Message ID | 20170215041751.11441-1-kenneth@whitecape.org (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On ti, 2017-02-14 at 20:17 -0800, Kenneth Graunke wrote: > This patch makes the I915_PARAM_HAS_EXEC_CONSTANTS getparam return 0 > (indicating the optional feature is not supported), and makes execbuf > always return -EINVAL if the flags are used. > > Apparently, no userspace ever shipped which used this optional feature: > I checked the git history of Mesa, xf86-video-intel, libva, and Beignet, > and there were zero commits showing a use of these flags. Kernel commit > 72bfa19c8deb4 apparently introduced the feature prematurely. > > 'relative_constants_mode' has always been tracked per-device, but this > has actually been wrong ever since hardware contexts were introduced, as > the INSTPM register is saved (and automatically restored) as part of the > render ring context. The software per-device value could therefore get > out of sync with the hardware per-context value. This meant that using > them is actually unsafe: a client which tried to use them could damage > the state of other clients, causing the GPU to interpret their BO > offsets as absolute pointers, leading to bogus memory reads. > > These flags were also never ported to execlist mode, making them no-ops > on Gen9+ (which requires execlists), and Gen8 in the default mode. > > On Gen8+, userspace can write these registers directly, achieving the > same effect. On Gen6-7.5, it likely makes sense to extend the command > parser to support them. I don't think anyone wants this on Gen4-5. > > Based on a patch by Dave Gordon. > > Cc: Dave Gordon <david.s.gordon@intel.com> > Cc: Chris Wilson <chris@chris-wilson.co.uk> > Cc: stable@vger.kernel.org > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92448 > Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> Explanation makes sense, only Google hits seem to be on Mesa detecting the feature and later, removing the detection when unused. Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Regards, Joonas
On Tue, Feb 14, 2017 at 08:17:51PM -0800, Kenneth Graunke wrote: > This patch makes the I915_PARAM_HAS_EXEC_CONSTANTS getparam return 0 > (indicating the optional feature is not supported), and makes execbuf > always return -EINVAL if the flags are used. > > Apparently, no userspace ever shipped which used this optional feature: > I checked the git history of Mesa, xf86-video-intel, libva, and Beignet, > and there were zero commits showing a use of these flags. Kernel commit > 72bfa19c8deb4 apparently introduced the feature prematurely. It was actually cairo where we had patches to use it first. And we then realised the use was broken for gen6. > 'relative_constants_mode' has always been tracked per-device, but this > has actually been wrong ever since hardware contexts were introduced, as > the INSTPM register is saved (and automatically restored) as part of the > render ring context. The software per-device value could therefore get > out of sync with the hardware per-context value. This meant that using > them is actually unsafe: a client which tried to use them could damage > the state of other clients, causing the GPU to interpret their BO > offsets as absolute pointers, leading to bogus memory reads. > > These flags were also never ported to execlist mode, making them no-ops > on Gen9+ (which requires execlists), and Gen8 in the default mode. > > On Gen8+, userspace can write these registers directly, achieving the > same effect. On Gen6-7.5, it likely makes sense to extend the command > parser to support them. I don't think anyone wants this on Gen4-5. > > Based on a patch by Dave Gordon. > > Cc: Dave Gordon <david.s.gordon@intel.com> > Cc: Chris Wilson <chris@chris-wilson.co.uk> > Cc: stable@vger.kernel.org > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92448 > Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> > --- > drivers/gpu/drm/i915/i915_drv.c | 2 +- > drivers/gpu/drm/i915/i915_drv.h | 2 -- > drivers/gpu/drm/i915/i915_gem.c | 2 -- > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 50 ++---------------------------- > 4 files changed, 3 insertions(+), 53 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c > index 309c29c84c54..1a53a4bb09c8 100644 > --- a/drivers/gpu/drm/i915/i915_drv.c > +++ b/drivers/gpu/drm/i915/i915_drv.c > @@ -276,7 +276,7 @@ static int i915_getparam(struct drm_device *dev, void *data, > value = !!dev_priv->engine[VCS2]; > break; > case I915_PARAM_HAS_EXEC_CONSTANTS: > - value = INTEL_GEN(dev_priv) >= 4; > + value = 0; The style for obsolete param is to return -ENODEV. If that makes sense we can move this to the start of this block with the other dead params. Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> -Chris
On Wednesday, February 15, 2017 12:12:50 AM PST Chris Wilson wrote: > On Tue, Feb 14, 2017 at 08:17:51PM -0800, Kenneth Graunke wrote: > > This patch makes the I915_PARAM_HAS_EXEC_CONSTANTS getparam return 0 > > (indicating the optional feature is not supported), and makes execbuf > > always return -EINVAL if the flags are used. > > > > Apparently, no userspace ever shipped which used this optional feature: > > I checked the git history of Mesa, xf86-video-intel, libva, and Beignet, > > and there were zero commits showing a use of these flags. Kernel commit > > 72bfa19c8deb4 apparently introduced the feature prematurely. > > It was actually cairo where we had patches to use it first. And we then > realised the use was broken for gen6. Ahhh, that makes sense. I knew I was forgetting about some userspace. Thanks for reminding me about cairo-drm. > > 'relative_constants_mode' has always been tracked per-device, but this > > has actually been wrong ever since hardware contexts were introduced, as > > the INSTPM register is saved (and automatically restored) as part of the > > render ring context. The software per-device value could therefore get > > out of sync with the hardware per-context value. This meant that using > > them is actually unsafe: a client which tried to use them could damage > > the state of other clients, causing the GPU to interpret their BO > > offsets as absolute pointers, leading to bogus memory reads. > > > > These flags were also never ported to execlist mode, making them no-ops > > on Gen9+ (which requires execlists), and Gen8 in the default mode. > > > > On Gen8+, userspace can write these registers directly, achieving the > > same effect. On Gen6-7.5, it likely makes sense to extend the command > > parser to support them. I don't think anyone wants this on Gen4-5. > > > > Based on a patch by Dave Gordon. > > > > Cc: Dave Gordon <david.s.gordon@intel.com> > > Cc: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: stable@vger.kernel.org > > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92448 > > Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> > > --- > > drivers/gpu/drm/i915/i915_drv.c | 2 +- > > drivers/gpu/drm/i915/i915_drv.h | 2 -- > > drivers/gpu/drm/i915/i915_gem.c | 2 -- > > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 50 ++---------------------------- > > 4 files changed, 3 insertions(+), 53 deletions(-) > > > > diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c > > index 309c29c84c54..1a53a4bb09c8 100644 > > --- a/drivers/gpu/drm/i915/i915_drv.c > > +++ b/drivers/gpu/drm/i915/i915_drv.c > > @@ -276,7 +276,7 @@ static int i915_getparam(struct drm_device *dev, void *data, > > value = !!dev_priv->engine[VCS2]; > > break; > > case I915_PARAM_HAS_EXEC_CONSTANTS: > > - value = INTEL_GEN(dev_priv) >= 4; > > + value = 0; > > The style for obsolete param is to return -ENODEV. If that makes sense > we can move this to the start of this block with the other dead params. > > Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk> > -Chris Thanks, I missed that! I'll change this and send a v3.
diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 309c29c84c54..1a53a4bb09c8 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -276,7 +276,7 @@ static int i915_getparam(struct drm_device *dev, void *data, value = !!dev_priv->engine[VCS2]; break; case I915_PARAM_HAS_EXEC_CONSTANTS: - value = INTEL_GEN(dev_priv) >= 4; + value = 0; break; case I915_PARAM_HAS_LLC: value = HAS_LLC(dev_priv); diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index 918118a268b8..8a6ba63d3f74 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -2077,8 +2077,6 @@ struct drm_i915_private { const struct intel_device_info info; - int relative_constants_mode; - void __iomem *regs; struct intel_uncore uncore; diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c index 71297920fdf4..c9be0285c7cf 100644 --- a/drivers/gpu/drm/i915/i915_gem.c +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -4656,8 +4656,6 @@ i915_gem_load_init(struct drm_i915_private *dev_priv) init_waitqueue_head(&dev_priv->gpu_error.wait_queue); init_waitqueue_head(&dev_priv->gpu_error.reset_queue); - dev_priv->relative_constants_mode = I915_EXEC_CONSTANTS_REL_GENERAL; - init_waitqueue_head(&dev_priv->pending_flip_queue); dev_priv->mm.interruptible = true; diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c index da0846fe2ad6..febb067903e9 100644 --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c @@ -1412,10 +1412,7 @@ execbuf_submit(struct i915_execbuffer_params *params, struct drm_i915_gem_execbuffer2 *args, struct list_head *vmas) { - struct drm_i915_private *dev_priv = params->request->i915; u64 exec_start, exec_len; - int instp_mode; - u32 instp_mask, *cs; int ret; ret = i915_gem_execbuffer_move_to_gpu(params->request, vmas); @@ -1426,54 +1423,11 @@ execbuf_submit(struct i915_execbuffer_params *params, if (ret) return ret; - instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; - instp_mask = I915_EXEC_CONSTANTS_MASK; - switch (instp_mode) { - case I915_EXEC_CONSTANTS_REL_GENERAL: - case I915_EXEC_CONSTANTS_ABSOLUTE: - case I915_EXEC_CONSTANTS_REL_SURFACE: - if (instp_mode != 0 && params->engine->id != RCS) { - DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); - return -EINVAL; - } - - if (instp_mode != dev_priv->relative_constants_mode) { - if (INTEL_INFO(dev_priv)->gen < 4) { - DRM_DEBUG("no rel constants on pre-gen4\n"); - return -EINVAL; - } - - if (INTEL_INFO(dev_priv)->gen > 5 && - instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { - DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); - return -EINVAL; - } - - /* The HW changed the meaning on this bit on gen6 */ - if (INTEL_INFO(dev_priv)->gen >= 6) - instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; - } - break; - default: - DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); + if (args->flags & I915_EXEC_CONSTANTS_MASK) { + DRM_DEBUG("I915_EXEC_CONSTANTS_* unsupported\n"); return -EINVAL; } - if (params->engine->id == RCS && - instp_mode != dev_priv->relative_constants_mode) { - cs = intel_ring_begin(params->request, 4); - if (IS_ERR(cs)) - return PTR_ERR(cs); - - *cs++ = MI_NOOP; - *cs++ = MI_LOAD_REGISTER_IMM(1); - *cs++ = i915_mmio_reg_offset(INSTPM); - *cs++ = instp_mask << 16 | instp_mode; - intel_ring_advance(params->request, cs); - - dev_priv->relative_constants_mode = instp_mode; - } - if (args->flags & I915_EXEC_GEN7_SOL_RESET) { ret = i915_reset_gen7_sol_offsets(params->request); if (ret)
This patch makes the I915_PARAM_HAS_EXEC_CONSTANTS getparam return 0 (indicating the optional feature is not supported), and makes execbuf always return -EINVAL if the flags are used. Apparently, no userspace ever shipped which used this optional feature: I checked the git history of Mesa, xf86-video-intel, libva, and Beignet, and there were zero commits showing a use of these flags. Kernel commit 72bfa19c8deb4 apparently introduced the feature prematurely. 'relative_constants_mode' has always been tracked per-device, but this has actually been wrong ever since hardware contexts were introduced, as the INSTPM register is saved (and automatically restored) as part of the render ring context. The software per-device value could therefore get out of sync with the hardware per-context value. This meant that using them is actually unsafe: a client which tried to use them could damage the state of other clients, causing the GPU to interpret their BO offsets as absolute pointers, leading to bogus memory reads. These flags were also never ported to execlist mode, making them no-ops on Gen9+ (which requires execlists), and Gen8 in the default mode. On Gen8+, userspace can write these registers directly, achieving the same effect. On Gen6-7.5, it likely makes sense to extend the command parser to support them. I don't think anyone wants this on Gen4-5. Based on a patch by Dave Gordon. Cc: Dave Gordon <david.s.gordon@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Cc: stable@vger.kernel.org Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92448 Signed-off-by: Kenneth Graunke <kenneth@whitecape.org> --- drivers/gpu/drm/i915/i915_drv.c | 2 +- drivers/gpu/drm/i915/i915_drv.h | 2 -- drivers/gpu/drm/i915/i915_gem.c | 2 -- drivers/gpu/drm/i915/i915_gem_execbuffer.c | 50 ++---------------------------- 4 files changed, 3 insertions(+), 53 deletions(-)