Re: [PATCH 13/32] drm/amdgpu: prepare map process for multi-process debug devices
On 2023-01-25 14:53, Jonathan Kim wrote:
Unlike single process debug devices, multi-process debug devices allow
debug mode setting per-VMID (non-device-global).
Because the HWS manages PASID-VMID mapping, the new MAP_PROCESS API allows
the KFD to forward the required SPI debug register write requests.
To request a new debug mode setting change, the KFD must be able to
preempt all queues then remap all queues with these new setting
requests for MAP_PROCESS to take effect.
Note that by default, trap enablement in non-debug mode must be disabled
for performance reasons for multi-process debug devices due to setup
overhead in FW.
v2: remove asic family code name comment in per vmid support check
Signed-off-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h| 7 +++
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 50 +++
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 3 ++
.../drm/amd/amdkfd/kfd_packet_manager_v9.c| 15 ++
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 9
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 5 ++
6 files changed, 89 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 8aa7a3ad4e97..53c5a3e55bd2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -32,5 +32,12 @@ int kfd_dbg_trap_disable(struct kfd_process *target);
int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
void __user *runtime_info,
uint32_t *runtime_info_size);
+
+static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_dev *dev)
+{
+ return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
+}
+
void debug_event_write_work_handler(struct work_struct *work);
+
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 0cd3a5e9ff25..2517716d7cbc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2554,6 +2554,56 @@ int release_debug_trap_vmid(struct device_queue_manager
*dqm,
return r;
}
+int debug_lock_and_unmap(struct device_queue_manager *dqm)
+{
+ int r;
+
+ if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+ pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+ return -EINVAL;
+ }
+
+ if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+ return 0;
+
+ dqm_lock(dqm);
+
+ r = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, 0,
false);
+ if (r)
+ dqm_unlock(dqm);
+
+ return r;
+}
+
+int debug_map_and_unlock(struct device_queue_manager *dqm)
+{
+ int r;
+
+ if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+ pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+ return -EINVAL;
+ }
+
+ if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+ return 0;
+
+ r = map_queues_cpsch(dqm);
+
+ dqm_unlock(dqm);
+
+ return r;
+}
+
+int debug_refresh_runlist(struct device_queue_manager *dqm)
+{
+ int r = debug_lock_and_unmap(dqm);
+
+ if (r)
+ return r;
+
+ return debug_map_and_unlock(dqm);
+}
+
#if defined(CONFIG_DEBUG_FS)
static void seq_reg_dump(struct seq_file *m,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index 0cb1504d24cf..bef3be84c5cc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -286,6 +286,9 @@ int reserve_debug_trap_vmid(struct device_queue_manager
*dqm,
struct qcm_process_device *qpd);
int release_debug_trap_vmid(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
+int debug_lock_and_unmap(struct device_queue_manager *dqm);
+int debug_map_and_unlock(struct device_queue_manager *dqm);
+int debug_refresh_runlist(struct device_queue_manager *dqm);
static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
{
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
index 363cf8e005cc..f19c506da23d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
@@ -88,6 +88,10 @@ static int pm_map_process_aldebaran(struct packet_manager
*pm,
{
struct pm4_mes_map_process_aldebaran *packet;
uint64_t vm_page_table_base_addr = qpd->page_table_base;
+ struct kfd_dev *kfd = pm->dqm->dev;
+ struct kfd_process_device *pdd =
+ container_of(qpd, struct kfd_process_device, qpd);
+ int i;
packet = (struct pm4_mes_map_process_aldebaran *)buffer;
[PATCH 13/32] drm/amdgpu: prepare map process for multi-process debug devices
Unlike single process debug devices, multi-process debug devices allow
debug mode setting per-VMID (non-device-global).
Because the HWS manages PASID-VMID mapping, the new MAP_PROCESS API allows
the KFD to forward the required SPI debug register write requests.
To request a new debug mode setting change, the KFD must be able to
preempt all queues then remap all queues with these new setting
requests for MAP_PROCESS to take effect.
Note that by default, trap enablement in non-debug mode must be disabled
for performance reasons for multi-process debug devices due to setup
overhead in FW.
v2: remove asic family code name comment in per vmid support check
Signed-off-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h| 7 +++
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 50 +++
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 3 ++
.../drm/amd/amdkfd/kfd_packet_manager_v9.c| 15 ++
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 9
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 5 ++
6 files changed, 89 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
index 8aa7a3ad4e97..53c5a3e55bd2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
@@ -32,5 +32,12 @@ int kfd_dbg_trap_disable(struct kfd_process *target);
int kfd_dbg_trap_enable(struct kfd_process *target, uint32_t fd,
void __user *runtime_info,
uint32_t *runtime_info_size);
+
+static inline bool kfd_dbg_is_per_vmid_supported(struct kfd_dev *dev)
+{
+ return KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2);
+}
+
void debug_event_write_work_handler(struct work_struct *work);
+
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 0cd3a5e9ff25..2517716d7cbc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2554,6 +2554,56 @@ int release_debug_trap_vmid(struct device_queue_manager
*dqm,
return r;
}
+int debug_lock_and_unmap(struct device_queue_manager *dqm)
+{
+ int r;
+
+ if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+ pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+ return -EINVAL;
+ }
+
+ if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+ return 0;
+
+ dqm_lock(dqm);
+
+ r = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, 0,
false);
+ if (r)
+ dqm_unlock(dqm);
+
+ return r;
+}
+
+int debug_map_and_unlock(struct device_queue_manager *dqm)
+{
+ int r;
+
+ if (dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
+ pr_err("Unsupported on sched_policy: %i\n", dqm->sched_policy);
+ return -EINVAL;
+ }
+
+ if (!kfd_dbg_is_per_vmid_supported(dqm->dev))
+ return 0;
+
+ r = map_queues_cpsch(dqm);
+
+ dqm_unlock(dqm);
+
+ return r;
+}
+
+int debug_refresh_runlist(struct device_queue_manager *dqm)
+{
+ int r = debug_lock_and_unmap(dqm);
+
+ if (r)
+ return r;
+
+ return debug_map_and_unlock(dqm);
+}
+
#if defined(CONFIG_DEBUG_FS)
static void seq_reg_dump(struct seq_file *m,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index 0cb1504d24cf..bef3be84c5cc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -286,6 +286,9 @@ int reserve_debug_trap_vmid(struct device_queue_manager
*dqm,
struct qcm_process_device *qpd);
int release_debug_trap_vmid(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
+int debug_lock_and_unmap(struct device_queue_manager *dqm);
+int debug_map_and_unlock(struct device_queue_manager *dqm);
+int debug_refresh_runlist(struct device_queue_manager *dqm);
static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd)
{
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
index 363cf8e005cc..f19c506da23d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
@@ -88,6 +88,10 @@ static int pm_map_process_aldebaran(struct packet_manager
*pm,
{
struct pm4_mes_map_process_aldebaran *packet;
uint64_t vm_page_table_base_addr = qpd->page_table_base;
+ struct kfd_dev *kfd = pm->dqm->dev;
+ struct kfd_process_device *pdd =
+ container_of(qpd, struct kfd_process_device, qpd);
+ int i;
packet = (struct pm4_mes_map_process_aldebaran *)buffer;
memset(buffer, 0, sizeof(struct pm4_mes_map_process_ald
