| Message ID | 1308670579-15138-1-git-send-email-laurent.pinchart@ideasonboard.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Headers | show |
Hi Geert, On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > On Tue, Jun 21, 2011 at 17:36, Laurent Pinchart wrote: > > +The following types and visuals are supported. > > + > > +- FB_TYPE_PACKED_PIXELS > > + > > +- FB_TYPE_PLANES > > You forgot FB_TYPE_INTERLEAVED_PLANES, FB_TYPE_TEXT, and > FB_TYPE_VGA_PLANES. Ah, that's the "feel free to extend the API doc" :-) To be honest, I don't know how they work. That's why I haven't documented them. > > +The FOURCC-based API replaces format descriptions by four character > > codes +(FOURCC). FOURCCs are abstract identifiers that uniquely define a > > format +without explicitly describing it. This is the only API that > > supports YUV +formats. Drivers are also encouraged to implement the > > FOURCC-based API for RGB +and grayscale formats. > > + > > +Drivers that support the FOURCC-based API report this capability by > > setting +the FB_CAP_FOURCC bit in the fb_fix_screeninfo capabilities > > field. + > > +FOURCC definitions are located in the linux/videodev2.h header. However, > > and +despite starting with the V4L2_PIX_FMT_prefix, they are not > > restricted to V4L2 +and don't require usage of the V4L2 subsystem. > > FOURCC documentation is +available in > > Documentation/DocBook/v4l/pixfmt.xml. > > + > > +To select a format, applications set the FB_VMODE_FOURCC bit in the > > +fb_var_screeninfo vmode field, and set the fourcc field to the desired > > FOURCC. +The bits_per_pixel, red, green, blue, transp and nonstd fields > > must be set to +0 by applications and ignored by drivers. Note that the > > grayscale and fourcc +fields share the same memory location. Application > > must thus not set the +grayscale field to 0. > > These are the only parts I don't like: (ab)using the vmode field (this > isn't really a vmode flag), and the union of grayscale and fourcc (avoid > unions where possible). I've proposed adding a FB_NONSTD_FORMAT bit to the nonstd field as a FOURCC mode indicator in my initial RFC. Florian Tobias Schandinat wasn't very happy with that, and proposed using the vmode field instead. Given that there's virtually no fbdev documentation, whether the vmode field and/or nonstd field are good fit for a FOURCC mode indicator is subject to interpretation. > What about storing the FOURCC value in nonstd instead? Wouldn't that be a union of nonstd and fourcc ? :-) FOURCC-based format setting will be a standard fbdev API, I'm not very keen on storing it in the nonstd field without a union. > As FOURCC values are always 4 ASCII characters (hence all 4 bytes must > be non-zero), I don't think there are any conflicts with existing values of > nonstd. To make it even safer and easier to parse, you could set bit 31 of > nonstd as a FOURCC indicator. I would then create a union between nonstd and fourcc, and document nonstd as being used for the legacy API only. Most existing drivers use a couple of nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and uses bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd & 0xff000000 != 0 could be used as a FOURCC mode test. This assumes that FOURCCs will never have their last character set to '\0'. Is that a safe assumption for the future ?
Hi Geert, Laurent, On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > Hi Geert, > > On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: >> On Tue, Jun 21, 2011 at 17:36, Laurent Pinchart wrote: >>> +The FOURCC-based API replaces format descriptions by four character >>> codes +(FOURCC). FOURCCs are abstract identifiers that uniquely define a >>> format +without explicitly describing it. This is the only API that >>> supports YUV +formats. Drivers are also encouraged to implement the >>> FOURCC-based API for RGB +and grayscale formats. >>> + >>> +Drivers that support the FOURCC-based API report this capability by >>> setting +the FB_CAP_FOURCC bit in the fb_fix_screeninfo capabilities >>> field. + >>> +FOURCC definitions are located in the linux/videodev2.h header. However, >>> and +despite starting with the V4L2_PIX_FMT_prefix, they are not >>> restricted to V4L2 +and don't require usage of the V4L2 subsystem. >>> FOURCC documentation is +available in >>> Documentation/DocBook/v4l/pixfmt.xml. >>> + >>> +To select a format, applications set the FB_VMODE_FOURCC bit in the >>> +fb_var_screeninfo vmode field, and set the fourcc field to the desired >>> FOURCC. +The bits_per_pixel, red, green, blue, transp and nonstd fields >>> must be set to +0 by applications and ignored by drivers. Note that the >>> grayscale and fourcc +fields share the same memory location. Application >>> must thus not set the +grayscale field to 0. >> >> These are the only parts I don't like: (ab)using the vmode field (this >> isn't really a vmode flag), and the union of grayscale and fourcc (avoid >> unions where possible). > > I've proposed adding a FB_NONSTD_FORMAT bit to the nonstd field as a FOURCC > mode indicator in my initial RFC. Florian Tobias Schandinat wasn't very happy > with that, and proposed using the vmode field instead. > > Given that there's virtually no fbdev documentation, whether the vmode field > and/or nonstd field are good fit for a FOURCC mode indicator is subject to > interpretation. The reason for my suggestion is that the vmode field is accepted to contain only flags and at least to me there is no hard line what is part of the video mode and what is not. In contrast the nonstd field is already used in a lot of different (incompatible) ways. I think if we only use the nonstd field for handling FOURCC it is likely that some problems will appear. >> What about storing the FOURCC value in nonstd instead? > > Wouldn't that be a union of nonstd and fourcc ? :-) FOURCC-based format > setting will be a standard fbdev API, I'm not very keen on storing it in the > nonstd field without a union. > >> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must >> be non-zero), I don't think there are any conflicts with existing values of >> nonstd. To make it even safer and easier to parse, you could set bit 31 of >> nonstd as a FOURCC indicator. > > I would then create a union between nonstd and fourcc, and document nonstd as > being used for the legacy API only. Most existing drivers use a couple of > nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and uses > bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& > 0xff000000 != 0 could be used as a FOURCC mode test. > > This assumes that FOURCCs will never have their last character set to '\0'. Is > that a safe assumption for the future ? Yes, I think. The information I found indicates that space should be used for padding, so a \0 shouldn't exist. I think using only the nonstd field and requiring applications to check the capabilities would be possible, although not fool proof ;) Great work, Laurent, do you have plans to modify fbset to allow using this format API from the command line? Regards, Florian Tobias Schandinat -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Florian, On Wednesday 22 June 2011 07:45:45 Florian Tobias Schandinat wrote: > On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > > On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > >> On Tue, Jun 21, 2011 at 17:36, Laurent Pinchart wrote: > >>> +The FOURCC-based API replaces format descriptions by four character > >>> codes +(FOURCC). FOURCCs are abstract identifiers that uniquely define > >>> a format +without explicitly describing it. This is the only API that > >>> supports YUV +formats. Drivers are also encouraged to implement the > >>> FOURCC-based API for RGB +and grayscale formats. > >>> + > >>> +Drivers that support the FOURCC-based API report this capability by > >>> setting +the FB_CAP_FOURCC bit in the fb_fix_screeninfo capabilities > >>> field. + > >>> +FOURCC definitions are located in the linux/videodev2.h header. > >>> However, and +despite starting with the V4L2_PIX_FMT_prefix, they are > >>> not restricted to V4L2 +and don't require usage of the V4L2 subsystem. > >>> FOURCC documentation is +available in > >>> Documentation/DocBook/v4l/pixfmt.xml. > >>> + > >>> +To select a format, applications set the FB_VMODE_FOURCC bit in the > >>> +fb_var_screeninfo vmode field, and set the fourcc field to the desired > >>> FOURCC. +The bits_per_pixel, red, green, blue, transp and nonstd fields > >>> must be set to +0 by applications and ignored by drivers. Note that the > >>> grayscale and fourcc +fields share the same memory location. > >>> Application must thus not set the +grayscale field to 0. > >> > >> These are the only parts I don't like: (ab)using the vmode field (this > >> isn't really a vmode flag), and the union of grayscale and fourcc (avoid > >> unions where possible). > > > > I've proposed adding a FB_NONSTD_FORMAT bit to the nonstd field as a > > FOURCC mode indicator in my initial RFC. Florian Tobias Schandinat > > wasn't very happy with that, and proposed using the vmode field instead. > > > > Given that there's virtually no fbdev documentation, whether the vmode > > field and/or nonstd field are good fit for a FOURCC mode indicator is > > subject to interpretation. > > The reason for my suggestion is that the vmode field is accepted to contain > only flags and at least to me there is no hard line what is part of the > video mode and what is not. Lacks of documentation indeed makes that line fuzzy. I really hope that api.txt will be extended to cover the full fbdev API :-) > In contrast the nonstd field is already used in a lot of different > (incompatible) ways. I think if we only use the nonstd field for handling > FOURCC it is likely that some problems will appear. > > >> What about storing the FOURCC value in nonstd instead? > > > > Wouldn't that be a union of nonstd and fourcc ? :-) FOURCC-based format > > setting will be a standard fbdev API, I'm not very keen on storing it in > > the nonstd field without a union. > > > >> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must > >> be non-zero), I don't think there are any conflicts with existing values > >> of nonstd. To make it even safer and easier to parse, you could set bit > >> 31 of nonstd as a FOURCC indicator. > > > > I would then create a union between nonstd and fourcc, and document > > nonstd as being used for the legacy API only. Most existing drivers use > > a couple of nonstd bits only. The driver that (ab)uses nonstd the most > > is pxafb and uses bits 22:0. Bits 31:24 are never used as far as I can > > tell, so nonstd& 0xff000000 != 0 could be used as a FOURCC mode test. > > > > This assumes that FOURCCs will never have their last character set to > > '\0'. Is that a safe assumption for the future ? > > Yes, I think. The information I found indicates that space should be used > for padding, so a \0 shouldn't exist. > I think using only the nonstd field and requiring applications to check the > capabilities would be possible, although not fool proof ;) > > Great work, Laurent, do you have plans to modify fbset to allow using this > format API from the command line? Once we agree on an API, I will implement it in a driver and update fbset.
On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat <FlorianSchandinat@gmx.de> wrote: > On 06/21/2011 10:31 PM, Laurent Pinchart wrote: >> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: >>> On Tue, Jun 21, 2011 at 17:36, Laurent Pinchart wrote: >>>> +The FOURCC-based API replaces format descriptions by four character >>>> codes +(FOURCC). FOURCCs are abstract identifiers that uniquely define a >>>> format +without explicitly describing it. This is the only API that >>>> supports YUV +formats. Drivers are also encouraged to implement the >>>> FOURCC-based API for RGB +and grayscale formats. >>>> + >>>> +Drivers that support the FOURCC-based API report this capability by >>>> setting +the FB_CAP_FOURCC bit in the fb_fix_screeninfo capabilities >>>> field. + >>>> +FOURCC definitions are located in the linux/videodev2.h header. >>>> However, >>>> and +despite starting with the V4L2_PIX_FMT_prefix, they are not >>>> restricted to V4L2 +and don't require usage of the V4L2 subsystem. >>>> FOURCC documentation is +available in >>>> Documentation/DocBook/v4l/pixfmt.xml. >>>> + >>>> +To select a format, applications set the FB_VMODE_FOURCC bit in the >>>> +fb_var_screeninfo vmode field, and set the fourcc field to the desired >>>> FOURCC. +The bits_per_pixel, red, green, blue, transp and nonstd fields >>>> must be set to +0 by applications and ignored by drivers. Note that the >>>> grayscale and fourcc +fields share the same memory location. Application >>>> must thus not set the +grayscale field to 0. >>> >>> These are the only parts I don't like: (ab)using the vmode field (this >>> isn't really a vmode flag), and the union of grayscale and fourcc (avoid >>> unions where possible). >> >> I've proposed adding a FB_NONSTD_FORMAT bit to the nonstd field as a >> FOURCC >> mode indicator in my initial RFC. Florian Tobias Schandinat wasn't very >> happy >> with that, and proposed using the vmode field instead. >> >> Given that there's virtually no fbdev documentation, whether the vmode >> field >> and/or nonstd field are good fit for a FOURCC mode indicator is subject to >> interpretation. > > The reason for my suggestion is that the vmode field is accepted to contain > only flags and at least to me there is no hard line what is part of the > video mode and what is not. In contrast the nonstd field is already used in > a lot of different (incompatible) ways. I think if we only use the nonstd > field for handling FOURCC it is likely that some problems will appear. > >>> What about storing the FOURCC value in nonstd instead? >> >> Wouldn't that be a union of nonstd and fourcc ? :-) FOURCC-based format >> setting will be a standard fbdev API, I'm not very keen on storing it in >> the >> nonstd field without a union. >> >>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must >>> be non-zero), I don't think there are any conflicts with existing values >>> of >>> nonstd. To make it even safer and easier to parse, you could set bit 31 >>> of >>> nonstd as a FOURCC indicator. >> >> I would then create a union between nonstd and fourcc, and document nonstd >> as >> being used for the legacy API only. Most existing drivers use a couple of >> nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and >> uses >> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& >> 0xff000000 != 0 could be used as a FOURCC mode test. >> >> This assumes that FOURCCs will never have their last character set to >> '\0'. Is >> that a safe assumption for the future ? > > Yes, I think. The information I found indicates that space should be used > for padding, so a \0 shouldn't exist. > I think using only the nonstd field and requiring applications to check the > capabilities would be possible, although not fool proof ;) So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC if any of them is set. Nicely backwards compatible, as sane drivers should reject nonstd values they don't support (apps _will_ start filling in FOURCC values ignoring capabilities, won't they?). 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 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: > On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat > <FlorianSchandinat@gmx.de> wrote: > > On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > >> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > >>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must > >>> be non-zero), I don't think there are any conflicts with existing values > >>> of > >>> nonstd. To make it even safer and easier to parse, you could set bit 31 > >>> of > >>> nonstd as a FOURCC indicator. > >> > >> I would then create a union between nonstd and fourcc, and document nonstd > >> as > >> being used for the legacy API only. Most existing drivers use a couple of > >> nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and > >> uses > >> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& > >> 0xff000000 != 0 could be used as a FOURCC mode test. > >> > >> This assumes that FOURCCs will never have their last character set to > >> '\0'. Is > >> that a safe assumption for the future ? > > > > Yes, I think. The information I found indicates that space should be used > > for padding, so a \0 shouldn't exist. > > I think using only the nonstd field and requiring applications to check the > > capabilities would be possible, although not fool proof ;) > > So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC if > any of them is set. > > Nicely backwards compatible, as sane drivers should reject nonstd values they > don't support (apps _will_ start filling in FOURCC values ignoring capabilities, > won't they?). > That seems like a reasonable case, but if we're going to do that then certainly the nonstd bit encoding needs to be documented and treated as a hard ABI. I'm not so sure about the if any bit in the upper byte is set assume FOURCC case though, there will presumably be other users in the future that will want bits for themselves, too. What exactly was the issue with having a FOURCC capability bit in the upper byte? -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Jun 24, 2011 at 08:19, Paul Mundt <lethal@linux-sh.org> wrote: > On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: >> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat >> <FlorianSchandinat@gmx.de> wrote: >> > On 06/21/2011 10:31 PM, Laurent Pinchart wrote: >> >> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: >> >>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must >> >>> be non-zero), I don't think there are any conflicts with existing values >> >>> of >> >>> nonstd. To make it even safer and easier to parse, you could set bit 31 >> >>> of >> >>> nonstd as a FOURCC indicator. >> >> >> >> I would then create a union between nonstd and fourcc, and document nonstd >> >> as >> >> being used for the legacy API only. Most existing drivers use a couple of >> >> nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and >> >> uses >> >> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& >> >> 0xff000000 != 0 could be used as a FOURCC mode test. >> >> >> >> This assumes that FOURCCs will never have their last character set to >> >> '\0'. Is >> >> that a safe assumption for the future ? >> > >> > Yes, I think. The information I found indicates that space should be used >> > for padding, so a \0 shouldn't exist. >> > I think using only the nonstd field and requiring applications to check the >> > capabilities would be possible, although not fool proof ;) >> >> So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC if >> any of them is set. >> >> Nicely backwards compatible, as sane drivers should reject nonstd values they >> don't support (apps _will_ start filling in FOURCC values ignoring capabilities, >> won't they?). >> > That seems like a reasonable case, but if we're going to do that then > certainly the nonstd bit encoding needs to be documented and treated as a > hard ABI. > > I'm not so sure about the if any bit in the upper byte is set assume > FOURCC case though, there will presumably be other users in the future > that will want bits for themselves, too. What exactly was the issue with > having a FOURCC capability bit in the upper byte? That indeed gives less issues (as long as you don't stuff 8-bit ASCII in the MSB) and more bits for tradiditional nonstd users. BTW, after giving it some more thought: what does the FB_CAP_FOURCC buys us? It's not like all drivers will support whatever random FOURCC mode you give them, so you have to check whether it's supported by doing a set_var first. 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 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 06/24/2011 06:55 PM, Geert Uytterhoeven wrote: > On Fri, Jun 24, 2011 at 08:19, Paul Mundt<lethal@linux-sh.org> wrote: >> On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: >>> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat >>> <FlorianSchandinat@gmx.de> wrote: >>>> On 06/21/2011 10:31 PM, Laurent Pinchart wrote: >>>>> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: >>>>>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes must >>>>>> be non-zero), I don't think there are any conflicts with existing values >>>>>> of >>>>>> nonstd. To make it even safer and easier to parse, you could set bit 31 >>>>>> of >>>>>> nonstd as a FOURCC indicator. >>>>> >>>>> I would then create a union between nonstd and fourcc, and document nonstd >>>>> as >>>>> being used for the legacy API only. Most existing drivers use a couple of >>>>> nonstd bits only. The driver that (ab)uses nonstd the most is pxafb and >>>>> uses >>>>> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& >>>>> 0xff000000 != 0 could be used as a FOURCC mode test. >>>>> >>>>> This assumes that FOURCCs will never have their last character set to >>>>> '\0'. Is >>>>> that a safe assumption for the future ? >>>> >>>> Yes, I think. The information I found indicates that space should be used >>>> for padding, so a \0 shouldn't exist. >>>> I think using only the nonstd field and requiring applications to check the >>>> capabilities would be possible, although not fool proof ;) >>> >>> So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC if >>> any of them is set. >>> >>> Nicely backwards compatible, as sane drivers should reject nonstd values they >>> don't support (apps _will_ start filling in FOURCC values ignoring capabilities, >>> won't they?). >>> >> That seems like a reasonable case, but if we're going to do that then >> certainly the nonstd bit encoding needs to be documented and treated as a >> hard ABI. >> >> I'm not so sure about the if any bit in the upper byte is set assume >> FOURCC case though, there will presumably be other users in the future >> that will want bits for themselves, too. What exactly was the issue with >> having a FOURCC capability bit in the upper byte? > > That indeed gives less issues (as long as you don't stuff 8-bit ASCII > in the MSB) > and more bits for tradiditional nonstd users. The only disadvantage I can see is that it requires adding this bit in the application and stripping it when interpreting it. But I think the 24 lower bits should be enough for driver specific behavior (as the current values really can only be interpreted per driver). > BTW, after giving it some more thought: what does the FB_CAP_FOURCC buys > us? It's not like all drivers will support whatever random FOURCC mode > you give them, > so you have to check whether it's supported by doing a set_var first. Because any solution purely based on the nonstd field is insane. The abuse of that field is just too widespread. Drivers that use nonstd usually only check whether it is zero or nonzero and others will just interpret parts of nonstd and ignore the rest. They will happily accept FOURCC values in the nonstd and just doing the wrong thing. Even if we would fix those the problems applications will run into with older kernels will remain. Greetings, Florian Tobias Schandinat -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Friday 24 June 2011 21:45:57 Florian Tobias Schandinat wrote: > On 06/24/2011 06:55 PM, Geert Uytterhoeven wrote: > > On Fri, Jun 24, 2011 at 08:19, Paul Mundt wrote: > >> On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: > >>> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat wrote: > >>>> On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > >>>>> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > >>>>>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes > >>>>>> must be non-zero), I don't think there are any conflicts with > >>>>>> existing values of > >>>>>> nonstd. To make it even safer and easier to parse, you could set bit > >>>>>> 31 of > >>>>>> nonstd as a FOURCC indicator. > >>>>> > >>>>> I would then create a union between nonstd and fourcc, and document > >>>>> nonstd as > >>>>> being used for the legacy API only. Most existing drivers use a > >>>>> couple of nonstd bits only. The driver that (ab)uses nonstd the most > >>>>> is pxafb and uses > >>>>> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& > >>>>> 0xff000000 != 0 could be used as a FOURCC mode test. > >>>>> > >>>>> This assumes that FOURCCs will never have their last character set to > >>>>> '\0'. Is > >>>>> that a safe assumption for the future ? > >>>> > >>>> Yes, I think. The information I found indicates that space should be > >>>> used for padding, so a \0 shouldn't exist. > >>>> I think using only the nonstd field and requiring applications to > >>>> check the capabilities would be possible, although not fool proof ;) > >>> > >>> So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC > >>> if any of them is set. > >>> > >>> Nicely backwards compatible, as sane drivers should reject nonstd > >>> values they don't support (apps _will_ start filling in FOURCC values > >>> ignoring capabilities, won't they?). > >> > >> That seems like a reasonable case, but if we're going to do that then > >> certainly the nonstd bit encoding needs to be documented and treated as > >> a hard ABI. > >> > >> I'm not so sure about the if any bit in the upper byte is set assume > >> FOURCC case though, there will presumably be other users in the future > >> that will want bits for themselves, too. What exactly was the issue with > >> having a FOURCC capability bit in the upper byte? > > > > That indeed gives less issues (as long as you don't stuff 8-bit ASCII > > in the MSB) and more bits for tradiditional nonstd users. > > The only disadvantage I can see is that it requires adding this bit in the > application and stripping it when interpreting it. But I think the 24 lower > bits should be enough for driver specific behavior (as the current values > really can only be interpreted per driver). I'm also not keen on adding/stripping the MSB. It would be easier for applications to use FOURCCs directly. > > BTW, after giving it some more thought: what does the FB_CAP_FOURCC buys > > us? It's not like all drivers will support whatever random FOURCC mode > > you give them, so you have to check whether it's supported by doing a > > set_var first. > > Because any solution purely based on the nonstd field is insane. The abuse > of that field is just too widespread. Drivers that use nonstd usually only > check whether it is zero or nonzero and others will just interpret parts > of nonstd and ignore the rest. They will happily accept FOURCC values in > the nonstd and just doing the wrong thing. Even if we would fix those the > problems applications will run into with older kernels will remain. I agree with Florian here. Many drivers currently check whether nonstd != 0. Who knows what kind of values applications feed them ? I'd like to reach an agreement on the API, and implement it. I'm fine with either grayscale or nonstd to store the FOURCC (with a slight preference for grayscale), and with either a vmode flag or using the most significant byte of the grayscale/nonstd field to detect FOURCC mode. I believe FB_CAP_FOURCC (or something similar) is needed.
Ping ? As stated in my previous mail, I'd like to reach an agreement on the API, and implement it. I'm fine with either grayscale or nonstd to store the FOURCC (with a slight preference for grayscale), and with either a vmode flag or using the most significant byte of the grayscale/nonstd field to detect FOURCC mode. I believe FB_CAP_FOURCC (or something similar) is needed. Paul, Geert, Florian, what are your opinions ? On Monday 11 July 2011 17:32:51 Laurent Pinchart wrote: > On Friday 24 June 2011 21:45:57 Florian Tobias Schandinat wrote: > > On 06/24/2011 06:55 PM, Geert Uytterhoeven wrote: > > > On Fri, Jun 24, 2011 at 08:19, Paul Mundt wrote: > > >> On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: > > >>> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat wrote: > > >>>> On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > > >>>>> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > > >>>>>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes > > >>>>>> must be non-zero), I don't think there are any conflicts with > > >>>>>> existing values of > > >>>>>> nonstd. To make it even safer and easier to parse, you could set > > >>>>>> bit 31 of > > >>>>>> nonstd as a FOURCC indicator. > > >>>>> > > >>>>> I would then create a union between nonstd and fourcc, and document > > >>>>> nonstd as > > >>>>> being used for the legacy API only. Most existing drivers use a > > >>>>> couple of nonstd bits only. The driver that (ab)uses nonstd the > > >>>>> most is pxafb and uses > > >>>>> bits 22:0. Bits 31:24 are never used as far as I can tell, so > > >>>>> nonstd& 0xff000000 != 0 could be used as a FOURCC mode test. > > >>>>> > > >>>>> This assumes that FOURCCs will never have their last character set > > >>>>> to '\0'. Is > > >>>>> that a safe assumption for the future ? > > >>>> > > >>>> Yes, I think. The information I found indicates that space should be > > >>>> used for padding, so a \0 shouldn't exist. > > >>>> I think using only the nonstd field and requiring applications to > > >>>> check the capabilities would be possible, although not fool proof ;) > > >>> > > >>> So we can declare the 8 msb bits of nonstd reserved, and assume > > >>> FOURCC if any of them is set. > > >>> > > >>> Nicely backwards compatible, as sane drivers should reject nonstd > > >>> values they don't support (apps _will_ start filling in FOURCC values > > >>> ignoring capabilities, won't they?). > > >> > > >> That seems like a reasonable case, but if we're going to do that then > > >> certainly the nonstd bit encoding needs to be documented and treated > > >> as a hard ABI. > > >> > > >> I'm not so sure about the if any bit in the upper byte is set assume > > >> FOURCC case though, there will presumably be other users in the future > > >> that will want bits for themselves, too. What exactly was the issue > > >> with having a FOURCC capability bit in the upper byte? > > > > > > That indeed gives less issues (as long as you don't stuff 8-bit ASCII > > > in the MSB) and more bits for tradiditional nonstd users. > > > > The only disadvantage I can see is that it requires adding this bit in > > the application and stripping it when interpreting it. But I think the > > 24 lower bits should be enough for driver specific behavior (as the > > current values really can only be interpreted per driver). > > I'm also not keen on adding/stripping the MSB. It would be easier for > applications to use FOURCCs directly. > > > > BTW, after giving it some more thought: what does the FB_CAP_FOURCC > > > buys us? It's not like all drivers will support whatever random FOURCC > > > mode you give them, so you have to check whether it's supported by > > > doing a set_var first. > > > > Because any solution purely based on the nonstd field is insane. The > > abuse of that field is just too widespread. Drivers that use nonstd > > usually only check whether it is zero or nonzero and others will just > > interpret parts of nonstd and ignore the rest. They will happily accept > > FOURCC values in the nonstd and just doing the wrong thing. Even if we > > would fix those the problems applications will run into with older > > kernels will remain. > > I agree with Florian here. Many drivers currently check whether nonstd != > 0. Who knows what kind of values applications feed them ? > > I'd like to reach an agreement on the API, and implement it. I'm fine with > either grayscale or nonstd to store the FOURCC (with a slight preference > for grayscale), and with either a vmode flag or using the most significant > byte of the grayscale/nonstd field to detect FOURCC mode. I believe > FB_CAP_FOURCC (or something similar) is needed.
On Mon, 11 Jul 2011, Laurent Pinchart wrote: > On Friday 24 June 2011 21:45:57 Florian Tobias Schandinat wrote: > > On 06/24/2011 06:55 PM, Geert Uytterhoeven wrote: > > > On Fri, Jun 24, 2011 at 08:19, Paul Mundt wrote: > > >> On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: > > >>> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat wrote: > > >>>> On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > > >>>>> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > > >>>>>> As FOURCC values are always 4 ASCII characters (hence all 4 bytes > > >>>>>> must be non-zero), I don't think there are any conflicts with > > >>>>>> existing values of > > >>>>>> nonstd. To make it even safer and easier to parse, you could set bit > > >>>>>> 31 of > > >>>>>> nonstd as a FOURCC indicator. > > >>>>> > > >>>>> I would then create a union between nonstd and fourcc, and document > > >>>>> nonstd as > > >>>>> being used for the legacy API only. Most existing drivers use a > > >>>>> couple of nonstd bits only. The driver that (ab)uses nonstd the most > > >>>>> is pxafb and uses > > >>>>> bits 22:0. Bits 31:24 are never used as far as I can tell, so nonstd& > > >>>>> 0xff000000 != 0 could be used as a FOURCC mode test. > > >>>>> > > >>>>> This assumes that FOURCCs will never have their last character set to > > >>>>> '\0'. Is > > >>>>> that a safe assumption for the future ? > > >>>> > > >>>> Yes, I think. The information I found indicates that space should be > > >>>> used for padding, so a \0 shouldn't exist. > > >>>> I think using only the nonstd field and requiring applications to > > >>>> check the capabilities would be possible, although not fool proof ;) > > >>> > > >>> So we can declare the 8 msb bits of nonstd reserved, and assume FOURCC > > >>> if any of them is set. > > >>> > > >>> Nicely backwards compatible, as sane drivers should reject nonstd > > >>> values they don't support (apps _will_ start filling in FOURCC values > > >>> ignoring capabilities, won't they?). > > >> > > >> That seems like a reasonable case, but if we're going to do that then > > >> certainly the nonstd bit encoding needs to be documented and treated as > > >> a hard ABI. > > >> > > >> I'm not so sure about the if any bit in the upper byte is set assume > > >> FOURCC case though, there will presumably be other users in the future > > >> that will want bits for themselves, too. What exactly was the issue with > > >> having a FOURCC capability bit in the upper byte? > > > > > > That indeed gives less issues (as long as you don't stuff 8-bit ASCII > > > in the MSB) and more bits for tradiditional nonstd users. > > > > The only disadvantage I can see is that it requires adding this bit in the > > application and stripping it when interpreting it. But I think the 24 lower > > bits should be enough for driver specific behavior (as the current values > > really can only be interpreted per driver). > > I'm also not keen on adding/stripping the MSB. It would be easier for > applications to use FOURCCs directly. > > > > BTW, after giving it some more thought: what does the FB_CAP_FOURCC buys > > > us? It's not like all drivers will support whatever random FOURCC mode > > > you give them, so you have to check whether it's supported by doing a > > > set_var first. > > > > Because any solution purely based on the nonstd field is insane. The abuse > > of that field is just too widespread. Drivers that use nonstd usually only > > check whether it is zero or nonzero and others will just interpret parts > > of nonstd and ignore the rest. They will happily accept FOURCC values in > > the nonstd and just doing the wrong thing. Even if we would fix those the > > problems applications will run into with older kernels will remain. > > I agree with Florian here. Many drivers currently check whether nonstd != 0. > Who knows what kind of values applications feed them ? FWIW, I prefer the original Laurent's proposal, with a slight uncertainty about whether we need the .capability field, or whether the try-and-check approach is sufficient. As for struct fb_var_screeninfo fields to support switching to a FOURCC mode, I also prefer an explicit dedicated flag to specify switching to it. Even though using FOURCC doesn't fit under the notion of a videomode, using one of .vmode bits is too tempting, so, I would actually take the plunge and use FB_VMODE_FOURCC. As for the actual location of the fourcc code, I would leave .nonstd alone: who knows, maybe drivers will need both, whereas using grayscale and fourcc certainly doesn't make any sense. And I personally don't see a problem with using a union: buggy applications are, well, buggy... Actually, since in FOURCC mode we don't need any of __u32 bits_per_pixel; /* guess what */ __u32 grayscale; /* != 0 Graylevels instead of colors */ struct fb_bitfield red; /* bitfield in fb mem if true color, */ struct fb_bitfield green; /* else only length is significant */ struct fb_bitfield blue; struct fb_bitfield transp; /* transparency */ so, we could put them all in the union for the case, if we need anything else for the fourcc configuration in the future. > I'd like to reach an agreement on the API, and implement it. I'm fine with > either grayscale or nonstd to store the FOURCC (with a slight preference for > grayscale), and with either a vmode flag or using the most significant byte of > the grayscale/nonstd field to detect FOURCC mode. I believe FB_CAP_FOURCC (or > something similar) is needed. Thanks Guennadi --- Guennadi Liakhovetski, Ph.D. Freelance Open-Source Software Developer http://www.open-technology.de/ -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Guennadi, Thanks for the review. On Thursday 28 July 2011 10:31:24 Guennadi Liakhovetski wrote: > On Mon, 11 Jul 2011, Laurent Pinchart wrote: > > On Friday 24 June 2011 21:45:57 Florian Tobias Schandinat wrote: > > > On 06/24/2011 06:55 PM, Geert Uytterhoeven wrote: > > > > On Fri, Jun 24, 2011 at 08:19, Paul Mundt wrote: > > > >> On Thu, Jun 23, 2011 at 06:08:03PM +0200, Geert Uytterhoeven wrote: > > > >>> On Wed, Jun 22, 2011 at 07:45, Florian Tobias Schandinat wrote: > > > >>>> On 06/21/2011 10:31 PM, Laurent Pinchart wrote: > > > >>>>> On Tuesday 21 June 2011 22:49:14 Geert Uytterhoeven wrote: > > > >>>>>> As FOURCC values are always 4 ASCII characters (hence all 4 > > > >>>>>> bytes must be non-zero), I don't think there are any conflicts > > > >>>>>> with existing values of > > > >>>>>> nonstd. To make it even safer and easier to parse, you could set > > > >>>>>> bit 31 of > > > >>>>>> nonstd as a FOURCC indicator. > > > >>>>> > > > >>>>> I would then create a union between nonstd and fourcc, and > > > >>>>> document nonstd as > > > >>>>> being used for the legacy API only. Most existing drivers use a > > > >>>>> couple of nonstd bits only. The driver that (ab)uses nonstd the > > > >>>>> most is pxafb and uses > > > >>>>> bits 22:0. Bits 31:24 are never used as far as I can tell, so > > > >>>>> nonstd& 0xff000000 != 0 could be used as a FOURCC mode test. > > > >>>>> > > > >>>>> This assumes that FOURCCs will never have their last character > > > >>>>> set to '\0'. Is > > > >>>>> that a safe assumption for the future ? > > > >>>> > > > >>>> Yes, I think. The information I found indicates that space should > > > >>>> be used for padding, so a \0 shouldn't exist. > > > >>>> I think using only the nonstd field and requiring applications to > > > >>>> check the capabilities would be possible, although not fool proof > > > >>>> ;) > > > >>> > > > >>> So we can declare the 8 msb bits of nonstd reserved, and assume > > > >>> FOURCC if any of them is set. > > > >>> > > > >>> Nicely backwards compatible, as sane drivers should reject nonstd > > > >>> values they don't support (apps _will_ start filling in FOURCC > > > >>> values ignoring capabilities, won't they?). > > > >> > > > >> That seems like a reasonable case, but if we're going to do that > > > >> then certainly the nonstd bit encoding needs to be documented and > > > >> treated as a hard ABI. > > > >> > > > >> I'm not so sure about the if any bit in the upper byte is set assume > > > >> FOURCC case though, there will presumably be other users in the > > > >> future that will want bits for themselves, too. What exactly was > > > >> the issue with having a FOURCC capability bit in the upper byte? > > > > > > > > That indeed gives less issues (as long as you don't stuff 8-bit ASCII > > > > in the MSB) and more bits for tradiditional nonstd users. > > > > > > The only disadvantage I can see is that it requires adding this bit in > > > the application and stripping it when interpreting it. But I think the > > > 24 lower bits should be enough for driver specific behavior (as the > > > current values really can only be interpreted per driver). > > > > I'm also not keen on adding/stripping the MSB. It would be easier for > > applications to use FOURCCs directly. > > > > > > BTW, after giving it some more thought: what does the FB_CAP_FOURCC > > > > buys us? It's not like all drivers will support whatever random > > > > FOURCC mode you give them, so you have to check whether it's > > > > supported by doing a set_var first. > > > > > > Because any solution purely based on the nonstd field is insane. The > > > abuse of that field is just too widespread. Drivers that use nonstd > > > usually only check whether it is zero or nonzero and others will just > > > interpret parts of nonstd and ignore the rest. They will happily > > > accept FOURCC values in the nonstd and just doing the wrong thing. > > > Even if we would fix those the problems applications will run into > > > with older kernels will remain. > > > > I agree with Florian here. Many drivers currently check whether nonstd != > > 0. Who knows what kind of values applications feed them ? > > FWIW, I prefer the original Laurent's proposal, with a slight uncertainty > about whether we need the .capability field, or whether the try-and-check > approach is sufficient. Try-and-check would be better, but unfortunately I don't think it would work. Existing drivers will either ignore the new FB_VMODE_FOURCC flag or fail. I expect many drivers that ignore it to still accept it blindly, and not clear it, so applications won't be able to find out whether the flag is supported. > As for struct fb_var_screeninfo fields to support switching to a FOURCC > mode, I also prefer an explicit dedicated flag to specify switching to it. > Even though using FOURCC doesn't fit under the notion of a videomode, using > one of .vmode bits is too tempting, so, I would actually take the plunge and > use FB_VMODE_FOURCC. Another option would be to consider any grayscale > 1 value as a FOURCC. I've briefly checked the in-tree drivers: they only assign grayscale with 0 or 1, and check whether grayscale is 0 or different than 0. If a userspace application only sets grayscale > 1 when talking to a driver that supports the FOURCC-based API, we could get rid of the flag. What can't be easily found out is whether existing applications set grayscale to a > 1 value. They would break when used with FOURCC-aware drivers if we consider any grayscale > 1 value as a FOURCC. Is that a risk we can take ? > As for the actual location of the fourcc code, I would leave .nonstd > alone: who knows, maybe drivers will need both, whereas using grayscale > and fourcc certainly doesn't make any sense. And I personally don't see a > problem with using a union: buggy applications are, well, buggy... I agree with this. > Actually, since in FOURCC mode we don't need any of > > __u32 bits_per_pixel; /* guess what */ > __u32 grayscale; /* != 0 Graylevels instead of colors */ > > struct fb_bitfield red; /* bitfield in fb mem if true color, */ > struct fb_bitfield green; /* else only length is significant */ > struct fb_bitfield blue; > struct fb_bitfield transp; /* transparency */ > > so, we could put them all in the union for the case, if we need anything > else for the fourcc configuration in the future. Good point. It might make sense to exclude bits_per_pixel from the union though, as that's interesting information for applications. The red, green, blue and transp fields are less useful. > > I'd like to reach an agreement on the API, and implement it. I'm fine > > with either grayscale or nonstd to store the FOURCC (with a slight > > preference for grayscale), and with either a vmode flag or using the > > most significant byte of the grayscale/nonstd field to detect FOURCC > > mode. I believe FB_CAP_FOURCC (or something similar) is needed.
On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote: >> As for struct fb_var_screeninfo fields to support switching to a FOURCC >> mode, I also prefer an explicit dedicated flag to specify switching to it. >> Even though using FOURCC doesn't fit under the notion of a videomode, using >> one of .vmode bits is too tempting, so, I would actually take the plunge and >> use FB_VMODE_FOURCC. > > Another option would be to consider any grayscale > 1 value as a FOURCC. I've > briefly checked the in-tree drivers: they only assign grayscale with 0 or 1, > and check whether grayscale is 0 or different than 0. If a userspace > application only sets grayscale > 1 when talking to a driver that supports the > FOURCC-based API, we could get rid of the flag. > > What can't be easily found out is whether existing applications set grayscale > to a > 1 value. They would break when used with FOURCC-aware drivers if we > consider any grayscale > 1 value as a FOURCC. Is that a risk we can take ? I think we can. I'd expect applications to use either 1 or -1 (i.e. all ones), both are invalid FOURCC values. Still, I prefer the nonstd way. And limiting traditional nonstd values to the lowest 24 bits (there are no in-tree drivers using the highest 8 bits, right?). 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 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: > On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart > <laurent.pinchart@ideasonboard.com> wrote: >>> As for struct fb_var_screeninfo fields to support switching to a FOURCC >>> mode, I also prefer an explicit dedicated flag to specify switching to it. >>> Even though using FOURCC doesn't fit under the notion of a videomode, using >>> one of .vmode bits is too tempting, so, I would actually take the plunge and >>> use FB_VMODE_FOURCC. >> >> Another option would be to consider any grayscale> 1 value as a FOURCC. I've >> briefly checked the in-tree drivers: they only assign grayscale with 0 or 1, >> and check whether grayscale is 0 or different than 0. If a userspace >> application only sets grayscale> 1 when talking to a driver that supports the >> FOURCC-based API, we could get rid of the flag. >> >> What can't be easily found out is whether existing applications set grayscale >> to a> 1 value. They would break when used with FOURCC-aware drivers if we >> consider any grayscale> 1 value as a FOURCC. Is that a risk we can take ? > > I think we can. I'd expect applications to use either 1 or -1 (i.e. > all ones), both are > invalid FOURCC values. > > Still, I prefer the nonstd way. > And limiting traditional nonstd values to the lowest 24 bits (there > are no in-tree > drivers using the highest 8 bits, right?). Okay, it would be okay for me to - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != 0 - not having an explicit flag to enable FOURCC - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC - making support of FOURCC visible to userspace by capabilites |= FB_CAP_FOURCC The capabilities is not strictly necessary but I think it's very useful as - it allows applications to make sure the extension is supported (for example to adjust the UI) - it allows applications to distinguish whether a particular format is not supported or FOURCC at all - it allows signaling further extensions of the API - it does not hurt, one line per driver and still some bytes in fixinfo free So using it would look like this: - the driver must have capabilities |= FB_CAP_FOURCC - the application may check capabilities to know whether FOURCC is supported - the application may write a raw FOURCC value in nonstd to request changing to FOURCC mode with this format - when the driver switches to a FOURCC mode it must have visual = FB_VISUAL_FOURCC and the current FOURCC format in nonstd - the application should check visual and nonstd to make sure it gets what it wanted So if there are no strong objections against this I think we should implement it. I do not really care whether we use a union or not but I think if we decide to have one it should cover all fields that are undefined/unused in FOURCC mode. Hope we can find anything that everyone considers acceptable, Florian Tobias Schandinat -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Florian, Thanks for the feedback. On Monday 01 August 2011 00:54:48 Florian Tobias Schandinat wrote: > On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: > > On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart wrote: > >>> As for struct fb_var_screeninfo fields to support switching to a FOURCC > >>> mode, I also prefer an explicit dedicated flag to specify switching to > >>> it. Even though using FOURCC doesn't fit under the notion of a > >>> videomode, using one of .vmode bits is too tempting, so, I would > >>> actually take the plunge and use FB_VMODE_FOURCC. > >> > >> Another option would be to consider any grayscale> 1 value as a FOURCC. > >> I've briefly checked the in-tree drivers: they only assign grayscale > >> with 0 or 1, and check whether grayscale is 0 or different than 0. If a > >> userspace application only sets grayscale> 1 when talking to a driver > >> that supports the FOURCC-based API, we could get rid of the flag. > >> > >> What can't be easily found out is whether existing applications set > >> grayscale to a> 1 value. They would break when used with FOURCC-aware > >> drivers if we consider any grayscale> 1 value as a FOURCC. Is that a > >> risk we can take ? > > > > I think we can. I'd expect applications to use either 1 or -1 (i.e. > > all ones), both are > > invalid FOURCC values. > > > > Still, I prefer the nonstd way. > > And limiting traditional nonstd values to the lowest 24 bits (there > > are no in-tree > > drivers using the highest 8 bits, right?). > > Okay, it would be okay for me to > - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != 0 > - not having an explicit flag to enable FOURCC > - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC > - making support of FOURCC visible to userspace by capabilites |= > FB_CAP_FOURCC > > The capabilities is not strictly necessary but I think it's very useful as > - it allows applications to make sure the extension is supported (for > example to adjust the UI) > - it allows applications to distinguish whether a particular format is not > supported or FOURCC at all > - it allows signaling further extensions of the API > - it does not hurt, one line per driver and still some bytes in fixinfo > free Without a FOURCC capability applications will need to try FOURCCs blindly. Drivers that are not FOURCC aware would then risk interpreting the FOURCC as something else. As you mention below applications will need that check that visual == FB_VISUAL_FOURCC, so it's less of an issue than I initially thought, but it doesn't become a non-issue. The display might still show glitches. > So using it would look like this: > - the driver must have capabilities |= FB_CAP_FOURCC > - the application may check capabilities to know whether FOURCC is > supported - the application may write a raw FOURCC value in nonstd to > request changing to FOURCC mode with this format > - when the driver switches to a FOURCC mode it must have visual = > FB_VISUAL_FOURCC and the current FOURCC format in nonstd > - the application should check visual and nonstd to make sure it gets what > it wanted > > > So if there are no strong objections against this I think we should > implement it. I do not really care whether we use a union or not but I > think if we decide to have one it should cover all fields that are > undefined/unused in FOURCC mode. > > > Hope we can find anything that everyone considers acceptable, This sounds good to me, except that I would use the grayscale field instead of the nonstd field. nonstd has pretty weird usecases, while grayscale is better defined. nonstd might also make sense combined with FOURCC-based modes, while grayscale would be completely redundant. What's your opinion on that ?
Hi Geert, Thanks for the feedback. On Sunday 31 July 2011 22:32:42 Geert Uytterhoeven wrote: > On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart wrote: > >> As for struct fb_var_screeninfo fields to support switching to a FOURCC > >> mode, I also prefer an explicit dedicated flag to specify switching to > >> it. Even though using FOURCC doesn't fit under the notion of a > >> videomode, using one of .vmode bits is too tempting, so, I would > >> actually take the plunge and use FB_VMODE_FOURCC. > > > > Another option would be to consider any grayscale > 1 value as a FOURCC. > > I've briefly checked the in-tree drivers: they only assign grayscale > > with 0 or 1, and check whether grayscale is 0 or different than 0. If a > > userspace application only sets grayscale > 1 when talking to a driver > > that supports the FOURCC-based API, we could get rid of the flag. > > > > What can't be easily found out is whether existing applications set > > grayscale to a > 1 value. They would break when used with FOURCC-aware > > drivers if we consider any grayscale > 1 value as a FOURCC. Is that a > > risk we can take ? > > I think we can. I'd expect applications to use either 1 or -1 (i.e. all > ones), both are invalid FOURCC values. OK. > Still, I prefer the nonstd way. > And limiting traditional nonstd values to the lowest 24 bits (there > are no in-tree drivers using the highest 8 bits, right?). None that I've found. I still have a preference for the grayscale field though. As mentioned by Guennadi, the grayscale field would become redundant for FOURCC-based formats. It's then a good candidate, and would let drivers (and applications) do any crazy stuff they want with the nonstd field.
Hi Laurent, On 07/31/2011 11:28 PM, Laurent Pinchart wrote: > Hi Florian, > > Thanks for the feedback. > > On Monday 01 August 2011 00:54:48 Florian Tobias Schandinat wrote: >> On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: >>> On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart wrote: >>>>> As for struct fb_var_screeninfo fields to support switching to a FOURCC >>>>> mode, I also prefer an explicit dedicated flag to specify switching to >>>>> it. Even though using FOURCC doesn't fit under the notion of a >>>>> videomode, using one of .vmode bits is too tempting, so, I would >>>>> actually take the plunge and use FB_VMODE_FOURCC. >>>> >>>> Another option would be to consider any grayscale> 1 value as a FOURCC. >>>> I've briefly checked the in-tree drivers: they only assign grayscale >>>> with 0 or 1, and check whether grayscale is 0 or different than 0. If a >>>> userspace application only sets grayscale> 1 when talking to a driver >>>> that supports the FOURCC-based API, we could get rid of the flag. >>>> >>>> What can't be easily found out is whether existing applications set >>>> grayscale to a> 1 value. They would break when used with FOURCC-aware >>>> drivers if we consider any grayscale> 1 value as a FOURCC. Is that a >>>> risk we can take ? >>> >>> I think we can. I'd expect applications to use either 1 or -1 (i.e. >>> all ones), both are >>> invalid FOURCC values. >>> >>> Still, I prefer the nonstd way. >>> And limiting traditional nonstd values to the lowest 24 bits (there >>> are no in-tree >>> drivers using the highest 8 bits, right?). >> >> Okay, it would be okay for me to >> - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != 0 >> - not having an explicit flag to enable FOURCC >> - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC >> - making support of FOURCC visible to userspace by capabilites |= >> FB_CAP_FOURCC >> >> The capabilities is not strictly necessary but I think it's very useful as >> - it allows applications to make sure the extension is supported (for >> example to adjust the UI) >> - it allows applications to distinguish whether a particular format is not >> supported or FOURCC at all >> - it allows signaling further extensions of the API >> - it does not hurt, one line per driver and still some bytes in fixinfo >> free > > Without a FOURCC capability applications will need to try FOURCCs blindly. > Drivers that are not FOURCC aware would then risk interpreting the FOURCC as > something else. As you mention below applications will need that check that > visual == FB_VISUAL_FOURCC, so it's less of an issue than I initially thought, > but it doesn't become a non-issue. The display might still show glitches. True. >> So using it would look like this: >> - the driver must have capabilities |= FB_CAP_FOURCC >> - the application may check capabilities to know whether FOURCC is >> supported - the application may write a raw FOURCC value in nonstd to >> request changing to FOURCC mode with this format >> - when the driver switches to a FOURCC mode it must have visual = >> FB_VISUAL_FOURCC and the current FOURCC format in nonstd >> - the application should check visual and nonstd to make sure it gets what >> it wanted >> >> >> So if there are no strong objections against this I think we should >> implement it. I do not really care whether we use a union or not but I >> think if we decide to have one it should cover all fields that are >> undefined/unused in FOURCC mode. >> >> >> Hope we can find anything that everyone considers acceptable, > > This sounds good to me, except that I would use the grayscale field instead of > the nonstd field. nonstd has pretty weird usecases, while grayscale is better > defined. nonstd might also make sense combined with FOURCC-based modes, while > grayscale would be completely redundant. > > What's your opinion on that ? I do not really care, either one would be okay for me. You're right that nonstd is used for a lot of things and perhaps some of those should still be possible in FOURCC mode. On the other hand I think applications are more likely to pass random values to grayscale as its meaning seems globally accepted (in contrast to nonstd where the application needs to know the driver to get any use of it). Perhaps you should also say that in FOURCC mode all unused pixel format fields should be set to 0 by the application and other values of those may get a meaning in later extensions or individual drivers. Regards, Florian Tobias Schandinat -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Aug 1, 2011 at 00:54, Florian Tobias Schandinat <FlorianSchandinat@gmx.de> wrote: > On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: >> On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart >> <laurent.pinchart@ideasonboard.com> wrote: >>>> >>>> As for struct fb_var_screeninfo fields to support switching to a FOURCC >>>> mode, I also prefer an explicit dedicated flag to specify switching to >>>> it. >>>> Even though using FOURCC doesn't fit under the notion of a videomode, >>>> using >>>> one of .vmode bits is too tempting, so, I would actually take the plunge >>>> and >>>> use FB_VMODE_FOURCC. >>> >>> Another option would be to consider any grayscale> 1 value as a FOURCC. >>> I've >>> briefly checked the in-tree drivers: they only assign grayscale with 0 or >>> 1, >>> and check whether grayscale is 0 or different than 0. If a userspace >>> application only sets grayscale> 1 when talking to a driver that >>> supports the >>> FOURCC-based API, we could get rid of the flag. >>> >>> What can't be easily found out is whether existing applications set >>> grayscale >>> to a> 1 value. They would break when used with FOURCC-aware drivers if >>> we >>> consider any grayscale> 1 value as a FOURCC. Is that a risk we can take >>> ? >> >> I think we can. I'd expect applications to use either 1 or -1 (i.e. >> all ones), both are >> invalid FOURCC values. >> >> Still, I prefer the nonstd way. >> And limiting traditional nonstd values to the lowest 24 bits (there >> are no in-tree >> drivers using the highest 8 bits, right?). > > Okay, it would be okay for me to > - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != 0 > - not having an explicit flag to enable FOURCC > - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC > - making support of FOURCC visible to userspace by capabilites |= > FB_CAP_FOURCC > > The capabilities is not strictly necessary but I think it's very useful as > - it allows applications to make sure the extension is supported (for > example to adjust the UI) > - it allows applications to distinguish whether a particular format is not > supported or FOURCC at all > - it allows signaling further extensions of the API > - it does not hurt, one line per driver and still some bytes in fixinfo free > > > So using it would look like this: > - the driver must have capabilities |= FB_CAP_FOURCC > - the application may check capabilities to know whether FOURCC is supported > - the application may write a raw FOURCC value in nonstd to request changing > to FOURCC mode with this format > - when the driver switches to a FOURCC mode it must have visual = > FB_VISUAL_FOURCC and the current FOURCC format in nonstd > - the application should check visual and nonstd to make sure it gets what > it wanted As several of the FOURCC formats duplicate formats you can already specify in some other way (e.g. the RGB and greyscale formats), and as FOURCC makes life easier for the application writer, I'm wondering whether it makes sense to add FOURCC support in the generic layer for drivers that don't support it? I.e. the generic layer would fill in fb_var_screeninfo depending on the requested FOURCC mode, if possible. 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 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Florian, On Monday 01 August 2011 01:58:27 Florian Tobias Schandinat wrote: > On 07/31/2011 11:28 PM, Laurent Pinchart wrote: > > On Monday 01 August 2011 00:54:48 Florian Tobias Schandinat wrote: > >> On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: > >>> On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart wrote: > >>>>> As for struct fb_var_screeninfo fields to support switching to a > >>>>> FOURCC mode, I also prefer an explicit dedicated flag to specify > >>>>> switching to it. Even though using FOURCC doesn't fit under the > >>>>> notion of a videomode, using one of .vmode bits is too tempting, so, > >>>>> I would actually take the plunge and use FB_VMODE_FOURCC. > >>>> > >>>> Another option would be to consider any grayscale> 1 value as a > >>>> FOURCC. I've briefly checked the in-tree drivers: they only assign > >>>> grayscale with 0 or 1, and check whether grayscale is 0 or different > >>>> than 0. If a userspace application only sets grayscale> 1 when > >>>> talking to a driver that supports the FOURCC-based API, we could get > >>>> rid of the flag. > >>>> > >>>> What can't be easily found out is whether existing applications set > >>>> grayscale to a> 1 value. They would break when used with > >>>> FOURCC-aware drivers if we consider any grayscale> 1 value as a > >>>> FOURCC. Is that a risk we can take ? > >>> > >>> I think we can. I'd expect applications to use either 1 or -1 (i.e. > >>> all ones), both are > >>> invalid FOURCC values. > >>> > >>> Still, I prefer the nonstd way. > >>> And limiting traditional nonstd values to the lowest 24 bits (there > >>> are no in-tree > >>> drivers using the highest 8 bits, right?). > >> > >> Okay, it would be okay for me to > >> - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != > >> 0 - not having an explicit flag to enable FOURCC > >> - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC > >> - making support of FOURCC visible to userspace by capabilites |= > >> FB_CAP_FOURCC > >> > >> The capabilities is not strictly necessary but I think it's very useful > >> as - it allows applications to make sure the extension is supported > >> (for example to adjust the UI) > >> - it allows applications to distinguish whether a particular format is > >> not supported or FOURCC at all > >> - it allows signaling further extensions of the API > >> - it does not hurt, one line per driver and still some bytes in fixinfo > >> free > > > > Without a FOURCC capability applications will need to try FOURCCs > > blindly. Drivers that are not FOURCC aware would then risk interpreting > > the FOURCC as something else. As you mention below applications will > > need that check that visual == FB_VISUAL_FOURCC, so it's less of an > > issue than I initially thought, but it doesn't become a non-issue. The > > display might still show glitches. > > True. > > >> So using it would look like this: > >> - the driver must have capabilities |= FB_CAP_FOURCC > >> - the application may check capabilities to know whether FOURCC is > >> supported - the application may write a raw FOURCC value in nonstd to > >> request changing to FOURCC mode with this format > >> - when the driver switches to a FOURCC mode it must have visual = > >> FB_VISUAL_FOURCC and the current FOURCC format in nonstd > >> - the application should check visual and nonstd to make sure it gets > >> what it wanted > >> > >> > >> So if there are no strong objections against this I think we should > >> implement it. I do not really care whether we use a union or not but I > >> think if we decide to have one it should cover all fields that are > >> undefined/unused in FOURCC mode. > >> > >> > >> Hope we can find anything that everyone considers acceptable, > > > > This sounds good to me, except that I would use the grayscale field > > instead of the nonstd field. nonstd has pretty weird usecases, while > > grayscale is better defined. nonstd might also make sense combined with > > FOURCC-based modes, while grayscale would be completely redundant. > > > > What's your opinion on that ? > > I do not really care, either one would be okay for me. > You're right that nonstd is used for a lot of things and perhaps some of > those should still be possible in FOURCC mode. On the other hand I think > applications are more likely to pass random values to grayscale as its > meaning seems globally accepted (in contrast to nonstd where the > application needs to know the driver to get any use of it). > Perhaps you should also say that in FOURCC mode all unused pixel format > fields should be set to 0 by the application and other values of those may > get a meaning in later extensions or individual drivers. Good point. I'll add that to the documentation.
Hi Geert, On Monday 01 August 2011 11:49:46 Geert Uytterhoeven wrote: > On Mon, Aug 1, 2011 at 00:54, Florian Tobias Schandinat wrote: > > On 07/31/2011 08:32 PM, Geert Uytterhoeven wrote: > >> On Thu, Jul 28, 2011 at 12:51, Laurent Pinchart wrote: > >>>> As for struct fb_var_screeninfo fields to support switching to a > >>>> FOURCC mode, I also prefer an explicit dedicated flag to specify > >>>> switching to it. > >>>> Even though using FOURCC doesn't fit under the notion of a videomode, > >>>> using > >>>> one of .vmode bits is too tempting, so, I would actually take the > >>>> plunge and > >>>> use FB_VMODE_FOURCC. > >>> > >>> Another option would be to consider any grayscale> 1 value as a > >>> FOURCC. I've > >>> briefly checked the in-tree drivers: they only assign grayscale with 0 > >>> or 1, > >>> and check whether grayscale is 0 or different than 0. If a userspace > >>> application only sets grayscale> 1 when talking to a driver that > >>> supports the > >>> FOURCC-based API, we could get rid of the flag. > >>> > >>> What can't be easily found out is whether existing applications set > >>> grayscale > >>> to a> 1 value. They would break when used with FOURCC-aware drivers if > >>> we > >>> consider any grayscale> 1 value as a FOURCC. Is that a risk we can > >>> take ? > >> > >> I think we can. I'd expect applications to use either 1 or -1 (i.e. > >> all ones), both are > >> invalid FOURCC values. > >> > >> Still, I prefer the nonstd way. > >> And limiting traditional nonstd values to the lowest 24 bits (there > >> are no in-tree > >> drivers using the highest 8 bits, right?). > > > > Okay, it would be okay for me to > > - write raw FOURCC values in nonstd, enable FOURCC mode if upper byte != > > 0 - not having an explicit flag to enable FOURCC > > - in FOURCC mode drivers must set visual to FB_VISUAL_FOURCC > > - making support of FOURCC visible to userspace by capabilites |= > > FB_CAP_FOURCC > > > > The capabilities is not strictly necessary but I think it's very useful > > as - it allows applications to make sure the extension is supported (for > > example to adjust the UI) > > - it allows applications to distinguish whether a particular format is > > not supported or FOURCC at all > > - it allows signaling further extensions of the API > > - it does not hurt, one line per driver and still some bytes in fixinfo > > free > > > > > > So using it would look like this: > > - the driver must have capabilities |= FB_CAP_FOURCC > > - the application may check capabilities to know whether FOURCC is > > supported - the application may write a raw FOURCC value in nonstd to > > request changing to FOURCC mode with this format > > - when the driver switches to a FOURCC mode it must have visual = > > FB_VISUAL_FOURCC and the current FOURCC format in nonstd > > - the application should check visual and nonstd to make sure it gets > > what it wanted > > As several of the FOURCC formats duplicate formats you can already specify > in some other way (e.g. the RGB and greyscale formats), and as FOURCC makes > life easier for the application writer, I'm wondering whether it makes sense > to add FOURCC support in the generic layer for drivers that don't support > it? I.e. the generic layer would fill in fb_var_screeninfo depending on the > requested FOURCC mode, if possible. That's a good idea, but I'd like to add that in a second step. I'm working on a proof-of-concept by porting a driver to the FOURCC-based API first.
Hi LAurent, On Thu, Aug 11, 2011 at 19:19, Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote: > On Monday 01 August 2011 11:49:46 Geert Uytterhoeven wrote: >> As several of the FOURCC formats duplicate formats you can already specify >> in some other way (e.g. the RGB and greyscale formats), and as FOURCC makes >> life easier for the application writer, I'm wondering whether it makes sense >> to add FOURCC support in the generic layer for drivers that don't support >> it? I.e. the generic layer would fill in fb_var_screeninfo depending on the >> requested FOURCC mode, if possible. > > That's a good idea, but I'd like to add that in a second step. I'm working on > a proof-of-concept by porting a driver to the FOURCC-based API first. Sure! I was just mentioning it, so we keep it in the back of our head and don't make decisions now that would make it impossible to add that later. 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 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/Documentation/fb/api.txt b/Documentation/fb/api.txt new file mode 100644 index 0000000..d4cd6ec --- /dev/null +++ b/Documentation/fb/api.txt @@ -0,0 +1,284 @@ + The Frame Buffer Device API + --------------------------- + +Last revised: June 21, 2011 + + +0. Introduction +--------------- + +This document describes the frame buffer API used by applications to interact +with frame buffer devices. In-kernel APIs between device drivers and the frame +buffer core are not described. + +Due to a lack of documentation in the original frame buffer API, drivers +behaviours differ in subtle (and not so subtle) ways. This document describes +the recommended API implementation, but applications should be prepared to +deal with different behaviours. + + +1. Capabilities +--------------- + +Device and driver capabilities are reported in the fixed screen information +capabilities field. + +struct fb_fix_screeninfo { + ... + __u16 capabilities; /* see FB_CAP_* */ + ... +}; + +Application should use those capabilities to find out what features they can +expect from the device and driver. + +- FB_CAP_FOURCC + +The driver supports the four character code (FOURCC) based format setting API. +When supported, formats are configured using a FOURCC instead of manually +specifying color components layout. + + +2. Types and visuals +-------------------- + +Pixels are stored in memory in hardware-dependent formats. Applications need +to be aware of the pixel storage format in order to write image data to the +frame buffer memory in the format expected by the hardware. + +Formats are described by frame buffer types and visuals. Some visuals require +additional information, which are stored in the variable screen information +bits_per_pixel, grayscale, fourcc, red, green, blue and transp fields. + +The following types and visuals are supported. + +- FB_TYPE_PACKED_PIXELS + +Color components (usually RGB or YUV) are packed together into macropixels +that are stored in a single plane. The exact color components layout is +described in a visual-dependent way. + +Frame buffer visuals that don't use multiple color components per pixel +(such as monochrome and pseudo-color visuals) are reported as packed frame +buffer types, even though they don't stricly speaking pack color components +into macropixels. + +- FB_TYPE_PLANES + +Color components are stored in separate planes. Planes are located +contiguously in memory. + +- FB_VISUAL_MONO01 + +Pixels are black or white and stored on one bit. A bit set to 1 represents a +black pixel and a bit set to 0 a white pixel. Pixels are packed together in +bytes with 8 pixels per byte. + +FB_VISUAL_MONO01 is used with FB_TYPE_PACKED_PIXELS only. + +- FB_VISUAL_MONO10 + +Pixels are black or white and stored on one bit. A bit set to 1 represents a +white pixel and a bit set to 0 a black pixel. Pixels are packed together in +bytes with 8 pixels per byte. + +FB_VISUAL_MONO01 is used with FB_TYPE_PACKED_PIXELS only. + +- FB_VISUAL_TRUECOLOR + +Pixels are broken into red, green and blue components, and each component +indexes a read-only lookup table for the corresponding value. Lookup tables +are device-dependent, and provide linear or non-linear ramps. + +Each component is stored in memory according to the variable screen +information red, green, blue and transp fields. + +- FB_VISUAL_PSEUDOCOLOR and FB_VISUAL_STATIC_PSEUDOCOLOR + +Pixel values are encoded as indices into a colormap that stores red, green and +blue components. The colormap is read-only for FB_VISUAL_STATIC_PSEUDOCOLOR +and read-write for FB_VISUAL_PSEUDOCOLOR. + +Each pixel value is stored in the number of bits reported by the variable +screen information bits_per_pixel field. Pixels are contiguous in memory. + +FB_VISUAL_PSEUDOCOLOR and FB_VISUAL_STATIC_PSEUDOCOLOR are used with +FB_TYPE_PACKED_PIXELS only. + +- FB_VISUAL_DIRECTCOLOR + +Pixels are broken into red, green and blue components, and each component +indexes a programmable lookup table for the corresponding value. + +Each component is stored in memory according to the variable screen +information red, green, blue and transp fields. + +- FB_VISUAL_FOURCC + +Pixels are stored in memory as described by the format FOURCC identifier +stored in the variable screen information fourcc field. + + +3. Screen information +--------------------- + +Screen information are queried by applications using the FBIOGET_FSCREENINFO +and FBIOGET_VSCREENINFO ioctls. Those ioctls take a pointer to a +fb_fix_screeninfo and fb_var_screeninfo structure respectively. + +struct fb_fix_screeninfo stores device independent unchangeable information +about the frame buffer device and the current format. Those information can't +be directly modified by applications, but can be changed by the driver when an +application modifies the format. + +struct fb_fix_screeninfo { + char id[16]; /* identification string eg "TT Builtin" */ + unsigned long smem_start; /* Start of frame buffer mem */ + /* (physical address) */ + __u32 smem_len; /* Length of frame buffer mem */ + __u32 type; /* see FB_TYPE_* */ + __u32 type_aux; /* Interleave for interleaved Planes */ + __u32 visual; /* see FB_VISUAL_* */ + __u16 xpanstep; /* zero if no hardware panning */ + __u16 ypanstep; /* zero if no hardware panning */ + __u16 ywrapstep; /* zero if no hardware ywrap */ + __u32 line_length; /* length of a line in bytes */ + unsigned long mmio_start; /* Start of Memory Mapped I/O */ + /* (physical address) */ + __u32 mmio_len; /* Length of Memory Mapped I/O */ + __u32 accel; /* Indicate to driver which */ + /* specific chip/card we have */ + __u16 capabilities; /* see FB_CAP_* */ + __u16 reserved[2]; /* Reserved for future compatibility */ +}; + +struct fb_var_screeninfo stores device independent changeable information +about a frame buffer device, its current format and video mode, as well as +other miscellaneous parameters. + +struct fb_var_screeninfo { + __u32 xres; /* visible resolution */ + __u32 yres; + __u32 xres_virtual; /* virtual resolution */ + __u32 yres_virtual; + __u32 xoffset; /* offset from virtual to visible */ + __u32 yoffset; /* resolution */ + + __u32 bits_per_pixel; /* guess what */ + union { + __u32 grayscale; /* != 0 Graylevels instead of colors */ + __u32 fourcc; /* video mode */ + }; + + struct fb_bitfield red; /* bitfield in fb mem if true color, */ + struct fb_bitfield green; /* else only length is significant */ + struct fb_bitfield blue; + struct fb_bitfield transp; /* transparency */ + + __u32 nonstd; /* != 0 Non standard pixel format */ + + __u32 activate; /* see FB_ACTIVATE_* */ + + __u32 height; /* height of picture in mm */ + __u32 width; /* width of picture in mm */ + + __u32 accel_flags; /* (OBSOLETE) see fb_info.flags */ + + /* Timing: All values in pixclocks, except pixclock (of course) */ + __u32 pixclock; /* pixel clock in ps (pico seconds) */ + __u32 left_margin; /* time from sync to picture */ + __u32 right_margin; /* time from picture to sync */ + __u32 upper_margin; /* time from sync to picture */ + __u32 lower_margin; + __u32 hsync_len; /* length of horizontal sync */ + __u32 vsync_len; /* length of vertical sync */ + __u32 sync; /* see FB_SYNC_* */ + __u32 vmode; /* see FB_VMODE_* */ + __u32 rotate; /* angle we rotate counter clockwise */ + __u32 reserved[5]; /* Reserved for future compatibility */ +}; + +To modify variable information, applications call the FBIOPUT_VSCREENINFO +ioctl with a pointer to a fb_var_screeninfo structure. If the call is +successful, the driver will update the fixed screen information accordingly. + +Instead of filling the complete fb_var_screeninfo structure manually, +applications should call the FBIOGET_VSCREENINFO ioctl and modify only the +fields they care about. + + +4. Format configuration +----------------------- + +Frame buffer devices offer two ways to configure the frame buffer format: the +legacy API and the FOURCC-based API. + + +The legacy API has been the only frame buffer format configuration API for a +long time and is thus widely used by application. It is the recommended API +for applications when using RGB and grayscale formats, as well as legacy +non-standard formats. + +To select a format, applications set the fb_var_screeninfo bits_per_pixel field +to the desired frame buffer depth. Values up to 8 will usually map to +monochrome, grayscale or pseudocolor visuals, although this is not required. + +- For grayscale formats, applications set the grayscale field to a non-zero + value. The red, blue, green and transp fields must be set to 0 by + applications and ignored by drivers. Drivers must fill the red, blue and + green offsets to 0 and lengths to the bits_per_pixel value. + +- For pseudocolor formats, applications set the grayscale field to a zero + value. The red, blue, green and transp fields must be set to 0 by + applications and ignored by drivers. Drivers must fill the red, blue and + green offsets to 0 and lengths to the bits_per_pixel value. + +- For truecolor and directcolor formats, applications set the grayscale field + to a zero value, and the red, blue, green and transp fields to describe the + layout of color components in memory. + +struct fb_bitfield { + __u32 offset; /* beginning of bitfield */ + __u32 length; /* length of bitfield */ + __u32 msb_right; /* != 0 : Most significant bit is */ + /* right */ +}; + + Pixel values are bits_per_pixel wide and are split in non-overlapping red, + green, blue and alpha (transparency) components. Location and size of each + component in the pixel value are described by the fb_bitfield offset and + length fields. Offset are computed from the right. + + Pixels are always stored in an integer number of bytes. If the number of + bits per pixel is not a multiple of 8, pixel values are padded to the next + multiple of 8 bits. + +Upon successful format configuration, drivers update the fb_fix_screeninfo +type, visual and line_length fields depending on the selected format. + + +The FOURCC-based API replaces format descriptions by four character codes +(FOURCC). FOURCCs are abstract identifiers that uniquely define a format +without explicitly describing it. This is the only API that supports YUV +formats. Drivers are also encouraged to implement the FOURCC-based API for RGB +and grayscale formats. + +Drivers that support the FOURCC-based API report this capability by setting +the FB_CAP_FOURCC bit in the fb_fix_screeninfo capabilities field. + +FOURCC definitions are located in the linux/videodev2.h header. However, and +despite starting with the V4L2_PIX_FMT_prefix, they are not restricted to V4L2 +and don't require usage of the V4L2 subsystem. FOURCC documentation is +available in Documentation/DocBook/v4l/pixfmt.xml. + +To select a format, applications set the FB_VMODE_FOURCC bit in the +fb_var_screeninfo vmode field, and set the fourcc field to the desired FOURCC. +The bits_per_pixel, red, green, blue, transp and nonstd fields must be set to +0 by applications and ignored by drivers. Note that the grayscale and fourcc +fields share the same memory location. Application must thus not set the +grayscale field to 0. + +Upon successful format configuration, drivers update the fb_fix_screeninfo +type, visual and line_length fields depending on the selected format. The +visual field is set to FB_VISUAL_FOURCC. + diff --git a/include/linux/fb.h b/include/linux/fb.h index 6a82748..359e98e 100644 --- a/include/linux/fb.h +++ b/include/linux/fb.h @@ -69,6 +69,7 @@ #define FB_VISUAL_PSEUDOCOLOR 3 /* Pseudo color (like atari) */ #define FB_VISUAL_DIRECTCOLOR 4 /* Direct color */ #define FB_VISUAL_STATIC_PSEUDOCOLOR 5 /* Pseudo color readonly */ +#define FB_VISUAL_FOURCC 6 /* Visual identified by a V4L2 FOURCC */ #define FB_ACCEL_NONE 0 /* no hardware accelerator */ #define FB_ACCEL_ATARIBLITT 1 /* Atari Blitter */ @@ -154,6 +155,8 @@ #define FB_ACCEL_PUV3_UNIGFX 0xa0 /* PKUnity-v3 Unigfx */ +#define FB_CAP_FOURCC 1 /* Device supports FOURCC-based formats */ + struct fb_fix_screeninfo { char id[16]; /* identification string eg "TT Builtin" */ unsigned long smem_start; /* Start of frame buffer mem */ @@ -171,7 +174,8 @@ struct fb_fix_screeninfo { __u32 mmio_len; /* Length of Memory Mapped I/O */ __u32 accel; /* Indicate to driver which */ /* specific chip/card we have */ - __u16 reserved[3]; /* Reserved for future compatibility */ + __u16 capabilities; /* see FB_CAP_* */ + __u16 reserved[2]; /* Reserved for future compatibility */ }; /* Interpretation of offset for color fields: All offsets are from the right, @@ -220,6 +224,7 @@ struct fb_bitfield { #define FB_VMODE_INTERLACED 1 /* interlaced */ #define FB_VMODE_DOUBLE 2 /* double scan */ #define FB_VMODE_ODD_FLD_FIRST 4 /* interlaced: top line first */ +#define FB_VMODE_FOURCC 8 /* video mode expressed as a FOURCC */ #define FB_VMODE_MASK 255 #define FB_VMODE_YWRAP 256 /* ywrap instead of panning */ @@ -246,7 +251,10 @@ struct fb_var_screeninfo { __u32 yoffset; /* resolution */ __u32 bits_per_pixel; /* guess what */ - __u32 grayscale; /* != 0 Graylevels instead of colors */ + union { + __u32 grayscale; /* != 0 Graylevels instead of colors */ + __u32 fourcc; /* FOURCC format */ + }; struct fb_bitfield red; /* bitfield in fb mem if true color, */ struct fb_bitfield green; /* else only length is significant */
This API will be used to support YUV frame buffer formats in a standard way. Last but not least, create a much needed fbdev API documentation and document the format setting APIs. Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> --- Documentation/fb/api.txt | 284 ++++++++++++++++++++++++++++++++++++++++++++++ include/linux/fb.h | 12 ++- 2 files changed, 294 insertions(+), 2 deletions(-) create mode 100644 Documentation/fb/api.txt A bit late, but here's a patch that implements an fbdev format configuration API to support YUV formats. My next step is to implement a prototype in an fbdev driver to make sure the API works well. Thanks to Florian Tobias Schandinat for providing feedback on the initial RFC. Comments are as usual more than welcome. If you feel like writing a couple of lines of documentation, feel free to extend the API doc. That's much needed.