RE: [PATCH 2/2] drm/amdgpu: Add parsing SQ_EDC_INFO to SQ IH.
> -Original Message-
> From: Christian König [mailto:ckoenig.leichtzumer...@gmail.com]
> Sent: Wednesday, June 20, 2018 2:37 AM
> To: Grodzovsky, Andrey ; amd-
> g...@lists.freedesktop.org
> Cc: Panariti, David ; Haehnle, Nicolai
>
> Subject: Re: [PATCH 2/2] drm/amdgpu: Add parsing SQ_EDC_INFO to SQ IH.
>
> Am 19.06.2018 um 18:09 schrieb Andrey Grodzovsky:
> > Access to SQ_EDC_INFO requires selecting register instance and hence
> > mutex lock when accessing GRBM_GFX_INDEX for which a work is
> > schedueled from IH. But SQ interrupt can be raised on many instances
> > at once which means queuing work will usually succeed for the first
> > one but fail for the reset since the work takes time to process. To
> > avoid losing info about other interrupt instances call the parsing
> > function directly from high IRQ when current work hasn't finished and
> > avoid accessing SQ_EDC_INFO in that case.
> >
> > Signed-off-by: Andrey Grodzovsky
> > ---
> > drivers/gpu/drm/amd/amdgpu/amdgpu.h | 7 +++
> > drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 97
> ++-
> > 2 files changed, 91 insertions(+), 13 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > index e8c6cc1..a7b9ef5 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> > @@ -930,6 +930,11 @@ struct amdgpu_ngg {
> > boolinit;
> > };
> >
> > +struct sq_work {
> > + struct work_struct work;
> > + unsigned ih_data;
> > +};
> > +
> > struct amdgpu_gfx {
> > struct mutexgpu_clock_mutex;
> > struct amdgpu_gfx_configconfig;
> > @@ -970,6 +975,8 @@ struct amdgpu_gfx {
> > struct amdgpu_irq_src priv_inst_irq;
> > struct amdgpu_irq_src cp_ecc_error_irq;
> > struct amdgpu_irq_src sq_irq;
> > + struct sq_work sq_work;
> > +
> > /* gfx status */
> > uint32_tgfx_current_status;
> > /* ce ram size*/
> > diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > index 93904a7..0add7fc 100644
> > --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> > @@ -704,6 +704,17 @@ static const u32 stoney_mgcg_cgcg_init[] =
> > mmCGTS_SM_CTRL_REG, 0x, 0x96940200,
> > };
> >
> > +
> > +static const char * const sq_edc_source_names[] = {
> > + "SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
> > + "SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
> > + "SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data
> return",
> > + "SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
> > + "SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
> > + "SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
> > + "SQ_EDC_INFO_SOURCE_TA: EDC source is TA", };
> > +
> > static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
> > static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
> > static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev); @@
> > -2003,6 +2014,8 @@ static int gfx_v8_0_compute_ring_init(struct
> amdgpu_device *adev, int ring_id,
> > return 0;
> > }
> >
> > +static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
> > +
> > static int gfx_v8_0_sw_init(void *handle)
> > {
> > int i, j, k, r, ring_id;
> > @@ -2066,6 +2079,8 @@ static int gfx_v8_0_sw_init(void *handle)
> > return r;
> > }
> >
> > + INIT_WORK(>gfx.sq_work.work,
> gfx_v8_0_sq_irq_work_func);
> > +
> > adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
> >
> > gfx_v8_0_scratch_init(adev);
> > @@ -6952,14 +6967,11 @@ static int gfx_v8_0_cp_ecc_error_irq(struct
> amdgpu_device *adev,
> > return 0;
> > }
> >
> > -static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
> > - struct amdgpu_irq_src *source,
> > - struct amdgpu_iv_entry *entry)
> > +static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev,
> > +unsigned ih_data)
> > {
> > - u8 enc, se_id;
> > + u32 enc, se_id, sh_id, cu_id;
> > char type[20];
> > - unsigned ih_data = entry->src_data[0];
> > -
> > + int sq_ed
Re: [PATCH 2/2] drm/amdgpu: Add parsing SQ_EDC_INFO to SQ IH.
Am 19.06.2018 um 18:09 schrieb Andrey Grodzovsky:
Access to SQ_EDC_INFO requires selecting register instance and
hence mutex lock when accessing GRBM_GFX_INDEX for which a work
is schedueled from IH. But SQ interrupt can be raised on many instances
at once which means queuing work will usually succeed for the first one
but fail for the reset since the work takes time to process. To avoid
losing info about other interrupt instances call the parsing function
directly from high IRQ when current work hasn't finished and avoid
accessing SQ_EDC_INFO in that case.
Signed-off-by: Andrey Grodzovsky
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 7 +++
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 97 ++-
2 files changed, 91 insertions(+), 13 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index e8c6cc1..a7b9ef5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -930,6 +930,11 @@ struct amdgpu_ngg {
boolinit;
};
+struct sq_work {
+ struct work_struct work;
+ unsigned ih_data;
+};
+
struct amdgpu_gfx {
struct mutexgpu_clock_mutex;
struct amdgpu_gfx_configconfig;
@@ -970,6 +975,8 @@ struct amdgpu_gfx {
struct amdgpu_irq_src priv_inst_irq;
struct amdgpu_irq_src cp_ecc_error_irq;
struct amdgpu_irq_src sq_irq;
+ struct sq_work sq_work;
+
/* gfx status */
uint32_tgfx_current_status;
/* ce ram size*/
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 93904a7..0add7fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -704,6 +704,17 @@ static const u32 stoney_mgcg_cgcg_init[] =
mmCGTS_SM_CTRL_REG, 0x, 0x96940200,
};
+
+static const char * const sq_edc_source_names[] = {
+ "SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
+ "SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
+ "SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
+ "SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
+ "SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
+ "SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
+ "SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
+};
+
static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
@@ -2003,6 +2014,8 @@ static int gfx_v8_0_compute_ring_init(struct
amdgpu_device *adev, int ring_id,
return 0;
}
+static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
+
static int gfx_v8_0_sw_init(void *handle)
{
int i, j, k, r, ring_id;
@@ -2066,6 +2079,8 @@ static int gfx_v8_0_sw_init(void *handle)
return r;
}
+ INIT_WORK(>gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
+
adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
gfx_v8_0_scratch_init(adev);
@@ -6952,14 +6967,11 @@ static int gfx_v8_0_cp_ecc_error_irq(struct
amdgpu_device *adev,
return 0;
}
-static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
- struct amdgpu_irq_src *source,
- struct amdgpu_iv_entry *entry)
+static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
{
- u8 enc, se_id;
+ u32 enc, se_id, sh_id, cu_id;
char type[20];
- unsigned ih_data = entry->src_data[0];
-
+ int sq_edc_source = -1;
enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
@@ -6985,6 +6997,24 @@ static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
case 1:
case 2:
+ cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
+ sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE,
SH_ID);
+
+ /*
+* This function can be called either directly from ISR
+* or from BH in which case we can access SQ_EDC_INFO
+* instance
+*/
+ if (in_task()) {
+ mutex_lock(>grbm_idx_mutex);
+ gfx_v8_0_select_se_sh(adev, se_id, sh_id,
cu_id);
+
+ sq_edc_source =
REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
+
+ gfx_v8_0_select_se_sh(adev, 0x,
0x, 0x);
+ mutex_unlock(>grbm_idx_mutex);
+ }
+
if (enc == 1)
sprintf(type,
[PATCH 2/2] drm/amdgpu: Add parsing SQ_EDC_INFO to SQ IH.
Access to SQ_EDC_INFO requires selecting register instance and
hence mutex lock when accessing GRBM_GFX_INDEX for which a work
is schedueled from IH. But SQ interrupt can be raised on many instances
at once which means queuing work will usually succeed for the first one
but fail for the reset since the work takes time to process. To avoid
losing info about other interrupt instances call the parsing function
directly from high IRQ when current work hasn't finished and avoid
accessing SQ_EDC_INFO in that case.
Signed-off-by: Andrey Grodzovsky
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 7 +++
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 97 ++-
2 files changed, 91 insertions(+), 13 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index e8c6cc1..a7b9ef5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -930,6 +930,11 @@ struct amdgpu_ngg {
boolinit;
};
+struct sq_work {
+ struct work_struct work;
+ unsigned ih_data;
+};
+
struct amdgpu_gfx {
struct mutexgpu_clock_mutex;
struct amdgpu_gfx_configconfig;
@@ -970,6 +975,8 @@ struct amdgpu_gfx {
struct amdgpu_irq_src priv_inst_irq;
struct amdgpu_irq_src cp_ecc_error_irq;
struct amdgpu_irq_src sq_irq;
+ struct sq_work sq_work;
+
/* gfx status */
uint32_tgfx_current_status;
/* ce ram size*/
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 93904a7..0add7fc 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -704,6 +704,17 @@ static const u32 stoney_mgcg_cgcg_init[] =
mmCGTS_SM_CTRL_REG, 0x, 0x96940200,
};
+
+static const char * const sq_edc_source_names[] = {
+ "SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
+ "SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
+ "SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
+ "SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
+ "SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
+ "SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
+ "SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
+};
+
static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
@@ -2003,6 +2014,8 @@ static int gfx_v8_0_compute_ring_init(struct
amdgpu_device *adev, int ring_id,
return 0;
}
+static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
+
static int gfx_v8_0_sw_init(void *handle)
{
int i, j, k, r, ring_id;
@@ -2066,6 +2079,8 @@ static int gfx_v8_0_sw_init(void *handle)
return r;
}
+ INIT_WORK(>gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
+
adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
gfx_v8_0_scratch_init(adev);
@@ -6952,14 +6967,11 @@ static int gfx_v8_0_cp_ecc_error_irq(struct
amdgpu_device *adev,
return 0;
}
-static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
- struct amdgpu_irq_src *source,
- struct amdgpu_iv_entry *entry)
+static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
{
- u8 enc, se_id;
+ u32 enc, se_id, sh_id, cu_id;
char type[20];
- unsigned ih_data = entry->src_data[0];
-
+ int sq_edc_source = -1;
enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
@@ -6985,6 +6997,24 @@ static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
case 1:
case 2:
+ cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE,
CU_ID);
+ sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE,
SH_ID);
+
+ /*
+* This function can be called either directly from ISR
+* or from BH in which case we can access SQ_EDC_INFO
+* instance
+*/
+ if (in_task()) {
+ mutex_lock(>grbm_idx_mutex);
+ gfx_v8_0_select_se_sh(adev, se_id, sh_id,
cu_id);
+
+ sq_edc_source =
REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
+
+ gfx_v8_0_select_se_sh(adev, 0x,
0x, 0x);
+ mutex_unlock(>grbm_idx_mutex);
+ }
+
if (enc == 1)
sprintf(type, "instruction intr");
else
