| Message ID | 20241219214909.104869-2-gustavo.sousa@intel.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | drm/i915/display: Reduce global state funcs boilerplate | expand |
-----Original Message----- From: Intel-xe <intel-xe-bounces@lists.freedesktop.org> On Behalf Of Gustavo Sousa Sent: Thursday, December 19, 2024 1:49 PM To: intel-gfx@lists.freedesktop.org; intel-xe@lists.freedesktop.org Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>; Nikula, Jani <jani.nikula@intel.com> Subject: [PATCH 1/3] drm/i915/display: Do not assume zero offset when duplicating global state > > The current intel_*_duplicate_state() functions assume the offset for > the base member of their state structures is zero when calling > kmemdup(). While that is true today, such assumption should not be made > and proper offset must be applied when calling kmemdup(), otherwise we > will be duplicating the wrong memory area if, for some reason, the > offset is changed in the future. > > As such, update each of those functions to use its respective > to_*_state() as the parameter to kmemdup(). > > Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com> LGTM. Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com> -Jonathan Cavitt > --- > drivers/gpu/drm/i915/display/intel_bw.c | 4 ++-- > drivers/gpu/drm/i915/display/intel_cdclk.c | 4 ++-- > drivers/gpu/drm/i915/display/intel_pmdemand.c | 4 ++-- > drivers/gpu/drm/i915/display/skl_watermark.c | 4 ++-- > 4 files changed, 8 insertions(+), 8 deletions(-) > > diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c > index 08e8a67ca74c..30236010e0ed 100644 > --- a/drivers/gpu/drm/i915/display/intel_bw.c > +++ b/drivers/gpu/drm/i915/display/intel_bw.c > @@ -1425,9 +1425,9 @@ int intel_bw_atomic_check(struct intel_atomic_state *state) > static struct intel_global_state * > intel_bw_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_bw_state *state; > + struct intel_bw_state *state = to_intel_bw_state(obj->state); > > - state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL); > + state = kmemdup(state, sizeof(*state), GFP_KERNEL); > if (!state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c > index 3506e576bf6b..fc084e2a4c6a 100644 > --- a/drivers/gpu/drm/i915/display/intel_cdclk.c > +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c > @@ -3130,9 +3130,9 @@ static int fixed_modeset_calc_cdclk(struct intel_atomic_state *state) > > static struct intel_global_state *intel_cdclk_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_cdclk_state *cdclk_state; > + struct intel_cdclk_state *cdclk_state = to_intel_cdclk_state(obj->state); > > - cdclk_state = kmemdup(obj->state, sizeof(*cdclk_state), GFP_KERNEL); > + cdclk_state = kmemdup(cdclk_state, sizeof(*cdclk_state), GFP_KERNEL); > if (!cdclk_state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/intel_pmdemand.c b/drivers/gpu/drm/i915/display/intel_pmdemand.c > index cdd314956a31..1f71efb7d04d 100644 > --- a/drivers/gpu/drm/i915/display/intel_pmdemand.c > +++ b/drivers/gpu/drm/i915/display/intel_pmdemand.c > @@ -18,9 +18,9 @@ > static struct intel_global_state * > intel_pmdemand_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_pmdemand_state *pmdemand_state; > + struct intel_pmdemand_state *pmdemand_state = to_intel_pmdemand_state(obj->state); > > - pmdemand_state = kmemdup(obj->state, sizeof(*pmdemand_state), GFP_KERNEL); > + pmdemand_state = kmemdup(pmdemand_state, sizeof(*pmdemand_state), GFP_KERNEL); > if (!pmdemand_state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c b/drivers/gpu/drm/i915/display/skl_watermark.c > index 29e8ea91c858..b3d38e09df5a 100644 > --- a/drivers/gpu/drm/i915/display/skl_watermark.c > +++ b/drivers/gpu/drm/i915/display/skl_watermark.c > @@ -3291,9 +3291,9 @@ static void skl_setup_wm_latency(struct drm_i915_private *i915) > > static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_dbuf_state *dbuf_state; > + struct intel_dbuf_state *dbuf_state = to_intel_dbuf_state(obj->state); > > - dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL); > + dbuf_state = kmemdup(dbuf_state, sizeof(*dbuf_state), GFP_KERNEL); > if (!dbuf_state) > return NULL; > > -- > 2.47.1 > >
On Thu, Dec 19, 2024 at 06:48:36PM -0300, Gustavo Sousa wrote: > The current intel_*_duplicate_state() functions assume the offset for > the base member of their state structures is zero when calling > kmemdup(). While that is true today, such assumption should not be made > and proper offset must be applied when calling kmemdup(), otherwise we > will be duplicating the wrong memory area if, for some reason, the > offset is changed in the future. All kms objects we use make that same assumption. I think the correct thing to do is to just throw in some BUILD_BUG_ON()/etc. to make the thing not build if that doesn't hold. I had a patch like that ages ago, but it's no doubt 110% stale by now. I suppose no real harm if avoiding that assumption in spots like this, but the &foo->base==NULL <-> foo==NULL assumptions we have all over the place are the far bigger issue. > > As such, update each of those functions to use its respective > to_*_state() as the parameter to kmemdup(). > > Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com> > --- > drivers/gpu/drm/i915/display/intel_bw.c | 4 ++-- > drivers/gpu/drm/i915/display/intel_cdclk.c | 4 ++-- > drivers/gpu/drm/i915/display/intel_pmdemand.c | 4 ++-- > drivers/gpu/drm/i915/display/skl_watermark.c | 4 ++-- > 4 files changed, 8 insertions(+), 8 deletions(-) > > diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c > index 08e8a67ca74c..30236010e0ed 100644 > --- a/drivers/gpu/drm/i915/display/intel_bw.c > +++ b/drivers/gpu/drm/i915/display/intel_bw.c > @@ -1425,9 +1425,9 @@ int intel_bw_atomic_check(struct intel_atomic_state *state) > static struct intel_global_state * > intel_bw_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_bw_state *state; > + struct intel_bw_state *state = to_intel_bw_state(obj->state); > > - state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL); > + state = kmemdup(state, sizeof(*state), GFP_KERNEL); > if (!state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c > index 3506e576bf6b..fc084e2a4c6a 100644 > --- a/drivers/gpu/drm/i915/display/intel_cdclk.c > +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c > @@ -3130,9 +3130,9 @@ static int fixed_modeset_calc_cdclk(struct intel_atomic_state *state) > > static struct intel_global_state *intel_cdclk_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_cdclk_state *cdclk_state; > + struct intel_cdclk_state *cdclk_state = to_intel_cdclk_state(obj->state); > > - cdclk_state = kmemdup(obj->state, sizeof(*cdclk_state), GFP_KERNEL); > + cdclk_state = kmemdup(cdclk_state, sizeof(*cdclk_state), GFP_KERNEL); > if (!cdclk_state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/intel_pmdemand.c b/drivers/gpu/drm/i915/display/intel_pmdemand.c > index cdd314956a31..1f71efb7d04d 100644 > --- a/drivers/gpu/drm/i915/display/intel_pmdemand.c > +++ b/drivers/gpu/drm/i915/display/intel_pmdemand.c > @@ -18,9 +18,9 @@ > static struct intel_global_state * > intel_pmdemand_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_pmdemand_state *pmdemand_state; > + struct intel_pmdemand_state *pmdemand_state = to_intel_pmdemand_state(obj->state); > > - pmdemand_state = kmemdup(obj->state, sizeof(*pmdemand_state), GFP_KERNEL); > + pmdemand_state = kmemdup(pmdemand_state, sizeof(*pmdemand_state), GFP_KERNEL); > if (!pmdemand_state) > return NULL; > > diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c b/drivers/gpu/drm/i915/display/skl_watermark.c > index 29e8ea91c858..b3d38e09df5a 100644 > --- a/drivers/gpu/drm/i915/display/skl_watermark.c > +++ b/drivers/gpu/drm/i915/display/skl_watermark.c > @@ -3291,9 +3291,9 @@ static void skl_setup_wm_latency(struct drm_i915_private *i915) > > static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj) > { > - struct intel_dbuf_state *dbuf_state; > + struct intel_dbuf_state *dbuf_state = to_intel_dbuf_state(obj->state); > > - dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL); > + dbuf_state = kmemdup(dbuf_state, sizeof(*dbuf_state), GFP_KERNEL); > if (!dbuf_state) > return NULL; > > -- > 2.47.1
Quoting Ville Syrjälä (2024-12-20 06:11:57-03:00) >On Thu, Dec 19, 2024 at 06:48:36PM -0300, Gustavo Sousa wrote: >> The current intel_*_duplicate_state() functions assume the offset for >> the base member of their state structures is zero when calling >> kmemdup(). While that is true today, such assumption should not be made >> and proper offset must be applied when calling kmemdup(), otherwise we >> will be duplicating the wrong memory area if, for some reason, the >> offset is changed in the future. > >All kms objects we use make that same assumption. I think the correct >thing to do is to just throw in some BUILD_BUG_ON()/etc. to make the >thing not build if that doesn't hold. I had a patch like that ages >ago, but it's no doubt 110% stale by now. Okay. In this case, if we were to follow this route, some questions: 1) Where would be the best place to put BUILD_BUG_ON() (or static_assert())? Would requiring the usage of some macro provided by intel_global_state.h be acceptable? Something like: struct my_state { struct intel_global_state base; ... }; INTEL_GLOBAL_STATE_SUBCLASS(struct my_state, base); Well, we are still requiring a macro call here when compared to the explicit assert on the offset, but now the semantics is that the developer advertising a subclass of intel_global_state instead of hand rolling the necessary checks herself. 2) If we require the embedded struct to be always the first member, would there be any point in defining and using the to_*_state() macros? I don't see much benefit of that over simply casting the pointer if the offset is always zero. I might be missing something though. > >I suppose no real harm if avoiding that assumption in spots like >this, but the &foo->base==NULL <-> foo==NULL assumptions we have >all over the place are the far bigger issue. Yeah, maybe requiring base to be the first member is a more manageable approach indeed. -- Gustavo Sousa > >> >> As such, update each of those functions to use its respective >> to_*_state() as the parameter to kmemdup(). >> >> Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com> >> --- >> drivers/gpu/drm/i915/display/intel_bw.c | 4 ++-- >> drivers/gpu/drm/i915/display/intel_cdclk.c | 4 ++-- >> drivers/gpu/drm/i915/display/intel_pmdemand.c | 4 ++-- >> drivers/gpu/drm/i915/display/skl_watermark.c | 4 ++-- >> 4 files changed, 8 insertions(+), 8 deletions(-) >> >> diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c >> index 08e8a67ca74c..30236010e0ed 100644 >> --- a/drivers/gpu/drm/i915/display/intel_bw.c >> +++ b/drivers/gpu/drm/i915/display/intel_bw.c >> @@ -1425,9 +1425,9 @@ int intel_bw_atomic_check(struct intel_atomic_state *state) >> static struct intel_global_state * >> intel_bw_duplicate_state(struct intel_global_obj *obj) >> { >> - struct intel_bw_state *state; >> + struct intel_bw_state *state = to_intel_bw_state(obj->state); >> >> - state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL); >> + state = kmemdup(state, sizeof(*state), GFP_KERNEL); >> if (!state) >> return NULL; >> >> diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c >> index 3506e576bf6b..fc084e2a4c6a 100644 >> --- a/drivers/gpu/drm/i915/display/intel_cdclk.c >> +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c >> @@ -3130,9 +3130,9 @@ static int fixed_modeset_calc_cdclk(struct intel_atomic_state *state) >> >> static struct intel_global_state *intel_cdclk_duplicate_state(struct intel_global_obj *obj) >> { >> - struct intel_cdclk_state *cdclk_state; >> + struct intel_cdclk_state *cdclk_state = to_intel_cdclk_state(obj->state); >> >> - cdclk_state = kmemdup(obj->state, sizeof(*cdclk_state), GFP_KERNEL); >> + cdclk_state = kmemdup(cdclk_state, sizeof(*cdclk_state), GFP_KERNEL); >> if (!cdclk_state) >> return NULL; >> >> diff --git a/drivers/gpu/drm/i915/display/intel_pmdemand.c b/drivers/gpu/drm/i915/display/intel_pmdemand.c >> index cdd314956a31..1f71efb7d04d 100644 >> --- a/drivers/gpu/drm/i915/display/intel_pmdemand.c >> +++ b/drivers/gpu/drm/i915/display/intel_pmdemand.c >> @@ -18,9 +18,9 @@ >> static struct intel_global_state * >> intel_pmdemand_duplicate_state(struct intel_global_obj *obj) >> { >> - struct intel_pmdemand_state *pmdemand_state; >> + struct intel_pmdemand_state *pmdemand_state = to_intel_pmdemand_state(obj->state); >> >> - pmdemand_state = kmemdup(obj->state, sizeof(*pmdemand_state), GFP_KERNEL); >> + pmdemand_state = kmemdup(pmdemand_state, sizeof(*pmdemand_state), GFP_KERNEL); >> if (!pmdemand_state) >> return NULL; >> >> diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c b/drivers/gpu/drm/i915/display/skl_watermark.c >> index 29e8ea91c858..b3d38e09df5a 100644 >> --- a/drivers/gpu/drm/i915/display/skl_watermark.c >> +++ b/drivers/gpu/drm/i915/display/skl_watermark.c >> @@ -3291,9 +3291,9 @@ static void skl_setup_wm_latency(struct drm_i915_private *i915) >> >> static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj) >> { >> - struct intel_dbuf_state *dbuf_state; >> + struct intel_dbuf_state *dbuf_state = to_intel_dbuf_state(obj->state); >> >> - dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL); >> + dbuf_state = kmemdup(dbuf_state, sizeof(*dbuf_state), GFP_KERNEL); >> if (!dbuf_state) >> return NULL; >> >> -- >> 2.47.1 > >-- >Ville Syrjälä >Intel
diff --git a/drivers/gpu/drm/i915/display/intel_bw.c b/drivers/gpu/drm/i915/display/intel_bw.c index 08e8a67ca74c..30236010e0ed 100644 --- a/drivers/gpu/drm/i915/display/intel_bw.c +++ b/drivers/gpu/drm/i915/display/intel_bw.c @@ -1425,9 +1425,9 @@ int intel_bw_atomic_check(struct intel_atomic_state *state) static struct intel_global_state * intel_bw_duplicate_state(struct intel_global_obj *obj) { - struct intel_bw_state *state; + struct intel_bw_state *state = to_intel_bw_state(obj->state); - state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL); + state = kmemdup(state, sizeof(*state), GFP_KERNEL); if (!state) return NULL; diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c index 3506e576bf6b..fc084e2a4c6a 100644 --- a/drivers/gpu/drm/i915/display/intel_cdclk.c +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c @@ -3130,9 +3130,9 @@ static int fixed_modeset_calc_cdclk(struct intel_atomic_state *state) static struct intel_global_state *intel_cdclk_duplicate_state(struct intel_global_obj *obj) { - struct intel_cdclk_state *cdclk_state; + struct intel_cdclk_state *cdclk_state = to_intel_cdclk_state(obj->state); - cdclk_state = kmemdup(obj->state, sizeof(*cdclk_state), GFP_KERNEL); + cdclk_state = kmemdup(cdclk_state, sizeof(*cdclk_state), GFP_KERNEL); if (!cdclk_state) return NULL; diff --git a/drivers/gpu/drm/i915/display/intel_pmdemand.c b/drivers/gpu/drm/i915/display/intel_pmdemand.c index cdd314956a31..1f71efb7d04d 100644 --- a/drivers/gpu/drm/i915/display/intel_pmdemand.c +++ b/drivers/gpu/drm/i915/display/intel_pmdemand.c @@ -18,9 +18,9 @@ static struct intel_global_state * intel_pmdemand_duplicate_state(struct intel_global_obj *obj) { - struct intel_pmdemand_state *pmdemand_state; + struct intel_pmdemand_state *pmdemand_state = to_intel_pmdemand_state(obj->state); - pmdemand_state = kmemdup(obj->state, sizeof(*pmdemand_state), GFP_KERNEL); + pmdemand_state = kmemdup(pmdemand_state, sizeof(*pmdemand_state), GFP_KERNEL); if (!pmdemand_state) return NULL; diff --git a/drivers/gpu/drm/i915/display/skl_watermark.c b/drivers/gpu/drm/i915/display/skl_watermark.c index 29e8ea91c858..b3d38e09df5a 100644 --- a/drivers/gpu/drm/i915/display/skl_watermark.c +++ b/drivers/gpu/drm/i915/display/skl_watermark.c @@ -3291,9 +3291,9 @@ static void skl_setup_wm_latency(struct drm_i915_private *i915) static struct intel_global_state *intel_dbuf_duplicate_state(struct intel_global_obj *obj) { - struct intel_dbuf_state *dbuf_state; + struct intel_dbuf_state *dbuf_state = to_intel_dbuf_state(obj->state); - dbuf_state = kmemdup(obj->state, sizeof(*dbuf_state), GFP_KERNEL); + dbuf_state = kmemdup(dbuf_state, sizeof(*dbuf_state), GFP_KERNEL); if (!dbuf_state) return NULL;
The current intel_*_duplicate_state() functions assume the offset for the base member of their state structures is zero when calling kmemdup(). While that is true today, such assumption should not be made and proper offset must be applied when calling kmemdup(), otherwise we will be duplicating the wrong memory area if, for some reason, the offset is changed in the future. As such, update each of those functions to use its respective to_*_state() as the parameter to kmemdup(). Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com> --- drivers/gpu/drm/i915/display/intel_bw.c | 4 ++-- drivers/gpu/drm/i915/display/intel_cdclk.c | 4 ++-- drivers/gpu/drm/i915/display/intel_pmdemand.c | 4 ++-- drivers/gpu/drm/i915/display/skl_watermark.c | 4 ++-- 4 files changed, 8 insertions(+), 8 deletions(-)