Michel Dänzer 写道:
> On Wed, 2007-12-26 at 08:00 +0800, minskey guo wrote:
>
>> Hi, guys,
>>
>> In intelWaitIrq() of intel_ioctl.c file, there is the following
>> code:
>>
>> void intelWaitIrq( intelContextPtr intel, int seq )
>> {
>> ...
>> do {
>> ret = drmCommandWrite( intel->driFd, DRM_I830_IRQ_WAIT, &intel->iw,
>> sizeof(intel->iw) );
>> } while (ret == -EAGAIN || ret == -EINTR);
>> ...
>> }
>>
>>
>> But the kernel i915 DRM driver doesn't return EAGAIN for
>> ioctl DRM_I830_IRQ_WAIT. Refer to i915_wait_irq() in file
>> i915_irq.c and DRM_WAIT_ON macro in drm_os_linux.h
>>
>> >From the macro of DRM_WAIT_ON, I think i915_wait_irq()/
>> ioctl DRM_I830_IRQ_WAIT can get return value of -EBUYS
>> or -EINTR only, no -EAGAIN.
>>
>>
>> So, I guess that the checking for -EAGAIN should be replaced
>> with the checking for -EBUSY in intelWaitIrq() of intel_ioctl.c,
>> Is my understanding right ?
>>
>
> No. The ioctl will only return -EBUSY after a 3 second timeout, which
> usually means the GPU has locked up, and retrying is pointless.
>
>
Generally, in my understanding, if GPU gets lock, there must be
something wrong, software or hardware. If it is software flaw,
we could improve it. If hardware flaw, maybe we could reset the
GPU. From ATI while paper, ATI radeon driver for windows Vista
has a reset unit, which will reset the GPU if necessary.
Is there any better ways to solve this issue ? If we are using
3D desktop, the whole desktop will exit when EBUSY is encountered,
it is difficult for end users to accept this.
thanks,
-minskey
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Mesa3d-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
