| Message ID | 20210727093015.1225107-1-javierm@redhat.com (mailing list archive) |
|---|---|
| State | Mainlined, archived |
| Headers | show |
| Series | [v2] drivers/firmware: fix SYSFB depends to prevent build failures | expand |
On Tue, Jul 27, 2021 at 11:30:15AM +0200, Javier Martinez Canillas wrote: > The Generic System Framebuffers support is built when the COMPILE_TEST > option is enabled. But this wrongly assumes that all the architectures > declare a struct screen_info. > > This is true for most architectures, but at least the following do not: > arc, m68k, microblaze, openrisc, parisc and s390. > > By attempting to make this compile testeable on all architectures, it > leads to linking errors as reported by the kernel test robot for parisc: > > All errors (new ones prefixed by >>): > > hppa-linux-ld: drivers/firmware/sysfb.o: in function `sysfb_init': > (.init.text+0x24): undefined reference to `screen_info' > >> hppa-linux-ld: (.init.text+0x28): undefined reference to `screen_info' > > To prevent these errors only allow sysfb to be built on systems that are > going to need it, which are x86 BIOS and EFI. > > The EFI Kconfig symbol is used instead of (ARM || ARM64 || RISC) because > some of these architectures only declare a struct screen_info if EFI is > enabled. And also, because the SYSFB code is only used for EFI on these > architectures. For !EFI the "simple-framebuffer" device is registered by > OF when parsing the Device Tree Blob (if a DT node for this was defined). > > Fixes: d391c5827107 ("drivers/firmware: move x86 Generic System Framebuffers support") > Reported-by: kernel test robot <lkp@intel.com> > Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> Whacked onto drm-next so we're welcome again in linux-next. -Daniel > --- > > Changes in v2: > - Add a Fixes tag to the changelog. > > drivers/firmware/Kconfig | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/firmware/Kconfig b/drivers/firmware/Kconfig > index af6719cc576b..897f5f25c64e 100644 > --- a/drivers/firmware/Kconfig > +++ b/drivers/firmware/Kconfig > @@ -254,7 +254,7 @@ config QCOM_SCM_DOWNLOAD_MODE_DEFAULT > config SYSFB > bool > default y > - depends on X86 || ARM || ARM64 || RISCV || COMPILE_TEST > + depends on X86 || EFI > > config SYSFB_SIMPLEFB > bool "Mark VGA/VBE/EFI FB as generic system framebuffer" > -- > 2.31.1 >
Hi Javier, On Tue, Jul 27, 2021 at 11:33 AM Javier Martinez Canillas <javierm@redhat.com> wrote: > The Generic System Framebuffers support is built when the COMPILE_TEST > option is enabled. But this wrongly assumes that all the architectures > declare a struct screen_info. > > This is true for most architectures, but at least the following do not: > arc, m68k, microblaze, openrisc, parisc and s390. > > By attempting to make this compile testeable on all architectures, it > leads to linking errors as reported by the kernel test robot for parisc: > > All errors (new ones prefixed by >>): > > hppa-linux-ld: drivers/firmware/sysfb.o: in function `sysfb_init': > (.init.text+0x24): undefined reference to `screen_info' > >> hppa-linux-ld: (.init.text+0x28): undefined reference to `screen_info' > > To prevent these errors only allow sysfb to be built on systems that are > going to need it, which are x86 BIOS and EFI. > > The EFI Kconfig symbol is used instead of (ARM || ARM64 || RISC) because > some of these architectures only declare a struct screen_info if EFI is > enabled. And also, because the SYSFB code is only used for EFI on these > architectures. For !EFI the "simple-framebuffer" device is registered by > OF when parsing the Device Tree Blob (if a DT node for this was defined). > > Fixes: d391c5827107 ("drivers/firmware: move x86 Generic System Framebuffers support") > Reported-by: kernel test robot <lkp@intel.com> > Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> Thanks for your patch! > --- a/drivers/firmware/Kconfig > +++ b/drivers/firmware/Kconfig > @@ -254,7 +254,7 @@ config QCOM_SCM_DOWNLOAD_MODE_DEFAULT > config SYSFB > bool > default y > - depends on X86 || ARM || ARM64 || RISCV || COMPILE_TEST > + depends on X86 || EFI Thanks, much better. Still, now this worm is crawling out of the X86 can, I'm wondering why this option is so important that it has to default to y? It is not just a dependency for SYSFB_SIMPLEFB, but also causes the inclusion of drivers/firmware/sysfb.c. > > config SYSFB_SIMPLEFB > bool "Mark VGA/VBE/EFI FB as generic system framebuffer" Gr{oetje,eeting}s, Geert
Hello Geert, On 7/27/21 12:03 PM, Geert Uytterhoeven wrote: [snip] > Thanks for your patch! > You are welcome. >> --- a/drivers/firmware/Kconfig >> +++ b/drivers/firmware/Kconfig >> @@ -254,7 +254,7 @@ config QCOM_SCM_DOWNLOAD_MODE_DEFAULT >> config SYSFB >> bool >> default y >> - depends on X86 || ARM || ARM64 || RISCV || COMPILE_TEST >> + depends on X86 || EFI > > Thanks, much better. > Still, now this worm is crawling out of the X86 can, I'm wondering > why this option is so important that it has to default to y? > It is not just a dependency for SYSFB_SIMPLEFB, but also causes the > inclusion of drivers/firmware/sysfb.c. > It defaults to yes because drivers/firmware/sysfb.c contains the logic to register a "simple-framebuffer" device (or "efi-framebuffer" if the CONFIG_SYSFB_SIMPLEFB Kconfig symbol is not enabled). Not enabling this, would mean that a platform device to match a driver supporting the EFI GOP framebuffer (e.g: simple{drm,fb} or efifb) will not be registered. Which will lead to not having an early framebuffer. The logic used to be in drivers/firmware/efi/efi-init.c, that's built in if CONFIG_EFI is enabled. We just consolidated both X86 and EFI: https://cgit.freedesktop.org/drm/drm-misc/commit/?id=8633ef82f101 Best regards,
Hi Javier, On Tue, Jul 27, 2021 at 12:22 PM Javier Martinez Canillas <javierm@redhat.com> wrote: > On 7/27/21 12:03 PM, Geert Uytterhoeven wrote: > >> --- a/drivers/firmware/Kconfig > >> +++ b/drivers/firmware/Kconfig > >> @@ -254,7 +254,7 @@ config QCOM_SCM_DOWNLOAD_MODE_DEFAULT > >> config SYSFB > >> bool > >> default y > >> - depends on X86 || ARM || ARM64 || RISCV || COMPILE_TEST > >> + depends on X86 || EFI > > > > Thanks, much better. > > Still, now this worm is crawling out of the X86 can, I'm wondering > > why this option is so important that it has to default to y? > > It is not just a dependency for SYSFB_SIMPLEFB, but also causes the > > inclusion of drivers/firmware/sysfb.c. > > > > It defaults to yes because drivers/firmware/sysfb.c contains the logic > to register a "simple-framebuffer" device (or "efi-framebuffer" if the > CONFIG_SYSFB_SIMPLEFB Kconfig symbol is not enabled). > > Not enabling this, would mean that a platform device to match a driver > supporting the EFI GOP framebuffer (e.g: simple{drm,fb} or efifb) will > not be registered. Which will lead to not having an early framebuffer. Do all (embedded) EFI systems have a frame buffer? Perhaps SYSFB should be selected by SYSFB_SIMPLEFB, FB_VESA, and FB_EFI? > The logic used to be in drivers/firmware/efi/efi-init.c, that's built > in if CONFIG_EFI is enabled. We just consolidated both X86 and EFI: > > https://cgit.freedesktop.org/drm/drm-misc/commit/?id=8633ef82f101 Thanks, I'm aware of that commit, as I was just about to reply to it, when I saw the patch is this thread ;-) Gr{oetje,eeting}s, Geert
On 7/27/21 1:17 PM, Geert Uytterhoeven wrote: [snip] >> Not enabling this, would mean that a platform device to match a driver >> supporting the EFI GOP framebuffer (e.g: simple{drm,fb} or efifb) will >> not be registered. Which will lead to not having an early framebuffer. > > Do all (embedded) EFI systems have a frame buffer? > That's a good question. I don't know if all EFI firmwares are expected to provide a GOP or not. But even the u-boot EFI stub provides one, if video output is supported by u-boot on that system. > Perhaps SYSFB should be selected by SYSFB_SIMPLEFB, FB_VESA, > and FB_EFI? > It's another option, yes. I just thought that the use of select was not encouraged and using depends was less fragile / error prone. >> The logic used to be in drivers/firmware/efi/efi-init.c, that's built >> in if CONFIG_EFI is enabled. We just consolidated both X86 and EFI: >> >> https://cgit.freedesktop.org/drm/drm-misc/commit/?id=8633ef82f101 > > Thanks, I'm aware of that commit, as I was just about to reply to it, > when I saw the patch is this thread ;-) > Ok :) Best regards,
Hi Javier, On Tue, Jul 27, 2021 at 1:32 PM Javier Martinez Canillas <javierm@redhat.com> wrote: > On 7/27/21 1:17 PM, Geert Uytterhoeven wrote: > >> Not enabling this, would mean that a platform device to match a driver > >> supporting the EFI GOP framebuffer (e.g: simple{drm,fb} or efifb) will > >> not be registered. Which will lead to not having an early framebuffer. > > > > Do all (embedded) EFI systems have a frame buffer? > > That's a good question. I don't know if all EFI firmwares are expected > to provide a GOP or not. But even the u-boot EFI stub provides one, if > video output is supported by u-boot on that system. > > > Perhaps SYSFB should be selected by SYSFB_SIMPLEFB, FB_VESA, > > and FB_EFI? > > It's another option, yes. I just thought that the use of select was not > encouraged and using depends was less fragile / error prone. Select is very useful for config symbols that are invisible to the user (i.e. cannot be enabled/disabled manually). Gr{oetje,eeting}s, Geert
On Tue, Jul 27, 2021 at 01:32:29PM +0200, Javier Martinez Canillas wrote: > On 7/27/21 1:17 PM, Geert Uytterhoeven wrote: > > Do all (embedded) EFI systems have a frame buffer? > That's a good question. I don't know if all EFI firmwares are expected > to provide a GOP or not. But even the u-boot EFI stub provides one, if > video output is supported by u-boot on that system. No, GOP is only mandatory for systems with a graphical console device even assuming things fully conform to the spec. Indeed I've got a non-embedded EFI based system sitting next to my desk here which only has a serial console, never mind anything embedded.
Hello Geert, On 7/27/21 1:39 PM, Geert Uytterhoeven wrote: [snip] >>> Perhaps SYSFB should be selected by SYSFB_SIMPLEFB, FB_VESA, >>> and FB_EFI? >> >> It's another option, yes. I just thought that the use of select was not >> encouraged and using depends was less fragile / error prone. > > Select is very useful for config symbols that are invisible to the user (i.e. > cannot be enabled/disabled manually). > Got it. I don't have a strong opinion on this really. In fact, the first version of the patch did use select for the arches but I got as feedback that should use depends instead: https://www.mail-archive.com/dri-devel@lists.freedesktop.org/msg351961.html Granted, that was for the arches but you are proposing to do it for the drivers that match against the platform devices registered by sysfb. So it does make more sense to what I did in v1. Best regards,
diff --git a/drivers/firmware/Kconfig b/drivers/firmware/Kconfig index af6719cc576b..897f5f25c64e 100644 --- a/drivers/firmware/Kconfig +++ b/drivers/firmware/Kconfig @@ -254,7 +254,7 @@ config QCOM_SCM_DOWNLOAD_MODE_DEFAULT config SYSFB bool default y - depends on X86 || ARM || ARM64 || RISCV || COMPILE_TEST + depends on X86 || EFI config SYSFB_SIMPLEFB bool "Mark VGA/VBE/EFI FB as generic system framebuffer"
The Generic System Framebuffers support is built when the COMPILE_TEST option is enabled. But this wrongly assumes that all the architectures declare a struct screen_info. This is true for most architectures, but at least the following do not: arc, m68k, microblaze, openrisc, parisc and s390. By attempting to make this compile testeable on all architectures, it leads to linking errors as reported by the kernel test robot for parisc: All errors (new ones prefixed by >>): hppa-linux-ld: drivers/firmware/sysfb.o: in function `sysfb_init': (.init.text+0x24): undefined reference to `screen_info' >> hppa-linux-ld: (.init.text+0x28): undefined reference to `screen_info' To prevent these errors only allow sysfb to be built on systems that are going to need it, which are x86 BIOS and EFI. The EFI Kconfig symbol is used instead of (ARM || ARM64 || RISC) because some of these architectures only declare a struct screen_info if EFI is enabled. And also, because the SYSFB code is only used for EFI on these architectures. For !EFI the "simple-framebuffer" device is registered by OF when parsing the Device Tree Blob (if a DT node for this was defined). Fixes: d391c5827107 ("drivers/firmware: move x86 Generic System Framebuffers support") Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Javier Martinez Canillas <javierm@redhat.com> --- Changes in v2: - Add a Fixes tag to the changelog. drivers/firmware/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)