Skip to contentSkip to conversation

drm/i915: Round up object allocations

PATCH In review Ben Widawsky 4 participants 10 messages Jan 24, 2014 – Jan 26, 2014 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH 1/5 drm/i915: Round up object allocations

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +2 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 	struct address_space *mapping;
 	gfp_t mask;
 
+	size = round_up(size, PAGE_SIZE);
+
 	obj = i915_gem_object_alloc(dev);
 	if (obj == NULL)
 		return NULL;
Signature · 1 line
1.8.5.3
#2 lore raw
Quoted 14 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 	struct address_space *mapping;
 	gfp_t mask;
 
+	size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 10 lines
drivers/gpu/drm/i915/i915_gem.c +0 -0
 	obj = i915_gem_object_alloc(dev);
 	if (obj == NULL)
 		return NULL;
Signature · 6 lines
1.8.5.3

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#3 lore raw
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
 	struct address_space *mapping;
 	gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+	size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Why?

Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
#4 lore raw
Quoted 27 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ... -Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#5 lore raw
Quoted 28 lines · Daniel Vetter
Quoted 27 lines · Ben Widawsky
Quoted 25 lines · Daniel Vetter
Quoted 14 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 5 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 7 lines
Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

We already have an -EINVAL guard on our create ioctl(s). -Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#6 lore raw
Quoted 31 lines · Chris Wilson
Quoted 30 lines · Daniel Vetter
Quoted 28 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Why?

Quoted 7 lines
Quoted 5 lines

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

We already have an -EINVAL guard on our create ioctl(s).

Only for size == 0, not for non-aligned size, at least afaics (no coffee yet ...).
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#7 lore raw
Quoted 38 lines · Daniel Vetter
Quoted 37 lines · Chris Wilson
Quoted 35 lines · Daniel Vetter
Quoted 27 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

Quoted 4 lines
Quoted 1 line

We already have an -EINVAL guard on our create ioctl(s).

Only for size == 0, not for non-aligned size, at least afaics (no coffee yet ...).

The lack of caffeine was mine. I say if (blah(PAGE_SIZE)) return -EINVAL and was happy.

But as it happens it still implies that the BUG_ON in drm/drm_gem.c is relevant as is and we shouldn't be papering over it.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#8 lore raw
Quoted 44 lines · Chris Wilson
Quoted 38 lines · Daniel Vetter
Quoted 37 lines · Chris Wilson
Quoted 35 lines · Daniel Vetter
Quoted 27 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

Quoted 4 lines
Quoted 1 line

We already have an -EINVAL guard on our create ioctl(s).

Only for size == 0, not for non-aligned size, at least afaics (no coffee yet ...).

Quoted 7 lines

The lack of caffeine was mine. I say if (blah(PAGE_SIZE)) return -EINVAL and was happy.

But as it happens it still implies that the BUG_ON in drm/drm_gem.c is relevant as is and we shouldn't be papering over it.
-Chris

I was referring to internal, driver callers. If all internal callers round up, we may as well do it for them. If the earlier assertion is that we block bad sizes already at the IOCTL hanlder... then what's the problem again?

Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
#9 lore raw
Quoted 53 lines · Ben Widawsky
Quoted 44 lines · Chris Wilson
Quoted 38 lines · Daniel Vetter
Quoted 37 lines · Chris Wilson
Quoted 35 lines · Daniel Vetter
Quoted 27 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

Quoted 4 lines
Quoted 1 line

We already have an -EINVAL guard on our create ioctl(s).

Only for size == 0, not for non-aligned size, at least afaics (no coffee yet ...).

Quoted 7 lines

The lack of caffeine was mine. I say if (blah(PAGE_SIZE)) return -EINVAL and was happy.

But as it happens it still implies that the BUG_ON in drm/drm_gem.c is relevant as is and we shouldn't be papering over it.
-Chris

Quoted 4 lines

I was referring to internal, driver callers. If all internal callers round up, we may as well do it for them. If the earlier assertion is that we block bad sizes already at the IOCTL hanlder... then what's the problem again?

create2_ioctl should imo reject non-page-aligned requests. I guess we could dig through libdrm history and check whether that's also possible for the current create ioctl.
-Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
#10 lore raw
In reply to Ben Widawsky
Quoted 53 lines · Ben Widawsky
Quoted 44 lines · Chris Wilson
Quoted 38 lines · Daniel Vetter
Quoted 37 lines · Chris Wilson
Quoted 35 lines · Daniel Vetter
Quoted 27 lines · Ben Widawsky
Quoted 18 lines · Daniel Vetter
Quoted 17 lines · Ben Widawsky

DRM gets very mad when you request an object which occupies a partial page. As a DRM driver, i915 never really wants to anger DRM, and would always just want the rounding done for us.

---
 drivers/gpu/drm/i915/i915_gem.c | 2 ++
 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/i915_gem.c +0 -0
index 024e454..8cd1134 100644
@@ -4168,6 +4168,8 @@ struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
    struct address_space *mapping;
    gfp_t mask;
 
Quoted 7 lines

drivers/gpu/drm/i915/i915_gem.c struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev, · line 4172

Quoted 2 lines
drivers/gpu/drm/i915/i915_gem.c +2 -0
+   size = round_up(size, PAGE_SIZE);
+

Nope, if there's some code that doesn't do page-aligend bo allocations it needs to be fixed there. If you want throw a WARN_ON and early return in here.
-Daniel

Quoted 1 line

Why?

Because allocating a non-page aligned gem bo is a bug. All current in-kernel allocations are already aligned. I've thought that we also reject unaligned request from userspace but apparently we help out since forever (i.e. gem was merged). Might be worth a shot to turn that into an -EINVAL if libdrm does the right rounding ...

Quoted 4 lines
Quoted 1 line

We already have an -EINVAL guard on our create ioctl(s).

Only for size == 0, not for non-aligned size, at least afaics (no coffee yet ...).

Quoted 7 lines

The lack of caffeine was mine. I say if (blah(PAGE_SIZE)) return -EINVAL and was happy.

But as it happens it still implies that the BUG_ON in drm/drm_gem.c is relevant as is and we shouldn't be papering over it.
-Chris

Quoted 4 lines

I was referring to internal, driver callers. If all internal callers round up, we may as well do it for them. If the earlier assertion is that we block bad sizes already at the IOCTL hanlder... then what's the problem again?

You are suggesting we write:
size = PAGE_ALIGN(size);
BUG_ON(size & ~PAGE_MASK);

If you want to change something remove the BUG_ON, do the rounding and update all callers to not worry about page alignment. Tell me how many call sites require alteration? It seems to be an almost universal assumption that the size is page-aligned (despite my own attempts to get subpage allocations into the ABI).
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre