@@ -508,14 +508,14 @@ struct drm_mode_crtc_page_flip {/* create a dumb scanout buffer */structdrm_mode_create_dumb{-uint32_theight;-uint32_twidth;-uint32_tbpp;-uint32_tflags;+__u32height;+__u32width;+__u32bpp;+__u32flags;/* handle, pitch, size will be returned */-uint32_thandle;-uint32_tpitch;-uint64_tsize;+__u32handle;+__u32pitch;+__u64size;};/* set up for mmap of a dumb scanout buffer */
@@ -508,14 +508,14 @@ struct drm_mode_crtc_page_flip {/* create a dumb scanout buffer */structdrm_mode_create_dumb{-uint32_theight;-uint32_twidth;-uint32_tbpp;-uint32_tflags;+__u32height;+__u32width;+__u32bpp;+__u32flags;/* handle, pitch, size will be returned */-uint32_thandle;-uint32_tpitch;-uint64_tsize;+__u32handle;+__u32pitch;+__u64size;};/* set up for mmap of a dumb scanout buffer */
From: Mikko Rapeli <hidden> Date: 2015-10-15 13:48:48
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
-Mikko
@@ -508,14 +508,14 @@ struct drm_mode_crtc_page_flip {/* create a dumb scanout buffer */structdrm_mode_create_dumb{-uint32_theight;-uint32_twidth;-uint32_tbpp;-uint32_tflags;+__u32height;+__u32width;+__u32bpp;+__u32flags;/* handle, pitch, size will be returned */-uint32_thandle;-uint32_tpitch;-uint64_tsize;+__u32handle;+__u32pitch;+__u64size;};/* set up for mmap of a dumb scanout buffer */
From: Emil Velikov <hidden> Date: 2015-10-21 15:09:33
Hi Alex,
On 15 October 2015 at 14:48, Mikko Rapeli [off-list ref] wrote:
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
quoted
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
Based of the reply from Mikko, can you please elaborate your concern ?
Are you thinking about some corner case where this may cause breakage,
or it's solely on stylistic point of view ?
Over the last few years we've been doing some ad-hoc 'synchronisation'
with the headers in libdrm, and this will get us one step closer to
doing things properly.
Fwiw I fully support these changes, as does Gustavo for exynos and Rob for msm.
Reviewed-by: Emil Velikov <redacted>
Thanks
Emil
From: Alex Deucher <hidden> Date: 2015-10-21 15:18:40
On Wed, Oct 21, 2015 at 11:09 AM, Emil Velikov [off-list ref] wrote:
Hi Alex,
On 15 October 2015 at 14:48, Mikko Rapeli [off-list ref] wrote:
quoted
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
quoted
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
Based of the reply from Mikko, can you please elaborate your concern ?
Are you thinking about some corner case where this may cause breakage,
or it's solely on stylistic point of view ?
Style.
Over the last few years we've been doing some ad-hoc 'synchronisation'
with the headers in libdrm, and this will get us one step closer to
doing things properly.
How does this affect libdrm one way or another?
Alex
Fwiw I fully support these changes, as does Gustavo for exynos and Rob for msm.
Reviewed-by: Emil Velikov <redacted>
Thanks
Emil
From: Emil Velikov <hidden> Date: 2015-10-21 16:21:32
On 21 October 2015 at 16:18, Alex Deucher [off-list ref] wrote:
On Wed, Oct 21, 2015 at 11:09 AM, Emil Velikov [off-list ref] wrote:
quoted
Hi Alex,
On 15 October 2015 at 14:48, Mikko Rapeli [off-list ref] wrote:
quoted
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
quoted
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
Based of the reply from Mikko, can you please elaborate your concern ?
Are you thinking about some corner case where this may cause breakage,
or it's solely on stylistic point of view ?
Style.
In that case shouldn't the kernel coding style (not to mention the
issues with stdint.h types Linus has brought a while back) be
considered/prevail ?
quoted
Over the last few years we've been doing some ad-hoc 'synchronisation'
with the headers in libdrm, and this will get us one step closer to
doing things properly.
How does this affect libdrm one way or another?
In a strange way. Some (most?) distributions do not ship the uapi
headers from the kernel, but rely on the ones in libdrm. Due to
various issues (as addressed in the series) we have not synchronised
the headers, thus users such as the gallium radeon winsys has ended up
redefining/duplicating internally.
Thanks
Emil
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
From: Alex Deucher <hidden> Date: 2015-10-21 16:27:31
On Wed, Oct 21, 2015 at 12:21 PM, Emil Velikov [off-list ref] wrote:
On 21 October 2015 at 16:18, Alex Deucher [off-list ref] wrote:
quoted
On Wed, Oct 21, 2015 at 11:09 AM, Emil Velikov [off-list ref] wrote:
quoted
Hi Alex,
On 15 October 2015 at 14:48, Mikko Rapeli [off-list ref] wrote:
quoted
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
quoted
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
Based of the reply from Mikko, can you please elaborate your concern ?
Are you thinking about some corner case where this may cause breakage,
or it's solely on stylistic point of view ?
Style.
In that case shouldn't the kernel coding style (not to mention the
issues with stdint.h types Linus has brought a while back) be
considered/prevail ?
I don't really have a strong opinion one way or another at this point.
The types just look ugly.
quoted
quoted
Over the last few years we've been doing some ad-hoc 'synchronisation'
with the headers in libdrm, and this will get us one step closer to
doing things properly.
How does this affect libdrm one way or another?
In a strange way. Some (most?) distributions do not ship the uapi
headers from the kernel, but rely on the ones in libdrm. Due to
various issues (as addressed in the series) we have not synchronised
the headers, thus users such as the gallium radeon winsys has ended up
redefining/duplicating internally.
The problem is that you generally need the latest headers to support
the latest features in the user mode stacks, so if the user tries to
build against an older kernel, they will be missing definitions.
Alex
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel
From: Emil Velikov <hidden> Date: 2015-10-21 17:42:13
On 21 October 2015 at 17:27, Alex Deucher [off-list ref] wrote:
On Wed, Oct 21, 2015 at 12:21 PM, Emil Velikov [off-list ref] wrote:
quoted
On 21 October 2015 at 16:18, Alex Deucher [off-list ref] wrote:
quoted
On Wed, Oct 21, 2015 at 11:09 AM, Emil Velikov [off-list ref] wrote:
quoted
Hi Alex,
On 15 October 2015 at 14:48, Mikko Rapeli [off-list ref] wrote:
quoted
On Thu, Oct 15, 2015 at 09:32:10AM -0400, Alex Deucher wrote:
quoted
On Thu, Oct 15, 2015 at 1:55 AM, Mikko Rapeli [off-list ref] wrote:
quoted
Fixes userspace compilation error:
drm/drm_mode.h:472:2: error: unknown type name ‘uint32_t’
Signed-off-by: Mikko Rapeli <redacted>
NACK on all these type conversions. This has not been a problem for
years and years and the result looks terrible.
Documentation/CodingStyle, section 5
(e) Types safe for use in userspace.
In certain structures which are visible to userspace, we cannot
require C99 types and cannot use the 'u32' form above. Thus, we
use __u32 and similar types in all structures which are shared
with userspace.
I have only been looking at kernel headers from userspace occationally in
the past 10 years and had a several cases where the provided headers did
not compile when included into trivial programs trying to use the structs
for an ioctl() for example. This long lasting problem triggered me to write
a test for this and provide these fixes too. In previous reviews usage
of <stdint.h> and its types in kernel headers was already NACK'ed
so I changed several places from uint32_t's to __u32.
With these changes it is btw trivial now to add a grep test the there
are no uint32_t's in include/uapi/ anymore, thus enforcing that coding style
rule.
Based of the reply from Mikko, can you please elaborate your concern ?
Are you thinking about some corner case where this may cause breakage,
or it's solely on stylistic point of view ?
Style.
In that case shouldn't the kernel coding style (not to mention the
issues with stdint.h types Linus has brought a while back) be
considered/prevail ?
I don't really have a strong opinion one way or another at this point.
The types just look ugly.
Fwiw I don't find them too appealing either. But the alternative did
not fare so well :(
quoted
quoted
quoted
Over the last few years we've been doing some ad-hoc 'synchronisation'
with the headers in libdrm, and this will get us one step closer to
doing things properly.
How does this affect libdrm one way or another?
In a strange way. Some (most?) distributions do not ship the uapi
headers from the kernel, but rely on the ones in libdrm. Due to
various issues (as addressed in the series) we have not synchronised
the headers, thus users such as the gallium radeon winsys has ended up
redefining/duplicating internally.
The problem is that you generally need the latest headers to support
the latest features in the user mode stacks, so if the user tries to
build against an older kernel, they will be missing definitions.
Pretty much that's why we have the loose release criteria for libdrm.
Get the functionality upstream (mainline or -next), pull the changes
(sync) and release. As the sync has been busted for a while, people
are forced to do unpleasant things.
-Emil