| Message ID | 20180212145533.30046-1-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, Feb 12, 2018 at 02:55:33PM +0000, Chris Wilson wrote: > Use the new idr_init_base() function to create an IDR that knows id==0 > is never allocated as it maps to an invalid identifier. By knowing that > id==0 is invalid, the IDR can start from id=1 instead avoiding the issue > of having to start each lookup from the zeroth leaf as id==0 is always > unused (and thus the tree-of-bitmaps indicate that is the first > available). > > References: 6ce711f27500 ("idr: Make 1-based IDRs more efficient") > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > Cc: Christian Konig <christian.koenig@amd.com> > Cc: Dave Airlie <airlied@redhat.com> Yep, looks like all of these pass start==1 to idr_alloc(). Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Looks like at least tile_idr and crtc_idr could use this as well, although I suppose they're not hit nearly as hard as the gem stuff. Also someone should really s/crtc_idr/obj_id_idr/ or something along those lines. > --- > Note this requires 4.16-rc1. > --- > drivers/gpu/drm/drm_gem.c | 4 ++-- > drivers/gpu/drm/drm_syncobj.c | 2 +- > 2 files changed, 3 insertions(+), 3 deletions(-) > > diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c > index 01f8d9481211..4975ba9a7bc8 100644 > --- a/drivers/gpu/drm/drm_gem.c > +++ b/drivers/gpu/drm/drm_gem.c > @@ -98,7 +98,7 @@ drm_gem_init(struct drm_device *dev) > struct drm_vma_offset_manager *vma_offset_manager; > > mutex_init(&dev->object_name_lock); > - idr_init(&dev->object_name_idr); > + idr_init_base(&dev->object_name_idr, 1); > > vma_offset_manager = kzalloc(sizeof(*vma_offset_manager), GFP_KERNEL); > if (!vma_offset_manager) { > @@ -776,7 +776,7 @@ drm_gem_open_ioctl(struct drm_device *dev, void *data, > void > drm_gem_open(struct drm_device *dev, struct drm_file *file_private) > { > - idr_init(&file_private->object_idr); > + idr_init_base(&file_private->object_idr, 1); > spin_lock_init(&file_private->table_lock); > } > > diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c > index 0b7b0d1ad2d5..d4f4ce484529 100644 > --- a/drivers/gpu/drm/drm_syncobj.c > +++ b/drivers/gpu/drm/drm_syncobj.c > @@ -546,7 +546,7 @@ static int drm_syncobj_export_sync_file(struct drm_file *file_private, > void > drm_syncobj_open(struct drm_file *file_private) > { > - idr_init(&file_private->syncobj_idr); > + idr_init_base(&file_private->syncobj_idr, 1); > spin_lock_init(&file_private->syncobj_table_lock); > } > > -- > 2.16.1 > > _______________________________________________ > dri-devel mailing list > dri-devel@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/dri-devel
Am 12.02.2018 um 18:14 schrieb Ville Syrjälä: > On Mon, Feb 12, 2018 at 02:55:33PM +0000, Chris Wilson wrote: >> Use the new idr_init_base() function to create an IDR that knows id==0 >> is never allocated as it maps to an invalid identifier. By knowing that >> id==0 is invalid, the IDR can start from id=1 instead avoiding the issue >> of having to start each lookup from the zeroth leaf as id==0 is always >> unused (and thus the tree-of-bitmaps indicate that is the first >> available). >> >> References: 6ce711f27500 ("idr: Make 1-based IDRs more efficient") >> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> >> Cc: Christian Konig <christian.koenig@amd.com> >> Cc: Dave Airlie <airlied@redhat.com> > Yep, looks like all of these pass start==1 to idr_alloc(). > Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Acked-by: Christian König <christian.koenig@amd.com> as well. Probably going to do this for the command submission context handles in amdgpu as well. Thanks, Christian. > > Looks like at least tile_idr and crtc_idr could use this as well, > although I suppose they're not hit nearly as hard as the gem stuff. > Also someone should really s/crtc_idr/obj_id_idr/ or something > along those lines. > >> --- >> Note this requires 4.16-rc1. >> --- >> drivers/gpu/drm/drm_gem.c | 4 ++-- >> drivers/gpu/drm/drm_syncobj.c | 2 +- >> 2 files changed, 3 insertions(+), 3 deletions(-) >> >> diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c >> index 01f8d9481211..4975ba9a7bc8 100644 >> --- a/drivers/gpu/drm/drm_gem.c >> +++ b/drivers/gpu/drm/drm_gem.c >> @@ -98,7 +98,7 @@ drm_gem_init(struct drm_device *dev) >> struct drm_vma_offset_manager *vma_offset_manager; >> >> mutex_init(&dev->object_name_lock); >> - idr_init(&dev->object_name_idr); >> + idr_init_base(&dev->object_name_idr, 1); >> >> vma_offset_manager = kzalloc(sizeof(*vma_offset_manager), GFP_KERNEL); >> if (!vma_offset_manager) { >> @@ -776,7 +776,7 @@ drm_gem_open_ioctl(struct drm_device *dev, void *data, >> void >> drm_gem_open(struct drm_device *dev, struct drm_file *file_private) >> { >> - idr_init(&file_private->object_idr); >> + idr_init_base(&file_private->object_idr, 1); >> spin_lock_init(&file_private->table_lock); >> } >> >> diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c >> index 0b7b0d1ad2d5..d4f4ce484529 100644 >> --- a/drivers/gpu/drm/drm_syncobj.c >> +++ b/drivers/gpu/drm/drm_syncobj.c >> @@ -546,7 +546,7 @@ static int drm_syncobj_export_sync_file(struct drm_file *file_private, >> void >> drm_syncobj_open(struct drm_file *file_private) >> { >> - idr_init(&file_private->syncobj_idr); >> + idr_init_base(&file_private->syncobj_idr, 1); >> spin_lock_init(&file_private->syncobj_table_lock); >> } >> >> -- >> 2.16.1 >> >> _______________________________________________ >> dri-devel mailing list >> dri-devel@lists.freedesktop.org >> https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoting Christian König (2018-02-13 13:48:24) > Am 12.02.2018 um 18:14 schrieb Ville Syrjälä: > > On Mon, Feb 12, 2018 at 02:55:33PM +0000, Chris Wilson wrote: > >> Use the new idr_init_base() function to create an IDR that knows id==0 > >> is never allocated as it maps to an invalid identifier. By knowing that > >> id==0 is invalid, the IDR can start from id=1 instead avoiding the issue > >> of having to start each lookup from the zeroth leaf as id==0 is always > >> unused (and thus the tree-of-bitmaps indicate that is the first > >> available). > >> > >> References: 6ce711f27500 ("idr: Make 1-based IDRs more efficient") > >> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > >> Cc: Christian Konig <christian.koenig@amd.com> > >> Cc: Dave Airlie <airlied@redhat.com> > > Yep, looks like all of these pass start==1 to idr_alloc(). > > Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> > > Acked-by: Christian König <christian.koenig@amd.com> as well. > > Probably going to do this for the command submission context handles in > amdgpu as well. Thanks for the review, pushed to drm-misc-next. As noted, there are probably quite a few more idr's that we can move over to idr_init_base(1), e.g. the KMS objects. If someone feels like a small task? -Chris
On Mon, Feb 19, 2018 at 02:35:38PM +0000, Chris Wilson wrote: > Quoting Christian König (2018-02-13 13:48:24) > > Am 12.02.2018 um 18:14 schrieb Ville Syrjälä: > > > On Mon, Feb 12, 2018 at 02:55:33PM +0000, Chris Wilson wrote: > > >> Use the new idr_init_base() function to create an IDR that knows id==0 > > >> is never allocated as it maps to an invalid identifier. By knowing that > > >> id==0 is invalid, the IDR can start from id=1 instead avoiding the issue > > >> of having to start each lookup from the zeroth leaf as id==0 is always > > >> unused (and thus the tree-of-bitmaps indicate that is the first > > >> available). > > >> > > >> References: 6ce711f27500 ("idr: Make 1-based IDRs more efficient") > > >> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > >> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> > > >> Cc: Christian Konig <christian.koenig@amd.com> > > >> Cc: Dave Airlie <airlied@redhat.com> > > > Yep, looks like all of these pass start==1 to idr_alloc(). > > > Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> > > > > Acked-by: Christian König <christian.koenig@amd.com> as well. > > > > Probably going to do this for the command submission context handles in > > amdgpu as well. > > Thanks for the review, pushed to drm-misc-next. > > As noted, there are probably quite a few more idr's that we can move > over to idr_init_base(1), e.g. the KMS objects. If someone feels like a > small task? I'll type a todo.rst patch. -Daniel
diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c index 01f8d9481211..4975ba9a7bc8 100644 --- a/drivers/gpu/drm/drm_gem.c +++ b/drivers/gpu/drm/drm_gem.c @@ -98,7 +98,7 @@ drm_gem_init(struct drm_device *dev) struct drm_vma_offset_manager *vma_offset_manager; mutex_init(&dev->object_name_lock); - idr_init(&dev->object_name_idr); + idr_init_base(&dev->object_name_idr, 1); vma_offset_manager = kzalloc(sizeof(*vma_offset_manager), GFP_KERNEL); if (!vma_offset_manager) { @@ -776,7 +776,7 @@ drm_gem_open_ioctl(struct drm_device *dev, void *data, void drm_gem_open(struct drm_device *dev, struct drm_file *file_private) { - idr_init(&file_private->object_idr); + idr_init_base(&file_private->object_idr, 1); spin_lock_init(&file_private->table_lock); } diff --git a/drivers/gpu/drm/drm_syncobj.c b/drivers/gpu/drm/drm_syncobj.c index 0b7b0d1ad2d5..d4f4ce484529 100644 --- a/drivers/gpu/drm/drm_syncobj.c +++ b/drivers/gpu/drm/drm_syncobj.c @@ -546,7 +546,7 @@ static int drm_syncobj_export_sync_file(struct drm_file *file_private, void drm_syncobj_open(struct drm_file *file_private) { - idr_init(&file_private->syncobj_idr); + idr_init_base(&file_private->syncobj_idr, 1); spin_lock_init(&file_private->syncobj_table_lock); }
Use the new idr_init_base() function to create an IDR that knows id==0 is never allocated as it maps to an invalid identifier. By knowing that id==0 is invalid, the IDR can start from id=1 instead avoiding the issue of having to start each lookup from the zeroth leaf as id==0 is always unused (and thus the tree-of-bitmaps indicate that is the first available). References: 6ce711f27500 ("idr: Make 1-based IDRs more efficient") Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Christian Konig <christian.koenig@amd.com> Cc: Dave Airlie <airlied@redhat.com> --- Note this requires 4.16-rc1. --- drivers/gpu/drm/drm_gem.c | 4 ++-- drivers/gpu/drm/drm_syncobj.c | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-)