On 2/6/2021 5:08 AM, Chris Wilson wrote:
Quoting Chris Wilson (2021-02-06 13:06:05)
Quoting Daniele Ceraolo Spurio (2021-02-06 02:09:21)
+ if (!ret) {
+ ret = wait_for(!pxp->termination_in_progress, 10);
This only works by chance. The compiler doesn't even have to reload the
variable. See struct completion.
This was a last minute addition when I was told that waiting on the
in_play state change was not enough to guarantee full invalidation and I
admit I didn't fully think it through because I want to get the RFC out.
It appears we already have a ready made one with the termination
i915_request. But that will require RCU pointer management...
-Chris
I've tried to keep this decoupled from the request because after the
request completion there is an MMIO write and only after that we start
waiting for the interrupt. I'll go with a struct completion.
Daniele
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
