[PATCH v3 0/4] Improve SVM migrate event report

[PATCH v3 0/4] Improve SVM migrate event report

[Philip Yang]

1. Document how to use SMI system management interface to receive SVM
events, define string format macro for user mode.
2. Increase the event kfifo size, so less chance to drop event.
3. Output migrate end event with error code if migration failed.
4. Report dropped event count if fifo is full.

v3:
  Simplify event drop count handling (James Zhu) 

Philip Yang (4):
  drm/amdkfd: Document and define SVM events message macro
  drm/amdkfd: Output migrate end event if migrate failed
  drm/amdkfd: Increase SMI event fifo size
  drm/amdkfd: SMI report dropped event count

 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c    |  14 ++-
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c |  79 +++++++++------
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h |   3 +-
 include/uapi/linux/kfd_ioctl.h              | 107 +++++++++++++++++---
 4 files changed, 150 insertions(+), 53 deletions(-)

-- 
2.43.2

[PATCH v3 1/4] drm/amdkfd: Document and define SVM events message macro

[Philip Yang]

Document how to use SMI system management interface to enable and
receive SVM events. Document SVM event triggers.

Define SVM events message string format macro that could be used by user
mode for sscanf to parse the event. Add it to uAPI header file to make
it obvious that is changing uAPI in future.

No functional changes.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c |  45 +++++----
 include/uapi/linux/kfd_ioctl.h              | 100 +++++++++++++++++---
 2 files changed, 109 insertions(+), 36 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index ea6a8e43bd5b..de8b9abf7afc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
@@ -235,17 +235,16 @@ void kfd_smi_event_update_gpu_reset(struct kfd_node *dev, bool post_reset,
 		amdgpu_reset_get_desc(reset_context, reset_cause,
 				      sizeof(reset_cause));
 
-	kfd_smi_event_add(0, dev, event, "%x %s\n",
-			  dev->reset_seq_num,
-			  reset_cause);
+	kfd_smi_event_add(0, dev, event, KFD_EVENT_FMT_UPDATE_GPU_RESET(
+			  dev->reset_seq_num, reset_cause));
 }
 
 void kfd_smi_event_update_thermal_throttling(struct kfd_node *dev,
 					     uint64_t throttle_bitmask)
 {
-	kfd_smi_event_add(0, dev, KFD_SMI_EVENT_THERMAL_THROTTLE, "%llx:%llx\n",
+	kfd_smi_event_add(0, dev, KFD_SMI_EVENT_THERMAL_THROTTLE, KFD_EVENT_FMT_THERMAL_THROTTLING(
 			  throttle_bitmask,
-			  amdgpu_dpm_get_thermal_throttling_counter(dev->adev));
+			  amdgpu_dpm_get_thermal_throttling_counter(dev->adev)));
 }
 
 void kfd_smi_event_update_vmfault(struct kfd_node *dev, uint16_t pasid)
@@ -256,8 +255,8 @@ void kfd_smi_event_update_vmfault(struct kfd_node *dev, uint16_t pasid)
 	if (task_info) {
 		/* Report VM faults from user applications, not retry from kernel */
 		if (task_info->pid)
-			kfd_smi_event_add(0, dev, KFD_SMI_EVENT_VMFAULT, "%x:%s\n",
-					 task_info->pid, task_info->task_name);
+			kfd_smi_event_add(0, dev, KFD_SMI_EVENT_VMFAULT, KFD_EVENT_FMT_VMFAULT(
+					  task_info->pid, task_info->task_name));
 		amdgpu_vm_put_task_info(task_info);
 	}
 }
@@ -267,16 +266,16 @@ void kfd_smi_event_page_fault_start(struct kfd_node *node, pid_t pid,
 				    ktime_t ts)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_PAGE_FAULT_START,
-			  "%lld -%d @%lx(%x) %c\n", ktime_to_ns(ts), pid,
-			  address, node->id, write_fault ? 'W' : 'R');
+			  KFD_EVENT_FMT_PAGEFAULT_START(ktime_to_ns(ts), pid,
+			  address, node->id, write_fault ? 'W' : 'R'));
 }
 
 void kfd_smi_event_page_fault_end(struct kfd_node *node, pid_t pid,
 				  unsigned long address, bool migration)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_PAGE_FAULT_END,
-			  "%lld -%d @%lx(%x) %c\n", ktime_get_boottime_ns(),
-			  pid, address, node->id, migration ? 'M' : 'U');
+			  KFD_EVENT_FMT_PAGEFAULT_END(ktime_get_boottime_ns(),
+			  pid, address, node->id, migration ? 'M' : 'U'));
 }
 
 void kfd_smi_event_migration_start(struct kfd_node *node, pid_t pid,
@@ -286,9 +285,9 @@ void kfd_smi_event_migration_start(struct kfd_node *node, pid_t pid,
 				   uint32_t trigger)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_MIGRATE_START,
-			  "%lld -%d @%lx(%lx) %x->%x %x:%x %d\n",
+			  KFD_EVENT_FMT_MIGRATE_START(
 			  ktime_get_boottime_ns(), pid, start, end - start,
-			  from, to, prefetch_loc, preferred_loc, trigger);
+			  from, to, prefetch_loc, preferred_loc, trigger));
 }
 
 void kfd_smi_event_migration_end(struct kfd_node *node, pid_t pid,
@@ -296,24 +295,24 @@ void kfd_smi_event_migration_end(struct kfd_node *node, pid_t pid,
 				 uint32_t from, uint32_t to, uint32_t trigger)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_MIGRATE_END,
-			  "%lld -%d @%lx(%lx) %x->%x %d\n",
+			  KFD_EVENT_FMT_MIGRATE_END(
 			  ktime_get_boottime_ns(), pid, start, end - start,
-			  from, to, trigger);
+			  from, to, trigger));
 }
 
 void kfd_smi_event_queue_eviction(struct kfd_node *node, pid_t pid,
 				  uint32_t trigger)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_QUEUE_EVICTION,
-			  "%lld -%d %x %d\n", ktime_get_boottime_ns(), pid,
-			  node->id, trigger);
+			  KFD_EVENT_FMT_QUEUE_EVICTION(ktime_get_boottime_ns(), pid,
+			  node->id, trigger));
 }
 
 void kfd_smi_event_queue_restore(struct kfd_node *node, pid_t pid)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_QUEUE_RESTORE,
-			  "%lld -%d %x\n", ktime_get_boottime_ns(), pid,
-			  node->id);
+			  KFD_EVENT_FMT_QUEUE_RESTORE(ktime_get_boottime_ns(), pid,
+			  node->id, 0));
 }
 
 void kfd_smi_event_queue_restore_rescheduled(struct mm_struct *mm)
@@ -330,8 +329,8 @@ void kfd_smi_event_queue_restore_rescheduled(struct mm_struct *mm)
 
 		kfd_smi_event_add(p->lead_thread->pid, pdd->dev,
 				  KFD_SMI_EVENT_QUEUE_RESTORE,
-				  "%lld -%d %x %c\n", ktime_get_boottime_ns(),
-				  p->lead_thread->pid, pdd->dev->id, 'R');
+				  KFD_EVENT_FMT_QUEUE_RESTORE(ktime_get_boottime_ns(),
+				  p->lead_thread->pid, pdd->dev->id, 'R'));
 	}
 	kfd_unref_process(p);
 }
@@ -341,8 +340,8 @@ void kfd_smi_event_unmap_from_gpu(struct kfd_node *node, pid_t pid,
 				  uint32_t trigger)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_UNMAP_FROM_GPU,
-			  "%lld -%d @%lx(%lx) %x %d\n", ktime_get_boottime_ns(),
-			  pid, address, last - address + 1, node->id, trigger);
+			  KFD_EVENT_FMT_UNMAP_FROM_GPU(ktime_get_boottime_ns(),
+			  pid, address, last - address + 1, node->id, trigger));
 }
 
 int kfd_smi_event_open(struct kfd_node *dev, uint32_t *fd)
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index 71a7ce5f2d4c..717307d6b5b7 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -540,26 +540,29 @@ enum kfd_smi_event {
 	KFD_SMI_EVENT_ALL_PROCESS = 64
 };
 
+/* The reason of the page migration event */
 enum KFD_MIGRATE_TRIGGERS {
-	KFD_MIGRATE_TRIGGER_PREFETCH,
-	KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU,
-	KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU,
-	KFD_MIGRATE_TRIGGER_TTM_EVICTION
+	KFD_MIGRATE_TRIGGER_PREFETCH,		/* Prefetch to GPU VRAM or system memory */
+	KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU,	/* GPU page fault recover */
+	KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU,	/* CPU page fault recover */
+	KFD_MIGRATE_TRIGGER_TTM_EVICTION	/* TTM eviction */
 };
 
+/* The reason of user queue evition event */
 enum KFD_QUEUE_EVICTION_TRIGGERS {
-	KFD_QUEUE_EVICTION_TRIGGER_SVM,
-	KFD_QUEUE_EVICTION_TRIGGER_USERPTR,
-	KFD_QUEUE_EVICTION_TRIGGER_TTM,
-	KFD_QUEUE_EVICTION_TRIGGER_SUSPEND,
-	KFD_QUEUE_EVICTION_CRIU_CHECKPOINT,
-	KFD_QUEUE_EVICTION_CRIU_RESTORE
+	KFD_QUEUE_EVICTION_TRIGGER_SVM,		/* SVM buffer migration */
+	KFD_QUEUE_EVICTION_TRIGGER_USERPTR,	/* userptr movement */
+	KFD_QUEUE_EVICTION_TRIGGER_TTM,		/* TTM move buffer */
+	KFD_QUEUE_EVICTION_TRIGGER_SUSPEND,	/* GPU suspend */
+	KFD_QUEUE_EVICTION_CRIU_CHECKPOINT,	/* CRIU checkpoint */
+	KFD_QUEUE_EVICTION_CRIU_RESTORE		/* CRIU restore */
 };
 
+/* The reason of unmap buffer from GPU event */
 enum KFD_SVM_UNMAP_TRIGGERS {
-	KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY,
-	KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY_MIGRATE,
-	KFD_SVM_UNMAP_TRIGGER_UNMAP_FROM_CPU
+	KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY,	/* MMU notifier CPU buffer movement */
+	KFD_SVM_UNMAP_TRIGGER_MMU_NOTIFY_MIGRATE,/* MMU notifier page migration */
+	KFD_SVM_UNMAP_TRIGGER_UNMAP_FROM_CPU	/* Unmap to free the buffer */
 };
 
 #define KFD_SMI_EVENT_MASK_FROM_INDEX(i) (1ULL << ((i) - 1))
@@ -570,6 +573,77 @@ struct kfd_ioctl_smi_events_args {
 	__u32 anon_fd;	/* from KFD */
 };
 
+/*
+ * SVM event tracing via SMI system management interface
+ *
+ * Open event file descriptor
+ *    use ioctl AMDKFD_IOC_SMI_EVENTS, pass in gpuid and return a anonymous file
+ *    descriptor to receive SMI events.
+ *    If calling with sudo permission, then file descriptor can be used to receive
+ *    SVM events from all processes, otherwise, to only receive SVM events of same
+ *    process.
+ *
+ * To enable the SVM event
+ *    Write event file descriptor with KFD_SMI_EVENT_MASK_FROM_INDEX(event) bitmap
+ *    mask to start record the event to the kfifo, use bitmap mask combination
+ *    for multiple events. New event mask will overwrite the previous event mask.
+ *    KFD_SMI_EVENT_MASK_FROM_INDEX(KFD_SMI_EVENT_ALL_PROCESS) bit requires sudo
+ *    permisson to receive SVM events from all process.
+ *
+ * To receive the event
+ *    Application can poll file descriptor to wait for the events, then read event
+ *    from the file into a buffer. Each event is one line string message, starting
+ *    with the event id, then the event specific information.
+ *
+ * To decode event information
+ *    The following event format string macro can be used with sscanf to decode
+ *    the specific event information.
+ *    event triggers: the reason to generate the event, defined as enum for unmap,
+ *    eviction and migrate events.
+ *    node, from, to, prefetch_loc, preferred_loc: GPU ID, or 0 for system memory.
+ *    addr: user mode address, in pages
+ *    size: in pages
+ *    pid: the process ID to generate the event
+ *    ns: timestamp in nanosecond-resolution, starts at system boot time but
+ *        stops during suspend
+ *    migrate_update: GPU page fault is recovered by 'M' for migrate, 'U' for update
+ *    rw: 'W' for write page fault, 'R' for read page fault
+ *    rescheduled: 'R' if the queue restore failed and rescheduled to try again
+ */
+#define KFD_EVENT_FMT_UPDATE_GPU_RESET(reset_seq_num, reset_cause)\
+		"%x %s\n", (reset_seq_num), (reset_cause)
+
+#define KFD_EVENT_FMT_THERMAL_THROTTLING(bitmask, counter)\
+		"%llx:%llx\n", (bitmask), (counter)
+
+#define KFD_EVENT_FMT_VMFAULT(pid, task_name)\
+		"%x:%s\n", (pid), (task_name)
+
+#define KFD_EVENT_FMT_PAGEFAULT_START(ns, pid, addr, node, rw)\
+		"%lld -%d @%lx(%x) %c\n", (ns), (pid), (addr), (node), (rw)
+
+#define KFD_EVENT_FMT_PAGEFAULT_END(ns, pid, addr, node, migrate_update)\
+		"%lld -%d @%lx(%x) %c\n", (ns), (pid), (addr), (node), (migrate_update)
+
+#define KFD_EVENT_FMT_MIGRATE_START(ns, pid, start, size, from, to, prefetch_loc,\
+		preferred_loc, migrate_trigger)\
+		"%lld -%d @%lx(%lx) %x->%x %x:%x %d\n", (ns), (pid), (start), (size),\
+		(from), (to), (prefetch_loc), (preferred_loc), (migrate_trigger)
+
+#define KFD_EVENT_FMT_MIGRATE_END(ns, pid, start, size, from, to, migrate_trigger)\
+		"%lld -%d @%lx(%lx) %x->%x %d\n", (ns), (pid), (start), (size),\
+		(from), (to), (migrate_trigger)
+
+#define KFD_EVENT_FMT_QUEUE_EVICTION(ns, pid, node, evict_trigger)\
+		"%lld -%d %x %d\n", (ns), (pid), (node), (evict_trigger)
+
+#define KFD_EVENT_FMT_QUEUE_RESTORE(ns, pid, node, rescheduled)\
+		"%lld -%d %x %c\n", (ns), (pid), (node), (rescheduled)
+
+#define KFD_EVENT_FMT_UNMAP_FROM_GPU(ns, pid, addr, size, node, unmap_trigger)\
+		"%lld -%d @%lx(%lx) %x %d\n", (ns), (pid), (addr), (size),\
+		(node), (unmap_trigger)
+
 /**************************************************************************************************
  * CRIU IOCTLs (Checkpoint Restore In Userspace)
  *
-- 
2.43.2

[PATCH v3 2/4] drm/amdkfd: Output migrate end event if migrate failed

[Philip Yang]

If page migration failed, also output migrate end event to match with
migrate start event, with failure error_code added to the end of the
migrate message macro. This will not break uAPI because application uses
old message macro sscanf drop and ignore the error_code.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c    | 14 ++++++--------
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c |  5 +++--
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h |  3 ++-
 include/uapi/linux/kfd_ioctl.h              |  7 ++++---
 4 files changed, 15 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index 8ee3d07ffbdf..eacfeb32f35d 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -445,14 +445,13 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange,
 	pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n",
 			 mpages, cpages, migrate.npages);
 
-	kfd_smi_event_migration_end(node, p->lead_thread->pid,
-				    start >> PAGE_SHIFT, end >> PAGE_SHIFT,
-				    0, node->id, trigger);
-
 	svm_range_dma_unmap_dev(adev->dev, scratch, 0, npages);
 
 out_free:
 	kvfree(buf);
+	kfd_smi_event_migration_end(node, p->lead_thread->pid,
+				    start >> PAGE_SHIFT, end >> PAGE_SHIFT,
+				    0, node->id, trigger, r);
 out:
 	if (!r && mpages) {
 		pdd = svm_range_get_pdd_by_node(prange, node);
@@ -751,14 +750,13 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange,
 	svm_migrate_copy_done(adev, mfence);
 	migrate_vma_finalize(&migrate);
 
-	kfd_smi_event_migration_end(node, p->lead_thread->pid,
-				    start >> PAGE_SHIFT, end >> PAGE_SHIFT,
-				    node->id, 0, trigger);
-
 	svm_range_dma_unmap_dev(adev->dev, scratch, 0, npages);
 
 out_free:
 	kvfree(buf);
+	kfd_smi_event_migration_end(node, p->lead_thread->pid,
+				    start >> PAGE_SHIFT, end >> PAGE_SHIFT,
+				    node->id, 0, trigger, r);
 out:
 	if (!r && cpages) {
 		mpages = cpages - upages;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index de8b9abf7afc..1d94b445a060 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
@@ -292,12 +292,13 @@ void kfd_smi_event_migration_start(struct kfd_node *node, pid_t pid,
 
 void kfd_smi_event_migration_end(struct kfd_node *node, pid_t pid,
 				 unsigned long start, unsigned long end,
-				 uint32_t from, uint32_t to, uint32_t trigger)
+				 uint32_t from, uint32_t to, uint32_t trigger,
+				 int error_code)
 {
 	kfd_smi_event_add(pid, node, KFD_SMI_EVENT_MIGRATE_END,
 			  KFD_EVENT_FMT_MIGRATE_END(
 			  ktime_get_boottime_ns(), pid, start, end - start,
-			  from, to, trigger));
+			  from, to, trigger, error_code));
 }
 
 void kfd_smi_event_queue_eviction(struct kfd_node *node, pid_t pid,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
index 85010b8307f8..503bff13d815 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h
@@ -44,7 +44,8 @@ void kfd_smi_event_migration_start(struct kfd_node *node, pid_t pid,
 			     uint32_t trigger);
 void kfd_smi_event_migration_end(struct kfd_node *node, pid_t pid,
 			     unsigned long start, unsigned long end,
-			     uint32_t from, uint32_t to, uint32_t trigger);
+			     uint32_t from, uint32_t to, uint32_t trigger,
+			     int error_code);
 void kfd_smi_event_queue_eviction(struct kfd_node *node, pid_t pid,
 				  uint32_t trigger);
 void kfd_smi_event_queue_restore(struct kfd_node *node, pid_t pid);
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index 717307d6b5b7..fa9f9846b88e 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -609,6 +609,7 @@ struct kfd_ioctl_smi_events_args {
  *    migrate_update: GPU page fault is recovered by 'M' for migrate, 'U' for update
  *    rw: 'W' for write page fault, 'R' for read page fault
  *    rescheduled: 'R' if the queue restore failed and rescheduled to try again
+ *    error_code: migrate failure error code, 0 if no error
  */
 #define KFD_EVENT_FMT_UPDATE_GPU_RESET(reset_seq_num, reset_cause)\
 		"%x %s\n", (reset_seq_num), (reset_cause)
@@ -630,9 +631,9 @@ struct kfd_ioctl_smi_events_args {
 		"%lld -%d @%lx(%lx) %x->%x %x:%x %d\n", (ns), (pid), (start), (size),\
 		(from), (to), (prefetch_loc), (preferred_loc), (migrate_trigger)
 
-#define KFD_EVENT_FMT_MIGRATE_END(ns, pid, start, size, from, to, migrate_trigger)\
-		"%lld -%d @%lx(%lx) %x->%x %d\n", (ns), (pid), (start), (size),\
-		(from), (to), (migrate_trigger)
+#define KFD_EVENT_FMT_MIGRATE_END(ns, pid, start, size, from, to, migrate_trigger, error_code) \
+		"%lld -%d @%lx(%lx) %x->%x %d %d\n", (ns), (pid), (start), (size),\
+		(from), (to), (migrate_trigger), (error_code)
 
 #define KFD_EVENT_FMT_QUEUE_EVICTION(ns, pid, node, evict_trigger)\
 		"%lld -%d %x %d\n", (ns), (pid), (node), (evict_trigger)
-- 
2.43.2

[PATCH v3 3/4] drm/amdkfd: Increase SMI event fifo size

[Philip Yang]

SMI event fifo size 1KB was enough to report GPU vm fault or reset
event, but could drop the more frequent SVM migration events. Increase
kfifo size to 8KB to store about 100 migrate events, less chance to drop
the migrate events if lots of migration happened in the short period of
time. Add KFD prefix to the macro name.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index 1d94b445a060..9b8169761ec5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
@@ -44,7 +44,7 @@ struct kfd_smi_client {
 	bool suser;
 };
 
-#define MAX_KFIFO_SIZE	1024
+#define KFD_MAX_KFIFO_SIZE	8192
 
 static __poll_t kfd_smi_ev_poll(struct file *, struct poll_table_struct *);
 static ssize_t kfd_smi_ev_read(struct file *, char __user *, size_t, loff_t *);
@@ -86,7 +86,7 @@ static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
 	struct kfd_smi_client *client = filep->private_data;
 	unsigned char *buf;
 
-	size = min_t(size_t, size, MAX_KFIFO_SIZE);
+	size = min_t(size_t, size, KFD_MAX_KFIFO_SIZE);
 	buf = kmalloc(size, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
@@ -355,7 +355,7 @@ int kfd_smi_event_open(struct kfd_node *dev, uint32_t *fd)
 		return -ENOMEM;
 	INIT_LIST_HEAD(&client->list);
 
-	ret = kfifo_alloc(&client->fifo, MAX_KFIFO_SIZE, GFP_KERNEL);
+	ret = kfifo_alloc(&client->fifo, KFD_MAX_KFIFO_SIZE, GFP_KERNEL);
 	if (ret) {
 		kfree(client);
 		return ret;
-- 
2.43.2

[PATCH v3 4/4] drm/amdkfd: SMI report dropped event count

[Philip Yang]

Add new SMI event to report the dropped event count.

When the event kfifo is full, drop count is not zero, or no enough space
left to store the event message, increase drop count.

After reading event out from kfifo, if event was dropped, drop_count is
not zero, generate a dropped event record and reset drop count to zero.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 25 +++++++++++++++++----
 include/uapi/linux/kfd_ioctl.h              |  6 +++++
 2 files changed, 27 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index 9b8169761ec5..db321b8ea127 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
@@ -42,6 +42,7 @@ struct kfd_smi_client {
 	struct rcu_head rcu;
 	pid_t pid;
 	bool suser;
+	u32 drop_count;
 };
 
 #define KFD_MAX_KFIFO_SIZE	8192
@@ -103,12 +104,26 @@ static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
 	}
 	to_copy = min(size, to_copy);
 	ret = kfifo_out(&client->fifo, buf, to_copy);
-	spin_unlock(&client->lock);
 	if (ret <= 0) {
+		spin_unlock(&client->lock);
 		ret = -EAGAIN;
 		goto ret_err;
 	}
 
+	if (client->drop_count) {
+		char msg[KFD_SMI_EVENT_MSG_SIZE];
+		int len;
+
+		len = snprintf(msg, sizeof(msg), "%x ", KFD_SMI_EVENT_DROPPED_EVENT);
+		len += snprintf(msg + len, sizeof(msg) - len,
+				KFD_EVENT_FMT_DROPPED_EVENT(ktime_get_boottime_ns(),
+				client->pid, client->drop_count));
+		kfifo_in(&client->fifo, msg, len);
+		client->drop_count = 0;
+	}
+
+	spin_unlock(&client->lock);
+
 	ret = copy_to_user(user, buf, to_copy);
 	if (ret) {
 		ret = -EFAULT;
@@ -182,13 +197,15 @@ static void add_event_to_kfifo(pid_t pid, struct kfd_node *dev,
 	list_for_each_entry_rcu(client, &dev->smi_clients, list) {
 		if (!kfd_smi_ev_enabled(pid, client, smi_event))
 			continue;
+
 		spin_lock(&client->lock);
-		if (kfifo_avail(&client->fifo) >= len) {
+		if (!client->drop_count && kfifo_avail(&client->fifo) >= len) {
 			kfifo_in(&client->fifo, event_msg, len);
 			wake_up_all(&client->wait_queue);
 		} else {
-			pr_debug("smi_event(EventID: %u): no space left\n",
-					smi_event);
+			client->drop_count++;
+			pr_debug("smi_event(EventID: %u): no space left drop_count %d\n",
+				 smi_event, client->drop_count);
 		}
 		spin_unlock(&client->lock);
 	}
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index fa9f9846b88e..7afd66d45313 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -530,6 +530,7 @@ enum kfd_smi_event {
 	KFD_SMI_EVENT_QUEUE_EVICTION = 9,
 	KFD_SMI_EVENT_QUEUE_RESTORE = 10,
 	KFD_SMI_EVENT_UNMAP_FROM_GPU = 11,
+	KFD_SMI_EVENT_DROPPED_EVENT = 12,
 
 	/*
 	 * max event number, as a flag bit to get events from all processes,
@@ -610,6 +611,7 @@ struct kfd_ioctl_smi_events_args {
  *    rw: 'W' for write page fault, 'R' for read page fault
  *    rescheduled: 'R' if the queue restore failed and rescheduled to try again
  *    error_code: migrate failure error code, 0 if no error
+ *    drop_count: how many events dropped when fifo is full
  */
 #define KFD_EVENT_FMT_UPDATE_GPU_RESET(reset_seq_num, reset_cause)\
 		"%x %s\n", (reset_seq_num), (reset_cause)
@@ -645,6 +647,10 @@ struct kfd_ioctl_smi_events_args {
 		"%lld -%d @%lx(%lx) %x %d\n", (ns), (pid), (addr), (size),\
 		(node), (unmap_trigger)
 
+#define KFD_EVENT_FMT_DROPPED_EVENT(ns, pid, drop_count)\
+		"%lld -%d %d\n", (ns), (pid), (drop_count)
+
+
 /**************************************************************************************************
  * CRIU IOCTLs (Checkpoint Restore In Userspace)
  *
-- 
2.43.2

[PATCH] drm/amdkfd: SMI report dropped event count

[Philip Yang]

Add new SMI event to report the dropped event count.

When the event kfifo is full, drop count is not zero, or no enough space
left to store the event message, increase drop count.

After reading event out from kfifo, if event was dropped, drop_count is
not zero, generate a dropped event record and reset drop count to zero.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 25 +++++++++++++++++----
 include/uapi/linux/kfd_ioctl.h              |  6 +++++
 2 files changed, 27 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index 9b8169761ec5..db321b8ea127 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
@@ -42,6 +42,7 @@ struct kfd_smi_client {
 	struct rcu_head rcu;
 	pid_t pid;
 	bool suser;
+	u32 drop_count;
 };
 
 #define KFD_MAX_KFIFO_SIZE	8192
@@ -103,12 +104,26 @@ static ssize_t kfd_smi_ev_read(struct file *filep, char __user *user,
 	}
 	to_copy = min(size, to_copy);
 	ret = kfifo_out(&client->fifo, buf, to_copy);
-	spin_unlock(&client->lock);
 	if (ret <= 0) {
+		spin_unlock(&client->lock);
 		ret = -EAGAIN;
 		goto ret_err;
 	}
 
+	if (client->drop_count) {
+		char msg[KFD_SMI_EVENT_MSG_SIZE];
+		int len;
+
+		len = snprintf(msg, sizeof(msg), "%x ", KFD_SMI_EVENT_DROPPED_EVENT);
+		len += snprintf(msg + len, sizeof(msg) - len,
+				KFD_EVENT_FMT_DROPPED_EVENT(ktime_get_boottime_ns(),
+				client->pid, client->drop_count));
+		kfifo_in(&client->fifo, msg, len);
+		client->drop_count = 0;
+	}
+
+	spin_unlock(&client->lock);
+
 	ret = copy_to_user(user, buf, to_copy);
 	if (ret) {
 		ret = -EFAULT;
@@ -182,13 +197,15 @@ static void add_event_to_kfifo(pid_t pid, struct kfd_node *dev,
 	list_for_each_entry_rcu(client, &dev->smi_clients, list) {
 		if (!kfd_smi_ev_enabled(pid, client, smi_event))
 			continue;
+
 		spin_lock(&client->lock);
-		if (kfifo_avail(&client->fifo) >= len) {
+		if (!client->drop_count && kfifo_avail(&client->fifo) >= len) {
 			kfifo_in(&client->fifo, event_msg, len);
 			wake_up_all(&client->wait_queue);
 		} else {
-			pr_debug("smi_event(EventID: %u): no space left\n",
-					smi_event);
+			client->drop_count++;
+			pr_debug("smi_event(EventID: %u): no space left drop_count %d\n",
+				 smi_event, client->drop_count);
 		}
 		spin_unlock(&client->lock);
 	}
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index fa9f9846b88e..7afd66d45313 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -530,6 +530,7 @@ enum kfd_smi_event {
 	KFD_SMI_EVENT_QUEUE_EVICTION = 9,
 	KFD_SMI_EVENT_QUEUE_RESTORE = 10,
 	KFD_SMI_EVENT_UNMAP_FROM_GPU = 11,
+	KFD_SMI_EVENT_DROPPED_EVENT = 12,
 
 	/*
 	 * max event number, as a flag bit to get events from all processes,
@@ -610,6 +611,7 @@ struct kfd_ioctl_smi_events_args {
  *    rw: 'W' for write page fault, 'R' for read page fault
  *    rescheduled: 'R' if the queue restore failed and rescheduled to try again
  *    error_code: migrate failure error code, 0 if no error
+ *    drop_count: how many events dropped when fifo is full
  */
 #define KFD_EVENT_FMT_UPDATE_GPU_RESET(reset_seq_num, reset_cause)\
 		"%x %s\n", (reset_seq_num), (reset_cause)
@@ -645,6 +647,10 @@ struct kfd_ioctl_smi_events_args {
 		"%lld -%d @%lx(%lx) %x %d\n", (ns), (pid), (addr), (size),\
 		(node), (unmap_trigger)
 
+#define KFD_EVENT_FMT_DROPPED_EVENT(ns, pid, drop_count)\
+		"%lld -%d %d\n", (ns), (pid), (drop_count)
+
+
 /**************************************************************************************************
  * CRIU IOCTLs (Checkpoint Restore In Userspace)
  *
-- 
2.43.2

Re: [PATCH v3 0/4] Improve SVM migrate event report

[James Zhu]

[-- Attachment #1: Type: text/plain, Size: 1016 bytes --]

Reviewed-by:JamesZhu<James.Zhu@amd.com>fortheseries.

On 2024-08-27 13:10, Philip Yang wrote:
> 1. Document how to use SMI system management interface to receive SVM
> events, define string format macro for user mode.
> 2. Increase the event kfifo size, so less chance to drop event.
> 3. Output migrate end event with error code if migration failed.
> 4. Report dropped event count if fifo is full.
>
> v3:
>    Simplify event drop count handling (James Zhu)
>
> Philip Yang (4):
>    drm/amdkfd: Document and define SVM events message macro
>    drm/amdkfd: Output migrate end event if migrate failed
>    drm/amdkfd: Increase SMI event fifo size
>    drm/amdkfd: SMI report dropped event count
>
>   drivers/gpu/drm/amd/amdkfd/kfd_migrate.c    |  14 ++-
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c |  79 +++++++++------
>   drivers/gpu/drm/amd/amdkfd/kfd_smi_events.h |   3 +-
>   include/uapi/linux/kfd_ioctl.h              | 107 +++++++++++++++++---
>   4 files changed, 150 insertions(+), 53 deletions(-)
>

[-- Attachment #2: Type: text/html, Size: 1861 bytes --]