From mboxrd@z Thu Jan 1 00:00:00 1970 From: John Harrison Subject: Re: [PATCH 2/3] drm/i915: s/seqno/request/ tracking inside objects Date: Thu, 11 Sep 2014 12:51:37 +0100 Message-ID: <54118CC9.2060007@Intel.com> References: <1409995709-17996-1-git-send-email-chris@chris-wilson.co.uk> <1409995709-17996-2-git-send-email-chris@chris-wilson.co.uk> <540F0B44.4040401@Intel.com> <20140911065108.GC28332@nuc-i3427.alporthouse.com> Mime-Version: 1.0 Content-Type: multipart/mixed; boundary="===============1909182329==" Return-path: Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by gabe.freedesktop.org (Postfix) with ESMTP id AFB0C89CAF for ; Thu, 11 Sep 2014 04:51:41 -0700 (PDT) In-Reply-To: <20140911065108.GC28332@nuc-i3427.alporthouse.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" To: Chris Wilson , intel-gfx@lists.freedesktop.org, Jesse Barnes , Daniel Vetter , Damien Lespiau , Oscar Mateo , Brad Volkin , "Kukanova, Svetlana" , Akash Goel , "Daniel, Thomas" , "Siluvery, Arun" List-Id: intel-gfx@lists.freedesktop.org This is a multi-part message in MIME format. --===============1909182329== Content-Type: multipart/alternative; boundary="------------040201090106080102070708" This is a multi-part message in MIME format. --------------040201090106080102070708 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Will the PPGTT fix be sent out soon? If not, can you just send me a quick fix directly? Also, noticed that in 'submit_execbuf()', if the 'copy_from_user()' call fails then it jumps to the error return but does not set 'ret' thus will actually return success. Also, there seems to be a mix of 'goto error' vs 'return E*'. In general it looks like you do the goto only after 'rq' has been allocated but there are a couple of places that do the jump with a null 'rq'. The code will cope with trying to put a null rq but it seems a bit inconsistent. Thanks, John. On 11/09/2014 07:51, Chris Wilson wrote: > On Tue, Sep 09, 2014 at 03:14:29PM +0100, John Harrison wrote: >> I pulled a fresh tree on Monday and applied this set of patches. >> There were two conflicts. It looks like nightly does not have >> 'i915_gem_context_setparam_ioctl' yet but the tree the patches came >> from does. Also, my tree has 'DRM_I915_CTX_BAN_PERIOD' instead of >> 'ctx->hang_stats.ban_period_seconds'. > A couple of patches earlier in the week with which I have been using for > stress testing the hangcheck and capture paths. > >> However, I can only boot if I have both execlists and PPGTT >> disabled. With just PPGTT enabled, I get continuous GPU hangs and >> nothing ever gets rendered. With execlists enabled, I get a null >> pointer dereference in 'execlists_get_ring'. With both disabled, I >> can boot to an Ubuntu desktop but shortly after it goes pop with >> 'BUG_ON(obj->active == 0)' in 'i915_gem_object_retire__read'. > Fixed full-ppgtt, rather trivial oversight. But for lack of hardware > and simulator, I haven't been able to run on bdw, so execlists is only > compile tested. > -Chris > --------------040201090106080102070708 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit

Will the PPGTT fix be sent out soon? If not, can you just send me a quick fix directly?

 

Also, noticed that in 'submit_execbuf()', if the 'copy_from_user()' call fails then it jumps to the error return but does not set 'ret' thus will actually return success. Also, there seems to be a mix of 'goto error' vs 'return E*'. In general it looks like you do the goto only after 'rq' has been allocated but there are a couple of places that do the jump with a null 'rq'. The code will cope with trying to put a null rq but it seems a bit inconsistent.

 


Thanks,
John.

On 11/09/2014 07:51, Chris Wilson wrote:
On Tue, Sep 09, 2014 at 03:14:29PM +0100, John Harrison wrote:

I pulled a fresh tree on Monday and applied this set of patches.
There were two conflicts. It looks like nightly does not have
'i915_gem_context_setparam_ioctl' yet but the tree the patches came
from does. Also, my tree has 'DRM_I915_CTX_BAN_PERIOD' instead of
'ctx->hang_stats.ban_period_seconds'.

A couple of patches earlier in the week with which I have been using for
stress testing the hangcheck and capture paths.
 

However, I can only boot if I have both execlists and PPGTT
disabled. With just PPGTT enabled, I get continuous GPU hangs and
nothing ever gets rendered. With execlists enabled, I get a null
pointer dereference in 'execlists_get_ring'. With both disabled, I
can boot to an Ubuntu desktop but shortly after it goes pop with
'BUG_ON(obj->active == 0)' in 'i915_gem_object_retire__read'.

Fixed full-ppgtt, rather trivial oversight. But for lack of hardware
and simulator, I haven't been able to run on bdw, so execlists is only
compile tested.
-Chris

--------------040201090106080102070708-- --===============1909182329== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/intel-gfx --===============1909182329==--