Message ID | 1448544702-5594-1-git-send-email-chris@chris-wilson.co.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On Thu, Nov 26, 2015 at 01:31:42PM +0000, Chris Wilson wrote: > We have relied upon the sole caller (wait_ioctl) validating the timeout > argument. However, when waiting for multiple requests I forgot to ensure > that the timeout was still positive on the later requests. This is more > simply done inside __i915_wait_request. > > Fixes regression introduced in > commit b47161858ba13c9c7e03333132230d66e008dd55 > Author: Chris Wilson <chris@chris-wilson.co.uk> > Date: Mon Apr 27 13:41:17 2015 +0100 > > drm/i915: Implement inter-engine read-read optimisations > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > Cc: stable@vger.kernel.org Commit message should explain what the actual problem is - we add 1 jiffy of delay for each wait_request, potentially waiting quite a bit longer than what userspace asked for. And not sure this really justifies for cc: stable, since all the wait syscalls reserve the right to wait longer. Of course we should fix it, just to keep validating this possible. So with the commit message amended and cc: stable dropped this is Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch> > --- > drivers/gpu/drm/i915/i915_gem.c | 12 ++++++++++-- > 1 file changed, 10 insertions(+), 2 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c > index 73c2c48729ec..8c19a980f5e6 100644 > --- a/drivers/gpu/drm/i915/i915_gem.c > +++ b/drivers/gpu/drm/i915/i915_gem.c > @@ -1210,8 +1210,16 @@ int __i915_wait_request(struct drm_i915_gem_request *req, > if (i915_gem_request_completed(req, true)) > return 0; > > - timeout_expire = timeout ? > - jiffies + nsecs_to_jiffies_timeout((u64)*timeout) : 0; > + timeout_expire = 0; > + if (timeout) { > + if (WARN_ON(*timeout < 0)) > + return -EINVAL; > + > + if (*timeout == 0) > + return -ETIME; > + > + timeout_expire = jiffies + nsecs_to_jiffies_timeout(*timeout); > + } > > if (INTEL_INFO(dev_priv)->gen >= 6) > gen6_rps_boost(dev_priv, rps, req->emitted_jiffies); > -- > 2.6.2 >
On Thu, Nov 26, 2015 at 03:49:00PM +0100, Daniel Vetter wrote: > On Thu, Nov 26, 2015 at 01:31:42PM +0000, Chris Wilson wrote: > > We have relied upon the sole caller (wait_ioctl) validating the timeout > > argument. However, when waiting for multiple requests I forgot to ensure > > that the timeout was still positive on the later requests. This is more > > simply done inside __i915_wait_request. > > > > Fixes regression introduced in > > commit b47161858ba13c9c7e03333132230d66e008dd55 > > Author: Chris Wilson <chris@chris-wilson.co.uk> > > Date: Mon Apr 27 13:41:17 2015 +0100 > > > > drm/i915: Implement inter-engine read-read optimisations > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> > > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > > Cc: stable@vger.kernel.org > > Commit message should explain what the actual problem is - we add 1 jiffy > of delay for each wait_request, potentially waiting quite a bit longer > than what userspace asked for. > > And not sure this really justifies for cc: stable, since all the wait > syscalls reserve the right to wait longer. Of course we should fix it, > just to keep validating this possible. Dropping stable is fine, that was just a knee jerk reaction to finding a regression. The impact is 1 jiffie for each extra active ring for a wait_ioctl with a timeout -- I don't think anyone has noticed. -Chris
On Thu, 26 Nov 2015, Chris Wilson <chris@chris-wilson.co.uk> wrote: > On Thu, Nov 26, 2015 at 03:49:00PM +0100, Daniel Vetter wrote: >> On Thu, Nov 26, 2015 at 01:31:42PM +0000, Chris Wilson wrote: >> > We have relied upon the sole caller (wait_ioctl) validating the timeout >> > argument. However, when waiting for multiple requests I forgot to ensure >> > that the timeout was still positive on the later requests. This is more >> > simply done inside __i915_wait_request. >> > >> > Fixes regression introduced in >> > commit b47161858ba13c9c7e03333132230d66e008dd55 >> > Author: Chris Wilson <chris@chris-wilson.co.uk> >> > Date: Mon Apr 27 13:41:17 2015 +0100 >> > >> > drm/i915: Implement inter-engine read-read optimisations >> > >> > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >> > Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> >> > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> >> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> >> > Cc: stable@vger.kernel.org >> >> Commit message should explain what the actual problem is - we add 1 jiffy >> of delay for each wait_request, potentially waiting quite a bit longer >> than what userspace asked for. >> >> And not sure this really justifies for cc: stable, since all the wait >> syscalls reserve the right to wait longer. Of course we should fix it, >> just to keep validating this possible. > > Dropping stable is fine, that was just a knee jerk reaction to finding a > regression. The impact is 1 jiffie for each extra active ring for a > wait_ioctl with a timeout -- I don't think anyone has noticed. Pushed to drm-intel-fixes with some random copy-paste added about the 1 jiffy. Thanks for the patch and review. BR, Jani.
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 73c2c48729ec..8c19a980f5e6 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -1210,8 +1210,16 @@ int __i915_wait_request(struct drm_i915_gem_request *req, if (i915_gem_request_completed(req, true)) return 0; - timeout_expire = timeout ? - jiffies + nsecs_to_jiffies_timeout((u64)*timeout) : 0; + timeout_expire = 0; + if (timeout) { + if (WARN_ON(*timeout < 0)) + return -EINVAL; + + if (*timeout == 0) + return -ETIME; + + timeout_expire = jiffies + nsecs_to_jiffies_timeout(*timeout); + } if (INTEL_INFO(dev_priv)->gen >= 6) gen6_rps_boost(dev_priv, rps, req->emitted_jiffies);
We have relied upon the sole caller (wait_ioctl) validating the timeout argument. However, when waiting for multiple requests I forgot to ensure that the timeout was still positive on the later requests. This is more simply done inside __i915_wait_request. Fixes regression introduced in commit b47161858ba13c9c7e03333132230d66e008dd55 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Apr 27 13:41:17 2015 +0100 drm/i915: Implement inter-engine read-read optimisations Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Lionel Landwerlin <lionel.g.landwerlin@linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: stable@vger.kernel.org --- drivers/gpu/drm/i915/i915_gem.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)