[PATCH] drm/xe/guc: In guc_ct_send_recv flush g2h worker if g2h resp times out

[Nilawar, Badal]

On 28-09-2024 02:57, Matthew Brost wrote:
> On Sat, Sep 28, 2024 at 12:54:28AM +0530, Badal Nilawar wrote:
>> It is observed that for GuC CT request G2H IRQ triggered and g2h_worker
>> queued, but it didn't get opportunity to execute and timeout occurred.
>> To address this the g2h_worker is being flushed.
>>
>> Cc: John Harrison <John.C.Harrison at Intel.com>
>> Signed-off-by: Badal Nilawar <badal.nilawar at intel.com>
>> ---
>>   drivers/gpu/drm/xe/xe_guc_ct.c | 11 +++++++++++
>>   1 file changed, 11 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
>> index 4b95f75b1546..4a5d7f85d1a0 100644
>> --- a/drivers/gpu/drm/xe/xe_guc_ct.c
>> +++ b/drivers/gpu/drm/xe/xe_guc_ct.c
>> @@ -903,6 +903,17 @@ static int guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
>>   	}
>>   
>>   	ret = wait_event_timeout(ct->g2h_fence_wq, g2h_fence.done, HZ);
>> +
>> +	/*
>> +	 * It is observed that for above GuC CT request G2H IRQ triggered
> 
> Where is this observed. 1 second is a long to wait for a worker...

Please see this log.

[  176.602482] xe 0000:00:02.0: [drm:xe_guc_pc_get_min_freq [xe]] GT0: 
GT[0] GuC PC status query
[  176.603019] xe 0000:00:02.0: [drm:xe_guc_irq_handler [xe]] GT0: G2H 
IRQ GT[0]
[  176.603449] xe 0000:00:02.0: [drm:g2h_worker_func [xe]] GT0: G2H work 
running GT[0]
[  176.604379] xe 0000:00:02.0: [drm:xe_guc_pc_get_max_freq [xe]] GT0: 
GT[0] GuC PC status query
[  176.605464] xe 0000:00:02.0: [drm:xe_guc_irq_handler [xe]] GT0: G2H 
IRQ GT[0]
[  176.605821] xe 0000:00:02.0: [drm:g2h_worker_func [xe]] GT0: G2H work 
running GT[0]
[  176.716699] xe 0000:00:02.0: [drm] GT0: trying reset
[  176.716718] xe 0000:00:02.0: [drm] GT0: GuC PC status query    //GuC 
PC check request
[  176.717648] xe 0000:00:02.0: [drm:xe_guc_irq_handler [xe]] GT0: G2H 
IRQ GT[0]  // IRQ
[  177.728637] xe 0000:00:02.0: [drm] *ERROR* GT0: Timed out wait for 
G2H, fence 1311, action 3003  //Timeout
[  177.737637] xe 0000:00:02.0: [drm] *ERROR* GT0: GuC PC query task 
state failed: -ETIME
[  177.745644] xe 0000:00:02.0: [drm] GT0: reset queued
[  177.849081] xe 0000:00:02.0: [drm:xe_guc_pc_get_min_freq [xe]] GT0: 
GT[0] GuC PC status query
[  177.849659] xe 0000:00:02.0: [drm:xe_guc_irq_handler [xe]] GT0: G2H 
IRQ GT[0]
[  178.632672] xe 0000:00:02.0: [drm] GT0: reset started
[  178.632639] xe 0000:00:02.0: [drm:g2h_worker_func [xe]] GT0: G2H work 
running GT[0] // Worker ran
[  178.632897] xe 0000:00:02.0: [drm] GT0: G2H fence (1311) not found!

> 
>> +	 * and g2h_worker queued, but it didn't get opportunity to execute
>> +	 * and timeout occurred. To address the g2h_worker is being flushed.
>> +	 */
>> +	if (!ret) {
>> +		flush_work(&ct->g2h_worker);
>> +		ret = wait_event_timeout(ct->g2h_fence_wq, g2h_fence.done, HZ);
> 
> If this is needed I wouldn't wait 1 second, if the flush worked
> 'g2h_fence.done' should immediately be signaled. Maybe wait 1 MS?

In config HZ is set to 250, which is 4 ms I think.

CONFIG_HZ_250=y
# CONFIG_HZ_300 is not set
# CONFIG_HZ_1000 is not set
CONFIG_HZ=250

Regards,
Badal

> 
> Matt
> 
>> +	}
>> +
>>   	if (!ret) {
>>   		xe_gt_err(gt, "Timed out wait for G2H, fence %u, action %04x",
>>   			  g2h_fence.seqno, action[0]);
>> -- 
>> 2.34.1
>>