Merge tag 'amd-drm-next-6.12-2024-09-13' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

amd-drm-next-6.12-2024-09-13:

amdgpu:
- GPUVM sync fixes
- kdoc fixes
- Misc spelling mistakes
- Add some raven GFXOFF quirks
- Use clamp helper
- DC fixes
- JPEG fixes
- Process isolation fix
- Queue reset fix
- W=1 cleanup
- SMU14 fixes
- JPEG fixes

amdkfd:
- Fetch cacheline info from IP discovery
- Queue reset fix
- RAS fix
- Document SVM events
- CRIU fixes
- Race fix in dma-buf handling

drm:
- dma-buf fd race fixes

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240913134139.2861073-1-alexander.deucher@amd.com

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)
  *