Re: [PATCH v2 5/8] drm/amdkfd: add migration SMI event

[Felix Kuehling <felix.kuehling@xxxxxxx>]

Am 2022-01-20 um 18:13 schrieb Philip Yang:
> After migration is finished, output timestamp when migration starts,
> duration of migration, svm range address and size, GPU id of
> migration source and destination and svm range attributes,
>
> Migration trigger could be prefetch, CPU or GPU page fault and TTM
> eviction.
>
> Signed-off-by: Philip Yang <Philip.Yang@xxxxxxx>
> ---
>   drivers/gpu/drm/amd/amdkfd/kfd_migrate.c    | 67 ++++++++++++++-------
>   drivers/gpu/drm/amd/amdkfd/kfd_migrate.h    |  5 +-
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 29 +++++++++
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h |  5 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_svm.c        | 16 +++--
>   5 files changed, 91 insertions(+), 31 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> index 88db82b3d443..06fb888f87aa 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
> @@ -32,6 +32,7 @@
>   #include "kfd_priv.h"
>   #include "kfd_svm.h"
>   #include "kfd_migrate.h"
> +#include "kfd_smi_events.h"
>   
>   #ifdef dev_fmt
>   #undef dev_fmt
> @@ -402,10 +403,11 @@ svm_migrate_copy_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>   static long
>   svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>   			struct vm_area_struct *vma, uint64_t start,
> -			uint64_t end)
> +			uint64_t end, uint32_t trigger)
>   {
>   	uint64_t npages = (end - start) >> PAGE_SHIFT;
> -	struct kfd_process_device *pdd;
> +	struct kfd_process_device *pdd = NULL;
> +	uint64_t timestamp = ktime_get_boottime_ns();
>   	struct dma_fence *mfence = NULL;
>   	struct migrate_vma migrate;
>   	unsigned long cpages = 0;
> @@ -431,6 +433,10 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>   	migrate.dst = migrate.src + npages;
>   	scratch = (dma_addr_t *)(migrate.dst + npages);
>   
> +	pdd = svm_range_get_pdd_by_adev(prange, adev);
> +	if (!pdd)
> +		goto out_free;
> +
>   	r = migrate_vma_setup(&migrate);
>   	if (r) {
>   		dev_err(adev->dev, "vma setup fail %d range [0x%lx 0x%lx]\n", r,
> @@ -459,6 +465,11 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>   	svm_migrate_copy_done(adev, mfence);
>   	migrate_vma_finalize(&migrate);
>   
> +	kfd_smi_event_migration(adev->kfd.dev, pdd->process->pasid,

The pasid is only used by kfd_smi_event_migration to determine the PID. 
But you already know the PID here:

	container_of(prange->svms, struct kfd_process, svms)->lead_thread->pid;


> +				start >> PAGE_SHIFT, end >> PAGE_SHIFT,
> +				0, adev->kfd.dev->id, prange->prefetch_loc,
> +				prange->preferred_loc, trigger, timestamp);
> +
>   	svm_range_dma_unmap(adev->dev, scratch, 0, npages);
>   	svm_range_free_dma_mappings(prange);
>   
> @@ -466,10 +477,7 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>   	kvfree(buf);
>   out:
>   	if (!r && cpages) {
> -		pdd = svm_range_get_pdd_by_adev(prange, adev);
> -		if (pdd)
> -			WRITE_ONCE(pdd->page_in, pdd->page_in + cpages);
> -
> +		WRITE_ONCE(pdd->page_in, pdd->page_in + cpages);
>   		return cpages;
>   	}
>   	return r;
> @@ -480,6 +488,7 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>    * @prange: range structure
>    * @best_loc: the device to migrate to
>    * @mm: the process mm structure
> + * @trigger: reason of migration
>    *
>    * Context: Process context, caller hold mmap read lock, svms lock, prange lock
>    *
> @@ -488,7 +497,7 @@ svm_migrate_vma_to_vram(struct amdgpu_device *adev, struct svm_range *prange,
>    */
>   static int
>   svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc,
> -			struct mm_struct *mm)
> +			struct mm_struct *mm, uint32_t trigger)
>   {
>   	unsigned long addr, start, end;
>   	struct vm_area_struct *vma;
> @@ -525,7 +534,7 @@ svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc,
>   			break;
>   
>   		next = min(vma->vm_end, end);
> -		r = svm_migrate_vma_to_vram(adev, prange, vma, addr, next);
> +		r = svm_migrate_vma_to_vram(adev, prange, vma, addr, next, trigger);
>   		if (r < 0) {
>   			pr_debug("failed %ld to migrate\n", r);
>   			break;
> @@ -641,12 +650,14 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
>   
>   static long
>   svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
> -		       struct vm_area_struct *vma, uint64_t start, uint64_t end)
> +		       struct vm_area_struct *vma, uint64_t start, uint64_t end,
> +		       uint32_t trigger)
>   {
>   	uint64_t npages = (end - start) >> PAGE_SHIFT;
> +	uint64_t timestamp = ktime_get_boottime_ns();
>   	unsigned long upages = npages;
>   	unsigned long cpages = 0;
> -	struct kfd_process_device *pdd;
> +	struct kfd_process_device *pdd = NULL;
>   	struct dma_fence *mfence = NULL;
>   	struct migrate_vma migrate;
>   	dma_addr_t *scratch;
> @@ -671,6 +682,10 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
>   	migrate.dst = migrate.src + npages;
>   	scratch = (dma_addr_t *)(migrate.dst + npages);
>   
> +	pdd = svm_range_get_pdd_by_adev(prange, adev);
> +	if (!pdd)
> +		goto out_free;
> +
>   	r = migrate_vma_setup(&migrate);
>   	if (r) {
>   		dev_err(adev->dev, "vma setup fail %d range [0x%lx 0x%lx]\n", r,
> @@ -701,16 +716,19 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
>   
>   	svm_migrate_copy_done(adev, mfence);
>   	migrate_vma_finalize(&migrate);
> +
> +	kfd_smi_event_migration(adev->kfd.dev, pdd->process->pasid,
> +				start >> PAGE_SHIFT, end >> PAGE_SHIFT,
> +				adev->kfd.dev->id, 0, prange->prefetch_loc,
> +				prange->preferred_loc, trigger, timestamp);
> +
>   	svm_range_dma_unmap(adev->dev, scratch, 0, npages);
>   
>   out_free:
>   	kvfree(buf);
>   out:
>   	if (!r && cpages) {
> -		pdd = svm_range_get_pdd_by_adev(prange, adev);
> -		if (pdd)
> -			WRITE_ONCE(pdd->page_out, pdd->page_out + cpages);
> -
> +		WRITE_ONCE(pdd->page_out, pdd->page_out + cpages);
>   		return upages;
>   	}
>   	return r ? r : upages;
> @@ -720,13 +738,15 @@ svm_migrate_vma_to_ram(struct amdgpu_device *adev, struct svm_range *prange,
>    * svm_migrate_vram_to_ram - migrate svm range from device to system
>    * @prange: range structure
>    * @mm: process mm, use current->mm if NULL
> + * @trigger: reason of migration
>    *
>    * Context: Process context, caller hold mmap read lock, svms lock, prange lock
>    *
>    * Return:
>    * 0 - OK, otherwise error code
>    */
> -int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
> +int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm,
> +			    uint32_t trigger)
>   {
>   	struct amdgpu_device *adev;
>   	struct vm_area_struct *vma;
> @@ -764,7 +784,7 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
>   			break;
>   
>   		next = min(vma->vm_end, end);
> -		r = svm_migrate_vma_to_ram(adev, prange, vma, addr, next);
> +		r = svm_migrate_vma_to_ram(adev, prange, vma, addr, next, trigger);
>   		if (r < 0) {
>   			pr_debug("failed %ld to migrate\n", r);
>   			break;
> @@ -787,6 +807,7 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
>    * @prange: range structure
>    * @best_loc: the device to migrate to
>    * @mm: process mm, use current->mm if NULL
> + * @trigger: reason of migration
>    *
>    * Context: Process context, caller hold mmap read lock, svms lock, prange lock
>    *
> @@ -795,7 +816,7 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm)
>    */
>   static int
>   svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc,
> -			 struct mm_struct *mm)
> +			 struct mm_struct *mm, uint32_t trigger)
>   {
>   	int r, retries = 3;
>   
> @@ -807,7 +828,7 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc,
>   	pr_debug("from gpu 0x%x to gpu 0x%x\n", prange->actual_loc, best_loc);
>   
>   	do {
> -		r = svm_migrate_vram_to_ram(prange, mm);
> +		r = svm_migrate_vram_to_ram(prange, mm, trigger);
>   		if (r)
>   			return r;
>   	} while (prange->actual_loc && --retries);
> @@ -815,17 +836,17 @@ svm_migrate_vram_to_vram(struct svm_range *prange, uint32_t best_loc,
>   	if (prange->actual_loc)
>   		return -EDEADLK;
>   
> -	return svm_migrate_ram_to_vram(prange, best_loc, mm);
> +	return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger);
>   }
>   
>   int
>   svm_migrate_to_vram(struct svm_range *prange, uint32_t best_loc,
> -		    struct mm_struct *mm)
> +		    struct mm_struct *mm, uint32_t trigger)
>   {
>   	if  (!prange->actual_loc)
> -		return svm_migrate_ram_to_vram(prange, best_loc, mm);
> +		return svm_migrate_ram_to_vram(prange, best_loc, mm, trigger);
>   	else
> -		return svm_migrate_vram_to_vram(prange, best_loc, mm);
> +		return svm_migrate_vram_to_vram(prange, best_loc, mm, trigger);
>   
>   }
>   
> @@ -894,7 +915,7 @@ static vm_fault_t svm_migrate_to_ram(struct vm_fault *vmf)
>   		goto out_unlock_prange;
>   	}
>   
> -	r = svm_migrate_vram_to_ram(prange, mm);
> +	r = svm_migrate_vram_to_ram(prange, mm, MIGRATION_TRIGGER_PAGEFAULT_CPU);
>   	if (r)
>   		pr_debug("failed %d migrate 0x%p [0x%lx 0x%lx] to ram\n", r,
>   			 prange, prange->start, prange->last);
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h
> index 2f5b3394c9ed..b3f0754b32fa 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.h
> @@ -41,8 +41,9 @@ enum MIGRATION_COPY_DIR {
>   };
>   
>   int svm_migrate_to_vram(struct svm_range *prange,  uint32_t best_loc,
> -			struct mm_struct *mm);
> -int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm);
> +			struct mm_struct *mm, uint32_t trigger);
> +int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm,
> +			    uint32_t trigger);
>   unsigned long
>   svm_migrate_addr_to_pfn(struct amdgpu_device *adev, unsigned long addr);
>   
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> index 080eba0d3be0..97393f4f3549 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
> @@ -314,6 +314,35 @@ void kfd_smi_event_page_fault_end(struct kfd_dev *dev, pid_t pid,
>   			   len);
>   }
>   
> +void kfd_smi_event_migration(struct kfd_dev *dev, uint16_t pasid,
> +			     unsigned long start, unsigned long end,
> +			     uint32_t from, uint32_t to,
> +			     uint32_t prefetch_loc, uint32_t preferred_loc,
> +			     uint32_t trigger, uint64_t ts)
> +{
> +	struct amdgpu_task_info task_info = {0};
> +	char fifo_in[128];
> +	uint64_t duration;
> +	pid_t pid;
> +	int len;
> +
> +	if (!kfd_smi_event_duration(dev, ts, &duration))
> +		return;
> +
> +	amdgpu_vm_get_task_info(dev->adev, pasid, &task_info);
> +	if (!task_info.pid)
> +		return;
> +	pid = task_info.pid;
> +
> +	len = snprintf(fifo_in, sizeof(fifo_in),
> +		       "%x %lld(%lld) -%d @%lx(%lx) %x->%x %x:%x %d\n",
> +		       KFD_SMI_EVENT_MIGRATION, ts, duration, pid, start,
> +		       end - start, from, to, prefetch_loc, preferred_loc,
> +		       trigger);

Would it make sense to have start and end events instead of a duration, 
for consistency with the page fault events?

Regards,
  Felix


> +
> +	add_event_to_kfifo(pid, dev, KFD_SMI_EVENT_MIGRATION, fifo_in, len);
> +}
> +
>   int kfd_smi_event_open(struct kfd_dev *dev, uint32_t *fd)
>   {
>   	struct kfd_smi_client *client;
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> index 7f70db914d2c..5788728f2879 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
> @@ -34,4 +34,9 @@ void kfd_smi_event_page_fault_start(struct kfd_dev *dev, pid_t pid,
>   void kfd_smi_event_page_fault_end(struct kfd_dev *dev, pid_t pid,
>   				  unsigned long address, bool migration,
>   				  uint64_t ts);
> +void kfd_smi_event_migration(struct kfd_dev *dev, uint16_t pasid,
> +			     unsigned long start, unsigned long end,
> +			     uint32_t from, uint32_t to,
> +			     uint32_t prefetch_loc, uint32_t preferred_loc,
> +			     uint32_t trigger, uint64_t ts);
>   #endif
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> index 08b21f9759ea..7dbc724364e6 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
> @@ -2791,7 +2791,8 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
>   	if (prange->actual_loc != best_loc) {
>   		migration = true;
>   		if (best_loc) {
> -			r = svm_migrate_to_vram(prange, best_loc, mm);
> +			r = svm_migrate_to_vram(prange, best_loc, mm,
> +						MIGRATION_TRIGGER_PAGEFAULT);
>   			if (r) {
>   				pr_debug("svm_migrate_to_vram failed (%d) at %llx, falling back to system memory\n",
>   					 r, addr);
> @@ -2799,12 +2800,14 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid,
>   				 * VRAM failed
>   				 */
>   				if (prange->actual_loc)
> -					r = svm_migrate_vram_to_ram(prange, mm);
> +					r = svm_migrate_vram_to_ram(prange, mm,
> +						MIGRATION_TRIGGER_PAGEFAULT);
>   				else
>   					r = 0;
>   			}
>   		} else {
> -			r = svm_migrate_vram_to_ram(prange, mm);
> +			r = svm_migrate_vram_to_ram(prange, mm,
> +						    MIGRATION_TRIGGER_PAGEFAULT);
>   		}
>   		if (r) {
>   			pr_debug("failed %d to migrate svms %p [0x%lx 0x%lx]\n",
> @@ -3126,12 +3129,12 @@ svm_range_trigger_migration(struct mm_struct *mm, struct svm_range *prange,
>   		return 0;
>   
>   	if (!best_loc) {
> -		r = svm_migrate_vram_to_ram(prange, mm);
> +		r = svm_migrate_vram_to_ram(prange, mm, MIGRATION_TRIGGER_PREFETCH);
>   		*migrated = !r;
>   		return r;
>   	}
>   
> -	r = svm_migrate_to_vram(prange, best_loc, mm);
> +	r = svm_migrate_to_vram(prange, best_loc, mm, MIGRATION_TRIGGER_PREFETCH);
>   	*migrated = !r;
>   
>   	return r;
> @@ -3188,7 +3191,8 @@ static void svm_range_evict_svm_bo_worker(struct work_struct *work)
>   		mutex_lock(&prange->migrate_mutex);
>   		do {
>   			svm_migrate_vram_to_ram(prange,
> -						svm_bo->eviction_fence->mm);
> +						svm_bo->eviction_fence->mm,
> +						MIGRATION_TRIGGER_TTM_EVICTION);
>   		} while (prange->actual_loc && --retries);
>   		WARN(prange->actual_loc, "Migration failed during eviction");
>