[PATCH 2.6.38-10-generic] device driver: fix oops in radeon driver due to incorrect value from hardware
alexdeucher at gmail.com
Wed Aug 10 06:27:58 PDT 2011
2011/8/10 Michel Dänzer <michel at daenzer.net>:
> On Die, 2011-08-09 at 23:52 +0530, Mayank Rungta wrote:
>> Added a check for the radeon ring buffer write index in r600.c which
>> reads 0xffffffff on resume. This results in an Oops during
>> radeon_ring_write. Masking the value averts this.
>> This problem is not seen to be fixed in 3.0 r600.c as well.
>> Detailed analysis of the problem can be found at -
>> BUG: unable to handle kernel paging request at fa501ffc - Oops at
>> r600_cp_start+0x48/0x380 in r600_cp_resume+0x345/0x580 [radeon]
>> --- linux-2.6.38/drivers/gpu/drm/radeon/r600.c.orig 2011-08-05
>> 15:39:40.824612700 +0530
>> +++ linux-2.6.38/drivers/gpu/drm/radeon/r600.c 2011-08-08
>> 05:29:21.744417857 +0530
>> @@ -2218,6 +2218,8 @@ int r600_cp_resume(struct radeon_device
>> rdev->cp.rptr = RREG32(CP_RB_RPTR);
>> rdev->cp.wptr = RREG32(CP_RB_WPTR);
>> + /* protect against crazy HW on resume */
>> + rdev->cp.wptr &= rdev->cp.ptr_mask;
> The indentation of the lines you're adding doesn't match the surrounding
> Although the same workaround is already in r100.c, I wonder if we
> shouldn't rather try and eliminate all reads from the CP_RB_WPTR
> register, at least other than for debugging purposes. Alex, what do you
Either this or reset the registers to 0 or a saved value on resume
rather than reading from them.
> Otherwise, this should probably be added in evergreen.c as well.
Yes, and ni.c as well.
>> Developer's Certificate of Origin 1.1
> No need to include all this text, just the *-by: tags are enough.
> Earthling Michel Dänzer | http://www.amd.com
> Libre software enthusiast | Debian, X and DRI developer
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
More information about the dri-devel