| Message ID | 1495521583-29151-1-git-send-email-jeffy.chen@rock-chips.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, May 23, 2017 at 02:39:43PM +0800, Jeffy Chen wrote: > The system would crash when trying to alloc zero sized gem buffer: > [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR > ... > [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec It's unfortunate that you didn't include the entire stack trace. From code inspection, it seems like the 0 size comes from the fb_probe path? Is there somewhere in the helpers that you could check the mode is sane so all drivers can benefit? Sean > > Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> > --- > > drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > index df9e570..8917922 100644 > --- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > +++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > @@ -315,6 +315,11 @@ struct rockchip_gem_object * > struct drm_gem_object *obj; > int ret; > > + if (!size) { > + DRM_ERROR("gem buffer size is zero\n"); > + return ERR_PTR(-EINVAL); > + } > + > size = round_up(size, PAGE_SIZE); > > rk_obj = kzalloc(sizeof(*rk_obj), GFP_KERNEL); > -- > 2.1.4 >
Hi sean, On 05/25/2017 11:30 PM, Sean Paul wrote: > On Tue, May 23, 2017 at 02:39:43PM +0800, Jeffy Chen wrote: >> The system would crash when trying to alloc zero sized gem buffer: >> [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR >> ... >> [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec > > It's unfortunate that you didn't include the entire stack trace. From code > inspection, it seems like the 0 size comes from the fb_probe path? Is there > somewhere in the helpers that you could check the mode is sane so all drivers > can benefit? hmm, sorry, i was testing it on chromeos 4.4 kernel, it turns out that we have a custom ioctl for userspace to create gem buffer(the same as exynos drm), which might get the the 0 size. but on upstream kernel, it could only be called by dump_create, and the drm_mode_create_dumb_ioctl already did the size check. will resent this patch, and rewrite the commit message, thanx. > > Sean > >> >> Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> >> --- >> >> drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 5 +++++ >> 1 file changed, 5 insertions(+) >> >> diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c >> index df9e570..8917922 100644 >> --- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c >> +++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c >> @@ -315,6 +315,11 @@ struct rockchip_gem_object * >> struct drm_gem_object *obj; >> int ret; >> >> + if (!size) { >> + DRM_ERROR("gem buffer size is zero\n"); >> + return ERR_PTR(-EINVAL); >> + } >> + >> size = round_up(size, PAGE_SIZE); >> >> rk_obj = kzalloc(sizeof(*rk_obj), GFP_KERNEL); >> -- >> 2.1.4 >> >
On Fri, May 26, 2017 at 10:30:09AM +0800, jeffy wrote: > Hi sean, > > On 05/25/2017 11:30 PM, Sean Paul wrote: > > On Tue, May 23, 2017 at 02:39:43PM +0800, Jeffy Chen wrote: > > > The system would crash when trying to alloc zero sized gem buffer: > > > [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR > > > ... > > > [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec > > > > It's unfortunate that you didn't include the entire stack trace. From code > > inspection, it seems like the 0 size comes from the fb_probe path? Is there > > somewhere in the helpers that you could check the mode is sane so all drivers > > can benefit? > > hmm, sorry, i was testing it on chromeos 4.4 kernel, it turns out that we > have a custom ioctl for userspace to create gem buffer(the same as exynos > drm), which might get the the 0 size. > > but on upstream kernel, it could only be called by dump_create, and the > drm_mode_create_dumb_ioctl already did the size check. > > will resent this patch, and rewrite the commit message, thanx. That suggests that this patch isn't needed at all.
On Fri, May 26, 2017 at 7:52 AM, Christoph Hellwig <hch@infradead.org> wrote: > On Fri, May 26, 2017 at 10:30:09AM +0800, jeffy wrote: >> Hi sean, >> >> On 05/25/2017 11:30 PM, Sean Paul wrote: >> > On Tue, May 23, 2017 at 02:39:43PM +0800, Jeffy Chen wrote: >> > > The system would crash when trying to alloc zero sized gem buffer: >> > > [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR >> > > ... >> > > [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec >> > >> > It's unfortunate that you didn't include the entire stack trace. From code >> > inspection, it seems like the 0 size comes from the fb_probe path? Is there >> > somewhere in the helpers that you could check the mode is sane so all drivers >> > can benefit? >> >> hmm, sorry, i was testing it on chromeos 4.4 kernel, it turns out that we >> have a custom ioctl for userspace to create gem buffer(the same as exynos >> drm), which might get the the 0 size. >> >> but on upstream kernel, it could only be called by dump_create, and the >> drm_mode_create_dumb_ioctl already did the size check. >> >> will resent this patch, and rewrite the commit message, thanx. > > That suggests that this patch isn't needed at all. Yes, not needed for upstream. But next time around pls include the entire backtrace (or at least the relevant parts), not just the last line, so that we can figure this out directly. Thanks, Daniel
On Fri, May 26, 2017 at 10:30:09AM +0800, jeffy wrote: > Hi sean, > > On 05/25/2017 11:30 PM, Sean Paul wrote: > > On Tue, May 23, 2017 at 02:39:43PM +0800, Jeffy Chen wrote: > > > The system would crash when trying to alloc zero sized gem buffer: > > > [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR > > > ... > > > [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec > > > > It's unfortunate that you didn't include the entire stack trace. From code > > inspection, it seems like the 0 size comes from the fb_probe path? Is there > > somewhere in the helpers that you could check the mode is sane so all drivers > > can benefit? > > hmm, sorry, i was testing it on chromeos 4.4 kernel, it turns out that we > have a custom ioctl for userspace to create gem buffer(the same as exynos > drm), which might get the the 0 size. > > but on upstream kernel, it could only be called by dump_create, and the > drm_mode_create_dumb_ioctl already did the size check. Ah, ok. In that case, fix the custom ioctl such that it ensures we never call this function with size == 0, and upload it downstream with a CHROMIUM prefix. Sean > > will resent this patch, and rewrite the commit message, thanx. > > > > > Sean > > > > > > > > Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> > > > --- > > > > > > drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 5 +++++ > > > 1 file changed, 5 insertions(+) > > > > > > diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > > > index df9e570..8917922 100644 > > > --- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > > > +++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c > > > @@ -315,6 +315,11 @@ struct rockchip_gem_object * > > > struct drm_gem_object *obj; > > > int ret; > > > > > > + if (!size) { > > > + DRM_ERROR("gem buffer size is zero\n"); > > > + return ERR_PTR(-EINVAL); > > > + } > > > + > > > size = round_up(size, PAGE_SIZE); > > > > > > rk_obj = kzalloc(sizeof(*rk_obj), GFP_KERNEL); > > > -- > > > 2.1.4 > > > > > >
diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c index df9e570..8917922 100644 --- a/drivers/gpu/drm/rockchip/rockchip_drm_gem.c +++ b/drivers/gpu/drm/rockchip/rockchip_drm_gem.c @@ -315,6 +315,11 @@ struct rockchip_gem_object * struct drm_gem_object *obj; int ret; + if (!size) { + DRM_ERROR("gem buffer size is zero\n"); + return ERR_PTR(-EINVAL); + } + size = round_up(size, PAGE_SIZE); rk_obj = kzalloc(sizeof(*rk_obj), GFP_KERNEL);
The system would crash when trying to alloc zero sized gem buffer: [ 6.712435] Unable to handle kernel NULL pointer dereference at virtual address 00000010 <--ZERO_SIZE_PTR ... [ 6.757502] PC is at sg_alloc_table_from_pages+0x170/0x1ec Signed-off-by: Jeffy Chen <jeffy.chen@rock-chips.com> --- drivers/gpu/drm/rockchip/rockchip_drm_gem.c | 5 +++++ 1 file changed, 5 insertions(+)