[v3 6/6] drm/amdgpu: get rid of false warnings caused by amdgpu_irq_put()

[Gerry Liu]

> 2025年1月8日 18:02，Lazar, Lijo <lijo.lazar at amd.com> 写道：
> 
> 
> 
> On 1/8/2025 2:26 PM, Jiang Liu wrote:
>> If error happens before amdgpu_fence_driver_hw_init() gets called during
>> device probe, it will trigger a false warning in amdgpu_irq_put() as
>> below:
>> [ 1209.300996] ------------[ cut here ]------------
>> [ 1209.301061] WARNING: CPU: 48 PID: 293 at /tmp/amd.Rc9jFrl7/amd/amdgpu/amdgpu_irq.c:633 amdgpu_irq_put+0x45/0x70 [amdgpu]
>> [ 1209.301062] Modules linked in: ...
>> [ 1209.301093] CPU: 48 PID: 293 Comm: kworker/48:1 Kdump: loaded Tainted: G        W  OE     5.10.134-17.2.al8.x86_64 #1
>> [ 1209.301094] Hardware name: Alibaba Alibaba Cloud ECS/Alibaba Cloud ECS, BIOS 3.0.ES.AL.P.087.05 04/07/2024
>> [ 1209.301095] Workqueue: events work_for_cpu_fn
>> [ 1209.301159] RIP: 0010:amdgpu_irq_put+0x45/0x70 [amdgpu]
>> [ 1209.301160] Code: 48 8b 4e 10 48 83 39 00 74 2c 89 d1 48 8d 04 88 8b 08 85 c9 74 14 f0 ff 08 b8 00 00 00 00 74 05 c3 cc cc cc cc e9 8b fd ff ff <0f> 0b b8 ea ff ff ff c3 cc cc cc cc b8 ea ff ff ff c3 cc cc cc cc
>> [ 1209.301162] RSP: 0018:ffffb08a99c8fd88 EFLAGS: 00010246
>> [ 1209.301162] RAX: ffff9efe1bcbf500 RBX: ffff9efe1cc3e400 RCX: 0000000000000000
>> [ 1209.301163] RDX: 0000000000000000 RSI: ffff9efe1cc3b108 RDI: ffff9efe1cc00000
>> [ 1209.301163] RBP: ffff9efe1cc10818 R08: 0000000000000001 R09: 000000000000000d
>> [ 1209.301164] R10: ffffb08a99c8fb48 R11: ffffffffa2068018 R12: ffff9efe1cc109d0
>> [ 1209.301164] R13: ffff9efe1cc00010 R14: ffff9efe1cc00000 R15: ffff9efe1cc3b108
>> [ 1209.301165] FS:  0000000000000000(0000) GS:ffff9ff9fce00000(0000) knlGS:0000000000000000
>> [ 1209.301165] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [ 1209.301165] CR2: 00007fd0f6e860d0 CR3: 0000010092baa003 CR4: 0000000002770ee0
>> [ 1209.301166] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>> [ 1209.301166] DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
>> [ 1209.301167] PKRU: 55555554
>> [ 1209.301167] Call Trace:
>> [ 1209.301225]  amdgpu_fence_driver_hw_fini+0xda/0x110 [amdgpu]
>> [ 1209.301284]  amdgpu_device_fini_hw+0xaf/0x200 [amdgpu]
>> [ 1209.301342]  amdgpu_driver_load_kms+0x7f/0xc0 [amdgpu]
>> [ 1209.301400]  amdgpu_pci_probe+0x1cd/0x4a0 [amdgpu]
>> [ 1209.301401]  local_pci_probe+0x40/0xa0
>> [ 1209.301402]  work_for_cpu_fn+0x13/0x20
>> [ 1209.301403]  process_one_work+0x1ad/0x380
>> [ 1209.301404]  worker_thread+0x1c8/0x310
>> [ 1209.301405]  ? process_one_work+0x380/0x380
>> [ 1209.301406]  kthread+0x118/0x140
>> [ 1209.301407]  ? __kthread_bind_mask+0x60/0x60
>> [ 1209.301408]  ret_from_fork+0x1f/0x30
>> [ 1209.301410] ---[ end trace 733f120fe2ab13e5 ]---
>> [ 1209.301418] ------------[ cut here ]------------
>> 
>> Signed-off-by: Jiang Liu <gerry at linux.alibaba.com>
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 9 +++++++--
>> drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h  | 1 +
>> 2 files changed, 8 insertions(+), 2 deletions(-)
>> 
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> index b5e87b515139..0e41a535e05f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
>> @@ -614,9 +614,11 @@ void amdgpu_fence_driver_hw_fini(struct amdgpu_device *adev)
>> 
>> 		if (!drm_dev_is_unplugged(adev_to_drm(adev)) &&
>> 		    ring->fence_drv.irq_src &&
>> -		    amdgpu_fence_need_ring_interrupt_restore(ring))
>> +		    ring->fence_drv.irq_enabled) {
>> 			amdgpu_irq_put(adev, ring->fence_drv.irq_src,
>> 				       ring->fence_drv.irq_type);
>> +		        ring->fence_drv.irq_enabled = false;
>> +		}
>> 
>> 		del_timer_sync(&ring->fence_drv.fallback_timer);
>> 	}
>> @@ -693,9 +695,12 @@ void amdgpu_fence_driver_hw_init(struct amdgpu_device *adev)
>> 
>> 		/* enable the interrupt */
>> 		if (ring->fence_drv.irq_src &&
>> -		    amdgpu_fence_need_ring_interrupt_restore(ring))
>> +		    !ring->fence_drv.irq_enabled &&
>> +		    amdgpu_fence_need_ring_interrupt_restore(ring)) {
>> 			amdgpu_irq_get(adev, ring->fence_drv.irq_src,
>> 				       ring->fence_drv.irq_type);
>> +		        ring->fence_drv.irq_enabled = true;
>> +		}
> 
> I guess the problem is more generic like calling fence driver hw_fini()
> when hw_init is not called.
> 
You are so smart:)
I’m working on another patch set to fix these generic issues by tweaking the ip block and ras block state machine.
Thanks,
Gerry

> Thanks,
> Lijo
> 
>> 	}
>> }
>> 
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> index dee5a1b4e572..959d474a0516 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
>> @@ -118,6 +118,7 @@ struct amdgpu_fence_driver {
>> 	uint32_t			sync_seq;
>> 	atomic_t			last_seq;
>> 	bool				initialized;
>> +	bool				irq_enabled;
>> 	struct amdgpu_irq_src		*irq_src;
>> 	unsigned			irq_type;
>> 	struct timer_list		fallback_timer;

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20250108/0e8abe34/attachment-0001.htm>