| Message ID | 20201117214029.591896-1-daniel.vetter@ffwll.ch (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | char/agp: Disable frontend without CONFIG_DRM_LEGACY | expand |
Hi Am 17.11.20 um 22:40 schrieb Daniel Vetter: > It's probably full of bugs ready for exploiting by userspace. And > there's not going to be any userspace for this without any of the drm > legacy drivers enabled too. So just couple it together. > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > Cc: David Airlie <airlied@linux.ie> > Cc: Adam Jackson <ajax@redhat.com> > --- > drivers/char/agp/Makefile | 6 +++++- > drivers/char/agp/agp.h | 5 +++++ > 2 files changed, 10 insertions(+), 1 deletion(-) > > diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile > index cb2497d157f6..90ed8c789e48 100644 > --- a/drivers/char/agp/Makefile > +++ b/drivers/char/agp/Makefile > @@ -1,7 +1,11 @@ > # SPDX-License-Identifier: GPL-2.0 > -agpgart-y := backend.o frontend.o generic.o isoch.o > +agpgart-y := backend.o generic.o isoch.o > > +ifeq ($(CONFIG_DRM_LEGACY),y) > agpgart-$(CONFIG_COMPAT) += compat_ioctl.o > +agpgart-y += frontend.o > +endif > + > > obj-$(CONFIG_AGP) += agpgart.o > obj-$(CONFIG_AGP_ALI) += ali-agp.o > diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h > index 4eb1c772ded7..bb09d64cd51e 100644 > --- a/drivers/char/agp/agp.h > +++ b/drivers/char/agp/agp.h > @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data *bridge); > void agp_remove_bridge(struct agp_bridge_data *bridge); > > /* Frontend routines. */ > +#if IS_ENABLED(CONFIG_DRM_LEGACY) > int agp_frontend_initialize(void); > void agp_frontend_cleanup(void); > +#else > +static inline int agp_frontend_initialize(void) { return 0; } > +static inline void agp_frontend_cleanup(void) {} > +#endif There's one non-legacy driver that uses these agp structures, which is radeon. Does this change affect radeon? Best regards Thomas > > /* Generic routines. */ > void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode); >
Am 18.11.20 um 09:02 schrieb Thomas Zimmermann: > Hi > > Am 17.11.20 um 22:40 schrieb Daniel Vetter: >> It's probably full of bugs ready for exploiting by userspace. And >> there's not going to be any userspace for this without any of the drm >> legacy drivers enabled too. So just couple it together. >> >> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> >> Cc: David Airlie <airlied@linux.ie> >> Cc: Adam Jackson <ajax@redhat.com> >> --- >> drivers/char/agp/Makefile | 6 +++++- >> drivers/char/agp/agp.h | 5 +++++ >> 2 files changed, 10 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile >> index cb2497d157f6..90ed8c789e48 100644 >> --- a/drivers/char/agp/Makefile >> +++ b/drivers/char/agp/Makefile >> @@ -1,7 +1,11 @@ >> # SPDX-License-Identifier: GPL-2.0 >> -agpgart-y := backend.o frontend.o generic.o isoch.o >> +agpgart-y := backend.o generic.o isoch.o >> >> +ifeq ($(CONFIG_DRM_LEGACY),y) >> agpgart-$(CONFIG_COMPAT) += compat_ioctl.o >> +agpgart-y += frontend.o >> +endif >> + >> >> obj-$(CONFIG_AGP) += agpgart.o >> obj-$(CONFIG_AGP_ALI) += ali-agp.o >> diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h >> index 4eb1c772ded7..bb09d64cd51e 100644 >> --- a/drivers/char/agp/agp.h >> +++ b/drivers/char/agp/agp.h >> @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data *bridge); >> void agp_remove_bridge(struct agp_bridge_data *bridge); >> >> /* Frontend routines. */ >> +#if IS_ENABLED(CONFIG_DRM_LEGACY) >> int agp_frontend_initialize(void); >> void agp_frontend_cleanup(void); >> +#else >> +static inline int agp_frontend_initialize(void) { return 0; } >> +static inline void agp_frontend_cleanup(void) {} >> +#endif > There's one non-legacy driver that uses these agp structures, which is > radeon. Does this change affect radeon? Nouveau uses AGP as well, but I'm not sure if both drivers use any of this stuff. Regards, Christian. > > Best regards > Thomas > >> >> /* Generic routines. */ >> void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode); >>
On Wed, Nov 18, 2020 at 9:24 AM Christian König <christian.koenig@amd.com> wrote: > > Am 18.11.20 um 09:02 schrieb Thomas Zimmermann: > > Hi > > > > Am 17.11.20 um 22:40 schrieb Daniel Vetter: > >> It's probably full of bugs ready for exploiting by userspace. And > >> there's not going to be any userspace for this without any of the drm > >> legacy drivers enabled too. So just couple it together. > >> > >> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> > >> Cc: David Airlie <airlied@linux.ie> > >> Cc: Adam Jackson <ajax@redhat.com> > >> --- > >> drivers/char/agp/Makefile | 6 +++++- > >> drivers/char/agp/agp.h | 5 +++++ > >> 2 files changed, 10 insertions(+), 1 deletion(-) > >> > >> diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile > >> index cb2497d157f6..90ed8c789e48 100644 > >> --- a/drivers/char/agp/Makefile > >> +++ b/drivers/char/agp/Makefile > >> @@ -1,7 +1,11 @@ > >> # SPDX-License-Identifier: GPL-2.0 > >> -agpgart-y := backend.o frontend.o generic.o isoch.o > >> +agpgart-y := backend.o generic.o isoch.o > >> > >> +ifeq ($(CONFIG_DRM_LEGACY),y) > >> agpgart-$(CONFIG_COMPAT) += compat_ioctl.o > >> +agpgart-y += frontend.o > >> +endif > >> + > >> > >> obj-$(CONFIG_AGP) += agpgart.o > >> obj-$(CONFIG_AGP_ALI) += ali-agp.o > >> diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h > >> index 4eb1c772ded7..bb09d64cd51e 100644 > >> --- a/drivers/char/agp/agp.h > >> +++ b/drivers/char/agp/agp.h > >> @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data *bridge); > >> void agp_remove_bridge(struct agp_bridge_data *bridge); > >> > >> /* Frontend routines. */ > >> +#if IS_ENABLED(CONFIG_DRM_LEGACY) > >> int agp_frontend_initialize(void); > >> void agp_frontend_cleanup(void); > >> +#else > >> +static inline int agp_frontend_initialize(void) { return 0; } > >> +static inline void agp_frontend_cleanup(void) {} > >> +#endif > > There's one non-legacy driver that uses these agp structures, which is > > radeon. Does this change affect radeon? > > Nouveau uses AGP as well, but I'm not sure if both drivers use any of > this stuff. frontend = /dev/agp chardev interface for userspace drivers. If you're looking at kernel drivers, you're looking at the wrong thing, the kernel-internal interface is in char/agp/backend.c and still enabled. So no impact at all on any kernel code. Now the impact this does have on kms drivers using agp is that there's no longer a userspace ioctl interface to change the agp setup and mappings and fight the kms driver (which assumes it's fully in control of agp configuration). -Daniel > > Regards, > Christian. > > > > > Best regards > > Thomas > > > >> > >> /* Generic routines. */ > >> void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode); > >> >
Hi Am 18.11.20 um 09:53 schrieb Daniel Vetter: > On Wed, Nov 18, 2020 at 9:24 AM Christian König > <christian.koenig@amd.com> wrote: >> >> Am 18.11.20 um 09:02 schrieb Thomas Zimmermann: >>> Hi >>> >>> Am 17.11.20 um 22:40 schrieb Daniel Vetter: >>>> It's probably full of bugs ready for exploiting by userspace. And >>>> there's not going to be any userspace for this without any of the drm >>>> legacy drivers enabled too. So just couple it together. >>>> >>>> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> >>>> Cc: David Airlie <airlied@linux.ie> >>>> Cc: Adam Jackson <ajax@redhat.com> >>>> --- >>>> drivers/char/agp/Makefile | 6 +++++- >>>> drivers/char/agp/agp.h | 5 +++++ >>>> 2 files changed, 10 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile >>>> index cb2497d157f6..90ed8c789e48 100644 >>>> --- a/drivers/char/agp/Makefile >>>> +++ b/drivers/char/agp/Makefile >>>> @@ -1,7 +1,11 @@ >>>> # SPDX-License-Identifier: GPL-2.0 >>>> -agpgart-y := backend.o frontend.o generic.o isoch.o >>>> +agpgart-y := backend.o generic.o isoch.o >>>> >>>> +ifeq ($(CONFIG_DRM_LEGACY),y) >>>> agpgart-$(CONFIG_COMPAT) += compat_ioctl.o >>>> +agpgart-y += frontend.o >>>> +endif >>>> + >>>> >>>> obj-$(CONFIG_AGP) += agpgart.o >>>> obj-$(CONFIG_AGP_ALI) += ali-agp.o >>>> diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h >>>> index 4eb1c772ded7..bb09d64cd51e 100644 >>>> --- a/drivers/char/agp/agp.h >>>> +++ b/drivers/char/agp/agp.h >>>> @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data *bridge); >>>> void agp_remove_bridge(struct agp_bridge_data *bridge); >>>> >>>> /* Frontend routines. */ >>>> +#if IS_ENABLED(CONFIG_DRM_LEGACY) >>>> int agp_frontend_initialize(void); >>>> void agp_frontend_cleanup(void); >>>> +#else >>>> +static inline int agp_frontend_initialize(void) { return 0; } >>>> +static inline void agp_frontend_cleanup(void) {} >>>> +#endif >>> There's one non-legacy driver that uses these agp structures, which is >>> radeon. Does this change affect radeon? >> >> Nouveau uses AGP as well, but I'm not sure if both drivers use any of >> this stuff. > > frontend = /dev/agp chardev interface for userspace drivers. If you're > looking at kernel drivers, you're looking at the wrong thing, the > kernel-internal interface is in char/agp/backend.c and still enabled. > So no impact at all on any kernel code. > > Now the impact this does have on kms drivers using agp is that there's > no longer a userspace ioctl interface to change the agp setup and > mappings and fight the kms driver (which assumes it's fully in control > of agp configuration). Thanks for clarifying. I'm certainly not qualified, but still Acked-by: Thomas Zimmermann <tzimmermann@suse.de> Best regards Thomas > -Daniel > >> >> Regards, >> Christian. >> >>> >>> Best regards >>> Thomas >>> >>>> >>>> /* Generic routines. */ >>>> void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode); >>>> >> > >
Am 18.11.20 um 10:02 schrieb Thomas Zimmermann: > Hi > > Am 18.11.20 um 09:53 schrieb Daniel Vetter: >> On Wed, Nov 18, 2020 at 9:24 AM Christian König >> <christian.koenig@amd.com> wrote: >>> >>> Am 18.11.20 um 09:02 schrieb Thomas Zimmermann: >>>> Hi >>>> >>>> Am 17.11.20 um 22:40 schrieb Daniel Vetter: >>>>> It's probably full of bugs ready for exploiting by userspace. And >>>>> there's not going to be any userspace for this without any of the drm >>>>> legacy drivers enabled too. So just couple it together. >>>>> >>>>> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> >>>>> Cc: David Airlie <airlied@linux.ie> >>>>> Cc: Adam Jackson <ajax@redhat.com> >>>>> --- >>>>> drivers/char/agp/Makefile | 6 +++++- >>>>> drivers/char/agp/agp.h | 5 +++++ >>>>> 2 files changed, 10 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile >>>>> index cb2497d157f6..90ed8c789e48 100644 >>>>> --- a/drivers/char/agp/Makefile >>>>> +++ b/drivers/char/agp/Makefile >>>>> @@ -1,7 +1,11 @@ >>>>> # SPDX-License-Identifier: GPL-2.0 >>>>> -agpgart-y := backend.o frontend.o generic.o isoch.o >>>>> +agpgart-y := backend.o generic.o isoch.o >>>>> >>>>> +ifeq ($(CONFIG_DRM_LEGACY),y) >>>>> agpgart-$(CONFIG_COMPAT) += compat_ioctl.o >>>>> +agpgart-y += frontend.o >>>>> +endif >>>>> + >>>>> >>>>> obj-$(CONFIG_AGP) += agpgart.o >>>>> obj-$(CONFIG_AGP_ALI) += ali-agp.o >>>>> diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h >>>>> index 4eb1c772ded7..bb09d64cd51e 100644 >>>>> --- a/drivers/char/agp/agp.h >>>>> +++ b/drivers/char/agp/agp.h >>>>> @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data >>>>> *bridge); >>>>> void agp_remove_bridge(struct agp_bridge_data *bridge); >>>>> >>>>> /* Frontend routines. */ >>>>> +#if IS_ENABLED(CONFIG_DRM_LEGACY) >>>>> int agp_frontend_initialize(void); >>>>> void agp_frontend_cleanup(void); >>>>> +#else >>>>> +static inline int agp_frontend_initialize(void) { return 0; } >>>>> +static inline void agp_frontend_cleanup(void) {} >>>>> +#endif >>>> There's one non-legacy driver that uses these agp structures, which is >>>> radeon. Does this change affect radeon? >>> >>> Nouveau uses AGP as well, but I'm not sure if both drivers use any of >>> this stuff. >> >> frontend = /dev/agp chardev interface for userspace drivers. If you're >> looking at kernel drivers, you're looking at the wrong thing, the >> kernel-internal interface is in char/agp/backend.c and still enabled. >> So no impact at all on any kernel code. >> >> Now the impact this does have on kms drivers using agp is that there's >> no longer a userspace ioctl interface to change the agp setup and >> mappings and fight the kms driver (which assumes it's fully in control >> of agp configuration). I was strongly assuming something like this, thanks for clarifying. > > Thanks for clarifying. I'm certainly not qualified, but still > > Acked-by: Thomas Zimmermann <tzimmermann@suse.de> Acked-by: Christian König <christian.koenig@amd.com> as well. Regards, Christian. > > Best regards > Thomas > >> -Daniel >> >>> >>> Regards, >>> Christian. >>> >>>> >>>> Best regards >>>> Thomas >>>> >>>>> >>>>> /* Generic routines. */ >>>>> void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode); >>>>> >>> >> >> >
On Tue, Nov 17, 2020 at 4:40 PM Daniel Vetter <daniel.vetter@ffwll.ch> wrote: > > It's probably full of bugs ready for exploiting by userspace. And > there's not going to be any userspace for this without any of the drm > legacy drivers enabled too. So just couple it together. Not quite true. The only UMS driver using this is i810, which needs it even with DRI disabled. I have difficulty caring though. In related news, since i810 is only ever attached to 32-bit x86 we could disable the frontend unconditionally on amd64 (and everything else tbh). Acked-by: Adam Jackson <ajax@redhat.com> - ajax
On Wed, Nov 18, 2020 at 10:55:27AM -0500, Adam Jackson wrote: > On Tue, Nov 17, 2020 at 4:40 PM Daniel Vetter <daniel.vetter@ffwll.ch> wrote: > > > > It's probably full of bugs ready for exploiting by userspace. And > > there's not going to be any userspace for this without any of the drm > > legacy drivers enabled too. So just couple it together. > > Not quite true. The only UMS driver using this is i810, which needs it > even with DRI disabled. I have difficulty caring though. TIL. > In related news, since i810 is only ever attached to 32-bit x86 we > could disable the frontend unconditionally on amd64 (and everything > else tbh). > > Acked-by: Adam Jackson <ajax@redhat.com> Pushed to drm-misc-next, thanks for taking a look. -Daniel
diff --git a/drivers/char/agp/Makefile b/drivers/char/agp/Makefile index cb2497d157f6..90ed8c789e48 100644 --- a/drivers/char/agp/Makefile +++ b/drivers/char/agp/Makefile @@ -1,7 +1,11 @@ # SPDX-License-Identifier: GPL-2.0 -agpgart-y := backend.o frontend.o generic.o isoch.o +agpgart-y := backend.o generic.o isoch.o +ifeq ($(CONFIG_DRM_LEGACY),y) agpgart-$(CONFIG_COMPAT) += compat_ioctl.o +agpgart-y += frontend.o +endif + obj-$(CONFIG_AGP) += agpgart.o obj-$(CONFIG_AGP_ALI) += ali-agp.o diff --git a/drivers/char/agp/agp.h b/drivers/char/agp/agp.h index 4eb1c772ded7..bb09d64cd51e 100644 --- a/drivers/char/agp/agp.h +++ b/drivers/char/agp/agp.h @@ -186,8 +186,13 @@ int agp_add_bridge(struct agp_bridge_data *bridge); void agp_remove_bridge(struct agp_bridge_data *bridge); /* Frontend routines. */ +#if IS_ENABLED(CONFIG_DRM_LEGACY) int agp_frontend_initialize(void); void agp_frontend_cleanup(void); +#else +static inline int agp_frontend_initialize(void) { return 0; } +static inline void agp_frontend_cleanup(void) {} +#endif /* Generic routines. */ void agp_generic_enable(struct agp_bridge_data *bridge, u32 mode);
It's probably full of bugs ready for exploiting by userspace. And there's not going to be any userspace for this without any of the drm legacy drivers enabled too. So just couple it together. Signed-off-by: Daniel Vetter <daniel.vetter@intel.com> Cc: David Airlie <airlied@linux.ie> Cc: Adam Jackson <ajax@redhat.com> --- drivers/char/agp/Makefile | 6 +++++- drivers/char/agp/agp.h | 5 +++++ 2 files changed, 10 insertions(+), 1 deletion(-)