| Message ID | 0744671ac096a12f0d538906bd324efa71b11400.1657300532.git.geert@linux-m68k.org |
|---|---|
| State | New |
| Headers | show |
| Series | drm: Endianness fixes | expand |
On 2022-07-08 20:21, Geert Uytterhoeven wrote: > As of commit eae06120f1974e1a ("drm: refuse ADDFB2 ioctl for broken > bigendian drivers"), drivers must set the > quirk_addfb_prefer_host_byte_order quirk to make the drm_mode_addfb() > compat code work correctly on big-endian machines. > > While that works fine for big-endian XRGB8888 and ARGB8888, which are > mapped to the existing little-endian BGRX8888 and BGRA8888 formats, it > does not work for big-endian XRGB1555 and RGB565, as the latter are not > listed in the format database. > > Fix this by adding the missing formats. Limit this to big-endian > platforms, as there is currently no need to support these formats on > little-endian platforms. > > Fixes: 6960e6da9cec3f66 ("drm: fix drm_mode_addfb() on big endian machines.") > Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org> > --- > Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order > and supporting RGB565 or XRGB1555, but no one tried that on big-endian? > Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN > in cirrus_fb_create, so you cannot get a graphical text console. > > Do we need these definitions on little-endian platforms, too? > Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | > DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in > formats[]? The intention of DRM_FORMAT_HOST_* is that they are macros in include/drm/drm_fourcc.h which just map to little endian formats defined in drivers/gpu/drm/drm_fourcc.c. Since this is not possible for big endian hosts for XRGB1555 or RGB565 (or any other format with non-8-bit components), this isn't applicable here. It's also doubtful that Cirrus hardware would access these formats as big endian (drivers/gpu/drm/tiny/cirrus.c has no endianness references at all, and the hardware was surely designed for x86 first and foremost). Instead, fbcon (and user space) needs to convert to little endian when using DRM_FORMAT_HOST_{XRGB1555,RGB565} with the cirrus driver on big endian hosts.
Hi Michel, On Mon, Jul 11, 2022 at 5:23 PM Michel Dänzer <michel.daenzer@mailbox.org> wrote: > On 2022-07-08 20:21, Geert Uytterhoeven wrote: > > As of commit eae06120f1974e1a ("drm: refuse ADDFB2 ioctl for broken > > bigendian drivers"), drivers must set the > > quirk_addfb_prefer_host_byte_order quirk to make the drm_mode_addfb() > > compat code work correctly on big-endian machines. > > > > While that works fine for big-endian XRGB8888 and ARGB8888, which are > > mapped to the existing little-endian BGRX8888 and BGRA8888 formats, it > > does not work for big-endian XRGB1555 and RGB565, as the latter are not > > listed in the format database. > > > > Fix this by adding the missing formats. Limit this to big-endian > > platforms, as there is currently no need to support these formats on > > little-endian platforms. > > > > Fixes: 6960e6da9cec3f66 ("drm: fix drm_mode_addfb() on big endian machines.") > > Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org> > > --- > > Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order > > and supporting RGB565 or XRGB1555, but no one tried that on big-endian? > > Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN > > in cirrus_fb_create, so you cannot get a graphical text console. > > > > Do we need these definitions on little-endian platforms, too? > > Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | > > DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in > > formats[]? > > The intention of DRM_FORMAT_HOST_* is that they are macros in include/drm/drm_fourcc.h which just map to little endian formats defined in drivers/gpu/drm/drm_fourcc.c. Since this is not possible for big endian hosts for XRGB1555 or RGB565 (or any other format with non-8-bit components), this isn't applicable here. I read that as that you prefer to write "DRM_FORMAT_{XRGB1555,RGB565} | DRM_FORMAT_BIG_ENDIAN" in formats[]? > It's also doubtful that Cirrus hardware would access these formats as big endian (drivers/gpu/drm/tiny/cirrus.c has no endianness references at all, and the hardware was surely designed for x86 first and foremost). > > Instead, fbcon (and user space) needs to convert to little endian when using DRM_FORMAT_HOST_{XRGB1555,RGB565} with the cirrus driver on big endian hosts. Yeah, probably the cirrus driver can use some fixes... Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
On 2022-07-11 17:30, Geert Uytterhoeven wrote: > Hi Michel, > > On Mon, Jul 11, 2022 at 5:23 PM Michel Dänzer > <michel.daenzer@mailbox.org> wrote: >> On 2022-07-08 20:21, Geert Uytterhoeven wrote: >>> As of commit eae06120f1974e1a ("drm: refuse ADDFB2 ioctl for broken >>> bigendian drivers"), drivers must set the >>> quirk_addfb_prefer_host_byte_order quirk to make the drm_mode_addfb() >>> compat code work correctly on big-endian machines. >>> >>> While that works fine for big-endian XRGB8888 and ARGB8888, which are >>> mapped to the existing little-endian BGRX8888 and BGRA8888 formats, it >>> does not work for big-endian XRGB1555 and RGB565, as the latter are not >>> listed in the format database. >>> >>> Fix this by adding the missing formats. Limit this to big-endian >>> platforms, as there is currently no need to support these formats on >>> little-endian platforms. >>> >>> Fixes: 6960e6da9cec3f66 ("drm: fix drm_mode_addfb() on big endian machines.") >>> Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org> >>> --- >>> Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order >>> and supporting RGB565 or XRGB1555, but no one tried that on big-endian? >>> Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN >>> in cirrus_fb_create, so you cannot get a graphical text console. >>> >>> Do we need these definitions on little-endian platforms, too? >>> Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | >>> DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in >>> formats[]? >> >> The intention of DRM_FORMAT_HOST_* is that they are macros in include/drm/drm_fourcc.h which just map to little endian formats defined in drivers/gpu/drm/drm_fourcc.c. Since this is not possible for big endian hosts for XRGB1555 or RGB565 (or any other format with non-8-bit components), this isn't applicable here. > > I read that as that you prefer to write > "DRM_FORMAT_{XRGB1555,RGB565} | DRM_FORMAT_BIG_ENDIAN" in formats[]? In other drivers for hardware which can access these formats as big endian, yes. Note that AFAIK little if any user-space code uses DRM_FORMAT_BIG_ENDIAN yet though. >> It's also doubtful that Cirrus hardware would access these formats as big endian (drivers/gpu/drm/tiny/cirrus.c has no endianness references at all, and the hardware was surely designed for x86 first and foremost). >> >> Instead, fbcon (and user space) needs to convert to little endian when using DRM_FORMAT_HOST_{XRGB1555,RGB565} with the cirrus driver on big endian hosts. > > Yeah, probably the cirrus driver can use some fixes... I suspect the fix here would rather need to be in the DRM glue code for fbcon than in the driver. Or maybe some kind of byte-swapping helper(s) which can be used by drivers.
On Mon, Jul 11, 2022 at 05:30:30PM +0200, Geert Uytterhoeven wrote: > Hi Michel, > > > > Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order > > > and supporting RGB565 or XRGB1555, but no one tried that on big-endian? > > > Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN > > > in cirrus_fb_create, so you cannot get a graphical text console. > > > > > > Do we need these definitions on little-endian platforms, too? > > > Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | > > > DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in > > > formats[]? > > > > The intention of DRM_FORMAT_HOST_* is that they are macros in > > include/drm/drm_fourcc.h which just map to little endian formats > > defined in drivers/gpu/drm/drm_fourcc.c. Since this is not possible > > for big endian hosts for XRGB1555 or RGB565 (or any other format > > with non-8-bit components), this isn't applicable here. It IMHO is not applicable to any physical hardware. It's used by virtio-gpu where the supported format depends on the byte order (it is argb8888 in native byte order). Only virtual hardware can have that kind of behavior. And we can probably drop the DRM_FORMAT_HOST_* variants for 1555 and 565, they are not used anywhere. > I read that as that you prefer to write "DRM_FORMAT_{XRGB1555,RGB565} > | DRM_FORMAT_BIG_ENDIAN" in formats[]? Agree. > > It's also doubtful that Cirrus hardware would access these formats > > as big endian (drivers/gpu/drm/tiny/cirrus.c has no endianness > > references at all, and the hardware was surely designed for x86 > > first and foremost). Yes. qemu mimics physical cirrus hardware which uses little endian. > > Instead, fbcon (and user space) needs to convert to little endian > > when using DRM_FORMAT_HOST_{XRGB1555,RGB565} with the cirrus driver > > on big endian hosts. Well, the cirrus driver uses shadow framebuffers anyway (the only workable approach given it has 4M vram only), and it also supports converting formats on-the-fly when copying from shadow to vram. So adding support for bigendian formats to the driver shouldn't be much of a problem. The vram will continue to run in little endian RGB565, the shadow will be big endian RGB565, and the driver must byteswap when copying. > Yeah, probably the cirrus driver can use some fixes... I'd call it improvements. It's not like the cirrus driver is broken. take care, Gerd
Hi Gerd, On Tue, Jul 12, 2022 at 9:47 AM Gerd Hoffmann <kraxel@redhat.com> wrote: > On Mon, Jul 11, 2022 at 05:30:30PM +0200, Geert Uytterhoeven wrote: > > > > Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order > > > > and supporting RGB565 or XRGB1555, but no one tried that on big-endian? > > > > Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN > > > > in cirrus_fb_create, so you cannot get a graphical text console. > > > > > > > > Do we need these definitions on little-endian platforms, too? > > > > Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | > > > > DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in > > > > formats[]? > > > > > > The intention of DRM_FORMAT_HOST_* is that they are macros in > > > include/drm/drm_fourcc.h which just map to little endian formats > > > defined in drivers/gpu/drm/drm_fourcc.c. Since this is not possible > > > for big endian hosts for XRGB1555 or RGB565 (or any other format > > > with non-8-bit components), this isn't applicable here. > > It IMHO is not applicable to any physical hardware. It's used by > virtio-gpu where the supported format depends on the byte order > (it is argb8888 in native byte order). Only virtual hardware can > have that kind of behavior. > > And we can probably drop the DRM_FORMAT_HOST_* variants for 1555 and > 565, they are not used anywhere. Atari DRM supports (big-endian) RGB565, so it uses DRM_FORMAT_HOST_RGB565. The alternative is to drop the quirk_addfb_prefer_host_byte_order requirement on big-endian, and always use a little-endian RGB565 shadow frame buffer, at the expense of never being able to get rid of the copying and byteswapping. [Cirrus discussion removed] Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
Hi, > So adding support for bigendian formats to the driver shouldn't be > much of a problem. The vram will continue to run in little endian > RGB565, the shadow will be big endian RGB565, and the driver must > byteswap when copying. For completeness: The other obvious option (for fbcon) would be to handle the byteswapping in the generic drm fbdev emulation, which would have the advantage that it would be more generic and would not depend on the drm driver supporting the bigendian rgb565 formats ... take care, Gerd
On Tue, Jul 12, 2022 at 10:01:15AM +0200, Geert Uytterhoeven wrote: > Hi Gerd, > > > It IMHO is not applicable to any physical hardware. It's used by > > virtio-gpu where the supported format depends on the byte order > > (it is argb8888 in native byte order). Only virtual hardware can > > have that kind of behavior. > > > > And we can probably drop the DRM_FORMAT_HOST_* variants for 1555 and > > 565, they are not used anywhere. > > Atari DRM supports (big-endian) RGB565, so it uses > DRM_FORMAT_HOST_RGB565. Fixed big endian should use 'DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN'. As described above DRM_FORMAT_HOST_RGB565 means bigendian on bigendian hosts and little endian on little endian hosts. Which is not correct when your hardware does big endian no matter what. take care, Gerd
Hi Gerd, On Tue, Jul 12, 2022 at 10:39 AM Gerd Hoffmann <kraxel@redhat.com> wrote: > On Tue, Jul 12, 2022 at 10:01:15AM +0200, Geert Uytterhoeven wrote: > > > It IMHO is not applicable to any physical hardware. It's used by > > > virtio-gpu where the supported format depends on the byte order > > > (it is argb8888 in native byte order). Only virtual hardware can > > > have that kind of behavior. > > > > > > And we can probably drop the DRM_FORMAT_HOST_* variants for 1555 and > > > 565, they are not used anywhere. > > > > Atari DRM supports (big-endian) RGB565, so it uses > > DRM_FORMAT_HOST_RGB565. > > Fixed big endian should use 'DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN'. True. > As described above DRM_FORMAT_HOST_RGB565 means bigendian on bigendian > hosts and little endian on little endian hosts. Which is not correct > when your hardware does big endian no matter what. But (a) drm_driver_legacy_fb_format() uses DRM_FORMAT_HOST_RGB565 if quirk_addfb_prefer_host_byte_order is set, and (b) quirk_addfb_prefer_host_byte_order must be set on big-endian as per commit eae06120f1974e1a ("drm: refuse ADDFB2 ioctl for broken bigendian drivers"). Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
> > As described above DRM_FORMAT_HOST_RGB565 means bigendian on bigendian > > hosts and little endian on little endian hosts. Which is not correct > > when your hardware does big endian no matter what. > > But (a) drm_driver_legacy_fb_format() uses DRM_FORMAT_HOST_RGB565 > if quirk_addfb_prefer_host_byte_order is set, Ah, right. Missed that in 'git grep' output. Given that traditional fbdev behavior is to expect native byte order using DRM_FORMAT_HOST_RGB565 there makes sense indeed. Scratch my comment about it being unused then ;) thanks, Gerd
On 2022-07-12 11:03, Gerd Hoffmann wrote: >>> As described above DRM_FORMAT_HOST_RGB565 means bigendian on bigendian >>> hosts and little endian on little endian hosts. Which is not correct >>> when your hardware does big endian no matter what. >> >> But (a) drm_driver_legacy_fb_format() uses DRM_FORMAT_HOST_RGB565 >> if quirk_addfb_prefer_host_byte_order is set, > > Ah, right. Missed that in 'git grep' output. Given that traditional > fbdev behavior is to expect native byte order using > DRM_FORMAT_HOST_RGB565 there makes sense indeed. > > Scratch my comment about it being unused then ;) DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN is still what the driver should use conceptually, and should match DRM_FORMAT_HOST_RGB565 in drm_driver_legacy_fb_format on a big endian host (which is presumably always the case for the atari driver).
Hi Michel, On Tue, Jul 12, 2022 at 11:09 AM Michel Dänzer <michel.daenzer@mailbox.org> wrote: > On 2022-07-12 11:03, Gerd Hoffmann wrote: > >>> As described above DRM_FORMAT_HOST_RGB565 means bigendian on bigendian > >>> hosts and little endian on little endian hosts. Which is not correct > >>> when your hardware does big endian no matter what. > >> > >> But (a) drm_driver_legacy_fb_format() uses DRM_FORMAT_HOST_RGB565 > >> if quirk_addfb_prefer_host_byte_order is set, > > > > Ah, right. Missed that in 'git grep' output. Given that traditional > > fbdev behavior is to expect native byte order using > > DRM_FORMAT_HOST_RGB565 there makes sense indeed. > > > > Scratch my comment about it being unused then ;) > > DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN is still what the driver should use conceptually, and should match DRM_FORMAT_HOST_RGB565 in drm_driver_legacy_fb_format on a big endian host (which is presumably always the case for the atari driver). Sure, I'll update the patch accordingly. Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds
diff --git a/drivers/gpu/drm/drm_fourcc.c b/drivers/gpu/drm/drm_fourcc.c index e09331bb3bc73f21..16a854c9e07fa4b0 100644 --- a/drivers/gpu/drm/drm_fourcc.c +++ b/drivers/gpu/drm/drm_fourcc.c @@ -190,6 +190,10 @@ const struct drm_format_info *__drm_format_info(u32 format) { .format = DRM_FORMAT_BGRA5551, .depth = 15, .num_planes = 1, .cpp = { 2, 0, 0 }, .hsub = 1, .vsub = 1, .has_alpha = true }, { .format = DRM_FORMAT_RGB565, .depth = 16, .num_planes = 1, .cpp = { 2, 0, 0 }, .hsub = 1, .vsub = 1 }, { .format = DRM_FORMAT_BGR565, .depth = 16, .num_planes = 1, .cpp = { 2, 0, 0 }, .hsub = 1, .vsub = 1 }, +#ifdef __BIG_ENDIAN + { .format = DRM_FORMAT_HOST_XRGB1555, .depth = 15, .num_planes = 1, .cpp = { 2, 0, 0 }, .hsub = 1, .vsub = 1 }, + { .format = DRM_FORMAT_HOST_RGB565, .depth = 16, .num_planes = 1, .cpp = { 2, 0, 0 }, .hsub = 1, .vsub = 1 }, +#endif { .format = DRM_FORMAT_RGB888, .depth = 24, .num_planes = 1, .cpp = { 3, 0, 0 }, .hsub = 1, .vsub = 1 }, { .format = DRM_FORMAT_BGR888, .depth = 24, .num_planes = 1, .cpp = { 3, 0, 0 }, .hsub = 1, .vsub = 1 }, { .format = DRM_FORMAT_XRGB8888, .depth = 24, .num_planes = 1, .cpp = { 4, 0, 0 }, .hsub = 1, .vsub = 1 },
As of commit eae06120f1974e1a ("drm: refuse ADDFB2 ioctl for broken bigendian drivers"), drivers must set the quirk_addfb_prefer_host_byte_order quirk to make the drm_mode_addfb() compat code work correctly on big-endian machines. While that works fine for big-endian XRGB8888 and ARGB8888, which are mapped to the existing little-endian BGRX8888 and BGRA8888 formats, it does not work for big-endian XRGB1555 and RGB565, as the latter are not listed in the format database. Fix this by adding the missing formats. Limit this to big-endian platforms, as there is currently no need to support these formats on little-endian platforms. Fixes: 6960e6da9cec3f66 ("drm: fix drm_mode_addfb() on big endian machines.") Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org> --- Cirrus is the only driver setting quirk_addfb_prefer_host_byte_order and supporting RGB565 or XRGB1555, but no one tried that on big-endian? Cirrus does not support DRM_FORMAT_RGB565 | DRM_FORMAT_BIG_ENDIAN in cirrus_fb_create, so you cannot get a graphical text console. Do we need these definitions on little-endian platforms, too? Would it be better to use "DRM_FORMAT_{XRGB1555,RGB565} | DRM_FORMAT_BIG_ENDIAN" instead of "DRM_FORMAT_HOST_{XRGB1555,RGB565}" in formats[]? Alternative, callers could filter out the DRM_FORMAT_BIG_ENDIAN flag when calling {,__}drm_format_info(). Advantages: - No need to have separate entries with DRM_FORMAT_BIG_ENDIAN set. Disadvantages: - {,__}drm_format_info() returns a pointer to a const object, whose .format field won't have the DRM_FORMAT_BIG_ENDIAN flag set, complicating callers, - All callers need to be updated, - It is difficult to know which big-endian formats are really supported, especially as only a few are needed. --- drivers/gpu/drm/drm_fourcc.c | 4 ++++ 1 file changed, 4 insertions(+)