Error message on RV710: reserve failed for wait

[Jean Delvare]

Hi again,

On Thu, 30 Sep 2010 15:18:27 +0200, Jean Delvare wrote:
> I am running kernel 2.6.36-rc6 on a Radeon HD4350 (RV710), and I see
> the following error messages in the logs:
> 
> Sep 30 14:09:27 endymion kernel: [21556.560593] radeon 0000:07:00.0: ffff88007c334000 reserve failed for wait
> Sep 30 14:09:29 endymion kernel: [21558.253859] radeon 0000:07:00.0: ffff88007c334000 reserve failed for wait
> 
> Sometimes that's just one or two of these, sometimes a bunch of them. I
> didn't notice any major problem so far. What kind of issue may result
> from this error? I see the message comes from function radeon_bo_wait()
> but I have no idea what this function does.

Problem still present in 2.6.36-rc7. I have investigated it myself, and
here's what I found.

The error code returned by ttm_bo_reserve() to radeon_bo_wait() when I
get this error message is EBUSY. This can only happen when
ttm_bo_reserve() isn't allowed to wait (no_wait=true). The only caller
passing no_wait=true is radeon_gem_busy_ioctl(). So, as a summary, I
see an error message in the kernel log whenever radeon_gem_busy_ioctl()
tries to process some ioctl but can't get access to some kernel data it
needs. Sorry for being vague here but I'm not familiar with the code at
all, so that's the best I can say.

I don't know who exactly is calling radeon_gem_busy_ioctl(), as it can
only be reached through a function pointer. I presume this is the X11
radeon driver, in user-space?

Jerome, you wrote that part of the code, maybe you could suggest a
course of action? Is it OK for the radeon kernel driver to return
-EBUSY to user-space in that case? If it is OK, then the error message
shouldn't be printed, or maybe it should be degraded to a debugging
message. If it isn't OK, what can be done to avoid it?

Thanks,
-- 
Jean Delvare