Changes in kfd_interrupt.c

e37ccf44 - drm/amdgpu: Show warning message if IH ring overflow

Tue, 03 Dec 2024 15:00:25 +0000

drm/amdgpu: Show warning message if IH ring overflowIf IH primary ring and KFD ih fifo overflows, we may miss CP, SDMAinterrupts and cause application soft hang. Show warning message withring name if overflow happens.Add function to get ih ring name to avoid duplicating it. To keepwarning message consistent between GPU generations, change all*_ih.c except ASICs older than Vega which has only one ih ring.Signed-off-by: Philip Yang Reviewed-by: Christian König Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

34db5a32 - drm/amdkfd: Queue interrupt work to different CPU

Tue, 26 Nov 2024 16:33:15 +0000

drm/amdkfd: Queue interrupt work to different CPUFor CPX mode, each KFD node has interrupt worker to process ih_fifo tosend events to user space. Currently all interrupt workers of same adevqueue to same CPU, all workers execution are actually serialized andthis cause KFD ih_fifo overflow when CPU usage is high.Use per-GPU unbounded highpri queue with number of workers equals tonumber of partitions, let queue_work select the next CPU round robinamong the local CPUs of same NUMA.Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

1b001432 - drm/amdgpu: Optimize gfx v9 GPU page fault handling

Wed, 13 Nov 2024 03:07:33 +0000

drm/amdgpu: Optimize gfx v9 GPU page fault handlingAfter GPU page fault, there are lots of page fault interrupts generatedat short period even with CAM filter enabled because the fault addressis different. Each page fault copy to KFD ih fifo to send event to userspace by KFD interrupt worker, this could cause KFD ih fifo overflowwhile other processes generate events at same time.KFD process is aborted after GPU page fault, we only need one GPU pagefault interrupt sent to KFD ih fifo to send memory exception event touser space.Incease KFD ih fifo size to 2 times of IH primary ring size, to handlethe burst events case.This patch handle the gfx v9 path, cover retry on/off and CAM filteron/off cases.Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

f607b2b8 - drm/amdkfd: KFD interrupt access ih_fifo data in-place

Fri, 22 Nov 2024 22:36:15 +0000

drm/amdkfd: KFD interrupt access ih_fifo data in-placeTo handle 40000 to 80000 interrupts per second running CPX mode with 4streams/queues per KFD node, KFD interrupt handler becomes theperformance bottleneck.Remove the kfifo_out memcpy overhead by accessing ih_fifo data in-placeand updating rptr with kfifo_skip_count.Signed-off-by: Philip Yang Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

417f78a2 - drm/amdkfd: Cleanup workqueue during module unload

Wed, 20 Mar 2024 22:43:14 +0000

drm/amdkfd: Cleanup workqueue during module unloadDestroy the high priority workqueue that handles interruptsduring KFD node cleanup.Signed-off-by: Mukul Joshi Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

8dc1db31 - drm/amdkfd: Introduce kfd_node struct (v5)

Wed, 14 Sep 2022 08:39:48 +0000

drm/amdkfd: Introduce kfd_node struct (v5)Introduce a new structure, kfd_node, which will now representa compute node. kfd_node is carved out of kfd_dev structure.kfd_dev struct now will become the parent of kfd_node, and willstore common resources such as doorbells, GTT sub-alloctor etc.kfd_node struct will store all resources specific to a computenode, such as device queue manager, interrupt handling etc.This is the first step in adding compute partition support in KFD.v2: introduce kfd_node struct to gc v11 (Hawking)v3: make reference to kfd_dev struct through kfd_node (Morris)v4: use kfd_node instead for kfd isr/mqd functions (Morris)v5: rebase (Alex)Signed-off-by: Mukul Joshi Tested-by: Amber Lin Reviewed-by: Felix Kuehling Signed-off-by: Hawking Zhang Signed-off-by: Morris Zhang Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

dcfe584b - drm/amdkfd: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"

Thu, 28 Jul 2022 03:30:26 +0000

drm/amdkfd: use time_is_before_jiffies(a + b) to replace "jiffies - a > b"time_is_before_jiffies deals with timer wrapping correctly.Signed-off-by: Yu Zhe Signed-off-by: Felix Kuehling Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

5273e82c - drm/amdkfd: Improve concurrency of event handling

Wed, 02 Mar 2022 01:40:45 +0000

drm/amdkfd: Improve concurrency of event handlingUse rcu_read_lock to read p->event_idr concurrently with other readersand writers. Use p->event_mutex only for creating and destroying eventsand in kfd_wait_on_events.Protect the contents of the kfd_event structure with a per-eventspinlock that can be taken inside the rcu_read_lock critical section.This eliminates contention of p->event_mutex in set_event, which tendsto be on the critical path for dispatch latency even when busy waitingis used. It also eliminates lock contention in event interrupt handlers.Since the p->event_mutex is now used much less, the impact of requiringit in kfd_wait_on_events should also be much smaller.This should improve event handling latency for processes using multipleGPUs concurrently.v2: Reschedule the worker periodically to avoid soft lockup warningsSigned-off-by: Felix Kuehling Reviewed-by: Sean Keely # v1Tested-by: Sanjay Tripathi Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

a0c5fd46 - drm/amdkfd: Use real device for messages

Fri, 18 Feb 2022 22:20:53 +0000

drm/amdkfd: Use real device for messageskfd_chardev() doesn't provide much useful information in dev_... messageson multi-GPU systems because there is only one KFD device, which doesn'tcorrespond to any particular GPU. Use the actual GPU device to indicatethe GPU that caused a message.Signed-off-by: Felix Kuehling Reviewed-by: Christian König Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

757f9e4d - drm/amdkfd: Drop IH ring overflow message to dbg

Fri, 18 Feb 2022 21:59:23 +0000

drm/amdkfd: Drop IH ring overflow message to dbgWhen this was first implemented, overflows weren't expected in regularoperations, and tests weren't in place to cause said overflow. Now thereare cases where overflows occur with real workloads, but we know thatthe kernel can handle this robustly, so move the message to a debugmessage.Signed-off-by: Kent Russell Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

d87f36a0 - drm/amdkfd: update SPDX license header

Fri, 11 Feb 2022 03:24:41 +0000

drm/amdkfd: update SPDX license headerUpdate the SPDX License header for all the KFD files.Reviewed-by: Felix Kuehling Signed-off-by: Rajneesh Bhardwaj Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

f0dc99a6 - drm/amdkfd: add kfd_device_info_init function

Wed, 17 Nov 2021 22:32:37 +0000

drm/amdkfd: add kfd_device_info_init functionInitializes kfd->device_info given either asic_type (enum) if GFXversion is less than GFX9, or GC IP version if greater. Also takes in vfand the target compiler gfx version. Uses SDMA version to determinenum_sdma_queues_per_engine.Convert device_info to a non-pointer member of kfd, change referencesaccordingly.Change unsupported asic condition to only probe f2g, move device_infoinitialization post-switch.Signed-off-by: Graham Sider Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

81de29d8 - drm/amdkfd: fix a potential NULL pointer dereference (v2)

Wed, 18 Sep 2019 16:30:31 +0000

drm/amdkfd: fix a potential NULL pointer dereference (v2)alloc_workqueue is not checked for errors and as a result,a potential NULL dereference could occur.v2 (Felix Kuehling):* Fix compile error (kfifo_free instead of fifo_free)* Return proper error codeSigned-off-by: Allen Pais Reviewed-by: Felix Kuehling Signed-off-by: Felix Kuehling Signed-off-by: Alex Deucher List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

58e69886 - drm/amdkfd: fix zero reading of VMID and PASID for Hawaii

Thu, 12 Jul 2018 02:32:51 +0000

drm/amdkfd: fix zero reading of VMID and PASID for HawaiiUpon VM Fault, the VMID and PASID written by HW are zeros inHawaii. Instead of reading from ih_ring_entry, read directlyfrom the registers. This workaround fix the soft hang issuescaused by mishandled VM Fault in Hawaii.Signed-off-by: Lan Xiao Signed-off-by: Felix Kuehling Acked-by: Christian König Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

af47b390 - drm/amdkfd: Remove vla

Fri, 13 Apr 2018 21:24:12 +0000

drm/amdkfd: Remove vlaThere's an ongoing effort to remove VLAs[1] from the kernel to eventuallyturn on -Wvla. Switch to a constant value that covers all hardware.[1] https://lkml.org/lkml/2018/3/7/621Reviewed-by: Felix Kuehling Acked-by: Christian König Signed-off-by: Laura Abbott Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

48e876a2 - drm/amdkfd: use a high priority workqueue for IH work

Fri, 27 Oct 2017 23:35:34 +0000

drm/amdkfd: use a high priority workqueue for IH workIn systems under heavy load the IH work may experience significantscheduling delays.Under load + system workqueue: Max Latency: 7.023695 ms Avg Latency: 0.263994 msUnder load + high priority workqueue: Max Latency: 1.162568 ms Avg Latency: 0.163213 msFurther work is required to measure the impact of per-cpu settings on IHperformance.Signed-off-by: Andres Rodriguez Signed-off-by: Felix Kuehling Acked-by: Oded Gabbay Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

0f875e3f - drm/amdkfd: wait only for IH work on IH exit

Fri, 27 Oct 2017 23:35:33 +0000

drm/amdkfd: wait only for IH work on IH exitWe don't need to wait for all work to complete in the IH exit function.We only need to make sure the interrupt_work has finished executing toguarantee that ih_kfifo is no longer in use.Signed-off-by: Andres Rodriguez Acked-by: Oded Gabbay Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

27232055 - drm/amdkfd: increase IH num entries to 8192

Fri, 27 Oct 2017 23:35:32 +0000

drm/amdkfd: increase IH num entries to 8192A larger buffer will let us accommodate applications with a large amountof semi-simultaneous event signals.Signed-off-by: Andres Rodriguez Signed-off-by: Felix Kuehling Acked-by: Oded Gabbay Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

04ad47bd - drm/amdkfd: use standard kernel kfifo for IH

Fri, 27 Oct 2017 23:35:31 +0000

drm/amdkfd: use standard kernel kfifo for IHReplace our implementation of a lockless ring buffer with the standardlinux kernel kfifo.We shouldn't maintain our own version of a standard data structure.Signed-off-by: Andres Rodriguez Signed-off-by: Felix Kuehling Acked-by: Oded Gabbay Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c

8eabaf54 - drm/amdkfd: Clean up KFD style errors and warnings v2

Wed, 16 Aug 2017 03:00:04 +0000

drm/amdkfd: Clean up KFD style errors and warnings v2Using checkpatch.pl -f showed a number of style issues. Thispatch addresses as many of them as possible. Some long lines have beenleft for readability, but attempts to minimize them have been made.v2: Broke long lines in gfx_v7 get_fw_versionSigned-off-by: Kent Russell Signed-off-by: Felix Kuehling Signed-off-by: Oded Gabbay List of files: /linux-6.15/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c