GW2765 reports that it supports higher (> 8) bpc modes, but
> when trying to set them we end up with a black screen. So, limit it to 8
> bpc modes.
> 
> Cc: sta...@vger.kernel.org # 6.5+
> Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2610
> Signed-off-by: Hamza Mahfooz 

Reviewed-by: Harry Wentland 

Harry

> ---
>  drivers/gpu/drm/drm_edid.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> index 0454da505687..bca2af4fe1fc 100644
> --- a/drivers/gpu/drm/drm_edid.c
> +++ b/drivers/gpu/drm/drm_edid.c
> @@ -123,6 +123,9 @@ static const struct edid_quirk {
>   /* AEO model 0 reports 8 bpc, but is a 6 bpc panel */
>   EDID_QUIRK('A', 'E', 'O', 0, EDID_QUIRK_FORCE_6BPC),
>  
> + /* BenQ GW2765 */
> + EDID_QUIRK('B', 'N', 'Q', 0x78d6, EDID_QUIRK_FORCE_8BPC),
> +
>   /* BOE model on HP Pavilion 15-n233sl reports 8 bpc, but is a 6 bpc 
> panel */
>   EDID_QUIRK('B', 'O', 'E', 0x78b, EDID_QUIRK_FORCE_6BPC),
>  

[PATCH] drm/amdgpu/pm: update SMU 13.0.0 PMFW version check

[Alex Deucher]

Update the PMFW version check the the ROCm optimizations.

Signed-off-by: Alex Deucher 
---
 drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c 
b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
index 56b11f6386df..cb7e1a22d7ef 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
@@ -2562,7 +2562,7 @@ static int smu_v13_0_0_set_power_profile_mode(struct 
smu_context *smu,
if (smu->power_profile_mode == PP_SMC_POWER_PROFILE_COMPUTE &&
(amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 0)) 
&&
((smu->adev->pm.fw_version == 0x004e6601) ||
-(smu->adev->pm.fw_version >= 0x004e7300))) {
+(smu->adev->pm.fw_version >= 0x004e7400))) {
workload_type = smu_cmn_to_asic_specific_index(smu,
   
CMN2ASIC_MAPPING_WORKLOAD,
   
PP_SMC_POWER_PROFILE_POWERSAVING);
-- 
2.41.0

[PATCH] drm/edid: add 8 bpc quirk to the BenQ GW2765

[Hamza Mahfooz]

The BenQ GW2765 reports that it supports higher (> 8) bpc modes, but
when trying to set them we end up with a black screen. So, limit it to 8
bpc modes.

Cc: sta...@vger.kernel.org # 6.5+
Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2610
Signed-off-by: Hamza Mahfooz 
---
 drivers/gpu/drm/drm_edid.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
index 0454da505687..bca2af4fe1fc 100644
--- a/drivers/gpu/drm/drm_edid.c
+++ b/drivers/gpu/drm/drm_edid.c
@@ -123,6 +123,9 @@ static const struct edid_quirk {
/* AEO model 0 reports 8 bpc, but is a 6 bpc panel */
EDID_QUIRK('A', 'E', 'O', 0, EDID_QUIRK_FORCE_6BPC),
 
+   /* BenQ GW2765 */
+   EDID_QUIRK('B', 'N', 'Q', 0x78d6, EDID_QUIRK_FORCE_8BPC),
+
/* BOE model on HP Pavilion 15-n233sl reports 8 bpc, but is a 6 bpc 
panel */
EDID_QUIRK('B', 'O', 'E', 0x78b, EDID_QUIRK_FORCE_6BPC),
 
-- 
2.42.0

Re: [PATCH v2] drm/amdgpu: Correctly use bo_va->ref_count in compute VMs

[Chen, Xiaogang]

On 10/12/2023 12:48 PM, Felix Kuehling wrote:

On 2023-10-12 12:34, Xiaogang.Chen wrote:

From: Xiaogang Chen 

This is needed to correctly handle BOs imported into compute VM from 
gfx.
Both kfd and gfx should use same bo_va and set bo_va->ref_count 
correctly

when map the Bos into same VM, otherwise we may trigger kernel general
protection when iterate mappings over bo_va's valids or invalids list.

Signed-off-by: Felix Kuehling 
Signed-off-by: Xiaogang Chen 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Tested-by: Xiaogang Chen 


Not sure if it makes sense to add my Reviewed-by, given that I mostly 
wrote this patch. But feel free to submit this to the branch.



ok, will submit it.

Thanks

Xiaogang


Thanks,
  Felix



---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  uint64_t va = mem->va;
  struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
  struct amdgpu_bo *bo[2] = {NULL, NULL};
+    struct amdgpu_bo_va *bo_va;
  bool same_hive = false;
  int i, ret;
  @@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  pr_debug("Unable to reserve BO during memory attach");
  goto unwind;
  }
-    attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+    if (!bo_va)
+    bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    else
+    ++bo_va->ref_count;
+    attachment[i]->bo_va = bo_va;
+
  amdgpu_bo_unreserve(bo[i]);
  if (unlikely(!attachment[i]->bo_va)) {
  ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  continue;
  if (attachment[i]->bo_va) {
  amdgpu_bo_reserve(bo[i], true);
-    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+    if (--attachment[i]->bo_va->ref_count == 0)
+    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
  amdgpu_bo_unreserve(bo[i]);
  list_del([i]->list);
  }

Re: [PATCH v2] drm/amdgpu: Correctly use bo_va->ref_count in compute VMs

[Felix Kuehling]

On 2023-10-12 12:34, Xiaogang.Chen wrote:

From: Xiaogang Chen 

This is needed to correctly handle BOs imported into compute VM from gfx.
Both kfd and gfx should use same bo_va and set bo_va->ref_count correctly
when map the Bos into same VM, otherwise we may trigger kernel general
protection when iterate mappings over bo_va's valids or invalids list.

Signed-off-by: Felix Kuehling 
Signed-off-by: Xiaogang Chen 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Tested-by: Xiaogang Chen 


Not sure if it makes sense to add my Reviewed-by, given that I mostly 
wrote this patch. But feel free to submit this to the branch.


Thanks,
  Felix



---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
uint64_t va = mem->va;
struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
struct amdgpu_bo *bo[2] = {NULL, NULL};
+   struct amdgpu_bo_va *bo_va;
bool same_hive = false;
int i, ret;
  
@@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,

pr_debug("Unable to reserve BO during memory attach");
goto unwind;
}
-   attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+   if (!bo_va)
+   bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   else
+   ++bo_va->ref_count;
+   attachment[i]->bo_va = bo_va;
+
amdgpu_bo_unreserve(bo[i]);
if (unlikely(!attachment[i]->bo_va)) {
ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
continue;
if (attachment[i]->bo_va) {
amdgpu_bo_reserve(bo[i], true);
-   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+   if (--attachment[i]->bo_va->ref_count == 0)
+   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
amdgpu_bo_unreserve(bo[i]);
list_del([i]->list);
}

Re: [PATCH] drm/amdgpu: Add EXT_COHERENT override for NUMA local nodes

[Felix Kuehling]

On 2023-10-12 12:39, David Francis wrote:

On NUMA systems


This is not applicable to all NUMA systems. It's specific to big APU 
systems with multiple NUMA nodes.




, local memory gets the local mtype, set by an
override callback. If EXT_COHERENT is set, memory will be set as
MTYPE_UC by default, with local memory MTYPE_CC.

Add an option in the override function for this case, and
add a check to ensure it is not used on UNCACHED memory.

Signed-off-by: David Francis 


With an update to the description as above, the patch looks good to me. 
Please allow time for Christian to review this as well. Then feel free 
to add my


Reviewed-by: Felix Kuehling 



---
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 13 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|  8 +-
  drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c |  2 +-
  drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 33 +++
  drivers/gpu/drm/amd/amdkfd/kfd_svm.c  |  6 ++---
  5 files changed, 41 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 46d27c87..189341944bf1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -761,6 +761,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
   * @immediate: immediate submission in a page fault
   * @unlocked: unlocked invalidation during MM callback
   * @flush_tlb: trigger tlb invalidation after update completed
+ * @allow_override: change MTYPE for local NUMA nodes
   * @resv: fences we need to sync to
   * @start: start of mapped range
   * @last: last mapped entry
@@ -777,7 +778,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
   * 0 for success, negative erro code for failure.
   */
  int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-  bool immediate, bool unlocked, bool flush_tlb,
+  bool immediate, bool unlocked, bool flush_tlb, bool 
allow_override,
   struct dma_resv *resv, uint64_t start, uint64_t last,
   uint64_t flags, uint64_t offset, uint64_t vram_base,
   struct ttm_resource *res, dma_addr_t *pages_addr,
@@ -815,6 +816,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
params.immediate = immediate;
params.pages_addr = pages_addr;
params.unlocked = unlocked;
+   params.allow_override = allow_override;
  
  	/* Implicitly sync to command submissions in the same VM before

 * unmapping. Sync to moving fences before mapping.
@@ -1062,6 +1064,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, 
struct amdgpu_bo_va *bo_va,
trace_amdgpu_vm_bo_update(mapping);
  
  		r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,

+  !(bo->flags & 
AMDGPU_GEM_CREATE_UNCACHED),
   resv, mapping->start, mapping->last,
   update_flags, mapping->offset,
   vram_base, mem, pages_addr,
@@ -1257,8 +1260,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
mapping->start < AMDGPU_GMC_HOLE_START)
init_pte_value = AMDGPU_PTE_DEFAULT_ATC;
  
-		r = amdgpu_vm_update_range(adev, vm, false, false, true, resv,

-  mapping->start, mapping->last,
+   r = amdgpu_vm_update_range(adev, vm, false, false, true, false,
+  resv, mapping->start, mapping->last,
   init_pte_value, 0, 0, NULL, NULL,
   );
amdgpu_vm_free_mapping(adev, vm, mapping, f);
@@ -2546,8 +2549,8 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, 
u32 pasid,
goto error_unlock;
}
  
-	r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr,

-  addr, flags, value, 0, NULL, NULL, NULL);
+   r = amdgpu_vm_update_range(adev, vm, true, false, false, false,
+  NULL, addr, addr, flags, value, 0, NULL, 
NULL, NULL);
if (r)
goto error_unlock;
  
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h

index 6e71978db13f..9ea8b5b644e3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -244,6 +244,12 @@ struct amdgpu_vm_update_params {
 * @table_freed: return true if page table is freed when updating
 */
bool table_freed;
+
+   /**
+* @allow_override: true for memory that is not uncached: allows MTYPE
+* to be overridden for NUMA local memory.
+*/
+   bool allow_override;
  };
  
  struct 

[PATCH v7 25/25] drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN

[Jim Cromie]

Lots of burn-in testing needed before signing, upstreaming.

NOTE: I set default Y to maximize testing by default.
Is there a better way to do this ?

Signed-off-by: Jim Cromie 
---
 drivers/gpu/drm/Kconfig | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index 3caa020391c7..708f5e8cb205 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -55,8 +55,7 @@ config DRM_DEBUG_MM
 
 config DRM_USE_DYNAMIC_DEBUG
bool "use dynamic debug to implement drm.debug"
-   default n
-   depends on BROKEN
+   default y
depends on DRM
depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE
depends on JUMP_LABEL
-- 
2.41.0

[PATCH v7 24/25] drm-drivers: DRM_CLASSMAP_USE in 2nd batch of drivers, helpers

[Jim Cromie]

Add a DRM_CLASSMAP_USE declaration to 2nd batch of helpers and *_drv.c
files.  For drivers, add the decl just above the module's PARAMs,
since it identifies the "inherited" drm.debug param.

Note: with CONFIG_DRM_USE_DYNAMIC_DEBUG=y, a module not also declaring
DRM_CLASSMAP_USE will have its class'd prdbgs stuck in the initial
(disabled, but for DEBUG) state.

The stuck sites are evident in /proc/dynamic_debug/control as:

   class:_UNKNOWN_ _id:N# control's last column

rather than a proper "enumeration":

   class:DRM_UT_CORE

This set of updates was found by choosing M for all DRM-config items I
found (not allmodconfig), building & modprobing them, and grepping
"class unknown," control.  There may yet be others.

Signed-off-by: Jim Cromie 
---
 drivers/gpu/drm/drm_gem_shmem_helper.c | 2 ++
 drivers/gpu/drm/gud/gud_drv.c  | 2 ++
 drivers/gpu/drm/mgag200/mgag200_drv.c  | 2 ++
 drivers/gpu/drm/qxl/qxl_drv.c  | 2 ++
 drivers/gpu/drm/radeon/radeon_drv.c| 2 ++
 drivers/gpu/drm/udl/udl_main.c | 2 ++
 drivers/gpu/drm/vkms/vkms_drv.c| 2 ++
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c| 2 ++
 8 files changed, 16 insertions(+)

diff --git a/drivers/gpu/drm/drm_gem_shmem_helper.c 
b/drivers/gpu/drm/drm_gem_shmem_helper.c
index e435f986cd13..066d906e3199 100644
--- a/drivers/gpu/drm/drm_gem_shmem_helper.c
+++ b/drivers/gpu/drm/drm_gem_shmem_helper.c
@@ -23,6 +23,8 @@
 #include 
 #include 
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 MODULE_IMPORT_NS(DMA_BUF);
 
 /**
diff --git a/drivers/gpu/drm/gud/gud_drv.c b/drivers/gpu/drm/gud/gud_drv.c
index 9d7bf8ee45f1..5b555045fce4 100644
--- a/drivers/gpu/drm/gud/gud_drv.c
+++ b/drivers/gpu/drm/gud/gud_drv.c
@@ -31,6 +31,8 @@
 
 #include "gud_internal.h"
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 /* Only used internally */
 static const struct drm_format_info gud_drm_format_r1 = {
.format = GUD_DRM_FORMAT_R1,
diff --git a/drivers/gpu/drm/mgag200/mgag200_drv.c 
b/drivers/gpu/drm/mgag200/mgag200_drv.c
index abddf37f0ea1..d678eb8e028d 100644
--- a/drivers/gpu/drm/mgag200/mgag200_drv.c
+++ b/drivers/gpu/drm/mgag200/mgag200_drv.c
@@ -24,6 +24,8 @@ static int mgag200_modeset = -1;
 MODULE_PARM_DESC(modeset, "Disable/Enable modesetting");
 module_param_named(modeset, mgag200_modeset, int, 0400);
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 int mgag200_init_pci_options(struct pci_dev *pdev, u32 option, u32 option2)
 {
struct device *dev = >dev;
diff --git a/drivers/gpu/drm/qxl/qxl_drv.c b/drivers/gpu/drm/qxl/qxl_drv.c
index b30ede1cf62d..91942ffcc2b4 100644
--- a/drivers/gpu/drm/qxl/qxl_drv.c
+++ b/drivers/gpu/drm/qxl/qxl_drv.c
@@ -65,6 +65,8 @@ module_param_named(modeset, qxl_modeset, int, 0400);
 MODULE_PARM_DESC(num_heads, "Number of virtual crtcs to expose (default 4)");
 module_param_named(num_heads, qxl_num_crtc, int, 0400);
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 static struct drm_driver qxl_driver;
 static struct pci_driver qxl_pci_driver;
 
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c 
b/drivers/gpu/drm/radeon/radeon_drv.c
index fa531493b111..ab29945af657 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -247,6 +247,8 @@ int radeon_cik_support = 1;
 MODULE_PARM_DESC(cik_support, "CIK support (1 = enabled (default), 0 = 
disabled)");
 module_param_named(cik_support, radeon_cik_support, int, 0444);
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 static struct pci_device_id pciidlist[] = {
radeon_PCI_IDS
 };
diff --git a/drivers/gpu/drm/udl/udl_main.c b/drivers/gpu/drm/udl/udl_main.c
index 3ebe2ce55dfd..ba57c14454e5 100644
--- a/drivers/gpu/drm/udl/udl_main.c
+++ b/drivers/gpu/drm/udl/udl_main.c
@@ -19,6 +19,8 @@
 
 #define NR_USB_REQUEST_CHANNEL 0x12
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 #define MAX_TRANSFER (PAGE_SIZE*16 - BULK_SIZE)
 #define WRITES_IN_FLIGHT (20)
 #define MAX_VENDOR_DESCRIPTOR_SIZE 256
diff --git a/drivers/gpu/drm/vkms/vkms_drv.c b/drivers/gpu/drm/vkms/vkms_drv.c
index dd0af086e7fa..086797c4b82b 100644
--- a/drivers/gpu/drm/vkms/vkms_drv.c
+++ b/drivers/gpu/drm/vkms/vkms_drv.c
@@ -39,6 +39,8 @@
 
 static struct vkms_config *default_config;
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 static bool enable_cursor = true;
 module_param_named(enable_cursor, enable_cursor, bool, 0444);
 MODULE_PARM_DESC(enable_cursor, "Enable/Disable cursor support");
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c 
b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
index 8b24ecf60e3e..9cb6be422621 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.c
@@ -275,6 +275,8 @@ static int vmw_probe(struct pci_dev *, const struct 
pci_device_id *);
 static int vmwgfx_pm_notifier(struct notifier_block *nb, unsigned long val,
  void *ptr);
 
+DRM_CLASSMAP_USE(drm_debug_classes);
+
 MODULE_PARM_DESC(restrict_iommu, "Try to limit IOMMU usage for TTM pages");
 module_param_named(restrict_iommu, vmw_restrict_iommu, int, 0600);
 

[PATCH v7 23/25] drm: use correct ccflags-y spelling

[Jim Cromie]

Incorrectly spelled CFLAGS- failed to add -DDYNAMIC_DEBUG_MODULE,
which broke builds with:

CONFIG_DRM_USE_DYNAMIC_DEBUG=y
CONFIG_DYNAMIC_DEBUG_CORE=y
CONFIG_DYNAMIC_DEBUG=n

Also add subdir-ccflags so that all drivers pick up the addition.

Fixes: 84ec67288c10 ("drm_print: wrap drm_*_dbg in dyndbg descriptor factory 
macro")
Signed-off-by: Jim Cromie 
---
 drivers/gpu/drm/Makefile | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile
index 215e78e79125..22b1984cc982 100644
--- a/drivers/gpu/drm/Makefile
+++ b/drivers/gpu/drm/Makefile
@@ -3,7 +3,8 @@
 # Makefile for the drm device driver.  This driver provides support for the
 # Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
 
-CFLAGS-$(CONFIG_DRM_USE_DYNAMIC_DEBUG) += -DDYNAMIC_DEBUG_MODULE
+ccflags-$(CONFIG_DRM_USE_DYNAMIC_DEBUG)+= 
-DDYNAMIC_DEBUG_MODULE
+subdir-ccflags-$(CONFIG_DRM_USE_DYNAMIC_DEBUG) += -DDYNAMIC_DEBUG_MODULE
 
 drm-y := \
drm_aperture.o \
-- 
2.41.0

[PATCH v7 21/25] dyndbg: refactor *dynamic_emit_prefix

[Jim Cromie]

Refactor the split of duties between outer & inner fns.

The outer fn was previously just an inline unlikely forward to inner,
which did all the work.

Now, outer handles +t and +l flags itself, and calls inner only when
_DPRINTK_FLAGS_INCL_LOOKUP is needed.

No functional change.

But it does make the results of the inner-fn more cache-friendly
(fewer entries, reused more often):

1- no spurious [TID] or  noise
2- no LINE-number to bloat the cache (avg 9 pr_debugs/fn)
3- only LOOKUP stuff

Currently LOOKUPs are descriptor-field refs but could be replaced by
accessor functions.  This would allow the __dyndbg_sites section to be
de-duplicated and reclaimed; currently module, filename fields are
~90% repeated.  As the accessors get more expensive, the value of
caching part of the prefix goes up.

Also change inner-fn to return count of extra chars written to the
buffer, and drop "inline" from outer, let the compiler decide.  Maybe
also change name accordingly.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 55 -
 1 file changed, 30 insertions(+), 25 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 17eefb35ac96..974395bf8a83 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -777,19 +777,8 @@ static int remaining(int wrote)
return 0;
 }
 
-static char *__dynamic_emit_prefix(const struct _ddebug *desc, char *buf)
+static int __dynamic_emit_prefix(const struct _ddebug *desc, char *buf, int 
pos)
 {
-   int pos_after_tid;
-   int pos = 0;
-
-   if (desc->flags & _DPRINTK_FLAGS_INCL_TID) {
-   if (in_interrupt())
-   pos += snprintf(buf + pos, remaining(pos), " ");
-   else
-   pos += snprintf(buf + pos, remaining(pos), "[%d] ",
-   task_pid_vnr(current));
-   }
-   pos_after_tid = pos;
if (desc->flags & _DPRINTK_FLAGS_INCL_MODNAME)
pos += snprintf(buf + pos, remaining(pos), "%s:",
desc->modname);
@@ -799,22 +788,38 @@ static char *__dynamic_emit_prefix(const struct _ddebug 
*desc, char *buf)
if (desc->flags & _DPRINTK_FLAGS_INCL_SOURCENAME)
pos += snprintf(buf + pos, remaining(pos), "%s:",
trim_prefix(desc->filename));
-   if (desc->flags & _DPRINTK_FLAGS_INCL_LINENO)
-   pos += snprintf(buf + pos, remaining(pos), "%d:",
-   desc->lineno);
-   if (pos - pos_after_tid)
-   pos += snprintf(buf + pos, remaining(pos), " ");
-   if (pos >= PREFIX_SIZE)
-   buf[PREFIX_SIZE - 1] = '\0';
-
-   return buf;
+   return pos;
 }
 
-static inline char *dynamic_emit_prefix(struct _ddebug *desc, char *buf)
+static char *dynamic_emit_prefix(struct _ddebug *desc, char *buf)
 {
-   if (unlikely(desc->flags & _DPRINTK_FLAGS_INCL_ANY))
-   return __dynamic_emit_prefix(desc, buf);
-   return buf;
+int pos_after_tid;
+int pos = 0;
+
+if (likely(!(desc->flags & _DPRINTK_FLAGS_INCL_ANY)))
+return buf;
+
+if (desc->flags & _DPRINTK_FLAGS_INCL_TID) {
+if (in_interrupt())
+pos += snprintf(buf + pos, remaining(pos), " ");
+else
+pos += snprintf(buf + pos, remaining(pos), "[%d] ",
+task_pid_vnr(current));
+}
+pos_after_tid = pos;
+
+if (unlikely(desc->flags & _DPRINTK_FLAGS_INCL_LOOKUP))
+pos += __dynamic_emit_prefix(desc, buf, pos);
+
+if (desc->flags & _DPRINTK_FLAGS_INCL_LINENO)
+pos += snprintf(buf + pos, remaining(pos), "%d:",
+desc->lineno);
+if (pos - pos_after_tid)
+pos += snprintf(buf + pos, remaining(pos), " ");
+if (pos >= PREFIX_SIZE)
+buf[PREFIX_SIZE - 1] = '\0';
+
+return buf;
 }
 
 void __dynamic_pr_debug(struct _ddebug *descriptor, const char *fmt, ...)
-- 
2.41.0

[PATCH v7 18/25] dyndbg-doc: add classmap info to howto

[Jim Cromie]

Add some basic info on classmap usage and api

Signed-off-by: Jim Cromie 
---
v5- adjustments per Randy Dunlap, me
---
 .../admin-guide/dynamic-debug-howto.rst   | 59 ++-
 1 file changed, 58 insertions(+), 1 deletion(-)

diff --git a/Documentation/admin-guide/dynamic-debug-howto.rst 
b/Documentation/admin-guide/dynamic-debug-howto.rst
index 0b3d39c610d9..d8813dcc394a 100644
--- a/Documentation/admin-guide/dynamic-debug-howto.rst
+++ b/Documentation/admin-guide/dynamic-debug-howto.rst
@@ -225,7 +225,6 @@ the ``p`` flag has meaning, other flags are ignored.
 Note the regexp ``^[-+=][fslmpt_]+$`` matches a flags specification.
 To clear all flags at once, use ``=_`` or ``-fslmpt``.
 
-
 Debug messages during Boot Process
 ==
 
@@ -375,3 +374,61 @@ just a shortcut for ``print_hex_dump(KERN_DEBUG)``.
 For ``print_hex_dump_debug()``/``print_hex_dump_bytes()``, format string is
 its ``prefix_str`` argument, if it is constant string; or ``hexdump``
 in case ``prefix_str`` is built dynamically.
+
+Dynamic Debug classmaps
+===
+
+Dyndbg allows selection/grouping of *prdbg* callsites using structural
+info: module, file, function, line.  Classmaps allow authors to add
+their own domain-oriented groupings using class-names.  Classmaps are
+exported, so they referencable from other modules.
+
+  # enable classes individually
+  :#> ddcmd class DRM_UT_CORE +p
+  :#> ddcmd class DRM_UT_KMS +p
+  # or more selectively
+  :#> ddcmd class DRM_UT_CORE module drm +p
+
+The "class FOO" syntax protects class'd prdbgs from generic overwrite::
+
+  # IOW this doesn't wipe any DRM.debug settings
+  :#> ddcmd -p
+
+To support the DRM.debug parameter, DYNDBG_CLASSMAP_PARAM* updates all
+classes in a classmap, mapping param-bits 0..N onto the classes:
+DRM_UT_<*> for the DRM use-case.
+
+Dynamic Debug Classmap API
+==
+
+DYNDBG_CLASSMAP_DEFINE - modules use this to create classmaps, naming
+each of the classes (stringified enum-symbols: "DRM_UT_<*>"), and
+type, and mapping the class-names to consecutive _class_ids.
+
+By doing so, modules tell dyndbg that they are have prdbgs with those
+class_ids, and they authorize dyndbg to accept "class FOO" for module
+defining that classname.
+
+There are 2 types of classmaps:
+
+ DD_CLASS_TYPE_DISJOINT_BITS: classes are independent, like DRM.debug
+ DD_CLASS_TYPE_LEVEL_NUM: classes are relative, ordered (V3 > V2)
+
+DYNDBG_CLASSMAP_PARAM - refers to a DEFINEd classmap, exposing the set
+of defined classes to manipulation as a group.  This interface
+enforces the relatedness of classes of DD_CLASS_TYPE_LEVEL_NUM typed
+classmaps; all classes are independent in the >control parser itself.
+
+DYNDBG_CLASSMAP_USE - drm drivers invoke this to ref the the CLASSMAP
+that drm DEFINEs.  This shares the classmap definition, and authorizes
+dyndbg to apply changes to the using modules class'd pr_debugs.  It
+also tells dyndbg how to initialize the user's prdbgs at modprobe.
+
+Modules or module-groups (drm & drivers) can define multiple
+classmaps, as long as they share the limited 0..62 per-module-group
+_class_id range, without overlap.
+
+``#define DEBUG`` will enable all pr_debugs in scope, including any
+class'd ones.  This won't be reflected in the PARAM readback value,
+but the pr_debug callsites can be toggled into agreement with the
+param.
-- 
2.41.0

[PATCH v7 22/25] dyndbg: improve err report in attach_user_module_classes

[Jim Cromie]

convert a WARN on 3 conditions, into BUG_ON 2 of them (which don't
happen), and an early return on (!cli->user_mod_name), which *was*
happening, so should be seen going forward.  Maybe this should be a
WARN.

NB: The underlying problem was a missing __align(8) in the
DYNDBG_CLASSMAP_USE, which manifested as a corrupt record with a map
pointer which segv'd.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 8 +---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 974395bf8a83..3dc512fb1d66 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1284,9 +1284,11 @@ static void ddebug_attach_user_module_classes(struct 
ddebug_table *dt,
 */
for_each_boxed_vector(di, class_users, num_class_users, i, cli) {
 
-   if (WARN_ON(!cli || !cli->map || !cli->user_mod_name))
-   continue;
-
+   BUG_ON(!cli || !cli->map);
+   if (!cli->user_mod_name) {
+   pr_warn("class_ref[%d] !user-mod-name looking for 
%s\n", i, dt->mod_name);
+   return;
+   }
if (!strcmp(cli->user_mod_name, dt->mod_name)) {
 
vpr_cm_info(cli->map, "class_ref[%d] %s -> %s", i,
-- 
2.41.0

[PATCH v7 20/25] dyndbg: add _DPRINTK_FLAGS_INCL_LOOKUP

[Jim Cromie]

dyndbg's dynamic prefixing (by +tmfsl flags) is needlessly expensive.

When an enabled (with +p) pr_debug is called, _DPRINTK_FLAGS_INCL_ANY
prefix decorations are sprintf'd into stack-mem for every call.

This string (or part of it) could be cached once its 1st generated,
and retreived thereafter, as long as its deleted any time the
callsite's flags are changed afterwards.

So consider the prefix/decoration flags: 'tmfsl', and what should be
in the cache:

-t  thread-id. not part of the "callsite" info, derived from current.
doesnt belong in the cache. it would be wrong.
can be done in outer: dynamic_emit_prefix()

-l  line number
this could be part of the prefix, but would bloat the cache
can also be done in outer: dynamic_emit_prefix()

-mfs  module, function, source-file
we cache these, composed into a sub-string.
they are "lookups", currently to descriptor fields,
could be accessor macros to "compressed" tables.
cache saves more access work.

All enabled together, they compose a prefix string like:

  # outer   -inner--   outer
  "[tid] module:function:sourcfile:line: "

So this patch extracts _DPRINTK_FLAGS_INCL_LOOKUP macro out of
_DPRINTK_FLAGS_INCL_ANY macro, then redefs latter.

Next re-refactor dynamic_emit_prefix inner/outer fns accordingly.

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h | 8 +---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index 927cb14f24e0..2237d454bc19 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -40,10 +40,12 @@ struct _ddebug {
 #define _DPRINTK_FLAGS_INCL_SOURCENAME (1<<5)
 #define _DPRINTK_FLAGS_PREFIX_CACHED   (1<<7)
 
-#define _DPRINTK_FLAGS_INCL_ANY\
-   (_DPRINTK_FLAGS_INCL_MODNAME | _DPRINTK_FLAGS_INCL_FUNCNAME |\
-_DPRINTK_FLAGS_INCL_LINENO  | _DPRINTK_FLAGS_INCL_TID |\
+#define _DPRINTK_FLAGS_INCL_LOOKUP \
+   (_DPRINTK_FLAGS_INCL_MODNAME | _DPRINTK_FLAGS_INCL_FUNCNAME |   \
 _DPRINTK_FLAGS_INCL_SOURCENAME)
+#define _DPRINTK_FLAGS_INCL_ANY
\
+   (_DPRINTK_FLAGS_INCL_LINENO | _DPRINTK_FLAGS_INCL_TID | \
+_DPRINTK_FLAGS_INCL_LOOKUP)
 
 #if defined DEBUG
 #define _DPRINTK_FLAGS_DEFAULT _DPRINTK_FLAGS_PRINT
-- 
2.41.0

[PATCH v7 17/25] dyndbg-API: promote DYNDBG_CLASSMAP_PARAM to API

[Jim Cromie]

move macro from test-dynamic-debug.c into header, and refine it.

Distinguish the 2 use cases of DYNDBG_CLASSMAP_PARAM*

1.DYNDBG_CLASSMAP_PARAM_REF
for DRM, to pass in extern __drm_debug by name.
dyndbg keeps bits in it, so drm can still use it as before

2.DYNDBG_CLASSMAP_PARAM
new user (test_dynamic_debug) doesn't need to share state,
decls a static long unsigned int to store the bitvec.

__DYNDBG_CLASSMAP_PARAM
   bottom layer - allocate,init a ddebug-class-param, module-param-cb.

Also clean up and improve comments in test-code, and add
MODULE_DESCRIPTIONs.

Signed-off-by: Jim Cromie 
---
v5b - parens-on-PARAM
---
 drivers/gpu/drm/drm_print.c |  8 ++---
 include/drm/drm_print.h |  6 ++--
 include/linux/dynamic_debug.h   | 37 -
 lib/test_dynamic_debug.c| 58 +
 lib/test_dynamic_debug_submod.c |  9 -
 5 files changed, 73 insertions(+), 45 deletions(-)

diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c
index dabcfa0dd279..8f4b609353a5 100644
--- a/drivers/gpu/drm/drm_print.c
+++ b/drivers/gpu/drm/drm_print.c
@@ -69,12 +69,8 @@ DRM_CLASSMAP_DEFINE(drm_debug_classes, 
DD_CLASS_TYPE_DISJOINT_BITS,
"DRM_UT_DP",
"DRM_UT_DRMRES");
 
-static struct ddebug_class_param drm_debug_bitmap = {
-   .bits = &__drm_debug,
-   .flags = "p",
-   .map = _debug_classes,
-};
-module_param_cb(debug, _ops_dyndbg_classes, _debug_bitmap, 0600);
+DRM_CLASSMAP_PARAM_REF(debug, __drm_debug, drm_debug_classes, p);
+
 #endif
 
 void __drm_puts_coredump(struct drm_printer *p, const char *str)
diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h
index 706afc97c79c..94d4f5500030 100644
--- a/include/drm/drm_print.h
+++ b/include/drm/drm_print.h
@@ -322,11 +322,13 @@ enum drm_debug_category {
 };
 
 #ifdef CONFIG_DRM_USE_DYNAMIC_DEBUG
-#define DRM_CLASSMAP_DEFINE(...) DYNDBG_CLASSMAP_DEFINE(__VA_ARGS__)
-#define DRM_CLASSMAP_USE(name)   DYNDBG_CLASSMAP_USE(name)
+#define DRM_CLASSMAP_DEFINE(...)   DYNDBG_CLASSMAP_DEFINE(__VA_ARGS__)
+#define DRM_CLASSMAP_USE(name) DYNDBG_CLASSMAP_USE(name)
+#define DRM_CLASSMAP_PARAM_REF(...)DYNDBG_CLASSMAP_PARAM_REF(__VA_ARGS__)
 #else
 #define DRM_CLASSMAP_DEFINE(...)
 #define DRM_CLASSMAP_USE(name)
+#define DRM_CLASSMAP_PARAM_REF(...)
 #endif
 
 static inline bool drm_debug_enabled_raw(enum drm_debug_category category)
diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index 1b027be2cdc4..f182f95caabb 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -91,7 +91,7 @@ struct ddebug_class_map {
  * used to validate a "class FOO .." >control command on the module
  */
 #define DYNDBG_CLASSMAP_DEFINE(_var, _maptype, _base, ...) \
-   const char *_var##_classnames[] = { __VA_ARGS__ };  \
+   static const char *_var##_classnames[] = { __VA_ARGS__ };   \
struct ddebug_class_map __aligned(8) __used \
__section("__dyndbg_classes") _var = {  \
.mod = THIS_MODULE, \
@@ -145,6 +145,41 @@ struct ddebug_class_param {
const struct ddebug_class_map *map;
 };
 
+/**
+ * DYNDBG_CLASSMAP_PARAM - wrap a dyndbg-classmap with a controlling sys-param
+ * @_name  sysfs node name
+ * @_var   name of the struct classmap var defining the controlled classes
+ * @_flags flags to be toggled, typically just 'p'
+ *
+ * Creates a sysfs-param to control the classes defined by the
+ * classmap.  Keeps bits in a private/static
+ */
+#define DYNDBG_CLASSMAP_PARAM(_name, _var, _flags) \
+   static unsigned long _name##_bvec;  \
+   __DYNDBG_CLASSMAP_PARAM(_name, _name##_bvec, _var, _flags)
+
+/**
+ * DYNDBG_CLASSMAP_PARAM_REF - wrap a dyndbg-classmap with a controlling 
sys-param
+ * @_name  sysfs node name
+ * @_bits  name of the module's unsigned long bit-vector, ex: __drm_debug
+ * @_var   name of the struct classmap var defining the controlled classes
+ * @_flags flags to be toggled, typically just 'p'
+ *
+ * Creates a sysfs-param to control the classmap, keeping bitvec in user 
@_bits.
+ * This lets drm use __drm_debug elsewhere too.
+ */
+#define DYNDBG_CLASSMAP_PARAM_REF(_name, _bits, _var, _flags)  \
+   __DYNDBG_CLASSMAP_PARAM(_name, _bits, _var, _flags)
+
+#define __DYNDBG_CLASSMAP_PARAM(_name, _bits, _var, _flags)\
+   static struct ddebug_class_param _name##_##_flags = {   \
+   .bits = &(_bits),   \
+   .flags = #_flags,   \
+   .map = &(_var), \
+   };  \
+   module_param_cb(_name, _ops_dyndbg_classes,   \
+ 

[PATCH v7 19/25] dyndbg: reserve flag bit _DPRINTK_FLAGS_PREFIX_CACHED

[Jim Cromie]

Reserve bit 7 to remember that a pr-debug callsite is/was:
- enabled, with +p
- wants a dynamic-prefix, with one+ of module:function:sourcfile
- was previously called
- was thus saved in the cache. NOT YET.

Its unclear whether any cache fetch would be faster than 2-3 field
fetches, but theres another factor; the 3 columns in the __dyndbg
section are highly redundant and compressible, but to get the
compression, we need field accessors, which will rebalance the
tradeoff.

So, for now, its just the bit reservation.

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index f182f95caabb..927cb14f24e0 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -38,6 +38,7 @@ struct _ddebug {
 #define _DPRINTK_FLAGS_INCL_LINENO (1<<3)
 #define _DPRINTK_FLAGS_INCL_TID(1<<4)
 #define _DPRINTK_FLAGS_INCL_SOURCENAME (1<<5)
+#define _DPRINTK_FLAGS_PREFIX_CACHED   (1<<7)
 
 #define _DPRINTK_FLAGS_INCL_ANY\
(_DPRINTK_FLAGS_INCL_MODNAME | _DPRINTK_FLAGS_INCL_FUNCNAME |\
-- 
2.41.0

[PATCH v7 11/25] dyndbg: tighten fn-sig of ddebug_apply_class_bitmap

[Jim Cromie]

old_bits arg is currently a pointer to the input bits, but this could
allow inadvertent changes to the input by the fn.  Disallow this.
And constify new_bits while here.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 21 +++--
 1 file changed, 11 insertions(+), 10 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 8158943b350d..8beb98a831f5 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -593,7 +593,8 @@ static int ddebug_exec_queries(char *query, const char 
*modname)
 
 /* apply a new class-param setting */
 static int ddebug_apply_class_bitmap(const struct ddebug_class_param *dcp,
-unsigned long *new_bits, unsigned long 
*old_bits,
+const unsigned long *new_bits,
+const unsigned long old_bits,
 const char *query_modname)
 {
 #define QUERY_SIZE 128
@@ -602,12 +603,12 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
int matches = 0;
int bi, ct;
 
-   if (*new_bits != *old_bits)
+   if (*new_bits != old_bits)
v2pr_info("apply bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits,
- *old_bits, query_modname ?: "'*'");
+ old_bits, query_modname ?: "'*'");
 
for (bi = 0; bi < map->length; bi++) {
-   if (test_bit(bi, new_bits) == test_bit(bi, old_bits))
+   if (test_bit(bi, new_bits) == test_bit(bi, _bits))
continue;
 
snprintf(query, QUERY_SIZE, "class %s %c%s", 
map->class_names[bi],
@@ -619,9 +620,9 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
v2pr_info("bit_%d: %d matches on class: %s -> 0x%lx\n", bi,
  ct, map->class_names[bi], *new_bits);
}
-   if (*new_bits != *old_bits)
+   if (*new_bits != old_bits)
v2pr_info("applied bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits,
- *old_bits, query_modname ?: "'*'");
+ old_bits, query_modname ?: "'*'");
 
return matches;
 }
@@ -678,7 +679,7 @@ static int param_set_dyndbg_classnames(const char *instr, 
const struct kernel_pa
continue;
}
curr_bits ^= BIT(cls_id);
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
dcp->bits, NULL);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, 
*dcp->bits, NULL);
*dcp->bits = curr_bits;
v2pr_info("%s: changed bit %d:%s\n", KP_NAME(kp), 
cls_id,
  map->class_names[cls_id]);
@@ -688,7 +689,7 @@ static int param_set_dyndbg_classnames(const char *instr, 
const struct kernel_pa
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
curr_bits = CLASSMAP_BITMASK(cls_id + (wanted ? 1 : 0 
));
 
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
_bits, NULL);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, 
old_bits, NULL);
*dcp->lvl = (cls_id + (wanted ? 1 : 0));
v2pr_info("%s: changed bit-%d: \"%s\" %lx->%lx\n", 
KP_NAME(kp), cls_id,
  map->class_names[cls_id], old_bits, 
curr_bits);
@@ -742,7 +743,7 @@ static int param_set_dyndbg_module_classes(const char 
*instr,
inrep &= CLASSMAP_BITMASK(map->length);
}
v2pr_info("bits:0x%lx > %s.%s\n", inrep, modnm ?: "*", 
KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, , dcp->bits, 
modnm);
+   totct += ddebug_apply_class_bitmap(dcp, , *dcp->bits, 
modnm);
*dcp->bits = inrep;
break;
case DD_CLASS_TYPE_LEVEL_NUM:
@@ -755,7 +756,7 @@ static int param_set_dyndbg_module_classes(const char 
*instr,
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
new_bits = CLASSMAP_BITMASK(inrep);
v2pr_info("lvl:%ld bits:0x%lx > %s\n", inrep, new_bits, 
KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, _bits, _bits, 
modnm);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, old_bits, 
modnm);
*dcp->lvl = inrep;
break;
default:
-- 
2.41.0

[PATCH v7 13/25] dyndbg-API: remove DD_CLASS_TYPE_(DISJOINT|LEVEL)_NAMES and code

[Jim Cromie]

Remove the NAMED class types; these 2 classmap types accept class
names at the PARAM interface, for example:

  echo +DRM_UT_CORE,-DRM_UT_KMS > /sys/module/drm/parameters/debug_names

The code works, but its only used by test-dynamic-debug, and wasn't
asked for by anyone else, so simplify things for now.

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h |  19 ++-
 lib/dynamic_debug.c   | 103 +++---
 lib/test_dynamic_debug.c  |  12 
 3 files changed, 12 insertions(+), 122 deletions(-)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index 8116d0a0d33a..8eaf8eabdc8d 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -61,24 +61,13 @@ struct _ddebug {
 enum class_map_type {
DD_CLASS_TYPE_DISJOINT_BITS,
/**
-* DD_CLASS_TYPE_DISJOINT_BITS: classes are independent, one per bit.
-* expecting hex input. Built for drm.debug, basis for other types.
+* DD_CLASS_TYPE_DISJOINT_BITS: classes are independent, mapped to 
bits[0..N].
+* Expects hex input. Built for drm.debug, basis for other types.
 */
DD_CLASS_TYPE_LEVEL_NUM,
/**
-* DD_CLASS_TYPE_LEVEL_NUM: input is numeric level, 0-N.
-* N turns on just bits N-1 .. 0, so N=0 turns all bits off.
-*/
-   DD_CLASS_TYPE_DISJOINT_NAMES,
-   /**
-* DD_CLASS_TYPE_DISJOINT_NAMES: input is a CSV of [+-]CLASS_NAMES,
-* classes are independent, like _DISJOINT_BITS.
-*/
-   DD_CLASS_TYPE_LEVEL_NAMES,
-   /**
-* DD_CLASS_TYPE_LEVEL_NAMES: input is a CSV of [+-]CLASS_NAMES,
-* intended for names like: INFO,DEBUG,TRACE, with a module prefix
-* avoid EMERG,ALERT,CRIT,ERR,WARNING: they're not debug
+* DD_CLASS_TYPE_LEVEL_NUM: input is numeric level, 0..N.
+* Input N turns on bits 0..N-1
 */
 };
 
diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 45870a699507..91c8b67fd8f8 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -632,77 +632,6 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
 
 #define CLASSMAP_BITMASK(width) ((1UL << (width)) - 1)
 
-/* accept comma-separated-list of [+-] classnames */
-static int param_set_dyndbg_classnames(const char *instr, const struct 
kernel_param *kp)
-{
-   const struct ddebug_class_param *dcp = kp->arg;
-   const struct ddebug_class_map *map = dcp->map;
-   unsigned long curr_bits, old_bits;
-   char *cl_str, *p, *tmp;
-   int cls_id, totct = 0;
-   bool wanted;
-
-   cl_str = tmp = kstrdup(instr, GFP_KERNEL);
-   p = strchr(cl_str, '\n');
-   if (p)
-   *p = '\0';
-
-   /* start with previously set state-bits, then modify */
-   curr_bits = old_bits = *dcp->bits;
-   vpr_info("\"%s\" > %s:0x%lx\n", cl_str, KP_NAME(kp), curr_bits);
-
-   for (; cl_str; cl_str = p) {
-   p = strchr(cl_str, ',');
-   if (p)
-   *p++ = '\0';
-
-   if (*cl_str == '-') {
-   wanted = false;
-   cl_str++;
-   } else {
-   wanted = true;
-   if (*cl_str == '+')
-   cl_str++;
-   }
-   cls_id = match_string(map->class_names, map->length, cl_str);
-   if (cls_id < 0) {
-   pr_err("%s unknown to %s\n", cl_str, KP_NAME(kp));
-   continue;
-   }
-
-   /* have one or more valid class_ids of one *_NAMES type */
-   switch (map->map_type) {
-   case DD_CLASS_TYPE_DISJOINT_NAMES:
-   /* the +/- pertains to a single bit */
-   if (test_bit(cls_id, _bits) == wanted) {
-   v3pr_info("no change on %s\n", cl_str);
-   continue;
-   }
-   curr_bits ^= BIT(cls_id);
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
*dcp->bits, NULL);
-   *dcp->bits = curr_bits;
-   v2pr_info("%s: changed bit %d:%s\n", KP_NAME(kp), 
cls_id,
- map->class_names[cls_id]);
-   break;
-   case DD_CLASS_TYPE_LEVEL_NAMES:
-   /* cls_id = N in 0..max. wanted +/- determines N or N-1 
*/
-   old_bits = CLASSMAP_BITMASK(*dcp->lvl);
-   curr_bits = CLASSMAP_BITMASK(cls_id + (wanted ? 1 : 0 
));
-
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
old_bits, NULL);
-   *dcp->lvl = (cls_id + (wanted ? 1 : 0));
-   v2pr_info("%s: changed bit-%d: \"%s\" %lx->%lx\n", 
KP_NAME(kp), cls_id,
- 

[PATCH v7 16/25] dyndbg: refactor ddebug_classparam_clamp_input

[Jim Cromie]

Extract input validation code, from param_set_dyndbg_module_classes()
(the sys-node >handler) to new: ddebug_classparam_clamp_input(kp),
call it from former.  It takes kernel-param arg, so it can complain
about "foo: bad input".

Reuse ddparam_clamp_input(kp) in ddebug_sync_classbits(),
to validate inputs from parent's params, just like our own.
To support that reuse, alter ddebug_sync_classbits() and caller to
pass kp instead of kp->arg.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 70 ++---
 1 file changed, 47 insertions(+), 23 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index c11feca70d6f..17eefb35ac96 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -656,6 +656,30 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
 
 #define CLASSMAP_BITMASK(width) ((1UL << (width)) - 1)
 
+static void ddebug_class_param_clamp_input(unsigned long *inrep, const struct 
kernel_param *kp)
+{
+   const struct ddebug_class_param *dcp = kp->arg;
+   const struct ddebug_class_map *map = dcp->map;
+
+   switch (map->map_type) {
+   case DD_CLASS_TYPE_DISJOINT_BITS:
+   /* expect bits. mask and warn if too many */
+   if (*inrep & ~CLASSMAP_BITMASK(map->length)) {
+   pr_warn("%s: input: 0x%lx exceeds mask: 0x%lx, 
masking\n",
+   KP_NAME(kp), *inrep, 
CLASSMAP_BITMASK(map->length));
+   *inrep &= CLASSMAP_BITMASK(map->length);
+   }
+   break;
+   case DD_CLASS_TYPE_LEVEL_NUM:
+   /* input is bitpos, of highest verbosity to be enabled */
+   if (*inrep > map->length) {
+   pr_warn("%s: level:%ld exceeds max:%d, clamping\n",
+   KP_NAME(kp), *inrep, map->length);
+   *inrep = map->length;
+   }
+   break;
+   }
+}
 static int param_set_dyndbg_module_classes(const char *instr,
   const struct kernel_param *kp,
   const char *modnm)
@@ -674,26 +698,15 @@ static int param_set_dyndbg_module_classes(const char 
*instr,
pr_err("expecting numeric input, not: %s > %s\n", instr, 
KP_NAME(kp));
return -EINVAL;
}
+   ddebug_class_param_clamp_input(, kp);
 
switch (map->map_type) {
case DD_CLASS_TYPE_DISJOINT_BITS:
-   /* expect bits. mask and warn if too many */
-   if (inrep & ~CLASSMAP_BITMASK(map->length)) {
-   pr_warn("%s: input: 0x%lx exceeds mask: 0x%lx, 
masking\n",
-   KP_NAME(kp), inrep, 
CLASSMAP_BITMASK(map->length));
-   inrep &= CLASSMAP_BITMASK(map->length);
-   }
v2pr_info("bits:0x%lx > %s.%s\n", inrep, modnm ?: "*", 
KP_NAME(kp));
totct += ddebug_apply_class_bitmap(dcp, , *dcp->bits, 
modnm);
*dcp->bits = inrep;
break;
case DD_CLASS_TYPE_LEVEL_NUM:
-   /* input is bitpos, of highest verbosity to be enabled */
-   if (inrep > map->length) {
-   pr_warn("%s: level:%ld exceeds max:%d, clamping\n",
-   KP_NAME(kp), inrep, map->length);
-   inrep = map->length;
-   }
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
new_bits = CLASSMAP_BITMASK(inrep);
v2pr_info("lvl:%ld bits:0x%lx > %s\n", inrep, new_bits, 
KP_NAME(kp));
@@ -1160,16 +1173,27 @@ static const char * const ddebug_classmap_typenames[] = 
{
  ddebug_classmap_typenames[_cm->map_type]);\
})
 
-static void ddebug_sync_classbits(const struct ddebug_class_param *dcp, const 
char *modname)
+static void ddebug_sync_classbits(const struct kernel_param *kp, const char 
*modname)
 {
-   /* clamp initial bitvec, mask off hi-bits */
-   if (*dcp->bits & ~CLASSMAP_BITMASK(dcp->map->length)) {
-   *dcp->bits &= CLASSMAP_BITMASK(dcp->map->length);
-   v2pr_info("preset classbits: %lx\n", *dcp->bits);
+   struct ddebug_class_param *dcp = kp->arg;
+   unsigned long new_bits;
+
+   ddebug_class_param_clamp_input(dcp->bits, kp);
+
+   switch (dcp->map->map_type) {
+   case DD_CLASS_TYPE_DISJOINT_BITS:
+   v2pr_info("  %s: classbits: 0x%lx\n", KP_NAME(kp), *dcp->bits);
+   ddebug_apply_class_bitmap(dcp, dcp->bits, 0UL, modname);
+   break;
+   case DD_CLASS_TYPE_LEVEL_NUM:
+   new_bits = CLASSMAP_BITMASK(*dcp->lvl);
+   v2pr_info("  %s: lvl:%ld bits:0x%lx\n", KP_NAME(kp), *dcp->lvl, 
new_bits);
+   ddebug_apply_class_bitmap(dcp, _bits, 0UL, modname);
+   break;
+   default:
+   

[PATCH v7 07/25] dyndbg: drop NUM_TYPE_ARRAY

[Jim Cromie]

ARRAY_SIZE works here, since array decl is complete.

no functional change

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index b53217e4b711..8116d0a0d33a 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -106,11 +106,9 @@ struct ddebug_class_map {
.mod_name = KBUILD_MODNAME, \
.base = _base,  \
.map_type = _maptype,   \
-   .length = NUM_TYPE_ARGS(char*, __VA_ARGS__),\
+   .length = ARRAY_SIZE(_var##_classnames),\
.class_names = _var##_classnames,   \
}
-#define NUM_TYPE_ARGS(eltype, ...) \
-(sizeof((eltype[]){__VA_ARGS__}) / sizeof(eltype))
 
 /* encapsulate linker provided built-in (or module) dyndbg data */
 struct _ddebug_info {
-- 
2.41.0

[PATCH v7 14/25] dyndbg-API: fix CONFIG_DRM_USE_DYNAMIC_DEBUG regression

[Jim Cromie]

DECLARE_DYNDBG_CLASSMAP() has a design error; it fails a basic K
rule: "define once, refer many times".

When DRM_USE_DYNAMIC_DEBUG=y, DECLARE_DYNDBG_CLASSMAP() is used across
DRM core & drivers; they all repeat the same classmap-defn args, which
must match for the modules to respond together when DRM.debug
categories are enabled.

Worse, it causes the CONFIG_DRM_USE_DYNAMIC_DEBUG=Y regression; 1st
drm.ko loads, and dyndbg initializes its DRM.debug callsites, then a
drm-driver loads, but too late - it missed the DRM.debug enablement.

So replace it with 2 macros:
  DYNDBG_CLASSMAP_DEFINE - invoked once from core - drm.ko
  DYNDBG_CLASSMAP_USE- from all drm drivers and helpers.

DYNDBG_CLASSMAP_DEFINE: based on DECLARE_DYNDBG_CLASSMAP, but now it
drops the static on the constructed classmap variable, and exports it
instead.

DYNDBG_CLASSMAP_USE: then refers to the exported var by name:
* used from drivers, helper-mods
* lets us drop the repetitive "classname" args
* fixes 2nd-defn problem
* creates a ddebug_class_user record in new __dyndbg_class_users section
  this allows ddebug_add_module(etal) to handle them per-module.

The distinction, and the usage record, allows dyndbg to initialize the
driver's DRM.debug callsites separately after it is modprobed.

Since DRM now needs updates to use the new macros, it also gets 2
wrappers: DRM_CLASSMAP_DEFINE, DRM_CLASSMAP_USE which declutter the
users by hiding the ifdef CONFIG_DRM_USE_DYNAMIC_DEBUG.

To review, dyndbg's existing __dyndbg_classes[] section does:

. catalogs the classmaps defined by the module (or builtin modules)
. authorizes dyndbg to >control those class'd prdbgs for the module.
. DYNDBG_CLASSMAP_DEFINE(and old one) creates classmaps in this section.

This patch adds __dyndbg_class_users[] section:

. catalogs uses/references to the classmap definitions.
. authorizes dyndbg to >control those class'd prdbgs in ref'g module.
. DYNDBG_CLASSMAP_USE() creates classmap-user records in this section.

Now ddebug_add_module(etal) can handle classmap-uses like (and after)
classmaps; when a dependent module is loaded, its parent's kernel
params are scanned to find the param wired to dyndbg-param-ops, whose
classmap matches the one ref'd by the client.

To support this, theres a few data/header changes:

. new struct ddebug_class_user
  contains: user-module-name, 
  it records drm-driver's use of a classmap in the section, allowing lookup

struct ddebug_info gets 2 new fields to encapsulate the new section:
  class_users, num_class_users.
  set by dynamic_debug_init() for builtins.
  or by kernel/module/main:load_info() for loadable modules.

vmlinux.lds.h: new BOUNDED_SECTION for __dyndbg_class_users

dynamic_debug.c has 2 changes in ddebug_add_module(), ddebug_change():

ddebug_add_module() already calls ddebug_attach_module_classes()
to handle classmaps DEFINEd by a module, now it also calls
ddebug_attach_user_module_classes() to handle USEd classmaps.  To
avoid this work when possible, 1st scan the module's descriptors and
count the number of class'd pr_debugs.

ddebug_attach_user_module_classes() scans the module's class_users
section, follows the refs to the parent's classmap, and calls
ddebug_apply_params() on each.  It also avoids work by checking the
module's class-ct.

ddebug_apply_params(new fn):

It scans module's/builtin kernel-params, calls ddebug_match_apply_kparam
for each to find the params/sysfs-nodes which may be wired to a classmap.

ddebug_match_apply_kparam(new fn):

1st, it tests the kernel-param.ops is dyndbg's; this guarantees that
the attached arg is a struct ddebug_class_param, which has a ref to
the param's state, and to the classmap defining the param's handling.

2nd, it requires that the classmap ref'd by the kparam is the one
we're called for; modules can use many separate classmaps (as
test_dynamic_debug does).

Then apply the "parent" kparam's setting to the dependent module,
using ddebug_apply_class_bitmap().

ddebug_change(and callees) also gets adjustments:

ddebug_find_valid_class(): This does a search over the module's
classmaps, looking for the class FOO echo'd to >control.  So now it
searches over __dyndbg_class_users[] after __dyndbg_classes[].

ddebug_class_name(): return class-names for defined AND used classes.

test_dynamic_debug.c, test_dynamic_debug_submod.c:

This (already) demonstrates the 2 types of classmaps & sysfs-params,
following the 4-part recipe:

1. define an enum for the classmap: DRM.debug has DRM_UT_{CORE,KMS,...}
   multiple classes must share 0-62 classid space.
2. DYNDBG_CLASSMAP_DEFINE(.. DRM_UT_{CORE,KMS,...})
3. DYNDBG_CLASSMAP_PARAM* (classmap)
4. DYNDBG_CLASSMAP_USE()
   by _submod only, skipping 2,3

Move all the enum declarations together, to better explain how they
share the 0..62 class-id space available to a module (non-overlapping
subranges).

reorg macros 2,3 by name.  This gives a tabular format, making it easy
to see the pattern of repetition, and the points of change.

And 

[PATCH v7 15/25] dyndbg: add for_each_boxed_vector

[Jim Cromie]

Add a for_each iterator to walk a counted vector member in a struct
(ie the box), and use it to replace 8 open-coded loops.

Signed-off-by: Jim Cromie 
---
v5- parens-on-box-force-precedence
---
 lib/dynamic_debug.c | 20 +++-
 1 file changed, 11 insertions(+), 9 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index be49e104ec76..c11feca70d6f 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -158,6 +158,9 @@ static void vpr_info_dq(const struct ddebug_query *query, 
const char *msg)
  _dt->num_class_users);\
})
 
+#define for_each_boxed_vector(_box, _vec, _len, _ct, _curs)\
+   for (_ct = 0, _curs = (_box)->_vec; _ct < (_box)->_len; _ct++, _curs++)
+
 #define __outvar /* filled by callee */
 static struct ddebug_class_map *ddebug_find_valid_class(struct ddebug_table 
const *dt,
const char 
*class_string,
@@ -167,7 +170,7 @@ static struct ddebug_class_map 
*ddebug_find_valid_class(struct ddebug_table cons
struct ddebug_class_user *cli;
int i, idx;
 
-   for (i = 0, map = dt->classes; i < dt->num_classes; i++, map++) {
+   for_each_boxed_vector(dt, classes, num_classes, i, map) {
idx = match_string(map->class_names, map->length, class_string);
if (idx >= 0) {
*class_id = idx + map->base;
@@ -175,7 +178,7 @@ static struct ddebug_class_map 
*ddebug_find_valid_class(struct ddebug_table cons
return map;
}
}
-   for (i = 0, cli = dt->class_users; i < dt->num_class_users; i++, cli++) 
{
+   for_each_boxed_vector(dt, class_users, num_class_users, i, cli) {
idx = match_string(cli->map->class_names, cli->map->length, 
class_string);
if (idx >= 0) {
*class_id = idx + cli->map->base;
@@ -1058,11 +1061,11 @@ static const char *ddebug_class_name(struct 
ddebug_table *dt, struct _ddebug *dp
struct ddebug_class_user *cli = dt->class_users;
int i;
 
-   for (i = 0; i < dt->num_classes; i++, map++)
+   for_each_boxed_vector(dt, classes, num_classes, i, map)
if (class_in_range(dp->class_id, map))
return map->class_names[dp->class_id - map->base];
 
-   for (i = 0; i < dt->num_class_users; i++, cli++)
+   for_each_boxed_vector(dt, class_users, num_class_users, i, cli)
if (class_in_range(dp->class_id, cli->map))
return cli->map->class_names[dp->class_id - 
cli->map->base];
 
@@ -1216,7 +1219,7 @@ static void ddebug_attach_module_classes(struct 
ddebug_table *dt, struct _ddebug
struct ddebug_class_map *cm;
int i, nc = 0;
 
-   for (i = 0, cm = di->classes; i < di->num_classes; i++, cm++) {
+   for_each_boxed_vector(di, classes, num_classes, i, cm) {
 
if (!strcmp(cm->mod_name, dt->mod_name)) {
vpr_cm_info(cm, "classes[%d]:", i);
@@ -1230,7 +1233,7 @@ static void ddebug_attach_module_classes(struct 
ddebug_table *dt, struct _ddebug
vpr_info("module:%s attached %d classes\n", dt->mod_name, nc);
dt->num_classes = nc;
 
-   for (i = 0, cm = dt->classes; i < dt->num_classes; i++, cm++)
+   for_each_boxed_vector(di, classes, num_classes, i, cm)
ddebug_apply_params(cm, cm->mod_name);
 }
 
@@ -1250,7 +1253,7 @@ static void ddebug_attach_user_module_classes(struct 
ddebug_table *dt,
 * module's refs, save to dt.  For loadables, this is the
 * whole array.
 */
-   for (i = 0, cli = di->class_users; i < di->num_class_users; i++, cli++) 
{
+   for_each_boxed_vector(di, class_users, num_class_users, i, cli) {
 
if (WARN_ON(!cli || !cli->map || !cli->user_mod_name))
continue;
@@ -1268,8 +1271,7 @@ static void ddebug_attach_user_module_classes(struct 
ddebug_table *dt,
 
dt->num_class_users = nc;
 
-   /* now iterate dt */
-   for (i = 0, cli = dt->class_users; i < dt->num_class_users; i++, cli++)
+   for_each_boxed_vector(di, class_users, num_class_users, i, cli)
ddebug_apply_params(cli->map, cli->user_mod_name);
 
vpr_dt_info(dt, "attach-client-module: ");
-- 
2.41.0

[PATCH v7 09/25] dyndbg: silence debugs with no-change updates

[Jim Cromie]

check for actual changes before announcing them, declutter logs.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 12 +++-
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index b0e11f6bfaa2..b07aab422604 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -591,7 +591,7 @@ static int ddebug_exec_queries(char *query, const char 
*modname)
return nfound;
 }
 
-/* apply a new bitmap to the sys-knob's current bit-state */
+/* apply a new class-param setting */
 static int ddebug_apply_class_bitmap(const struct ddebug_class_param *dcp,
 unsigned long *new_bits, unsigned long 
*old_bits,
 const char *query_modname)
@@ -602,8 +602,9 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
int matches = 0;
int bi, ct;
 
-   v2pr_info("apply bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits, 
*old_bits,
- query_modname ?: "");
+   if (*new_bits != *old_bits)
+   v2pr_info("apply bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits,
+ *old_bits, query_modname ?: "'*'");
 
for (bi = 0; bi < map->length; bi++) {
if (test_bit(bi, new_bits) == test_bit(bi, old_bits))
@@ -618,8 +619,9 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
v2pr_info("bit_%d: %d matches on class: %s -> 0x%lx\n", bi,
  ct, map->class_names[bi], *new_bits);
}
-   v2pr_info("applied bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits, 
*old_bits,
- query_modname ?: "");
+   if (*new_bits != *old_bits)
+   v2pr_info("applied bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits,
+ *old_bits, query_modname ?: "'*'");
 
return matches;
 }
-- 
2.41.0

[PATCH v7 12/25] dyndbg: reduce verbose=3 messages in ddebug_add_module

[Jim Cromie]

The fn currently says "add-module", then "skipping" if the module has
no prdbgs.  Just check 1st and return quietly.

no functional change

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 7 +++
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 8beb98a831f5..45870a699507 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1242,11 +1242,10 @@ static int ddebug_add_module(struct _ddebug_info *di, 
const char *modname)
 {
struct ddebug_table *dt;
 
-   v3pr_info("add-module: %s.%d sites\n", modname, di->num_descs);
-   if (!di->num_descs) {
-   v3pr_info(" skip %s\n", modname);
+   if (!di->num_descs)
return 0;
-   }
+
+   v3pr_info("add-module: %s %d sites\n", modname, di->num_descs);
 
dt = kzalloc(sizeof(*dt), GFP_KERNEL);
if (dt == NULL) {
-- 
2.41.0

[PATCH v7 10/25] dyndbg: tighten ddebug_class_name() 1st arg type

[Jim Cromie]

Change function's 1st arg-type, and deref in the caller.
The fn doesn't need any other fields in the struct.

no functional change.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 10 +-
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index b07aab422604..8158943b350d 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1117,12 +1117,12 @@ static void *ddebug_proc_next(struct seq_file *m, void 
*p, loff_t *pos)
 #define class_in_range(class_id, map)  \
(class_id >= map->base && class_id < map->base + map->length)
 
-static const char *ddebug_class_name(struct ddebug_iter *iter, struct _ddebug 
*dp)
+static const char *ddebug_class_name(struct ddebug_table *dt, struct _ddebug 
*dp)
 {
-   struct ddebug_class_map *map = iter->table->classes;
-   int i, nc = iter->table->num_classes;
+   struct ddebug_class_map *map = dt->classes;
+   int i;
 
-   for (i = 0; i < nc; i++, map++)
+   for (i = 0; i < dt->num_classes; i++, map++)
if (class_in_range(dp->class_id, map))
return map->class_names[dp->class_id - map->base];
 
@@ -1156,7 +1156,7 @@ static int ddebug_proc_show(struct seq_file *m, void *p)
seq_puts(m, "\"");
 
if (dp->class_id != _DPRINTK_CLASS_DFLT) {
-   class = ddebug_class_name(iter, dp);
+   class = ddebug_class_name(iter->table, dp);
if (class)
seq_printf(m, " class:%s", class);
else
-- 
2.41.0

[PATCH v7 05/25] dyndbg: ddebug_apply_class_bitmap - add module arg, select on it

[Jim Cromie]

Add query_module param to ddebug_apply_class_bitmap().  This allows
its caller to update just one module, or all (as currently).  We'll
use this later to propagate drm.debug to each USEr as they're
modprobed.

No functional change.

Signed-off-by: Jim Cromie 
---

after `modprobe i915`, heres the module dependencies,
though not all on drm.debug.

bash-5.2# lsmod
Module  Size  Used by
i915 3133440  0
drm_buddy  20480  1 i915
ttm90112  1 i915
i2c_algo_bit   16384  1 i915
video  61440  1 i915
wmi32768  1 video
drm_display_helper200704  1 i915
drm_kms_helper208896  2 drm_display_helper,i915
drm   606208  5 
drm_kms_helper,drm_display_helper,drm_buddy,i915,ttm
cec57344  2 drm_display_helper,i915
---
 lib/dynamic_debug.c | 19 ---
 1 file changed, 12 insertions(+), 7 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index a3be2e7c8c84..ba41fdeaaf98 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -601,7 +601,8 @@ static int ddebug_exec_queries(char *query, const char 
*modname)
 
 /* apply a new bitmap to the sys-knob's current bit-state */
 static int ddebug_apply_class_bitmap(const struct ddebug_class_param *dcp,
-unsigned long *new_bits, unsigned long 
*old_bits)
+unsigned long *new_bits, unsigned long 
*old_bits,
+const char *query_modname)
 {
 #define QUERY_SIZE 128
char query[QUERY_SIZE];
@@ -609,7 +610,8 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
int matches = 0;
int bi, ct;
 
-   v2pr_info("apply: 0x%lx to: 0x%lx\n", *new_bits, *old_bits);
+   v2pr_info("apply bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits, 
*old_bits,
+ query_modname ?: "");
 
for (bi = 0; bi < map->length; bi++) {
if (test_bit(bi, new_bits) == test_bit(bi, old_bits))
@@ -618,12 +620,15 @@ static int ddebug_apply_class_bitmap(const struct 
ddebug_class_param *dcp,
snprintf(query, QUERY_SIZE, "class %s %c%s", 
map->class_names[bi],
 test_bit(bi, new_bits) ? '+' : '-', dcp->flags);
 
-   ct = ddebug_exec_queries(query, NULL);
+   ct = ddebug_exec_queries(query, query_modname);
matches += ct;
 
v2pr_info("bit_%d: %d matches on class: %s -> 0x%lx\n", bi,
  ct, map->class_names[bi], *new_bits);
}
+   v2pr_info("applied bitmap: 0x%lx to: 0x%lx for %s\n", *new_bits, 
*old_bits,
+ query_modname ?: "");
+
return matches;
 }
 
@@ -679,7 +684,7 @@ static int param_set_dyndbg_classnames(const char *instr, 
const struct kernel_pa
continue;
}
curr_bits ^= BIT(cls_id);
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
dcp->bits);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, 
dcp->bits, NULL);
*dcp->bits = curr_bits;
v2pr_info("%s: changed bit %d:%s\n", KP_NAME(kp), 
cls_id,
  map->class_names[cls_id]);
@@ -689,7 +694,7 @@ static int param_set_dyndbg_classnames(const char *instr, 
const struct kernel_pa
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
curr_bits = CLASSMAP_BITMASK(cls_id + (wanted ? 1 : 0 
));
 
-   totct += ddebug_apply_class_bitmap(dcp, _bits, 
_bits);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, 
_bits, NULL);
*dcp->lvl = (cls_id + (wanted ? 1 : 0));
v2pr_info("%s: changed bit-%d: \"%s\" %lx->%lx\n", 
KP_NAME(kp), cls_id,
  map->class_names[cls_id], old_bits, 
curr_bits);
@@ -752,7 +757,7 @@ int param_set_dyndbg_classes(const char *instr, const 
struct kernel_param *kp)
inrep &= CLASSMAP_BITMASK(map->length);
}
v2pr_info("bits:%lx > %s\n", inrep, KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, , dcp->bits);
+   totct += ddebug_apply_class_bitmap(dcp, , dcp->bits, 
NULL);
*dcp->bits = inrep;
break;
case DD_CLASS_TYPE_LEVEL_NUM:
@@ -765,7 +770,7 @@ int param_set_dyndbg_classes(const char *instr, const 
struct kernel_param *kp)
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
new_bits = CLASSMAP_BITMASK(inrep);
v2pr_info("lvl:%ld bits:0x%lx > %s\n", inrep, new_bits, 
KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, _bits, _bits);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, _bits, 
NULL);

[PATCH v7 06/25] dyndbg: split param_set_dyndbg_classes to module/wrapper fns

[Jim Cromie]

rename param_set_dyndbg_classes: add _module_ name & arg, old name is
wrapper to new.  New arg allows caller to specify that only one module
is affected by a prdbgs update.

Outer fn preserves kernel_param interface, passing NULL to inner fn.
This selectivity will be used later to narrow the scope of changes
made.

no functional change.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 37 ++---
 1 file changed, 22 insertions(+), 15 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index ba41fdeaaf98..b67c9b137447 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -708,18 +708,9 @@ static int param_set_dyndbg_classnames(const char *instr, 
const struct kernel_pa
return 0;
 }
 
-/**
- * param_set_dyndbg_classes - class FOO >control
- * @instr: string echo>d to sysfs, input depends on map_type
- * @kp:kp->arg has state: bits/lvl, map, map_type
- *
- * Enable/disable prdbgs by their class, as given in the arguments to
- * DECLARE_DYNDBG_CLASSMAP.  For LEVEL map-types, enforce relative
- * levels by bitpos.
- *
- * Returns: 0 or <0 if error.
- */
-int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp)
+static int param_set_dyndbg_module_classes(const char *instr,
+  const struct kernel_param *kp,
+  const char *modnm)
 {
const struct ddebug_class_param *dcp = kp->arg;
const struct ddebug_class_map *map = dcp->map;
@@ -756,8 +747,8 @@ int param_set_dyndbg_classes(const char *instr, const 
struct kernel_param *kp)
KP_NAME(kp), inrep, 
CLASSMAP_BITMASK(map->length));
inrep &= CLASSMAP_BITMASK(map->length);
}
-   v2pr_info("bits:%lx > %s\n", inrep, KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, , dcp->bits, 
NULL);
+   v2pr_info("bits:0x%lx > %s.%s\n", inrep, modnm ?: "*", 
KP_NAME(kp));
+   totct += ddebug_apply_class_bitmap(dcp, , dcp->bits, 
modnm);
*dcp->bits = inrep;
break;
case DD_CLASS_TYPE_LEVEL_NUM:
@@ -770,7 +761,7 @@ int param_set_dyndbg_classes(const char *instr, const 
struct kernel_param *kp)
old_bits = CLASSMAP_BITMASK(*dcp->lvl);
new_bits = CLASSMAP_BITMASK(inrep);
v2pr_info("lvl:%ld bits:0x%lx > %s\n", inrep, new_bits, 
KP_NAME(kp));
-   totct += ddebug_apply_class_bitmap(dcp, _bits, _bits, 
NULL);
+   totct += ddebug_apply_class_bitmap(dcp, _bits, _bits, 
modnm);
*dcp->lvl = inrep;
break;
default:
@@ -779,6 +770,22 @@ int param_set_dyndbg_classes(const char *instr, const 
struct kernel_param *kp)
vpr_info("%s: total matches: %d\n", KP_NAME(kp), totct);
return 0;
 }
+
+/**
+ * param_set_dyndbg_classes - class FOO >control
+ * @instr: string echo>d to sysfs, input depends on map_type
+ * @kp:kp->arg has state: bits/lvl, map, map_type
+ *
+ * Enable/disable prdbgs by their class, as given in the arguments to
+ * DECLARE_DYNDBG_CLASSMAP.  For LEVEL map-types, enforce relative
+ * levels by bitpos.
+ *
+ * Returns: 0 or <0 if error.
+ */
+int param_set_dyndbg_classes(const char *instr, const struct kernel_param *kp)
+{
+   return param_set_dyndbg_module_classes(instr, kp, NULL);
+}
 EXPORT_SYMBOL(param_set_dyndbg_classes);
 
 /**
-- 
2.41.0

[PATCH v7 08/25] dyndbg: reduce verbose/debug clutter

[Jim Cromie]

currently, for verbose=3, these are logged (blank lines for clarity):

 dyndbg: query 0: "class DRM_UT_CORE +p" mod:*
 dyndbg: split into words: "class" "DRM_UT_CORE" "+p"

 dyndbg: op='+'
 dyndbg: flags=0x1
 dyndbg: *flagsp=0x1 *maskp=0x

 dyndbg: parsed: func="" file="" module="" format="" lineno=0-0 class=...
 dyndbg: no matches for query
 dyndbg: no-match: func="" file="" module="" format="" lineno=0-0 class=...
 dyndbg: processed 1 queries, with 0 matches, 0 errs

That is excessive, so this patch:
 - shrinks 3 lines of 2nd stanza to single line
 - drops 1st 2 lines of 3rd stanza
   3rd is like 1st, with result, not procedure.
   2nd is just status, retold in 4th, with more info.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 14 +++---
 1 file changed, 3 insertions(+), 11 deletions(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index b67c9b137447..b0e11f6bfaa2 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -266,9 +266,6 @@ static int ddebug_change(const struct ddebug_query *query,
}
mutex_unlock(_lock);
 
-   if (!nfound && verbose)
-   pr_info("no matches for query\n");
-
return nfound;
 }
 
@@ -497,7 +494,6 @@ static int ddebug_parse_flags(const char *str, struct 
flag_settings *modifiers)
pr_err("bad flag-op %c, at start of %s\n", *str, str);
return -EINVAL;
}
-   v3pr_info("op='%c'\n", op);
 
for (; *str ; ++str) {
for (i = ARRAY_SIZE(opt_array) - 1; i >= 0; i--) {
@@ -511,7 +507,6 @@ static int ddebug_parse_flags(const char *str, struct 
flag_settings *modifiers)
return -EINVAL;
}
}
-   v3pr_info("flags=0x%x\n", modifiers->flags);
 
/* calculate final flags, mask based upon op */
switch (op) {
@@ -527,7 +522,7 @@ static int ddebug_parse_flags(const char *str, struct 
flag_settings *modifiers)
modifiers->flags = 0;
break;
}
-   v3pr_info("*flagsp=0x%x *maskp=0x%x\n", modifiers->flags, 
modifiers->mask);
+   v3pr_info("op='%c' flags=0x%x maskp=0x%x\n", op, modifiers->flags, 
modifiers->mask);
 
return 0;
 }
@@ -537,7 +532,7 @@ static int ddebug_exec_query(char *query_string, const char 
*modname)
struct flag_settings modifiers = {};
struct ddebug_query query = {};
 #define MAXWORDS 9
-   int nwords, nfound;
+   int nwords;
char *words[MAXWORDS];
 
nwords = ddebug_tokenize(query_string, words, MAXWORDS);
@@ -555,10 +550,7 @@ static int ddebug_exec_query(char *query_string, const 
char *modname)
return -EINVAL;
}
/* actually go and implement the change */
-   nfound = ddebug_change(, );
-   vpr_info_dq(, nfound ? "applied" : "no-match");
-
-   return nfound;
+   return ddebug_change(, );
 }
 
 /* handle multiple queries in query string, continue on error, return
-- 
2.41.0

[PATCH v7 03/25] dyndbg: make ddebug_class_param union members same size

[Jim Cromie]

struct ddebug_class_param keeps a ref to the state-storage of the
param, make both flavors use the same unsigned long under-type.
ISTM this is simpler and safer.

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h | 2 +-
 lib/dynamic_debug.c   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index 4fcbf4d4fd0a..5231aaf361c4 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -124,7 +124,7 @@ struct _ddebug_info {
 struct ddebug_class_param {
union {
unsigned long *bits;
-   unsigned int *lvl;
+   unsigned long *lvl;
};
char flags[8];
const struct ddebug_class_map *map;
diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index ceb3067a5c83..b984ce338921 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -796,7 +796,7 @@ int param_get_dyndbg_classes(char *buffer, const struct 
kernel_param *kp)
 
case DD_CLASS_TYPE_LEVEL_NAMES:
case DD_CLASS_TYPE_LEVEL_NUM:
-   return scnprintf(buffer, PAGE_SIZE, "%d\n", *dcp->lvl);
+   return scnprintf(buffer, PAGE_SIZE, "%ld\n", *dcp->lvl);
default:
return -1;
}
-- 
2.41.0

[PATCH v7 04/25] dyndbg: replace classmap list with a vector

[Jim Cromie]

Classmaps are stored/linked in a section/array, but are each added to
the module's ddebug_table.maps list-head.

This is unnecessary; even when ddebug_attach_classmap() is handling
the builtin section (with classmaps for multiple builtin modules), its
contents are ordered, so a module's possibly multiple classmaps will
be consecutive in the section, and could be treated as a vector/block,
since both start-addy and subrange length are in the ddebug_info arg.

So this changes:

struct ddebug_class_map drops list-head link.

struct ddebug_table drops the list-head maps, and gets: classes &
num_classes for the start-addy and num_classes, placed to improve
struct packing.

The loading: in ddebug_attach_module_classes(), replace the
for-the-modname list-add loop, with a forloop that finds the module's
subrange (start,length) of matching classmaps within the possibly
builtin classmaps vector, and saves those to the ddebug_table.

The reading/using: change list-foreach loops in ddebug_class_name() &
ddebug_find_valid_class() to walk the array from start to length.

Also:
Move #define __outvar up, above an added use in a fn-prototype.
Simplify ddebug_attach_module_classes args, ref has both addy,len.

no functional changes

Signed-off-by: Jim Cromie 
---
 include/linux/dynamic_debug.h |  1 -
 lib/dynamic_debug.c   | 61 ++-
 2 files changed, 32 insertions(+), 30 deletions(-)

diff --git a/include/linux/dynamic_debug.h b/include/linux/dynamic_debug.h
index 5231aaf361c4..b53217e4b711 100644
--- a/include/linux/dynamic_debug.h
+++ b/include/linux/dynamic_debug.h
@@ -83,7 +83,6 @@ enum class_map_type {
 };
 
 struct ddebug_class_map {
-   struct list_head link;
struct module *mod;
const char *mod_name;   /* needed for builtins */
const char **class_names;
diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index b984ce338921..a3be2e7c8c84 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -45,10 +45,11 @@ extern struct ddebug_class_map __start___dyndbg_classes[];
 extern struct ddebug_class_map __stop___dyndbg_classes[];
 
 struct ddebug_table {
-   struct list_head link, maps;
+   struct list_head link;
const char *mod_name;
-   unsigned int num_ddebugs;
struct _ddebug *ddebugs;
+   struct ddebug_class_map *classes;
+   unsigned int num_ddebugs, num_classes;
 };
 
 struct ddebug_query {
@@ -147,13 +148,15 @@ static void vpr_info_dq(const struct ddebug_query *query, 
const char *msg)
  query->first_lineno, query->last_lineno, query->class_string);
 }
 
+#define __outvar /* filled by callee */
 static struct ddebug_class_map *ddebug_find_valid_class(struct ddebug_table 
const *dt,
- const char 
*class_string, int *class_id)
+   const char 
*class_string,
+   __outvar int *class_id)
 {
struct ddebug_class_map *map;
-   int idx;
+   int i, idx;
 
-   list_for_each_entry(map, >maps, link) {
+   for (map = dt->classes, i = 0; i < dt->num_classes; i++, map++) {
idx = match_string(map->class_names, map->length, class_string);
if (idx >= 0) {
*class_id = idx + map->base;
@@ -164,7 +167,6 @@ static struct ddebug_class_map 
*ddebug_find_valid_class(struct ddebug_table cons
return NULL;
 }
 
-#define __outvar /* filled by callee */
 /*
  * Search the tables for _ddebug's which match the given `query' and
  * apply the `flags' and `mask' to them.  Returns number of matching
@@ -,9 +1113,10 @@ static void *ddebug_proc_next(struct seq_file *m, void 
*p, loff_t *pos)
 
 static const char *ddebug_class_name(struct ddebug_iter *iter, struct _ddebug 
*dp)
 {
-   struct ddebug_class_map *map;
+   struct ddebug_class_map *map = iter->table->classes;
+   int i, nc = iter->table->num_classes;
 
-   list_for_each_entry(map, >table->maps, link)
+   for (i = 0; i < nc; i++, map++)
if (class_in_range(dp->class_id, map))
return map->class_names[dp->class_id - map->base];
 
@@ -1197,30 +1200,31 @@ static const struct proc_ops proc_fops = {
.proc_write = ddebug_proc_write
 };
 
-static void ddebug_attach_module_classes(struct ddebug_table *dt,
-struct ddebug_class_map *classes,
-int num_classes)
+static void ddebug_attach_module_classes(struct ddebug_table *dt, struct 
_ddebug_info *di)
 {
struct ddebug_class_map *cm;
-   int i, j, ct = 0;
+   int i, nc = 0;
 
-   for (cm = classes, i = 0; i < num_classes; i++, cm++) {
+   /*
+* Find this module's classmaps in a subrange/wholerange of
+* the builtin/modular classmap vector/section.  Save the start
+* and length of the subrange 

[PATCH v7 01/25] test-dyndbg: fixup CLASSMAP usage error

[Jim Cromie]

more careful reading of test output reveals:

lib/test_dynamic_debug.c:103 [test_dynamic_debug]do_cats =pmf "doing 
categories\n"
lib/test_dynamic_debug.c:105 [test_dynamic_debug]do_cats =p "LOW msg\n" 
class:MID
lib/test_dynamic_debug.c:106 [test_dynamic_debug]do_cats =p "MID msg\n" class:HI
lib/test_dynamic_debug.c:107 [test_dynamic_debug]do_cats =_ "HI msg\n" class 
unknown, _id:13

That last line is wrong, the HI class is declared.

But the enum's 1st val (explicitly initialized) was wrong; it must be
_base, not _base+1 (a DECLARE_DYNDBG_CLASSMAP[1] param).  So the last
enumeration exceeded the range of mapped class-id's, which triggered
the "class unknown" report.  I intentionally coded in an error, but
forgot to verify its detection and remove it.

RFC:

This patch fixes a bad usage of DECLARE_DYNDBG_CLASSMAP(), showing
that it is too error-prone.  As noted in test-mod comments:

 * Using the CLASSMAP api:
 * - classmaps must have corresponding enum
 * - enum symbols must match/correlate with class-name strings in the map.
 * - base must equal enum's 1st value
 * - multiple maps must set their base to share the 0-62 class_id space !!
 *   (build-bug-on tips welcome)

Those shortcomings could largely be fixed with a __stringify_list
(which doesn't exist,) used in DECLARE_DYNDBG_CLASSMAP to stringify
__VA_ARGS__.  Then, API would accept DRM_UT_* values literally; all
the categories, in order, and not their stringifications, which
created all the usage complications above.

[1] name changes later to DYNDBG_CLASSMAP_DEFINE

Signed-off-by: Jim Cromie 
---
 lib/test_dynamic_debug.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/test_dynamic_debug.c b/lib/test_dynamic_debug.c
index 8dd250ad022b..a01f0193a419 100644
--- a/lib/test_dynamic_debug.c
+++ b/lib/test_dynamic_debug.c
@@ -75,7 +75,7 @@ DD_SYS_WRAP(disjoint_bits, p);
 DD_SYS_WRAP(disjoint_bits, T);
 
 /* symbolic input, independent bits */
-enum cat_disjoint_names { LOW = 11, MID, HI };
+enum cat_disjoint_names { LOW = 10, MID, HI };
 DECLARE_DYNDBG_CLASSMAP(map_disjoint_names, DD_CLASS_TYPE_DISJOINT_NAMES, 10,
"LOW", "MID", "HI");
 DD_SYS_WRAP(disjoint_names, p);
-- 
2.41.0

[PATCH v7 02/25] dyndbg: reword "class unknown, " to "class:_UNKNOWN_"

[Jim Cromie]

This appears in the control-file to report an unknown class-name, which
indicates that the class_id is not authorized, and dyndbg will ignore
changes to it.  Generally, this means that a DYNDBG_CLASSMAP_DEFINE or
DYNDBG_CLASSMAP_USE is missing.

But the word "unknown" appears in quite a few prdbg formats, so thats
a suboptimal search term to find occurrences of the problem.  Thus
change it to "_UNKNOWN_" which properly shouts the condition.

Signed-off-by: Jim Cromie 
---
 lib/dynamic_debug.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 6fba6423cc10..ceb3067a5c83 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -1151,7 +1151,7 @@ static int ddebug_proc_show(struct seq_file *m, void *p)
if (class)
seq_printf(m, " class:%s", class);
else
-   seq_printf(m, " class unknown, _id:%d", dp->class_id);
+   seq_printf(m, " class:_UNKNOWN_ _id:%d", dp->class_id);
}
seq_puts(m, "\n");
 
-- 
2.41.0

[PATCH v7 00/25] fix DRM_USE_DYNAMIC_DEBUG=y regression

[Jim Cromie]

hi Jason, DRM-folk

This patchest fixes the chicken-egg initialization problem in the 1st
version of ddebug-class-maps, that DRM-CI uncovered.

The root-problem was DECLARE_DYNDBG_CLASSMAP, which broke the K rule:
"define once, refer many".  In patch 14 it is replaced by:

 DYNDBG_CLASSMAP_DEFINE - define and export a struct ddebug_class_map
 DYNDBG_CLASSMAP_USE - ref the exported struct

test-dynamic-debug is also extended with a -submod.ko, in order to
recapitulate the drm & drivers initialization scenario.

They're on v6.6-rc5 now, and apply cleanly to drm-tip/drm-tip.

Ive been running recent revs on rc3+, on my desktop and laptop.

The final blocker was a missing __align(8) on the ddebug_class_user
record inserted by DYNDBG_CLASSMAP_USE.  This caused DRM=y (builtin
only) to have a corrupt record for drm_kms_helper (builtin dependent).
Curiously, a clang build did not exhibit this problem.

Heres a part of dmesg, for a DRM=y kernel, booted with
 dynamic_debug.verbose=3 drm.debug=0x10

[0.466747] dyndbg: add-module: drm 406 sites
[0.467569] dyndbg: classes[0]: module:drm base:0 len:10 type:DISJOINT_BITS
[0.467743] dyndbg: module:drm attached 1 classes
[0.468557] dyndbg: builtin class: module:drm base:0 len:10 
type:DISJOINT_BITS
[0.468742] dyndbg:  found kp:drm.debug =0x10
[0.468743] dyndbg:   mapped to: module:drm base:0 len:10 type:DISJOINT_BITS
[0.469742] dyndbg:   drm.debug: classbits: 0x10
[0.470573] dyndbg: apply bitmap: 0x10 to: 0x0 for drm
[0.470743] dyndbg: query 0: "class DRM_UT_ATOMIC +p" mod:drm
[0.471743] dyndbg: split into words: "class" "DRM_UT_ATOMIC" "+p"
[0.472743] dyndbg: op='+' flags=0x1 maskp=0x
[0.473679] dyndbg: parsed: func="" file="" module="drm" format="" 
lineno=0-0 class=DRM_UT_ATOMIC
[0.473749] dyndbg: processed 1 queries, with 0 matches, 0 errs
[0.474742] dyndbg: bit_4: 0 matches on class: DRM_UT_ATOMIC -> 0x10
[0.475742] dyndbg: applied bitmap: 0x10 to: 0x0 for drm
[0.476686] dyndbg: 406 debug prints in module drm
[0.476743] dyndbg: add-module: drm_kms_helper 93 sites
[0.477727] dyndbg: class_ref[0] drm_kms_helper -> drm module:drm base:0 
len:10 type:DISJOINT_BITS
[0.477743] dyndbg: builtin class: module:drm base:0 len:10 
type:DISJOINT_BITS
[0.478742] dyndbg:  found kp:drm.debug =0x10
[0.478743] dyndbg:   mapped to: module:drm base:0 len:10 type:DISJOINT_BITS
[0.479743] dyndbg:   drm.debug: classbits: 0x10
[0.480592] dyndbg: apply bitmap: 0x10 to: 0x0 for drm_kms_helper
[0.480743] dyndbg: query 0: "class DRM_UT_ATOMIC +p" mod:drm_kms_helper
[0.481743] dyndbg: split into words: "class" "DRM_UT_ATOMIC" "+p"
[0.482743] dyndbg: op='+' flags=0x1 maskp=0x
[0.483743] dyndbg: parsed: func="" file="" module="drm_kms_helper" 
format="" lineno=0-0 class=DRM_UT_ATOMIC
[0.484750] dyndbg: class-ref: drm_kms_helper.DRM_UT_ATOMIC  
module:drm_kms_helper nd:93 nc:0 nu:1
[0.485809] dyndbg: processed 1 queries, with 44 matches, 0 errs
[0.486742] dyndbg: bit_4: 44 matches on class: DRM_UT_ATOMIC -> 0x10
[0.487742] dyndbg: applied bitmap: 0x10 to: 0x0 for drm_kms_helper
[0.488743] dyndbg: attach-client-module:  module:drm_kms_helper nd:93 nc:0 
nu:1
[0.489742] dyndbg:  93 debug prints in module drm_kms_helper

Widespread testing is appreciated.
I have scripts if anyone wants them.

I'll forward lkp-robot reports here when I get them.

Jim Cromie (25):
  test-dyndbg: fixup CLASSMAP usage error
  dyndbg: reword "class unknown," to "class:_UNKNOWN_"
  dyndbg: make ddebug_class_param union members same size
  dyndbg: replace classmap list with a vector
  dyndbg: ddebug_apply_class_bitmap - add module arg, select on it
  dyndbg: split param_set_dyndbg_classes to module/wrapper fns
  dyndbg: drop NUM_TYPE_ARRAY
  dyndbg: reduce verbose/debug clutter
  dyndbg: silence debugs with no-change updates
  dyndbg: tighten ddebug_class_name() 1st arg type
  dyndbg: tighten fn-sig of ddebug_apply_class_bitmap
  dyndbg: reduce verbose=3 messages in ddebug_add_module
  dyndbg-API: remove DD_CLASS_TYPE_(DISJOINT|LEVEL)_NAMES and code
  dyndbg-API: fix CONFIG_DRM_USE_DYNAMIC_DEBUG regression
  dyndbg: add for_each_boxed_vector
  dyndbg: refactor ddebug_classparam_clamp_input
  dyndbg-API: promote DYNDBG_CLASSMAP_PARAM to API
  dyndbg-doc: add classmap info to howto
  dyndbg: reserve flag bit _DPRINTK_FLAGS_PREFIX_CACHED
  dyndbg: add _DPRINTK_FLAGS_INCL_LOOKUP
  dyndbg: refactor *dynamic_emit_prefix
  dyndbg: improve err report in attach_user_module_classes
  drm: use correct ccflags-y spelling
  drm-drivers: DRM_CLASSMAP_USE in 2nd batch of drivers, helpers
  drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN

 .../admin-guide/dynamic-debug-howto.rst   |  59 ++-
 MAINTAINERS   |   2 +-
 drivers/gpu/drm/Kconfig   |   3 +-
 drivers/gpu/drm/Makefile  |   3 +-
 

[PATCH v7 00/25] fix DRM_USE_DYNAMIC_DEBUG=y regression

[Jim Cromie]

hi Jason, DRM-folk

This patchest fixes the chicken-egg initialization problem in the 1st
version of ddebug-class-maps, that DRM-CI uncovered.

The root-problem was DECLARE_DYNDBG_CLASSMAP, which broke the K rule:
"define once, refer many".  In patch 14 it is replaced by:

 DYNDBG_CLASSMAP_DEFINE - define and export a struct ddebug_class_map
 DYNDBG_CLASSMAP_USE - ref the exported struct

test-dynamic-debug is also extended with a -submod.ko, in order to
recapitulate the drm & drivers initialization scenario.

They're on v6.6-rc5 now, and apply cleanly to drm-tip/drm-tip.

Ive been running recent revs on rc3+, on my desktop and laptop.

The final blocker was a missing __align(8) on the ddebug_class_user
record inserted by DYNDBG_CLASSMAP_USE.  This caused DRM=y (builtin
only) to have a corrupt record for drm_kms_helper (builtin dependent).
Curiously, a clang build did not exhibit this problem.

Heres a part of dmesg, for a DRM=y kernel, booted with
 dynamic_debug.verbose=3 drm.debug=0x10

[0.466747] dyndbg: add-module: drm 406 sites
[0.467569] dyndbg: classes[0]: module:drm base:0 len:10 type:DISJOINT_BITS
[0.467743] dyndbg: module:drm attached 1 classes
[0.468557] dyndbg: builtin class: module:drm base:0 len:10 
type:DISJOINT_BITS
[0.468742] dyndbg:  found kp:drm.debug =0x10
[0.468743] dyndbg:   mapped to: module:drm base:0 len:10 type:DISJOINT_BITS
[0.469742] dyndbg:   drm.debug: classbits: 0x10
[0.470573] dyndbg: apply bitmap: 0x10 to: 0x0 for drm
[0.470743] dyndbg: query 0: "class DRM_UT_ATOMIC +p" mod:drm
[0.471743] dyndbg: split into words: "class" "DRM_UT_ATOMIC" "+p"
[0.472743] dyndbg: op='+' flags=0x1 maskp=0x
[0.473679] dyndbg: parsed: func="" file="" module="drm" format="" 
lineno=0-0 class=DRM_UT_ATOMIC
[0.473749] dyndbg: processed 1 queries, with 0 matches, 0 errs
[0.474742] dyndbg: bit_4: 0 matches on class: DRM_UT_ATOMIC -> 0x10
[0.475742] dyndbg: applied bitmap: 0x10 to: 0x0 for drm
[0.476686] dyndbg: 406 debug prints in module drm
[0.476743] dyndbg: add-module: drm_kms_helper 93 sites
[0.477727] dyndbg: class_ref[0] drm_kms_helper -> drm module:drm base:0 
len:10 type:DISJOINT_BITS
[0.477743] dyndbg: builtin class: module:drm base:0 len:10 
type:DISJOINT_BITS
[0.478742] dyndbg:  found kp:drm.debug =0x10
[0.478743] dyndbg:   mapped to: module:drm base:0 len:10 type:DISJOINT_BITS
[0.479743] dyndbg:   drm.debug: classbits: 0x10
[0.480592] dyndbg: apply bitmap: 0x10 to: 0x0 for drm_kms_helper
[0.480743] dyndbg: query 0: "class DRM_UT_ATOMIC +p" mod:drm_kms_helper
[0.481743] dyndbg: split into words: "class" "DRM_UT_ATOMIC" "+p"
[0.482743] dyndbg: op='+' flags=0x1 maskp=0x
[0.483743] dyndbg: parsed: func="" file="" module="drm_kms_helper" 
format="" lineno=0-0 class=DRM_UT_ATOMIC
[0.484750] dyndbg: class-ref: drm_kms_helper.DRM_UT_ATOMIC  
module:drm_kms_helper nd:93 nc:0 nu:1
[0.485809] dyndbg: processed 1 queries, with 44 matches, 0 errs
[0.486742] dyndbg: bit_4: 44 matches on class: DRM_UT_ATOMIC -> 0x10
[0.487742] dyndbg: applied bitmap: 0x10 to: 0x0 for drm_kms_helper
[0.488743] dyndbg: attach-client-module:  module:drm_kms_helper nd:93 nc:0 
nu:1
[0.489742] dyndbg:  93 debug prints in module drm_kms_helper

Widespread testing is appreciated.
I have scripts if anyone wants them.

I'll forward lkp-robot reports here when I get them.

Jim Cromie (25):
  test-dyndbg: fixup CLASSMAP usage error
  dyndbg: reword "class unknown," to "class:_UNKNOWN_"
  dyndbg: make ddebug_class_param union members same size
  dyndbg: replace classmap list with a vector
  dyndbg: ddebug_apply_class_bitmap - add module arg, select on it
  dyndbg: split param_set_dyndbg_classes to module/wrapper fns
  dyndbg: drop NUM_TYPE_ARRAY
  dyndbg: reduce verbose/debug clutter
  dyndbg: silence debugs with no-change updates
  dyndbg: tighten ddebug_class_name() 1st arg type
  dyndbg: tighten fn-sig of ddebug_apply_class_bitmap
  dyndbg: reduce verbose=3 messages in ddebug_add_module
  dyndbg-API: remove DD_CLASS_TYPE_(DISJOINT|LEVEL)_NAMES and code
  dyndbg-API: fix CONFIG_DRM_USE_DYNAMIC_DEBUG regression
  dyndbg: add for_each_boxed_vector
  dyndbg: refactor ddebug_classparam_clamp_input
  dyndbg-API: promote DYNDBG_CLASSMAP_PARAM to API
  dyndbg-doc: add classmap info to howto
  dyndbg: reserve flag bit _DPRINTK_FLAGS_PREFIX_CACHED
  dyndbg: add _DPRINTK_FLAGS_INCL_LOOKUP
  dyndbg: refactor *dynamic_emit_prefix
  dyndbg: improve err report in attach_user_module_classes
  drm: use correct ccflags-y spelling
  drm-drivers: DRM_CLASSMAP_USE in 2nd batch of drivers, helpers
  drm: restore CONFIG_DRM_USE_DYNAMIC_DEBUG un-BROKEN

 .../admin-guide/dynamic-debug-howto.rst   |  59 ++-
 MAINTAINERS   |   2 +-
 drivers/gpu/drm/Kconfig   |   3 +-
 drivers/gpu/drm/Makefile  |   3 +-
 

[PATCH] drm/amdgpu: Add EXT_COHERENT override for NUMA local nodes

[David Francis]

On NUMA systems, local memory gets the local mtype, set by an
override callback. If EXT_COHERENT is set, memory will be set as
MTYPE_UC by default, with local memory MTYPE_CC.

Add an option in the override function for this case, and
add a check to ensure it is not used on UNCACHED memory.

Signed-off-by: David Francis 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c| 13 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h|  8 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 33 +++
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c  |  6 ++---
 5 files changed, 41 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 46d27c87..189341944bf1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -761,6 +761,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * @immediate: immediate submission in a page fault
  * @unlocked: unlocked invalidation during MM callback
  * @flush_tlb: trigger tlb invalidation after update completed
+ * @allow_override: change MTYPE for local NUMA nodes
  * @resv: fences we need to sync to
  * @start: start of mapped range
  * @last: last mapped entry
@@ -777,7 +778,7 @@ static void amdgpu_vm_tlb_seq_cb(struct dma_fence *fence,
  * 0 for success, negative erro code for failure.
  */
 int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-  bool immediate, bool unlocked, bool flush_tlb,
+  bool immediate, bool unlocked, bool flush_tlb, bool 
allow_override,
   struct dma_resv *resv, uint64_t start, uint64_t last,
   uint64_t flags, uint64_t offset, uint64_t vram_base,
   struct ttm_resource *res, dma_addr_t *pages_addr,
@@ -815,6 +816,7 @@ int amdgpu_vm_update_range(struct amdgpu_device *adev, 
struct amdgpu_vm *vm,
params.immediate = immediate;
params.pages_addr = pages_addr;
params.unlocked = unlocked;
+   params.allow_override = allow_override;
 
/* Implicitly sync to command submissions in the same VM before
 * unmapping. Sync to moving fences before mapping.
@@ -1062,6 +1064,7 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev, 
struct amdgpu_bo_va *bo_va,
trace_amdgpu_vm_bo_update(mapping);
 
r = amdgpu_vm_update_range(adev, vm, false, false, flush_tlb,
+  !(bo->flags & 
AMDGPU_GEM_CREATE_UNCACHED),
   resv, mapping->start, mapping->last,
   update_flags, mapping->offset,
   vram_base, mem, pages_addr,
@@ -1257,8 +1260,8 @@ int amdgpu_vm_clear_freed(struct amdgpu_device *adev,
mapping->start < AMDGPU_GMC_HOLE_START)
init_pte_value = AMDGPU_PTE_DEFAULT_ATC;
 
-   r = amdgpu_vm_update_range(adev, vm, false, false, true, resv,
-  mapping->start, mapping->last,
+   r = amdgpu_vm_update_range(adev, vm, false, false, true, false,
+  resv, mapping->start, mapping->last,
   init_pte_value, 0, 0, NULL, NULL,
   );
amdgpu_vm_free_mapping(adev, vm, mapping, f);
@@ -2546,8 +2549,8 @@ bool amdgpu_vm_handle_fault(struct amdgpu_device *adev, 
u32 pasid,
goto error_unlock;
}
 
-   r = amdgpu_vm_update_range(adev, vm, true, false, false, NULL, addr,
-  addr, flags, value, 0, NULL, NULL, NULL);
+   r = amdgpu_vm_update_range(adev, vm, true, false, false, false,
+  NULL, addr, addr, flags, value, 0, NULL, 
NULL, NULL);
if (r)
goto error_unlock;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 6e71978db13f..9ea8b5b644e3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -244,6 +244,12 @@ struct amdgpu_vm_update_params {
 * @table_freed: return true if page table is freed when updating
 */
bool table_freed;
+
+   /**
+* @allow_override: true for memory that is not uncached: allows MTYPE
+* to be overridden for NUMA local memory.
+*/
+   bool allow_override;
 };
 
 struct amdgpu_vm_update_funcs {
@@ -436,7 +442,7 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base,
struct amdgpu_vm *vm, struct amdgpu_bo *bo);
 int amdgpu_vm_update_range(struct amdgpu_device *adev, struct amdgpu_vm *vm,
-  

[PATCH v2] drm/amdgpu: Correctly use bo_va->ref_count in compute VMs

[Xiaogang . Chen]

From: Xiaogang Chen 

This is needed to correctly handle BOs imported into compute VM from gfx.
Both kfd and gfx should use same bo_va and set bo_va->ref_count correctly
when map the Bos into same VM, otherwise we may trigger kernel general
protection when iterate mappings over bo_va's valids or invalids list.

Signed-off-by: Felix Kuehling 
Signed-off-by: Xiaogang Chen 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Tested-by: Xiaogang Chen 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
uint64_t va = mem->va;
struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
struct amdgpu_bo *bo[2] = {NULL, NULL};
+   struct amdgpu_bo_va *bo_va;
bool same_hive = false;
int i, ret;
 
@@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
pr_debug("Unable to reserve BO during memory attach");
goto unwind;
}
-   attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+   if (!bo_va)
+   bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   else
+   ++bo_va->ref_count;
+   attachment[i]->bo_va = bo_va;
+
amdgpu_bo_unreserve(bo[i]);
if (unlikely(!attachment[i]->bo_va)) {
ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
continue;
if (attachment[i]->bo_va) {
amdgpu_bo_reserve(bo[i], true);
-   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+   if (--attachment[i]->bo_va->ref_count == 0)
+   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
amdgpu_bo_unreserve(bo[i]);
list_del([i]->list);
}
-- 
2.25.1

Re: [PATCH] drm/radeon: fix a possible null pointer dereference

[Alex Deucher]

On Tue, Oct 10, 2023 at 9:32 PM Ma Ke  wrote:
>
> In radeon_tv_get_modes(), the return value of drm_cvt_mode()
> is assigned to mode, which will lead to a NULL pointer
> dereference on failure of drm_cvt_mode(). Add a check to
> avoid null point dereference.
>
> Signed-off-by: Ma Ke 

Applied.  Thanks!

Alex

> ---
>  drivers/gpu/drm/radeon/radeon_connectors.c | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_connectors.c 
> b/drivers/gpu/drm/radeon/radeon_connectors.c
> index d2f02c3dfce2..b84b58926106 100644
> --- a/drivers/gpu/drm/radeon/radeon_connectors.c
> +++ b/drivers/gpu/drm/radeon/radeon_connectors.c
> @@ -1119,6 +1119,8 @@ static int radeon_tv_get_modes(struct drm_connector 
> *connector)
> else {
> /* only 800x600 is supported right now on pre-avivo chips */
> tv_mode = drm_cvt_mode(dev, 800, 600, 60, false, false, 
> false);
> +   if (!tv_mode)
> +   return 0;
> tv_mode->type = DRM_MODE_TYPE_DRIVER | 
> DRM_MODE_TYPE_PREFERRED;
> drm_mode_probed_add(connector, tv_mode);
> }
> --
> 2.37.2
>

Re: [PATCH] Find bo_va before create it when map bo into compute VM

[Chen, Xiaogang]

On 10/12/2023 10:20 AM, Felix Kuehling wrote:

On 2023-10-11 19:36, Xiaogang.Chen wrote:

From: Xiaogang Chen 


Since you are the author of this updated patch, you should also add 
your Signed-off-by below.

ok, thanks.





This is needed to correctly handle BOs imported into compute VM from 
gfx.
Both kfd and gfx should use same bo_va when map the Bos into same VM, 
otherwise
we may trigger kernel general protection when iterate mappings from 
bo_va.


Signed-off-by: Felix Kuehling 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Reviewed-By: Xiaogang Chen 
Tested-By: Xiaogang Chen 


You lost any mention of reference counting from the commit headline 
and description. I think that's still an important concept that should 
be mentioned. I'd change the headline to something like this:


drm/amdgpu: Correctly use bo_va->ref_count in compute VMs


ok, will update commit headline and message and resend.

Regards

Xiaogang


Regards,
  Felix



---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  uint64_t va = mem->va;
  struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
  struct amdgpu_bo *bo[2] = {NULL, NULL};
+    struct amdgpu_bo_va *bo_va;
  bool same_hive = false;
  int i, ret;
  @@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  pr_debug("Unable to reserve BO during memory attach");
  goto unwind;
  }
-    attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+    if (!bo_va)
+    bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    else
+    ++bo_va->ref_count;
+    attachment[i]->bo_va = bo_va;
+
  amdgpu_bo_unreserve(bo[i]);
  if (unlikely(!attachment[i]->bo_va)) {
  ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  continue;
  if (attachment[i]->bo_va) {
  amdgpu_bo_reserve(bo[i], true);
-    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+    if (--attachment[i]->bo_va->ref_count == 0)
+    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
  amdgpu_bo_unreserve(bo[i]);
  list_del([i]->list);
  }

Re: [PATCH] Find bo_va before create it when map bo into compute VM

[Felix Kuehling]

On 2023-10-11 19:36, Xiaogang.Chen wrote:

From: Xiaogang Chen 


Since you are the author of this updated patch, you should also add your 
Signed-off-by below.





This is needed to correctly handle BOs imported into compute VM from gfx.
Both kfd and gfx should use same bo_va when map the Bos into same VM, otherwise
we may trigger kernel general protection when iterate mappings from bo_va.

Signed-off-by: Felix Kuehling 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Reviewed-By: Xiaogang Chen 
Tested-By: Xiaogang Chen 


You lost any mention of reference counting from the commit headline and 
description. I think that's still an important concept that should be 
mentioned. I'd change the headline to something like this:


drm/amdgpu: Correctly use bo_va->ref_count in compute VMs

Regards,
  Felix



---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
uint64_t va = mem->va;
struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
struct amdgpu_bo *bo[2] = {NULL, NULL};
+   struct amdgpu_bo_va *bo_va;
bool same_hive = false;
int i, ret;
  
@@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,

pr_debug("Unable to reserve BO during memory attach");
goto unwind;
}
-   attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+   if (!bo_va)
+   bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   else
+   ++bo_va->ref_count;
+   attachment[i]->bo_va = bo_va;
+
amdgpu_bo_unreserve(bo[i]);
if (unlikely(!attachment[i]->bo_va)) {
ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
continue;
if (attachment[i]->bo_va) {
amdgpu_bo_reserve(bo[i], true);
-   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+   if (--attachment[i]->bo_va->ref_count == 0)
+   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
amdgpu_bo_unreserve(bo[i]);
list_del([i]->list);
}

Re: [PATCH] Find bo_va before create it when map bo into compute VM

[Chen, Xiaogang]

On 10/12/2023 2:35 AM, Christian König wrote:
The subject line somehow got messed up. There should be an drm/amdgpu: 
or drm/amdkfd: prefix.


yes, will resend it.

Regards

Xiaogang.



Regards,
Christian.

Am 12.10.23 um 01:36 schrieb Xiaogang.Chen:

From: Xiaogang Chen 

This is needed to correctly handle BOs imported into compute VM from 
gfx.
Both kfd and gfx should use same bo_va when map the Bos into same VM, 
otherwise
we may trigger kernel general protection when iterate mappings from 
bo_va.


Signed-off-by: Felix Kuehling 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Reviewed-By: Xiaogang Chen 
Tested-By: Xiaogang Chen 
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  uint64_t va = mem->va;
  struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
  struct amdgpu_bo *bo[2] = {NULL, NULL};
+    struct amdgpu_bo_va *bo_va;
  bool same_hive = false;
  int i, ret;
  @@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  pr_debug("Unable to reserve BO during memory attach");
  goto unwind;
  }
-    attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+    if (!bo_va)
+    bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+    else
+    ++bo_va->ref_count;
+    attachment[i]->bo_va = bo_va;
+
  amdgpu_bo_unreserve(bo[i]);
  if (unlikely(!attachment[i]->bo_va)) {
  ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device 
*adev, struct kgd_mem *mem,

  continue;
  if (attachment[i]->bo_va) {
  amdgpu_bo_reserve(bo[i], true);
-    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+    if (--attachment[i]->bo_va->ref_count == 0)
+    amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
  amdgpu_bo_unreserve(bo[i]);
  list_del([i]->list);
  }

Re: [PATCH] drm/amdgpu: disable GFXOFF and PG during compute for GFX9

[Alex Deucher]

On Wed, Oct 11, 2023 at 9:49 PM Zhang, Jesse(Jie)  wrote:
>
> [AMD Official Use Only - General]
>
> -Original Message-
> From: Alex Deucher 
> Sent: Wednesday, October 11, 2023 10:20 PM
> To: Zhang, Jesse(Jie) 
> Cc: amd-gfx@lists.freedesktop.org; Deucher, Alexander 
> ; Yang, Philip ; Kuehling, 
> Felix ; Zhang, Yifan 
> Subject: Re: [PATCH] drm/amdgpu: disable GFXOFF and PG during compute for GFX9
>
> On Wed, Oct 11, 2023 at 3:52 AM Jesse Zhang  wrote:
> >
> > Temporary workaround to fix issues observed in some compute
> > applications when GFXOFF is enabled on GFX9.
> >
> > Signed-off-by: Jesse Zhang 
> > ---
> >  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 4 
> >  1 file changed, 4 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> > index d3805d6f..fef93d4edcbc 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> > @@ -683,12 +683,16 @@ int amdgpu_amdkfd_submit_ib(struct amdgpu_device
> > *adev,
> >
> >  void amdgpu_amdkfd_set_compute_idle(struct amdgpu_device *adev, bool
> > idle)  {
> > +   enum amd_powergating_state state = idle ? AMD_PG_STATE_GATE :
> > + AMD_PG_STATE_UNGATE;
> > /* Temporary workaround to fix issues observed in some
> >  * compute applications when GFXOFF is enabled on GFX11.
> >  */
> > if (IP_VERSION_MAJ(amdgpu_ip_version(adev, GC_HWIP, 0)) == 11) {
> > pr_debug("GFXOFF is %s\n", idle ? "enabled" : "disabled");
> > amdgpu_gfx_off_ctrl(adev, idle);
> > +   } else if ((IP_VERSION_MAJ(amdgpu_ip_version(adev, GC_HWIP, 0)) == 
> > 9) &&
> > +   (adev->flags & AMD_IS_APU)) {
> > +
> > + adev->ip_blocks[AMD_IP_BLOCK_TYPE_GFX].version->funcs->set_powergati
> > + ng_state((void *)adev, state);
>
> Why not use amdgpu_gfx_off_ctrl(adev, idle); for consistency?
> [Zhang, Jesse(Jie)]  Hi Alex,
>   amdgpu_gfx_off_ctrl only disable gfxoff. It also need to disable 
> gfx9's powergating to workaround.
>  So use the set_powergating_state to disable gfxoff and PG.

Thanks for clarifying.  Maybe add a comment to that effect.  With that fixed:
Acked-by: Alex Deucher 

Alex

>
> Jesse
> Alex
>
> > }
> > amdgpu_dpm_switch_power_profile(adev,
> > PP_SMC_POWER_PROFILE_COMPUTE,
> > --
> > 2.25.1
> >

RE: [PATCH 2/6] drm/amdgpu: add ras_err_info to identify RAS error source

[Zhang, Hawking]

[AMD Official Use Only - General]

+   dev_info(adev->dev, "Total %ld uncorrectable hardware errors 
detected in %s block\n",
+ras_mgr->err_data.ue_count, blk_name);

Please remove "Total" from the kernel message. With that addressed, the series 
is

Reviewed-by: Hawking Zhang 

Regards,
Hawking
-Original Message-
From: amd-gfx  On Behalf Of Yang Wang
Sent: Thursday, October 12, 2023 21:14
To: amd-gfx@lists.freedesktop.org
Cc: Zhou1, Tao ; Wang, Yang(Kevin) 
Subject: [PATCH 2/6] drm/amdgpu: add ras_err_info to identify RAS error source

introduced "ras_err_info" to better identify a RAS ERROR source.

NOTE:
For legacy chips, keep the original RAS error print format.

v1:
RAS errors may come from different dies during a RAS error query, therefore, 
need a new data structure to identify the source of RAS ERROR.

v2:
- use new data structure 'amdgpu_smuio_mcm_config_info' instead of
  ras_err_id (in v1 patch)
- refine ras error dump function name
- refine ras error dump log format

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c   | 289 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h   |  28 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_smuio.h |   5 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c   |  27 +-
 drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c|   7 +-
 drivers/gpu/drm/amd/amdgpu/nbio_v7_9.c|   7 +-
 6 files changed, 310 insertions(+), 53 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 5fb57419ef77..2522d29806dc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -152,8 +152,9 @@ static bool amdgpu_ras_get_error_query_ready(struct 
amdgpu_device *adev)

 static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t 
address)  {
-   struct ras_err_data err_data = {0, 0, 0, NULL};
+   struct ras_err_data err_data;
struct eeprom_table_record err_rec;
+   int ret;

if ((address >= adev->gmc.mc_vram_size) ||
(address >= RAS_UMC_INJECT_ADDR_LIMIT)) { @@ -170,6 +171,10 @@ 
static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t addre
return 0;
}

+   ret = amdgpu_ras_error_data_init(_data);
+   if (ret)
+   return ret;
+
memset(_rec, 0x0, sizeof(struct eeprom_table_record));
err_data.err_addr = _rec;
amdgpu_umc_fill_error_record(_data, address, address, 0, 0); @@ 
-180,6 +185,8 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device 
*adev, uint64_t addre
amdgpu_ras_save_bad_pages(adev, NULL);
}

+   amdgpu_ras_error_data_fini(_data);
+
dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL 
CORRUPT RAS EEPROM\n");
dev_warn(adev->dev, "Clear EEPROM:\n");
dev_warn(adev->dev, "echo 1 > 
/sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n");
@@ -1015,17 +1022,113 @@ static void amdgpu_ras_get_ecc_info(struct 
amdgpu_device *adev, struct ras_err_d
}
 }

+static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, 
struct ras_query_if *query_if,
+ struct ras_err_data *err_data, 
bool is_ue) {
+   struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, 
_if->head);
+   const char *blk_name = get_ras_block_str(_if->head);
+   struct amdgpu_smuio_mcm_config_info *mcm_info;
+   struct ras_err_node *err_node;
+   struct ras_err_info *err_info;
+
+   if (is_ue)
+   dev_info(adev->dev, "Total %ld uncorrectable hardware errors 
detected in %s block\n",
+ras_mgr->err_data.ue_count, blk_name);
+   else
+   dev_info(adev->dev, "Total %ld correctable hardware errors 
detected in %s block\n",
+ras_mgr->err_data.ue_count, blk_name);
+
+   for_each_ras_error(err_node, err_data) {
+   err_info = _node->err_info;
+   mcm_info = _info->mcm_info;
+   if (is_ue && err_info->ue_count) {
+   dev_info(adev->dev, "socket: %d, die: %d, "
+"%lld uncorrectable hardware errors detected 
in %s block\n",
+mcm_info->socket_id,
+mcm_info->die_id,
+err_info->ue_count,
+blk_name);
+   } else if (!is_ue && err_info->ce_count) {
+   dev_info(adev->dev, "socket: %d, die: %d, "
+"%lld correctable hardware errors detected in 
%s block\n",
+mcm_info->socket_id,
+mcm_info->die_id,
+err_info->ue_count,
+blk_name);
+   }
+   }
+}
+
+static void 

[PATCH 5/6] drm/amdgpu: add RAS error info support for gfx_v9_4_3

[Yang Wang]

add RAS error info support for gfx_v9_4_3.

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 10 --
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c 
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
index fbfe0a1c4b19..db179d085efa 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
@@ -3766,6 +3766,12 @@ static void gfx_v9_4_3_inst_query_ras_err_count(struct 
amdgpu_device *adev,
unsigned long ce_count = 0, ue_count = 0;
uint32_t i, j, k;
 
+   /* NOTE: convert xcc_id to physical XCD ID (XCD0 or XCD1) */
+   struct amdgpu_smuio_mcm_config_info mcm_info = {
+   .socket_id = adev->smuio.funcs->get_socket_id(adev),
+   .die_id = xcc_id & 0x01 ? 1 : 0,
+   };
+
mutex_lock(>grbm_idx_mutex);
 
for (i = 0; i < ARRAY_SIZE(gfx_v9_4_3_ce_reg_list); i++) {
@@ -3804,8 +3810,8 @@ static void gfx_v9_4_3_inst_query_ras_err_count(struct 
amdgpu_device *adev,
/* the caller should make sure initialize value of
 * err_data->ue_count and err_data->ce_count
 */
-   err_data->ce_count += ce_count;
-   err_data->ue_count += ue_count;
+   amdgpu_ras_error_statistic_ue_count(err_data, _info, ue_count);
+   amdgpu_ras_error_statistic_ce_count(err_data, _info, ce_count);
 }
 
 static void gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device *adev,
-- 
2.34.1

[PATCH 6/6] drm/amdgpu: add RAS error info support for mmhub_v1_8

[Yang Wang]

add RAS error info support for mmhub_v1_8.

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c | 15 +--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c 
b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c
index 2c0419faf8d4..aa00483e7b37 100644
--- a/drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c
+++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v1_8.c
@@ -626,6 +626,14 @@ static void mmhub_v1_8_inst_query_ras_error_count(struct 
amdgpu_device *adev,
  void *ras_err_status)
 {
struct ras_err_data *err_data = (struct ras_err_data *)ras_err_status;
+   unsigned long ue_count = 0, ce_count = 0;
+
+   /* NOTE: mmhub is converted by aid_mask and the range is 0-3,
+* which can be used as die ID directly */
+   struct amdgpu_smuio_mcm_config_info mcm_info = {
+   .socket_id = adev->smuio.funcs->get_socket_id(adev),
+   .die_id = mmhub_inst,
+   };
 
amdgpu_ras_inst_query_ras_error_count(adev,
mmhub_v1_8_ce_reg_list,
@@ -634,7 +642,7 @@ static void mmhub_v1_8_inst_query_ras_error_count(struct 
amdgpu_device *adev,
ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
mmhub_inst,
AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE,
-   _data->ce_count);
+   _count);
amdgpu_ras_inst_query_ras_error_count(adev,
mmhub_v1_8_ue_reg_list,
ARRAY_SIZE(mmhub_v1_8_ue_reg_list),
@@ -642,7 +650,10 @@ static void mmhub_v1_8_inst_query_ras_error_count(struct 
amdgpu_device *adev,
ARRAY_SIZE(mmhub_v1_8_ras_memory_list),
mmhub_inst,
AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
-   _data->ue_count);
+   _count);
+
+   amdgpu_ras_error_statistic_ce_count(err_data, _info, ce_count);
+   amdgpu_ras_error_statistic_ue_count(err_data, _info, ue_count);
 }
 
 static void mmhub_v1_8_query_ras_error_count(struct amdgpu_device *adev,
-- 
2.34.1

[PATCH 2/6] drm/amdgpu: add ras_err_info to identify RAS error source

[Yang Wang]

introduced "ras_err_info" to better identify a RAS ERROR source.

NOTE:
For legacy chips, keep the original RAS error print format.

v1:
RAS errors may come from different dies during a RAS error query,
therefore, need a new data structure to identify the source of RAS ERROR.

v2:
- use new data structure 'amdgpu_smuio_mcm_config_info' instead of
  ras_err_id (in v1 patch)
- refine ras error dump function name
- refine ras error dump log format

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c   | 289 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h   |  28 +++
 drivers/gpu/drm/amd/amdgpu/amdgpu_smuio.h |   5 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c   |  27 +-
 drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c|   7 +-
 drivers/gpu/drm/amd/amdgpu/nbio_v7_9.c|   7 +-
 6 files changed, 310 insertions(+), 53 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 5fb57419ef77..2522d29806dc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -152,8 +152,9 @@ static bool amdgpu_ras_get_error_query_ready(struct 
amdgpu_device *adev)
 
 static int amdgpu_reserve_page_direct(struct amdgpu_device *adev, uint64_t 
address)
 {
-   struct ras_err_data err_data = {0, 0, 0, NULL};
+   struct ras_err_data err_data;
struct eeprom_table_record err_rec;
+   int ret;
 
if ((address >= adev->gmc.mc_vram_size) ||
(address >= RAS_UMC_INJECT_ADDR_LIMIT)) {
@@ -170,6 +171,10 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device 
*adev, uint64_t addre
return 0;
}
 
+   ret = amdgpu_ras_error_data_init(_data);
+   if (ret)
+   return ret;
+
memset(_rec, 0x0, sizeof(struct eeprom_table_record));
err_data.err_addr = _rec;
amdgpu_umc_fill_error_record(_data, address, address, 0, 0);
@@ -180,6 +185,8 @@ static int amdgpu_reserve_page_direct(struct amdgpu_device 
*adev, uint64_t addre
amdgpu_ras_save_bad_pages(adev, NULL);
}
 
+   amdgpu_ras_error_data_fini(_data);
+
dev_warn(adev->dev, "WARNING: THIS IS ONLY FOR TEST PURPOSES AND WILL 
CORRUPT RAS EEPROM\n");
dev_warn(adev->dev, "Clear EEPROM:\n");
dev_warn(adev->dev, "echo 1 > 
/sys/kernel/debug/dri/0/ras/ras_eeprom_reset\n");
@@ -1015,17 +1022,113 @@ static void amdgpu_ras_get_ecc_info(struct 
amdgpu_device *adev, struct ras_err_d
}
 }
 
+static void amdgpu_ras_error_print_error_data(struct amdgpu_device *adev, 
struct ras_query_if *query_if,
+ struct ras_err_data *err_data, 
bool is_ue)
+{
+   struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, 
_if->head);
+   const char *blk_name = get_ras_block_str(_if->head);
+   struct amdgpu_smuio_mcm_config_info *mcm_info;
+   struct ras_err_node *err_node;
+   struct ras_err_info *err_info;
+
+   if (is_ue)
+   dev_info(adev->dev, "Total %ld uncorrectable hardware errors 
detected in %s block\n",
+ras_mgr->err_data.ue_count, blk_name);
+   else
+   dev_info(adev->dev, "Total %ld correctable hardware errors 
detected in %s block\n",
+ras_mgr->err_data.ue_count, blk_name);
+
+   for_each_ras_error(err_node, err_data) {
+   err_info = _node->err_info;
+   mcm_info = _info->mcm_info;
+   if (is_ue && err_info->ue_count) {
+   dev_info(adev->dev, "socket: %d, die: %d, "
+"%lld uncorrectable hardware errors detected 
in %s block\n",
+mcm_info->socket_id,
+mcm_info->die_id,
+err_info->ue_count,
+blk_name);
+   } else if (!is_ue && err_info->ce_count) {
+   dev_info(adev->dev, "socket: %d, die: %d, "
+"%lld correctable hardware errors detected in 
%s block\n",
+mcm_info->socket_id,
+mcm_info->die_id,
+err_info->ue_count,
+blk_name);
+   }
+   }
+}
+
+static void amdgpu_ras_error_generate_report(struct amdgpu_device *adev, 
struct ras_query_if *query_if,
+struct ras_err_data *err_data)
+{
+   struct ras_manager *ras_mgr = amdgpu_ras_find_obj(adev, 
_if->head);
+   const char *blk_name = get_ras_block_str(_if->head);
+
+   if (err_data->ce_count) {
+   if (!list_empty(_data->err_node_list)) {
+   amdgpu_ras_error_print_error_data(adev, query_if, 
err_data, false);
+   } else if (!adev->aid_mask &&
+  adev->smuio.funcs &&
+ 

[PATCH 3/6] drm/amdgpu: add RAS error info support for umc_v12_0

[Yang Wang]

add RAS error info support for umc_v12_0.

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/umc_v12_0.c | 16 ++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c 
b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
index 7714c2ef2cdc..cc66d44211fd 100644
--- a/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/umc_v12_0.c
@@ -137,15 +137,27 @@ static int umc_v12_0_query_error_count(struct 
amdgpu_device *adev,
uint32_t ch_inst, void *data)
 {
struct ras_err_data *err_data = (struct ras_err_data *)data;
+   unsigned long ue_count = 0, ce_count = 0;
+
+   /* NOTE: node_inst is converted by adev->umc.active_mask and the range 
is [0-3],
+* which can be used as die ID directly */
+   struct amdgpu_smuio_mcm_config_info mcm_info = {
+   .socket_id = adev->smuio.funcs->get_socket_id(adev),
+   .die_id = node_inst,
+   };
+
uint64_t umc_reg_offset =
get_umc_v12_0_reg_offset(adev, node_inst, umc_inst, ch_inst);
 
umc_v12_0_query_correctable_error_count(adev,
umc_reg_offset,
-   &(err_data->ce_count));
+   _count);
umc_v12_0_query_uncorrectable_error_count(adev,
umc_reg_offset,
-   &(err_data->ue_count));
+   _count);
+
+   amdgpu_ras_error_statistic_ue_count(err_data, _info, ue_count);
+   amdgpu_ras_error_statistic_ce_count(err_data, _info, ce_count);
 
return 0;
 }
-- 
2.34.1

[PATCH 4/6] drm/amdgpu: add RAS error info support for sdma_v4_4_2.

[Yang Wang]

add RAS error info support for sdma_v4_4_2.

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c | 9 -
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c 
b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
index 3b6990ef403a..31aa245552d6 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c
@@ -2130,6 +2130,11 @@ static void 
sdma_v4_4_2_inst_query_ras_error_count(struct amdgpu_device *adev,
 {
struct ras_err_data *err_data = (struct ras_err_data *)ras_err_status;
uint32_t sdma_dev_inst = GET_INST(SDMA0, sdma_inst);
+   unsigned long ue_count = 0;
+   struct amdgpu_smuio_mcm_config_info mcm_info = {
+   .socket_id = adev->smuio.funcs->get_socket_id(adev),
+   .die_id = adev->sdma.instance[sdma_inst].aid_id,
+   };
 
/* sdma v4_4_2 doesn't support query ce counts */
amdgpu_ras_inst_query_ras_error_count(adev,
@@ -2139,7 +2144,9 @@ static void sdma_v4_4_2_inst_query_ras_error_count(struct 
amdgpu_device *adev,
ARRAY_SIZE(sdma_v4_4_2_ras_memory_list),
sdma_dev_inst,
AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
-   _data->ue_count);
+   _count);
+
+   amdgpu_ras_error_statistic_ue_count(err_data, _info, ue_count);
 }
 
 static void sdma_v4_4_2_query_ras_error_count(struct amdgpu_device *adev,
-- 
2.34.1

[PATCH 1/6] drm/amdgpu: make err_data structure built-in for ras_manager

[Yang Wang]

(No effect outside the ras_mgr data structure)

Since a new member was added to the ras_err_data data structure,
it becomes unreasonable for the ras_mgr instance to contain this data,
because ras mgr only uses the 2 member information of ue_count/ce_count in 
err_data.

This patch changes the code err_data into built-in structure members,
making the code directly compatible.

Signed-off-by: Yang Wang 
Reviewed-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 5 -
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 728f98c6fc1c..46b0dcf846dc 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -496,7 +496,10 @@ struct ras_manager {
/* IH data */
struct ras_ih_data ih_data;
 
-   struct ras_err_data err_data;
+   struct {
+   unsigned long ue_count;
+   unsigned long ce_count;
+   } err_data;
 };
 
 struct ras_badpage {
-- 
2.34.1

RE: [PATCH 4/5] drm/amdgpu: bypass RAS error reset in some conditions

[Zhang, Hawking]

[AMD Official Use Only - General]

-   if (!amdgpu_ras_is_supported(adev, block))
+   /* skip ras error reset in gpu reset */
+   if (amdgpu_in_reset(adev) &&
+   mca_funcs && mca_funcs->mca_set_debug_mode)
+   return 0;

We should check RAS in_recovery flag in such case. Reset domain is locked in 
relative late phase, at least *after* error counter harvest. Please double 
check.

Regards,
Hawking
-Original Message-
From: Zhou1, Tao 
Sent: Thursday, October 12, 2023 17:01
To: amd-gfx@lists.freedesktop.org; Yang, Stanley ; Zhang, 
Hawking ; Li, Candice ; Chai, Thomas 
; Wang, Yang(Kevin) 
Cc: Zhou1, Tao 
Subject: [PATCH 4/5] drm/amdgpu: bypass RAS error reset in some conditions

PMFW is responsible for RAS error reset in some conditions, driver can skip the 
operation.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 18 --
 1 file changed, 16 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 91ed4fd96ee1..6dddb0423411 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -1105,11 +1105,18 @@ int amdgpu_ras_reset_error_count(struct amdgpu_device 
*adev,
enum amdgpu_ras_block block)
 {
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
+   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;

if (!block_obj || !block_obj->hw_ops)
return 0;

-   if (!amdgpu_ras_is_supported(adev, block))
+   /* skip ras error reset in gpu reset */
+   if (amdgpu_in_reset(adev) &&
+   mca_funcs && mca_funcs->mca_set_debug_mode)
+   return 0;
+
+   if (!amdgpu_ras_is_supported(adev, block) ||
+   !amdgpu_ras_get_mca_debug_mode(adev))
return 0;

if (block_obj->hw_ops->reset_ras_error_count)
@@ -1122,6 +1129,7 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device 
*adev,
enum amdgpu_ras_block block)
 {
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
+   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;

if (!block_obj || !block_obj->hw_ops) {
dev_dbg_once(adev->dev, "%s doesn't config RAS function\n", @@ 
-1129,7 +1137,13 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device 
*adev,
return 0;
}

-   if (!amdgpu_ras_is_supported(adev, block))
+   /* skip ras error reset in gpu reset */
+   if (amdgpu_in_reset(adev) &&
+   mca_funcs && mca_funcs->mca_set_debug_mode)
+   return 0;
+
+   if (!amdgpu_ras_is_supported(adev, block) ||
+   !amdgpu_ras_get_mca_debug_mode(adev))
return 0;

if (block_obj->hw_ops->reset_ras_error_count)
--
2.35.1

RE: [PATCH v2] drm/amdgpu: Expose ras version & schema info

[Wang, Yang(Kevin)]

Tip:

ffs(AMDGPU_RAS_ERROR__POISON) - 1) == __ffs(AMDGPU_RAS_ERROR__POISON) 
if you want to get a value which start from 0, you can use __ffs() directly.

Best Regards,
Kevin

-Original Message-
From: amd-gfx  On Behalf Of Lazar, Lijo
Sent: Wednesday, October 11, 2023 5:58 PM
To: Kamal, Asad ; amd-gfx@lists.freedesktop.org
Cc: Ma, Le ; Zhang, Morris ; Zhang, Hawking 

Subject: Re: [PATCH v2] drm/amdgpu: Expose ras version & schema info



On 10/11/2023 2:55 PM, Asad Kamal wrote:
> Expose ras table version & schema info to sysfs
> 
> v2: Updated schema to get poison support info from ras context, 
> removed asic specific checks
> 
> Signed-off-by: Asad Kamal 

One nit inline. With/without that change,

Reviewed-by: Lijo Lazar 

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 51 +++--
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h |  3 ++
>   2 files changed, 51 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
> index 9c8203e87859..cb9e48fb40d2 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
> @@ -1370,6 +1370,22 @@ static ssize_t amdgpu_ras_sysfs_features_read(struct 
> device *dev,
>   return sysfs_emit(buf, "feature mask: 0x%x\n", con->features);
>   }
>   
> +static ssize_t amdgpu_ras_sysfs_version_show(struct device *dev,
> + struct device_attribute *attr, char *buf) {
> + struct amdgpu_ras *con =
> + container_of(attr, struct amdgpu_ras, version_attr);
> + return sysfs_emit(buf, "table version: 0x%x\n", 
> +con->eeprom_control.tbl_hdr.version);
> +}
> +
> +static ssize_t amdgpu_ras_sysfs_schema_show(struct device *dev,
> + struct device_attribute *attr, char *buf) {
> + struct amdgpu_ras *con =
> + container_of(attr, struct amdgpu_ras, schema_attr);
> + return sysfs_emit(buf, "schema: 0x%x\n", con->schema); }
> +
>   static void amdgpu_ras_sysfs_remove_bad_page_node(struct amdgpu_device 
> *adev)
>   {
>   struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -1379,11 
> +1395,13 @@ static void amdgpu_ras_sysfs_remove_bad_page_node(struct 
> amdgpu_device *adev)
>   RAS_FS_NAME);
>   }
>   
> -static int amdgpu_ras_sysfs_remove_feature_node(struct amdgpu_device 
> *adev)
> +static int amdgpu_ras_sysfs_remove_dev_attr_node(struct amdgpu_device 
> +*adev)
>   {
>   struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
>   struct attribute *attrs[] = {
>   >features_attr.attr,
> + >version_attr.attr,

> + >schema_attr.attr,
>   NULL
>   };
>   struct attribute_group group = {
> @@ -1459,7 +1477,7 @@ static int amdgpu_ras_sysfs_remove_all(struct 
> amdgpu_device *adev)
>   if (amdgpu_bad_page_threshold != 0)
>   amdgpu_ras_sysfs_remove_bad_page_node(adev);
>   
> - amdgpu_ras_sysfs_remove_feature_node(adev);
> + amdgpu_ras_sysfs_remove_dev_attr_node(adev);
>   
>   return 0;
>   }
> @@ -1582,6 +1600,10 @@ static BIN_ATTR(gpu_vram_bad_pages, S_IRUGO,
>   amdgpu_ras_sysfs_badpages_read, NULL, 0);
>   static DEVICE_ATTR(features, S_IRUGO,
>   amdgpu_ras_sysfs_features_read, NULL);
> +static DEVICE_ATTR(version, 0444,
> + amdgpu_ras_sysfs_version_show, NULL); static 
> DEVICE_ATTR(schema, 
> +0444,
> + amdgpu_ras_sysfs_schema_show, NULL);
>   static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
>   {
>   struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -1590,6 
> +1612,8 @@ static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
>   };
>   struct attribute *attrs[] = {
>   >features_attr.attr,
> + >version_attr.attr,
> + >schema_attr.attr,
>   NULL
>   };
>   struct bin_attribute *bin_attrs[] = { @@ -1598,11 +1622,20 @@ 
> static int amdgpu_ras_fs_init(struct amdgpu_device *adev)
>   };
>   int r;
>   
> + group.attrs = attrs;
> +
>   /* add features entry */
>   con->features_attr = dev_attr_features;
> - group.attrs = attrs;
>   sysfs_attr_init(attrs[0]);
>   
> + /* add version entry */
> + con->version_attr = dev_attr_version;
> + sysfs_attr_init(attrs[1]);
> +
> + /* add schema entry */
> + con->schema_attr = dev_attr_schema;
> + sysfs_attr_init(attrs[2]);
> +
>   if (amdgpu_bad_page_threshold != 0) {
>   /* add bad_page_features entry */
>   bin_attr_gpu_vram_bad_pages.private = NULL; @@ -2594,6 +2627,14 
> @@ 
> static void amdgpu_ras_query_poison_mode(struct amdgpu_device *adev)
>   }
>   }
>   
> +static int amdgpu_get_ras_schema(struct amdgpu_device *adev) {
> + return  (amdgpu_ras_is_poison_mode_supported(adev) << 
> +(ffs(AMDGPU_RAS_ERROR__POISON) - 1)) |

It's simpler and more readable with a ternary operator.

Thanks,

RE: [PATCH 2/2] drm/amdgpu/umsch: fix missing stuff during rebase

[Gopalakrishnan, Veerabadhran (Veera)]

[AMD Official Use Only - General]

The series is Reviewed-by: Veerabadhran Gopalakrishnan 


-Veera

-Original Message-
From: Yu, Lang 
Sent: Thursday, October 12, 2023 1:05 PM
To: amd-gfx@lists.freedesktop.org
Cc: Deucher, Alexander ; Gopalakrishnan, 
Veerabadhran (Veera) ; Yu, Lang 

Subject: [PATCH 2/2] drm/amdgpu/umsch: fix missing stuff during rebase

These are missed during rebase.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
index 771ef8017a98..3713d0749813 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
@@ -674,6 +674,8 @@ const char *amdgpu_ucode_name(enum AMDGPU_UCODE_ID ucode_id)
return "UMSCH_MM_UCODE";
case AMDGPU_UCODE_ID_UMSCH_MM_DATA:
return "UMSCH_MM_DATA";
+   case AMDGPU_UCODE_ID_UMSCH_MM_CMD_BUFFER:
+   return "UMSCH_MM_CMD_BUFFER";
default:
return "UNKNOWN UCODE";
}
@@ -806,6 +808,7 @@ static int amdgpu_ucode_init_single_fw(struct amdgpu_device 
*adev,
sdma_hdr = (const struct sdma_firmware_header_v2_0 *)ucode->fw->data;
imu_hdr = (const struct imu_firmware_header_v1_0 *)ucode->fw->data;
vpe_hdr = (const struct vpe_firmware_header_v1_0 *)ucode->fw->data;
+   umsch_mm_hdr = (const struct umsch_mm_firmware_header_v1_0 
*)ucode->fw->data;

if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
switch (ucode->ucode_id) {
--
2.25.1

[PATCH 5/5] drm/amdgpu: reuse amdgpu_ras_reset_error_count code

[Tao Zhou]

To simplify the code of amdgpu_ras_reset_error_status without logical
change.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 29 +++--
 1 file changed, 8 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 6dddb0423411..3698be22 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -1107,17 +1107,20 @@ int amdgpu_ras_reset_error_count(struct amdgpu_device 
*adev,
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
 
-   if (!block_obj || !block_obj->hw_ops)
-   return 0;
+   if (!block_obj || !block_obj->hw_ops) {
+   dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",
+ras_block_str(block));
+   return -EOPNOTSUPP;
+   }
 
/* skip ras error reset in gpu reset */
if (amdgpu_in_reset(adev) &&
mca_funcs && mca_funcs->mca_set_debug_mode)
-   return 0;
+   return -EOPNOTSUPP;
 
if (!amdgpu_ras_is_supported(adev, block) ||
!amdgpu_ras_get_mca_debug_mode(adev))
-   return 0;
+   return -EOPNOTSUPP;
 
if (block_obj->hw_ops->reset_ras_error_count)
block_obj->hw_ops->reset_ras_error_count(adev);
@@ -1129,25 +1132,9 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device 
*adev,
enum amdgpu_ras_block block)
 {
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
-   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
 
-   if (!block_obj || !block_obj->hw_ops) {
-   dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",
-ras_block_str(block));
+   if (amdgpu_ras_reset_error_count(adev, block) == -EOPNOTSUPP)
return 0;
-   }
-
-   /* skip ras error reset in gpu reset */
-   if (amdgpu_in_reset(adev) &&
-   mca_funcs && mca_funcs->mca_set_debug_mode)
-   return 0;
-
-   if (!amdgpu_ras_is_supported(adev, block) ||
-   !amdgpu_ras_get_mca_debug_mode(adev))
-   return 0;
-
-   if (block_obj->hw_ops->reset_ras_error_count)
-   block_obj->hw_ops->reset_ras_error_count(adev);
 
if ((block == AMDGPU_RAS_BLOCK__GFX) ||
(block == AMDGPU_RAS_BLOCK__MMHUB)) {
-- 
2.35.1

[PATCH 4/5] drm/amdgpu: bypass RAS error reset in some conditions

[Tao Zhou]

PMFW is responsible for RAS error reset in some conditions, driver can
skip the operation.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 18 --
 1 file changed, 16 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 91ed4fd96ee1..6dddb0423411 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -1105,11 +1105,18 @@ int amdgpu_ras_reset_error_count(struct amdgpu_device 
*adev,
enum amdgpu_ras_block block)
 {
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
+   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
 
if (!block_obj || !block_obj->hw_ops)
return 0;
 
-   if (!amdgpu_ras_is_supported(adev, block))
+   /* skip ras error reset in gpu reset */
+   if (amdgpu_in_reset(adev) &&
+   mca_funcs && mca_funcs->mca_set_debug_mode)
+   return 0;
+
+   if (!amdgpu_ras_is_supported(adev, block) ||
+   !amdgpu_ras_get_mca_debug_mode(adev))
return 0;
 
if (block_obj->hw_ops->reset_ras_error_count)
@@ -1122,6 +1129,7 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device 
*adev,
enum amdgpu_ras_block block)
 {
struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
+   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
 
if (!block_obj || !block_obj->hw_ops) {
dev_dbg_once(adev->dev, "%s doesn't config RAS function\n",
@@ -1129,7 +1137,13 @@ int amdgpu_ras_reset_error_status(struct amdgpu_device 
*adev,
return 0;
}
 
-   if (!amdgpu_ras_is_supported(adev, block))
+   /* skip ras error reset in gpu reset */
+   if (amdgpu_in_reset(adev) &&
+   mca_funcs && mca_funcs->mca_set_debug_mode)
+   return 0;
+
+   if (!amdgpu_ras_is_supported(adev, block) ||
+   !amdgpu_ras_get_mca_debug_mode(adev))
return 0;
 
if (block_obj->hw_ops->reset_ras_error_count)
-- 
2.35.1

[PATCH 3/5] drm/amd/pm: record mca debug mode in RAS

[Tao Zhou]

Call amdgpu_ras_set_mca_debug_mode when we set mca debug mode in smu
v13_0_6.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c 
b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
index 8220bdcbd927..55b0846337a7 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
@@ -1414,6 +1414,7 @@ static int smu_v13_0_6_mca_set_debug_mode(struct 
smu_context *smu, bool enable)
if (smu_version < 0x554800)
return 0;
 
+   amdgpu_ras_set_mca_debug_mode(smu->adev, enable);
return smu_cmn_send_smc_msg_with_param(smu, SMU_MSG_ClearMcaOnRead,
   enable ? 0 : 
ClearMcaOnRead_UE_FLAG_MASK | ClearMcaOnRead_CE_POLL_MASK,
   NULL);
-- 
2.35.1

[PATCH 2/5] drm/amdgpu: add set/get mca debug mode operations

[Tao Zhou]

Record the debug mode status in RAS.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 21 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h |  5 +
 2 files changed, 26 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index 82ee6f3d12da..91ed4fd96ee1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -3250,6 +3250,27 @@ int amdgpu_ras_reset_gpu(struct amdgpu_device *adev)
return 0;
 }
 
+void amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable)
+{
+   struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+
+   if (con)
+   con->is_mca_debug_mode = enable;
+}
+
+bool amdgpu_ras_get_mca_debug_mode(struct amdgpu_device *adev)
+{
+   struct amdgpu_ras *con = amdgpu_ras_get_context(adev);
+   const struct amdgpu_mca_smu_funcs *mca_funcs = adev->mca.mca_funcs;
+
+   if (!con)
+   return false;
+
+   if (mca_funcs && mca_funcs->mca_set_debug_mode)
+   return con->is_mca_debug_mode;
+   else
+   return true;
+}
 
 /* Register each ip ras block into amdgpu ras */
 int amdgpu_ras_register_ras_block(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index deea10b6c184..c60688dc73ea 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -433,6 +433,8 @@ struct amdgpu_ras {
 
/* Indicates smu whether need update bad channel info */
bool update_channel_flag;
+   /* Record status of smu mca debug mode */
+   bool is_mca_debug_mode;
 
/* Record special requirements of gpu reset caller */
uint32_t  gpu_reset_flags;
@@ -748,6 +750,9 @@ struct amdgpu_ras* amdgpu_ras_get_context(struct 
amdgpu_device *adev);
 
 int amdgpu_ras_set_context(struct amdgpu_device *adev, struct amdgpu_ras 
*ras_con);
 
+void amdgpu_ras_set_mca_debug_mode(struct amdgpu_device *adev, bool enable);
+bool amdgpu_ras_get_mca_debug_mode(struct amdgpu_device *adev);
+
 int amdgpu_ras_register_ras_block(struct amdgpu_device *adev,
struct amdgpu_ras_block_object *ras_block_obj);
 void amdgpu_ras_interrupt_fatal_error_handler(struct amdgpu_device *adev);
-- 
2.35.1

[PATCH 1/5] drm/amdgpu: define ras_reset_error_count function

[Tao Zhou]

Make the code architecture more simple.

Signed-off-by: Tao Zhou 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  8 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c| 17 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h|  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c   |  4 ++--
 4 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 31f8c3ead161..04cfd67a37a0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -3594,9 +3594,7 @@ static void amdgpu_device_xgmi_reset_func(struct 
work_struct *__work)
if (adev->asic_reset_res)
goto fail;
 
-   if (adev->mmhub.ras && adev->mmhub.ras->ras_block.hw_ops &&
-   adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count)
-   
adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count(adev);
+   amdgpu_ras_reset_error_count(adev, AMDGPU_RAS_BLOCK__MMHUB);
} else {
 
task_barrier_full(>tb);
@@ -5242,9 +5240,7 @@ int amdgpu_do_asic_reset(struct list_head 
*device_list_handle,
 
if (!r && amdgpu_ras_intr_triggered()) {
list_for_each_entry(tmp_adev, device_list_handle, reset_list) {
-   if (tmp_adev->mmhub.ras && 
tmp_adev->mmhub.ras->ras_block.hw_ops &&
-   
tmp_adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count)
-   
tmp_adev->mmhub.ras->ras_block.hw_ops->reset_ras_error_count(tmp_adev);
+   amdgpu_ras_reset_error_count(tmp_adev, 
AMDGPU_RAS_BLOCK__MMHUB);
}
 
amdgpu_ras_intr_cleared();
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
index dacce5f2bfa3..82ee6f3d12da 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
@@ -1101,6 +1101,23 @@ int amdgpu_ras_query_error_status(struct amdgpu_device 
*adev,
return 0;
 }
 
+int amdgpu_ras_reset_error_count(struct amdgpu_device *adev,
+   enum amdgpu_ras_block block)
+{
+   struct amdgpu_ras_block_object *block_obj = 
amdgpu_ras_get_ras_block(adev, block, 0);
+
+   if (!block_obj || !block_obj->hw_ops)
+   return 0;
+
+   if (!amdgpu_ras_is_supported(adev, block))
+   return 0;
+
+   if (block_obj->hw_ops->reset_ras_error_count)
+   block_obj->hw_ops->reset_ras_error_count(adev);
+
+   return 0;
+}
+
 int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
enum amdgpu_ras_block block)
 {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
index 728f98c6fc1c..deea10b6c184 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
@@ -694,6 +694,8 @@ void amdgpu_ras_debugfs_create_all(struct amdgpu_device 
*adev);
 int amdgpu_ras_query_error_status(struct amdgpu_device *adev,
struct ras_query_if *info);
 
+int amdgpu_ras_reset_error_count(struct amdgpu_device *adev,
+   enum amdgpu_ras_block block);
 int amdgpu_ras_reset_error_status(struct amdgpu_device *adev,
enum amdgpu_ras_block block);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
index 70e38b013309..2b7dc490ba6b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c
@@ -914,7 +914,7 @@ static int amdgpu_xgmi_ras_late_init(struct amdgpu_device 
*adev, struct ras_comm
adev->gmc.xgmi.num_physical_nodes == 0)
return 0;
 
-   adev->gmc.xgmi.ras->ras_block.hw_ops->reset_ras_error_count(adev);
+   amdgpu_ras_reset_error_count(adev, AMDGPU_RAS_BLOCK__XGMI_WAFL);
 
return amdgpu_ras_block_late_init(adev, ras_block);
 }
@@ -1081,7 +1081,7 @@ static void amdgpu_xgmi_query_ras_error_count(struct 
amdgpu_device *adev,
break;
}
 
-   adev->gmc.xgmi.ras->ras_block.hw_ops->reset_ras_error_count(adev);
+   amdgpu_ras_reset_error_count(adev, AMDGPU_RAS_BLOCK__XGMI_WAFL);
 
err_data->ue_count += ue_cnt;
err_data->ce_count += ce_cnt;
-- 
2.35.1

Re: [PATCH] drm/amdgpu: flush the correct vmid tlb for specific pasid

[Christian König]

Am 12.10.23 um 09:41 schrieb Yifan Zhang:

flush the correct vmid tlb for specific pasid on gmc 11.

Signed-off-by: Yifan Zhang 


Good catch, Reviewed-by: Christian König 


---
  drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
index f4bb3886d1bc..19eaada35ede 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
@@ -314,7 +314,7 @@ static void gmc_v11_0_flush_gpu_tlb_pasid(struct 
amdgpu_device *adev,
  
  		valid = gmc_v11_0_get_vmid_pasid_mapping_info(adev, vmid,

  );
-   if (!valid || queried == pasid)
+   if (!valid || queried != pasid)
continue;
  
  		if (all_hub) {

RE: [PATCH 2/2] drm/amdgpu: don't use legacy invalidation on MMHUB v3.3

[Zhang, Yifan]

[AMD Official Use Only - General]

This series is:

Reviewed-by: Yifan Zhang 

-Original Message-
From: Yu, Lang 
Sent: Thursday, October 12, 2023 3:32 PM
To: amd-gfx@lists.freedesktop.org
Cc: Deucher, Alexander ; Zhang, Yifan 
; Yu, Lang 
Subject: [PATCH 2/2] drm/amdgpu: don't use legacy invalidation on MMHUB v3.3

Legacy invalidation is not supported.
This is missed during rebase.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c 
b/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
index 76b12f015d1d..dc4812ecc98d 100644
--- a/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
@@ -68,7 +68,7 @@ static uint32_t mmhub_v3_3_get_invalidate_req(unsigned int 
vmid,
/* invalidate using legacy mode on vmid*/
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ,
PER_VMID_INVALIDATE_REQ, 1 << vmid);
-   req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 
flush_type);
+   req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 
flush_type ? : 1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 
1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 
1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 
1);
--
2.25.1

Re: [PATCH] drm/amdgpu: Avoid FRU EEPROM access on APU

[Wang, Yang(Kevin)]

[AMD Official Use Only - General]

Reviewed-by: Yang Wang 

Best Regards
Kevin

发件人: Lazar, Lijo 
发送时间: 星期四, 十月 12, 2023 15:36
收件人: amd-gfx@lists.freedesktop.org 
抄送: Zhang, Hawking ; Deucher, Alexander 
; Wang, Yang(Kevin) 
主题: [PATCH] drm/amdgpu: Avoid FRU EEPROM access on APU

FRU EEPROM access is not valid for APU devices.

Signed-off-by: Lijo Lazar 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
index d635e61805ea..a08c148b13f9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
@@ -42,8 +42,9 @@ static bool is_fru_eeprom_supported(struct amdgpu_device 
*adev, u32 *fru_addr)

 /* The i2c access is blocked on VF
  * TODO: Need other way to get the info
+* Also, FRU not valid for APU devices.
  */
-   if (amdgpu_sriov_vf(adev))
+   if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
 return false;

 /* The default I2C EEPROM address of the FRU.
--
2.25.1

[PATCH] drm/amdgpu: flush the correct vmid tlb for specific pasid

[Yifan Zhang]

flush the correct vmid tlb for specific pasid on gmc 11.

Signed-off-by: Yifan Zhang 
---
 drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c 
b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
index f4bb3886d1bc..19eaada35ede 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c
@@ -314,7 +314,7 @@ static void gmc_v11_0_flush_gpu_tlb_pasid(struct 
amdgpu_device *adev,
 
valid = gmc_v11_0_get_vmid_pasid_mapping_info(adev, vmid,
  );
-   if (!valid || queried == pasid)
+   if (!valid || queried != pasid)
continue;
 
if (all_hub) {
-- 
2.37.3

[PATCH] drm/amdgpu: Avoid FRU EEPROM access on APU

[Lijo Lazar]

FRU EEPROM access is not valid for APU devices.

Signed-off-by: Lijo Lazar 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
index d635e61805ea..a08c148b13f9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fru_eeprom.c
@@ -42,8 +42,9 @@ static bool is_fru_eeprom_supported(struct amdgpu_device 
*adev, u32 *fru_addr)
 
/* The i2c access is blocked on VF
 * TODO: Need other way to get the info
+* Also, FRU not valid for APU devices.
 */
-   if (amdgpu_sriov_vf(adev))
+   if (amdgpu_sriov_vf(adev) || (adev->flags & AMD_IS_APU))
return false;
 
/* The default I2C EEPROM address of the FRU.
-- 
2.25.1

[PATCH 2/2] drm/amdgpu/umsch: fix missing stuff during rebase

[Lang Yu]

These are missed during rebase.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
index 771ef8017a98..3713d0749813 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c
@@ -674,6 +674,8 @@ const char *amdgpu_ucode_name(enum AMDGPU_UCODE_ID ucode_id)
return "UMSCH_MM_UCODE";
case AMDGPU_UCODE_ID_UMSCH_MM_DATA:
return "UMSCH_MM_DATA";
+   case AMDGPU_UCODE_ID_UMSCH_MM_CMD_BUFFER:
+   return "UMSCH_MM_CMD_BUFFER";
default:
return "UNKNOWN UCODE";
}
@@ -806,6 +808,7 @@ static int amdgpu_ucode_init_single_fw(struct amdgpu_device 
*adev,
sdma_hdr = (const struct sdma_firmware_header_v2_0 *)ucode->fw->data;
imu_hdr = (const struct imu_firmware_header_v1_0 *)ucode->fw->data;
vpe_hdr = (const struct vpe_firmware_header_v1_0 *)ucode->fw->data;
+   umsch_mm_hdr = (const struct umsch_mm_firmware_header_v1_0 
*)ucode->fw->data;
 
if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
switch (ucode->ucode_id) {
-- 
2.25.1

Re: [PATCH] Find bo_va before create it when map bo into compute VM

[Christian König]

The subject line somehow got messed up. There should be an drm/amdgpu: 
or drm/amdkfd: prefix.


Regards,
Christian.

Am 12.10.23 um 01:36 schrieb Xiaogang.Chen:

From: Xiaogang Chen 

This is needed to correctly handle BOs imported into compute VM from gfx.
Both kfd and gfx should use same bo_va when map the Bos into same VM, otherwise
we may trigger kernel general protection when iterate mappings from bo_va.

Signed-off-by: Felix Kuehling 
Acked-by: Christian König 
Reviewed-by: Ramesh Errabolu 
Reviewed-By: Xiaogang Chen 
Tested-By: Xiaogang Chen 
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 12 ++--
  1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c 
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index a15e59abe70a..c1ec93cc50ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -832,6 +832,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
uint64_t va = mem->va;
struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
struct amdgpu_bo *bo[2] = {NULL, NULL};
+   struct amdgpu_bo_va *bo_va;
bool same_hive = false;
int i, ret;
  
@@ -919,7 +920,13 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,

pr_debug("Unable to reserve BO during memory attach");
goto unwind;
}
-   attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   bo_va = amdgpu_vm_bo_find(vm, bo[i]);
+   if (!bo_va)
+   bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+   else
+   ++bo_va->ref_count;
+   attachment[i]->bo_va = bo_va;
+
amdgpu_bo_unreserve(bo[i]);
if (unlikely(!attachment[i]->bo_va)) {
ret = -ENOMEM;
@@ -943,7 +950,8 @@ static int kfd_mem_attach(struct amdgpu_device *adev, 
struct kgd_mem *mem,
continue;
if (attachment[i]->bo_va) {
amdgpu_bo_reserve(bo[i], true);
-   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
+   if (--attachment[i]->bo_va->ref_count == 0)
+   amdgpu_vm_bo_del(adev, attachment[i]->bo_va);
amdgpu_bo_unreserve(bo[i]);
list_del([i]->list);
}

[PATCH 1/2] drm/amdgpu/umsch: correct IP version format

[Lang Yu]

FW uses IP_VERSION_MAJ_MIN_REV format.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/umsch_mm_v4_0.c | 11 +++
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/umsch_mm_v4_0.c 
b/drivers/gpu/drm/amd/amdgpu/umsch_mm_v4_0.c
index 63917e2b51b4..8e7b763cfdb7 100644
--- a/drivers/gpu/drm/amd/amdgpu/umsch_mm_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/umsch_mm_v4_0.c
@@ -285,15 +285,18 @@ static int umsch_mm_v4_0_set_hw_resources(struct 
amdgpu_umsch_mm *umsch)
 
memcpy(set_hw_resources.mmhub_base, adev->reg_offset[MMHUB_HWIP][0],
   sizeof(uint32_t) * 5);
-   set_hw_resources.mmhub_version = amdgpu_ip_version(adev, MMHUB_HWIP, 0);
+   set_hw_resources.mmhub_version =
+   IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, MMHUB_HWIP, 0));
 
memcpy(set_hw_resources.osssys_base, adev->reg_offset[OSSSYS_HWIP][0],
   sizeof(uint32_t) * 5);
set_hw_resources.osssys_version =
-   amdgpu_ip_version(adev, OSSSYS_HWIP, 0);
+   IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, OSSSYS_HWIP, 0));
 
-   set_hw_resources.vcn_version = amdgpu_ip_version(adev, VCN_HWIP, 0);
-   set_hw_resources.vpe_version = amdgpu_ip_version(adev, VPE_HWIP, 0);
+   set_hw_resources.vcn_version =
+   IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, VCN_HWIP, 0));
+   set_hw_resources.vpe_version =
+   IP_VERSION_MAJ_MIN_REV(amdgpu_ip_version(adev, VPE_HWIP, 0));
 
set_hw_resources.api_status.api_completion_fence_addr = 
umsch->ring.fence_drv.gpu_addr;
set_hw_resources.api_status.api_completion_fence_value = 
++umsch->ring.fence_drv.sync_seq;
-- 
2.25.1

[PATCH 1/2] drm/amdgpu: correct NBIO v7.11 programing

[Lang Yu]

Use v7.7 before, switch to v7.11 now.
Fix incorrect programing.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c   | 56 +--
 .../asic_reg/nbio/nbio_7_11_0_offset.h|  9 ++-
 2 files changed, 33 insertions(+), 32 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c 
b/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
index 6873eead1e19..3a94f249929e 100644
--- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
+++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_11.c
@@ -66,19 +66,19 @@ static void nbio_v7_11_sdma_doorbell_range(struct 
amdgpu_device *adev, int insta
  bool use_doorbell, int doorbell_index,
  int doorbell_size)
 {
-   u32 reg = SOC15_REG_OFFSET(NBIO, 0, regGDC0_BIF_SDMA0_DOORBELL_RANGE);
+   u32 reg = SOC15_REG_OFFSET(NBIO, 0, regGDC0_BIF_CSDMA_DOORBELL_RANGE);
u32 doorbell_range = RREG32_PCIE_PORT(reg);
 
if (use_doorbell) {
doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
   OFFSET, doorbell_index);
doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
   SIZE, doorbell_size);
} else {
doorbell_range = REG_SET_FIELD(doorbell_range,
-  GDC0_BIF_SDMA0_DOORBELL_RANGE,
+  GDC0_BIF_CSDMA_DOORBELL_RANGE,
   SIZE, 0);
}
 
@@ -145,27 +145,25 @@ static void nbio_v7_11_enable_doorbell_aperture(struct 
amdgpu_device *adev,
 static void nbio_v7_11_enable_doorbell_selfring_aperture(struct amdgpu_device 
*adev,
bool enable)
 {
-/* u32 tmp = 0;
+   u32 tmp = 0;
 
if (enable) {
-   tmp = REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+   tmp = REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
DOORBELL_SELFRING_GPA_APER_EN, 1) |
-   REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+ REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
DOORBELL_SELFRING_GPA_APER_MODE, 1) |
-   REG_SET_FIELD(tmp, 
BIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
+ REG_SET_FIELD(tmp, 
BIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL,
DOORBELL_SELFRING_GPA_APER_SIZE, 0);
 
WREG32_SOC15(NBIO, 0,
-   regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_BASE_LOW,
+   regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_BASE_LOW,
lower_32_bits(adev->doorbell.base));
WREG32_SOC15(NBIO, 0,
-   regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_BASE_HIGH,
+   regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_BASE_HIGH,
upper_32_bits(adev->doorbell.base));
}
 
-   WREG32_SOC15(NBIO, 0, regBIF_BX_PF0_DOORBELL_SELFRING_GPA_APER_CNTL,
-   tmp);
-*/
+   WREG32_SOC15(NBIO, 0, regBIF_BX_PF1_DOORBELL_SELFRING_GPA_APER_CNTL, 
tmp);
 }
 
 
@@ -216,12 +214,12 @@ static void nbio_v7_11_ih_control(struct amdgpu_device 
*adev)
 
 static u32 nbio_v7_11_get_hdp_flush_req_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_GPU_HDP_FLUSH_REQ);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_GPU_HDP_FLUSH_REQ);
 }
 
 static u32 nbio_v7_11_get_hdp_flush_done_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_GPU_HDP_FLUSH_DONE);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_GPU_HDP_FLUSH_DONE);
 }
 
 static u32 nbio_v7_11_get_pcie_index_offset(struct amdgpu_device *adev)
@@ -236,27 +234,27 @@ static u32 nbio_v7_11_get_pcie_data_offset(struct 
amdgpu_device *adev)
 
 static u32 nbio_v7_11_get_pcie_port_index_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_RSMU_INDEX);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_RSMU_INDEX);
 }
 
 static u32 nbio_v7_11_get_pcie_port_data_offset(struct amdgpu_device *adev)
 {
-   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF0_RSMU_DATA);
+   return SOC15_REG_OFFSET(NBIO, 0, regBIF_BX_PF1_RSMU_DATA);
 }
 
 const struct nbio_hdp_flush_reg nbio_v7_11_hdp_flush_reg = {
-   .ref_and_mask_cp0 = BIF_BX_PF0_GPU_HDP_FLUSH_DONE__CP0_MASK,
-   .ref_and_mask_cp1 = 

[PATCH 2/2] drm/amdgpu: don't use legacy invalidation on MMHUB v3.3

[Lang Yu]

Legacy invalidation is not supported.
This is missed during rebase.

Signed-off-by: Lang Yu 
---
 drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c 
b/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
index 76b12f015d1d..dc4812ecc98d 100644
--- a/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/mmhub_v3_3.c
@@ -68,7 +68,7 @@ static uint32_t mmhub_v3_3_get_invalidate_req(unsigned int 
vmid,
/* invalidate using legacy mode on vmid*/
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ,
PER_VMID_INVALIDATE_REQ, 1 << vmid);
-   req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 
flush_type);
+   req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 
flush_type ? : 1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 
1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 
1);
req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 
1);
-- 
2.25.1