RE: [PATCH 2/2] drm/amdkfd: implement restore_mqd callbacks for GFX12/12.1
"Marioukhine, Vladimir" <[email protected]>
| Newsgroups | org.freedesktop.lists.amd-gfx |
|---|---|
| Message-ID | <SA1PR12MB8600CBE6F2B9B8D6BBEBEA349FA72@SA1PR12MB8600.namprd12.prod.outlook.com> |
AMD General Ping, Can I get a review on this change please? Kind regards, Vladimir From: Marioukhine, Vladimir Sent: Thursday, August 13, 2026 12:59 PM To: [email protected] Cc: Deucher, Alexander <[email protected]> Subject: [PATCH 2/2] drm/amdkfd: implement restore_mqd callbacks for GFX12/12.1 kfd_mqd_manager_v12.c (GFX 12.0) and kfd_mqd_manager_v12_1.c (GFX 12.1) do not implement restore_mqd or checkpoint_mqd callbacks, leaving the function pointers NULL and making CRIU restore non-functional on GFX12. Implement restore_mqd and checkpoint_mqd for both compute and SDMA queues in kfd_mqd_manager_v12.c and kfd_mqd_manager_v12_1.c, modeled after the GFX 11 implementation. Only restore_mqd is registered for now; checkpoint_mqd registration is deferred until get_checkpoint_info is implemented to avoid silently dropping the CWSR control stack. Fixes: 48f0bdf4e38e ("drm/amdkfd: Added MQD manager files for GFX12.") Fixes: 01bbc4a4b947 ("drm/amdkfd: Add MQD manager for GFX 12.1.0") Signed-off-by: Vladimir Marioukhine <[email protected]<mailto:[email protected]>> --- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12.c | 87 +++++++++++++++++++++ drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12_1.c | 87 +++++++++++++++++++++ 2 files changed, 174 insertions(+) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12.c index 7c387fa90076..a9a2dc4a0a75 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12.c @@ -380,6 +380,87 @@ static int debugfs_show_mqd_sdma(struct seq_file *m, void *data) #endif +static void checkpoint_mqd(struct mqd_manager *mm, void *mqd, + void *mqd_dst, void *ctl_stack_dst) +{ + /* + * GFXv12 control stack checkpoint is not yet implemented; + * get_checkpoint_info is unset, so ctl_stack_dst is always unused. + */ + struct v12_compute_mqd *m; + + m = get_mqd(mqd); + + memcpy(mqd_dst, m, sizeof(struct v12_compute_mqd)); +} + +static void restore_mqd(struct mqd_manager *mm, void **mqd, + struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr, + struct queue_properties *qp, const void *mqd_src, + const void *ctl_stack_src, const u32 ctl_stack_size) +{ + u64 addr; + struct v12_compute_mqd *m; + + if (ctl_stack_size) + pr_debug("GFX12 ctl stack restore not supported\n"); + + m = (struct v12_compute_mqd *)mqd_mem_obj->cpu_ptr; + addr = mqd_mem_obj->gpu_addr; + + memset(m, 0, AMDGPU_MQD_SIZE_ALIGN(mm->mqd_size)); + memcpy(m, mqd_src, sizeof(*m)); + + *mqd = m; + if (gart_addr) + *gart_addr = addr; + + m->cp_hqd_pq_doorbell_control = + qp->doorbell_off << CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT; + pr_debug("cp_hqd_pq_doorbell_control 0x%x\n", m->cp_hqd_pq_doorbell_control); + + qp->is_active = 0; +} + +static void checkpoint_mqd_sdma(struct mqd_manager *mm, void *mqd, + void *mqd_dst, void *ctl_stack_dst) +{ + struct v12_sdma_mqd *m; + + m = get_sdma_mqd(mqd); + + memcpy(mqd_dst, m, sizeof(struct v12_sdma_mqd)); +} + +static void restore_mqd_sdma(struct mqd_manager *mm, void **mqd, + struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr, + struct queue_properties *qp, + const void *mqd_src, + const void *ctl_stack_src, + const u32 ctl_stack_size) +{ + u64 addr; + struct v12_sdma_mqd *m; + + if (ctl_stack_size) + pr_debug("GFX12 ctl stack restore not supported\n"); + + m = (struct v12_sdma_mqd *)mqd_mem_obj->cpu_ptr; + addr = mqd_mem_obj->gpu_addr; + + memset(m, 0, AMDGPU_MQD_SIZE_ALIGN(mm->mqd_size)); + memcpy(m, mqd_src, sizeof(*m)); + + m->sdmax_rlcx_doorbell_offset = + qp->doorbell_off << SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT; + + *mqd = m; + if (gart_addr) + *gart_addr = addr; + + qp->is_active = 0; +} + struct mqd_manager *mqd_manager_init_v12(enum KFD_MQD_TYPE type, struct kfd_node *dev) { @@ -407,6 +488,7 @@ struct mqd_manager *mqd_manager_init_v12(enum KFD_MQD_TYPE type, mqd->mqd_size = sizeof(struct v12_compute_mqd); mqd->get_wave_state = get_wave_state; mqd->mqd_stride = kfd_mqd_stride; + mqd->restore_mqd = restore_mqd; #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif @@ -453,6 +535,7 @@ struct mqd_manager *mqd_manager_init_v12(enum KFD_MQD_TYPE type, mqd->is_occupied = kfd_is_occupied_sdma; mqd->mqd_size = sizeof(struct v12_sdma_mqd); mqd->mqd_stride = kfd_mqd_stride; + mqd->restore_mqd = restore_mqd_sdma; #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd_sdma; #endif diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12_1.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12_1.c index 431a940f91f3..daca9e39468e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12_1.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v12_1.c @@ -641,6 +641,87 @@ static int debugfs_show_mqd_sdma(struct seq_file *m, void *data) #endif +static void checkpoint_mqd_v12_1(struct mqd_manager *mm, void *mqd, + void *mqd_dst, void *ctl_stack_dst) +{ + /* + * GFXv12 control stack checkpoint is not yet implemented; + * get_checkpoint_info is unset, so ctl_stack_dst is always unused. + */ + struct v12_1_compute_mqd *m; + + m = get_mqd(mqd); + + memcpy(mqd_dst, m, sizeof(struct v12_1_compute_mqd)); +} + +static void restore_mqd_v12_1(struct mqd_manager *mm, void **mqd, + struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr, + struct queue_properties *qp, const void *mqd_src, + const void *ctl_stack_src, const u32 ctl_stack_size) +{ + u64 addr; + struct v12_1_compute_mqd *m; + + if (ctl_stack_size) + pr_debug("GFX12 ctl stack restore not supported\n"); + + m = (struct v12_1_compute_mqd *)mqd_mem_obj->cpu_ptr; + addr = mqd_mem_obj->gpu_addr; + + memset(m, 0, AMDGPU_MQD_SIZE_ALIGN(mm->mqd_size)); + memcpy(m, mqd_src, sizeof(*m)); + + *mqd = m; + if (gart_addr) + *gart_addr = addr; + + m->cp_hqd_pq_doorbell_control = + qp->doorbell_off << CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT; + pr_debug("cp_hqd_pq_doorbell_control 0x%x\n", m->cp_hqd_pq_doorbell_control); + + qp->is_active = 0; +} + +static void checkpoint_mqd_sdma_v12_1(struct mqd_manager *mm, void *mqd, + void *mqd_dst, void *ctl_stack_dst) +{ + struct v12_sdma_mqd *m; + + m = get_sdma_mqd(mqd); + + memcpy(mqd_dst, m, sizeof(struct v12_sdma_mqd)); +} + +static void restore_mqd_sdma_v12_1(struct mqd_manager *mm, void **mqd, + struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr, + struct queue_properties *qp, + const void *mqd_src, + const void *ctl_stack_src, + const u32 ctl_stack_size) +{ + u64 addr; + struct v12_sdma_mqd *m; + + if (ctl_stack_size) + pr_debug("GFX12 ctl stack restore not supported\n"); + + m = (struct v12_sdma_mqd *)mqd_mem_obj->cpu_ptr; + addr = mqd_mem_obj->gpu_addr; + + memset(m, 0, AMDGPU_MQD_SIZE_ALIGN(mm->mqd_size)); + memcpy(m, mqd_src, sizeof(*m)); + + m->sdmax_rlcx_doorbell_offset = + qp->doorbell_off << SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT; + + *mqd = m; + if (gart_addr) + *gart_addr = addr; + + qp->is_active = 0; +} + struct mqd_manager *mqd_manager_init_v12_1(enum KFD_MQD_TYPE type, struct kfd_node *dev) { @@ -668,6 +749,7 @@ struct mqd_manager *mqd_manager_init_v12_1(enum KFD_MQD_TYPE type, mqd->mqd_size = sizeof(struct v12_1_compute_mqd); mqd->get_wave_state = get_wave_state_v12_1; mqd->mqd_stride = kfd_mqd_stride; + mqd->restore_mqd = restore_mqd_v12_1; #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif @@ -714,6 +796,7 @@ struct mqd_manager *mqd_manager_init_v12_1(enum KFD_MQD_TYPE type, mqd->is_occupied = kfd_is_occupied_sdma; mqd->mqd_size = sizeof(struct v12_sdma_mqd); mqd->mqd_stride = kfd_mqd_stride; + mqd->restore_mqd = restore_mqd_sdma_v12_1; #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd_sdma; #endif