| Message ID | 1348727406-4191-1-git-send-email-sw0312.kim@samsung.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
fwiw, I had a similar patch: https://patchwork.kernel.org/patch/1229161/ although it was on top of some locking fixes from Daniel (which I think are also needed): https://patchwork.kernel.org/patch/1227251/ although that seemed to cause/trigger some explosions which I think still need to be debugged.. BR, -R On Thu, Sep 27, 2012 at 8:30 AM, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: > Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem > makes memory leak. release function of dma-buf cannot be called because f_count > of dma-buf increased by importing gem and gem ref count cannot be decrease > because of exported dma-buf. > > So I add dma_buf_put() for imported gem come from its own gem into each drivers > having prime_import and prime_export capabilities. With this, only gem ref > count is increased if importing gem exported from gem of same driver. > > Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com> > Signed-off-by: Kyungmin.park <kyungmin.park@samsung.com> > Cc: Inki Dae <inki.dae@samsung.com> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > Cc: Rob Clark <rob.clark@linaro.org> > Cc: Alex Deucher <alexander.deucher@amd.com> > --- > drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 5 +++++ > drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 +++++ > drivers/gpu/drm/nouveau/nouveau_prime.c | 1 + > drivers/gpu/drm/radeon/radeon_prime.c | 1 + > drivers/staging/omapdrm/omap_gem_dmabuf.c | 5 +++++ > 5 files changed, 17 insertions(+), 0 deletions(-) > > diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > index ae13feb..b0897c9 100644 > --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, > > /* is it from our device? */ > if (obj->dev == drm_dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(obj); > + dma_buf_put(dma_buf); > return obj; > } > } > diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c > index aa308e1..32e6287 100644 > --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c > +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c > @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, > obj = dma_buf->priv; > /* is it from our device? */ > if (obj->base.dev == dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(&obj->base); > + dma_buf_put(dma_buf); > return &obj->base; > } > } > diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c > index a25cf2c..bb653c6 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_prime.c > +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c > @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, > if (nvbo->gem) { > if (nvbo->gem->dev == dev) { > drm_gem_object_reference(nvbo->gem); > + dma_buf_put(dma_buf); > return nvbo->gem; > } > } > diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c > index 6bef46a..d344a3be 100644 > --- a/drivers/gpu/drm/radeon/radeon_prime.c > +++ b/drivers/gpu/drm/radeon/radeon_prime.c > @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, > bo = dma_buf->priv; > if (bo->gem_base.dev == dev) { > drm_gem_object_reference(&bo->gem_base); > + dma_buf_put(dma_buf); > return &bo->gem_base; > } > } > diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c > index 42728e0..5b50eb6 100644 > --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c > +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c > @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, > obj = buffer->priv; > /* is it from our device? */ > if (obj->dev == dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(obj); > + dma_buf_put(buffer); > return obj; > } > } > -- > 1.7.4.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel
Hi Rob, On 2012? 09? 27? 15:52, Rob Clark wrote: > fwiw, I had a similar patch: > > https://patchwork.kernel.org/patch/1229161/ Yes, I already check your patch and even my patch's title is a bit from your patch. I thought locking issue blocks your patch, so I sent simple fixes on current state. How do you think merging bug-fixes at first? Best Regards, - Seung-Woo Kim > > although it was on top of some locking fixes from Daniel (which I > think are also needed): > > https://patchwork.kernel.org/patch/1227251/ > > although that seemed to cause/trigger some explosions which I think > still need to be debugged.. > > BR, > -R > > On Thu, Sep 27, 2012 at 8:30 AM, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: >> Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem >> makes memory leak. release function of dma-buf cannot be called because f_count >> of dma-buf increased by importing gem and gem ref count cannot be decrease >> because of exported dma-buf. >> >> So I add dma_buf_put() for imported gem come from its own gem into each drivers >> having prime_import and prime_export capabilities. With this, only gem ref >> count is increased if importing gem exported from gem of same driver. >> >> Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com> >> Signed-off-by: Kyungmin.park <kyungmin.park@samsung.com> >> Cc: Inki Dae <inki.dae@samsung.com> >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> >> Cc: Rob Clark <rob.clark@linaro.org> >> Cc: Alex Deucher <alexander.deucher@amd.com> >> --- >> drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 5 +++++ >> drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 +++++ >> drivers/gpu/drm/nouveau/nouveau_prime.c | 1 + >> drivers/gpu/drm/radeon/radeon_prime.c | 1 + >> drivers/staging/omapdrm/omap_gem_dmabuf.c | 5 +++++ >> 5 files changed, 17 insertions(+), 0 deletions(-) >> >> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> index ae13feb..b0897c9 100644 >> --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, >> >> /* is it from our device? */ >> if (obj->dev == drm_dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(obj); >> + dma_buf_put(dma_buf); >> return obj; >> } >> } >> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> index aa308e1..32e6287 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, >> obj = dma_buf->priv; >> /* is it from our device? */ >> if (obj->base.dev == dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(&obj->base); >> + dma_buf_put(dma_buf); >> return &obj->base; >> } >> } >> diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c >> index a25cf2c..bb653c6 100644 >> --- a/drivers/gpu/drm/nouveau/nouveau_prime.c >> +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c >> @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, >> if (nvbo->gem) { >> if (nvbo->gem->dev == dev) { >> drm_gem_object_reference(nvbo->gem); >> + dma_buf_put(dma_buf); >> return nvbo->gem; >> } >> } >> diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c >> index 6bef46a..d344a3be 100644 >> --- a/drivers/gpu/drm/radeon/radeon_prime.c >> +++ b/drivers/gpu/drm/radeon/radeon_prime.c >> @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, >> bo = dma_buf->priv; >> if (bo->gem_base.dev == dev) { >> drm_gem_object_reference(&bo->gem_base); >> + dma_buf_put(dma_buf); >> return &bo->gem_base; >> } >> } >> diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c >> index 42728e0..5b50eb6 100644 >> --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c >> +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c >> @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, >> obj = buffer->priv; >> /* is it from our device? */ >> if (obj->dev == dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(obj); >> + dma_buf_put(buffer); >> return obj; >> } >> } >> -- >> 1.7.4.1 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> http://lists.freedesktop.org/mailman/listinfo/dri-devel >
On Thu, Sep 27, 2012 at 9:14 AM, ??? <sw0312.kim@samsung.com> wrote: > Hi Rob, > > On 2012? 09? 27? 15:52, Rob Clark wrote: >> fwiw, I had a similar patch: >> >> https://patchwork.kernel.org/patch/1229161/ > > Yes, I already check your patch and even my patch's title is a bit from > your patch. > > I thought locking issue blocks your patch, so I sent simple fixes on > current state. How do you think merging bug-fixes at first? well, the lack of locking is a real problem too.. although I didn't see quite what the locking changes could have to do with the crash that Dave had seen. But I don't know if anyone has looked at it more, and I only re-import my own buffers in omapdrm so I don't think I can trigger the same scenario myself. I guess if no one else has a chance to take at least a quick look at the locking crash, maybe we should take this patch. But I would be interested get the locking patch too. We are needing both patches on our product kernels to fix some issues. BR, -R > Best Regards, > - Seung-Woo Kim > >> >> although it was on top of some locking fixes from Daniel (which I >> think are also needed): >> >> https://patchwork.kernel.org/patch/1227251/ >> >> although that seemed to cause/trigger some explosions which I think >> still need to be debugged.. >> >> BR, >> -R >> >> On Thu, Sep 27, 2012 at 8:30 AM, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: >>> Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem >>> makes memory leak. release function of dma-buf cannot be called because f_count >>> of dma-buf increased by importing gem and gem ref count cannot be decrease >>> because of exported dma-buf. >>> >>> So I add dma_buf_put() for imported gem come from its own gem into each drivers >>> having prime_import and prime_export capabilities. With this, only gem ref >>> count is increased if importing gem exported from gem of same driver. >>> >>> Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com> >>> Signed-off-by: Kyungmin.park <kyungmin.park@samsung.com> >>> Cc: Inki Dae <inki.dae@samsung.com> >>> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> >>> Cc: Rob Clark <rob.clark@linaro.org> >>> Cc: Alex Deucher <alexander.deucher@amd.com> >>> --- >>> drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 5 +++++ >>> drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 +++++ >>> drivers/gpu/drm/nouveau/nouveau_prime.c | 1 + >>> drivers/gpu/drm/radeon/radeon_prime.c | 1 + >>> drivers/staging/omapdrm/omap_gem_dmabuf.c | 5 +++++ >>> 5 files changed, 17 insertions(+), 0 deletions(-) >>> >>> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >>> index ae13feb..b0897c9 100644 >>> --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >>> +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >>> @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, >>> >>> /* is it from our device? */ >>> if (obj->dev == drm_dev) { >>> + /* >>> + * Importing dmabuf exported from out own gem increases >>> + * refcount on gem itself instead of f_count of dmabuf. >>> + */ >>> drm_gem_object_reference(obj); >>> + dma_buf_put(dma_buf); >>> return obj; >>> } >>> } >>> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >>> index aa308e1..32e6287 100644 >>> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c >>> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >>> @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, >>> obj = dma_buf->priv; >>> /* is it from our device? */ >>> if (obj->base.dev == dev) { >>> + /* >>> + * Importing dmabuf exported from out own gem increases >>> + * refcount on gem itself instead of f_count of dmabuf. >>> + */ >>> drm_gem_object_reference(&obj->base); >>> + dma_buf_put(dma_buf); >>> return &obj->base; >>> } >>> } >>> diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c >>> index a25cf2c..bb653c6 100644 >>> --- a/drivers/gpu/drm/nouveau/nouveau_prime.c >>> +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c >>> @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, >>> if (nvbo->gem) { >>> if (nvbo->gem->dev == dev) { >>> drm_gem_object_reference(nvbo->gem); >>> + dma_buf_put(dma_buf); >>> return nvbo->gem; >>> } >>> } >>> diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c >>> index 6bef46a..d344a3be 100644 >>> --- a/drivers/gpu/drm/radeon/radeon_prime.c >>> +++ b/drivers/gpu/drm/radeon/radeon_prime.c >>> @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, >>> bo = dma_buf->priv; >>> if (bo->gem_base.dev == dev) { >>> drm_gem_object_reference(&bo->gem_base); >>> + dma_buf_put(dma_buf); >>> return &bo->gem_base; >>> } >>> } >>> diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c >>> index 42728e0..5b50eb6 100644 >>> --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c >>> +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c >>> @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, >>> obj = buffer->priv; >>> /* is it from our device? */ >>> if (obj->dev == dev) { >>> + /* >>> + * Importing dmabuf exported from out own gem increases >>> + * refcount on gem itself instead of f_count of dmabuf. >>> + */ >>> drm_gem_object_reference(obj); >>> + dma_buf_put(buffer); >>> return obj; >>> } >>> } >>> -- >>> 1.7.4.1 >>> >>> _______________________________________________ >>> dri-devel mailing list >>> dri-devel@lists.freedesktop.org >>> http://lists.freedesktop.org/mailman/listinfo/dri-devel >> > > -- > Seung-Woo Kim > Samsung Software R&D Center > -- > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel
On Thu, 27 Sep 2012, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: > Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem > makes memory leak. release function of dma-buf cannot be called because f_count > of dma-buf increased by importing gem and gem ref count cannot be decrease > because of exported dma-buf. > > So I add dma_buf_put() for imported gem come from its own gem into each drivers > having prime_import and prime_export capabilities. With this, only gem ref > count is increased if importing gem exported from gem of same driver. There's a reference leak bug [1] related to prime self import found by intel-gpu-tools tests. Without much thinking, I just applied this patch to see if it makes a difference. Instead, it now fails with: prime_self_import: prime_self_import.c:61: check_bo: Assertion `ptr1' failed. The assert is at [2]. I haven't looked into why this happens at all, so I'm just sharing this in case you guys find it helpful. BR, Jani. [1] https://bugs.freedesktop.org/show_bug.cgi?id=54111 [2] http://cgit.freedesktop.org/xorg/app/intel-gpu-tools/tree/tests/prime_self_import.c#n61 > > Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com> > Signed-off-by: Kyungmin.park <kyungmin.park@samsung.com> > Cc: Inki Dae <inki.dae@samsung.com> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > Cc: Rob Clark <rob.clark@linaro.org> > Cc: Alex Deucher <alexander.deucher@amd.com> > --- > drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 5 +++++ > drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 +++++ > drivers/gpu/drm/nouveau/nouveau_prime.c | 1 + > drivers/gpu/drm/radeon/radeon_prime.c | 1 + > drivers/staging/omapdrm/omap_gem_dmabuf.c | 5 +++++ > 5 files changed, 17 insertions(+), 0 deletions(-) > > diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > index ae13feb..b0897c9 100644 > --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c > @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, > > /* is it from our device? */ > if (obj->dev == drm_dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(obj); > + dma_buf_put(dma_buf); > return obj; > } > } > diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c > index aa308e1..32e6287 100644 > --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c > +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c > @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, > obj = dma_buf->priv; > /* is it from our device? */ > if (obj->base.dev == dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(&obj->base); > + dma_buf_put(dma_buf); > return &obj->base; > } > } > diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c > index a25cf2c..bb653c6 100644 > --- a/drivers/gpu/drm/nouveau/nouveau_prime.c > +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c > @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, > if (nvbo->gem) { > if (nvbo->gem->dev == dev) { > drm_gem_object_reference(nvbo->gem); > + dma_buf_put(dma_buf); > return nvbo->gem; > } > } > diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c > index 6bef46a..d344a3be 100644 > --- a/drivers/gpu/drm/radeon/radeon_prime.c > +++ b/drivers/gpu/drm/radeon/radeon_prime.c > @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, > bo = dma_buf->priv; > if (bo->gem_base.dev == dev) { > drm_gem_object_reference(&bo->gem_base); > + dma_buf_put(dma_buf); > return &bo->gem_base; > } > } > diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c > index 42728e0..5b50eb6 100644 > --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c > +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c > @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, > obj = buffer->priv; > /* is it from our device? */ > if (obj->dev == dev) { > + /* > + * Importing dmabuf exported from out own gem increases > + * refcount on gem itself instead of f_count of dmabuf. > + */ > drm_gem_object_reference(obj); > + dma_buf_put(buffer); > return obj; > } > } > -- > 1.7.4.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/dri-devel
Hi Jani, Sorry for late reply. On 2012? 09? 27? 22:43, Jani Nikula wrote: > On Thu, 27 Sep 2012, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: >> Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem >> makes memory leak. release function of dma-buf cannot be called because f_count >> of dma-buf increased by importing gem and gem ref count cannot be decrease >> because of exported dma-buf. >> >> So I add dma_buf_put() for imported gem come from its own gem into each drivers >> having prime_import and prime_export capabilities. With this, only gem ref >> count is increased if importing gem exported from gem of same driver. > > There's a reference leak bug [1] related to prime self import found by > intel-gpu-tools tests. Without much thinking, I just applied this patch > to see if it makes a difference. Instead, it now fails with: > > prime_self_import: prime_self_import.c:61: check_bo: Assertion `ptr1' > failed. > > The assert is at [2]. I think it was possible to re-use a handle imported_buf list which is already deleted, and it was fixed after Dave's commit "drm/prime: add exported buffers to current fprivs imported buffer list (v2)". > > I haven't looked into why this happens at all, so I'm just sharing this > in case you guys find it helpful. If you still have same assert, I think it will be very helpful to let me know exactly where the assert occurs from 5 check_bo() in test application in your link. Thanks and Best Regards, - Seung-Woo Kim > > BR, > Jani. > > > > [1] https://bugs.freedesktop.org/show_bug.cgi?id=54111 > [2] http://cgit.freedesktop.org/xorg/app/intel-gpu-tools/tree/tests/prime_self_import.c#n61 > > >> >> Signed-off-by: Seung-Woo Kim <sw0312.kim@samsung.com> >> Signed-off-by: Kyungmin.park <kyungmin.park@samsung.com> >> Cc: Inki Dae <inki.dae@samsung.com> >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> >> Cc: Rob Clark <rob.clark@linaro.org> >> Cc: Alex Deucher <alexander.deucher@amd.com> >> --- >> drivers/gpu/drm/exynos/exynos_drm_dmabuf.c | 5 +++++ >> drivers/gpu/drm/i915/i915_gem_dmabuf.c | 5 +++++ >> drivers/gpu/drm/nouveau/nouveau_prime.c | 1 + >> drivers/gpu/drm/radeon/radeon_prime.c | 1 + >> drivers/staging/omapdrm/omap_gem_dmabuf.c | 5 +++++ >> 5 files changed, 17 insertions(+), 0 deletions(-) >> >> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> index ae13feb..b0897c9 100644 >> --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c >> @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, >> >> /* is it from our device? */ >> if (obj->dev == drm_dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(obj); >> + dma_buf_put(dma_buf); >> return obj; >> } >> } >> diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> index aa308e1..32e6287 100644 >> --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c >> @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, >> obj = dma_buf->priv; >> /* is it from our device? */ >> if (obj->base.dev == dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(&obj->base); >> + dma_buf_put(dma_buf); >> return &obj->base; >> } >> } >> diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c >> index a25cf2c..bb653c6 100644 >> --- a/drivers/gpu/drm/nouveau/nouveau_prime.c >> +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c >> @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, >> if (nvbo->gem) { >> if (nvbo->gem->dev == dev) { >> drm_gem_object_reference(nvbo->gem); >> + dma_buf_put(dma_buf); >> return nvbo->gem; >> } >> } >> diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c >> index 6bef46a..d344a3be 100644 >> --- a/drivers/gpu/drm/radeon/radeon_prime.c >> +++ b/drivers/gpu/drm/radeon/radeon_prime.c >> @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, >> bo = dma_buf->priv; >> if (bo->gem_base.dev == dev) { >> drm_gem_object_reference(&bo->gem_base); >> + dma_buf_put(dma_buf); >> return &bo->gem_base; >> } >> } >> diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c >> index 42728e0..5b50eb6 100644 >> --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c >> +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c >> @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, >> obj = buffer->priv; >> /* is it from our device? */ >> if (obj->dev == dev) { >> + /* >> + * Importing dmabuf exported from out own gem increases >> + * refcount on gem itself instead of f_count of dmabuf. >> + */ >> drm_gem_object_reference(obj); >> + dma_buf_put(buffer); >> return obj; >> } >> } >> -- >> 1.7.4.1 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> http://lists.freedesktop.org/mailman/listinfo/dri-devel >
On Thu, Sep 27, 2012 at 4:30 PM, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: > Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem > makes memory leak. release function of dma-buf cannot be called because f_count > of dma-buf increased by importing gem and gem ref count cannot be decrease > because of exported dma-buf. > > So I add dma_buf_put() for imported gem come from its own gem into each drivers > having prime_import and prime_export capabilities. With this, only gem ref > count is increased if importing gem exported from gem of same driver. Okay its taken me a while to circle around and get back to this, but yes I admit this is needed, but I hate implementing it like this But I think I'll push it and work out a cleaner solution, I should also go back and look at the older patches. Dave.
Hi Dave, On 2012? 12? 18? 15:30, Dave Airlie wrote: > On Thu, Sep 27, 2012 at 4:30 PM, Seung-Woo Kim <sw0312.kim@samsung.com> wrote: >> Increasing ref counts of both dma-buf and gem for imported dma-buf come from gem >> makes memory leak. release function of dma-buf cannot be called because f_count >> of dma-buf increased by importing gem and gem ref count cannot be decrease >> because of exported dma-buf. >> >> So I add dma_buf_put() for imported gem come from its own gem into each drivers >> having prime_import and prime_export capabilities. With this, only gem ref >> count is increased if importing gem exported from gem of same driver. > > Okay its taken me a while to circle around and get back to this, but > yes I admit this is needed, but I hate implementing it like this > > But I think I'll push it and work out a cleaner solution, I should > also go back and look at the older patches. I want to also report some strange thing in dma-buf prime export. In drm_prime_handle_to_fd_ioctl(), flags is cleared to only support DRM_CLOEXEC but in gem_prime_export() callbacks of each driver, it uses 0600 as flags for dma_buf_export() like following. return dma_buf_export(obj, &i915_dmabuf_ops, obj->base.size, 0600); Best Regards, - Seung-Woo Kim > > Dave. >
> In drm_prime_handle_to_fd_ioctl(), flags is cleared to only support > DRM_CLOEXEC but in gem_prime_export() callbacks of each driver, it uses > 0600 as flags for dma_buf_export() like following. > > return dma_buf_export(obj, &i915_dmabuf_ops, obj->base.size, 0600); Oops, nice catch, radeon and nouveau did the correct thing here, I'll send a patch to fix that. Dave.
diff --git a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c index ae13feb..b0897c9 100644 --- a/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c +++ b/drivers/gpu/drm/exynos/exynos_drm_dmabuf.c @@ -211,7 +211,12 @@ struct drm_gem_object *exynos_dmabuf_prime_import(struct drm_device *drm_dev, /* is it from our device? */ if (obj->dev == drm_dev) { + /* + * Importing dmabuf exported from out own gem increases + * refcount on gem itself instead of f_count of dmabuf. + */ drm_gem_object_reference(obj); + dma_buf_put(dma_buf); return obj; } } diff --git a/drivers/gpu/drm/i915/i915_gem_dmabuf.c b/drivers/gpu/drm/i915/i915_gem_dmabuf.c index aa308e1..32e6287 100644 --- a/drivers/gpu/drm/i915/i915_gem_dmabuf.c +++ b/drivers/gpu/drm/i915/i915_gem_dmabuf.c @@ -188,7 +188,12 @@ struct drm_gem_object *i915_gem_prime_import(struct drm_device *dev, obj = dma_buf->priv; /* is it from our device? */ if (obj->base.dev == dev) { + /* + * Importing dmabuf exported from out own gem increases + * refcount on gem itself instead of f_count of dmabuf. + */ drm_gem_object_reference(&obj->base); + dma_buf_put(dma_buf); return &obj->base; } } diff --git a/drivers/gpu/drm/nouveau/nouveau_prime.c b/drivers/gpu/drm/nouveau/nouveau_prime.c index a25cf2c..bb653c6 100644 --- a/drivers/gpu/drm/nouveau/nouveau_prime.c +++ b/drivers/gpu/drm/nouveau/nouveau_prime.c @@ -199,6 +199,7 @@ struct drm_gem_object *nouveau_gem_prime_import(struct drm_device *dev, if (nvbo->gem) { if (nvbo->gem->dev == dev) { drm_gem_object_reference(nvbo->gem); + dma_buf_put(dma_buf); return nvbo->gem; } } diff --git a/drivers/gpu/drm/radeon/radeon_prime.c b/drivers/gpu/drm/radeon/radeon_prime.c index 6bef46a..d344a3be 100644 --- a/drivers/gpu/drm/radeon/radeon_prime.c +++ b/drivers/gpu/drm/radeon/radeon_prime.c @@ -195,6 +195,7 @@ struct drm_gem_object *radeon_gem_prime_import(struct drm_device *dev, bo = dma_buf->priv; if (bo->gem_base.dev == dev) { drm_gem_object_reference(&bo->gem_base); + dma_buf_put(dma_buf); return &bo->gem_base; } } diff --git a/drivers/staging/omapdrm/omap_gem_dmabuf.c b/drivers/staging/omapdrm/omap_gem_dmabuf.c index 42728e0..5b50eb6 100644 --- a/drivers/staging/omapdrm/omap_gem_dmabuf.c +++ b/drivers/staging/omapdrm/omap_gem_dmabuf.c @@ -207,7 +207,12 @@ struct drm_gem_object * omap_gem_prime_import(struct drm_device *dev, obj = buffer->priv; /* is it from our device? */ if (obj->dev == dev) { + /* + * Importing dmabuf exported from out own gem increases + * refcount on gem itself instead of f_count of dmabuf. + */ drm_gem_object_reference(obj); + dma_buf_put(buffer); return obj; } }