OK, according to https://www.kernel.org/doc/Documentation/memory-barriers.txt, SLEEP AND WAKE-UP FUNCTIONS sections implies that. Will respin the patch. Andrey On 06/20/2018 11:02 AM, Andrey Grodzovsky wrote: > > Are you referring to insert_work->smp_mb(); ? > > Andrey > > > On 06/20/2018 10:50 AM, Christian König wrote: >>> +    * Try to submit work so SQ_EDC_INFO can be accessed from >>> +    * BH. If previous work submission hasn't finished yet >>> +    * just print whatever info is possible directly from the ISR. >>> +    */ >>> +   if (work_pending(&adev->gfx.sq_work.work)) { >>> +       gfx_v8_0_parse_sq_irq(adev, ih_data); >>> +   } else { >>> +       adev->gfx.sq_work.ih_data = ih_data; >>> +       /* Verify the new value visible in BH handler */ >>> +       smp_wmb(); >> >> You can drop the barrier here and in gfx_v8_0_sq_irq_work_func(), the >> schedule_work() function is a barrier itself for both reads and >> writes anyway. >> >> Apart from that the both patches are Acked-by: Christian König >> <christian.koenig at amd.com>. >> >> Regards, >> Christian. > -------------- next part -------------- An HTML attachment was scrubbed... URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20180620/e78bea15/attachment.html>