| Message ID | 1437582854-3320-1-git-send-email-borneo.antonio@gmail.com |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
From: Antonio Borneo <borneo.antonio@gmail.com> Date: Thu, 23 Jul 2015 00:34:14 +0800 > - depends on (PCI || OF_IRQ) > + depends on (PCI || (OF_IRQ && GENERIC_PCI_IOMAP)) Having GENERIC_PCI_IOMAP withut PCI makes absolutely no sense to me. I can't see why a platform would provide GENERIC_PCI_IOMAP when it does not have PCI enabled. You need to explain this. When PCI is not enabled, you should not be using any interface pci_*() whatsoever. Especially such a core interface as pci_iomap(). ALTERNATIVELY, NOP versions of pci_iomap() and friends should be provided when PCI is disabled, which just return errors and non-success statuses. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jul 23, 2015 at 2:08 AM, David Miller <davem@davemloft.net> wrote: > From: Antonio Borneo <borneo.antonio@gmail.com> > Date: Thu, 23 Jul 2015 00:34:14 +0800 > >> - depends on (PCI || OF_IRQ) >> + depends on (PCI || (OF_IRQ && GENERIC_PCI_IOMAP)) > > Having GENERIC_PCI_IOMAP withut PCI makes absolutely no sense to me. > > I can't see why a platform would provide GENERIC_PCI_IOMAP when it > does not have PCI enabled. My understanding for GENERIC_PCI_IOMAP is to enable providing empty pci_iomap/pci_iounmap. Digging in git log, arch that used to provide such empty instances moved to a centralized implementation by using GENERIC_PCI_IOMAP. For me such empty functions should be available by default when PCI is not set, but this is not the case today. So there are cases that set GENERIC_PCI_IOMAP without setting PCI. E.g. make ARCH=mips workpad_defconfig make ARCH=microblaze allnoconfig Moreover, in lib/Kconfig GENERIC_PCI_IOMAP is selected when GENERIC_IOMAP is set, without dependencies from PCI. You can also check https://lkml.org/lkml/2013/9/11/76 already upstream in commit 78857614104a26cdada4c53eea104752042bf5a1 Anyway, no other Kconfig uses GENERIC_PCI_IOMAP within conditionals, which makes this patch quite unusual. If you think this is not the way to go, we can consider to revert both 0f8b6cea1f3271ccc37dd2847b39e235461e0ced b7d3282a245f44286b88e436b1d08c518cdf4569 and take some more time for a different solution. Regards, Antonio > You need to explain this. > > When PCI is not enabled, you should not be using any interface > pci_*() whatsoever. Especially such a core interface as pci_iomap(). > > ALTERNATIVELY, NOP versions of pci_iomap() and friends should be > provided when PCI is disabled, which just return errors and > non-success statuses. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
2015-07-23 4:35 GMT+03:00 Antonio Borneo <borneo.antonio@gmail.com>: > On Thu, Jul 23, 2015 at 2:08 AM, David Miller <davem@davemloft.net> wrote: >> From: Antonio Borneo <borneo.antonio@gmail.com> >> Date: Thu, 23 Jul 2015 00:34:14 +0800 >> >>> - depends on (PCI || OF_IRQ) >>> + depends on (PCI || (OF_IRQ && GENERIC_PCI_IOMAP)) >> >> Having GENERIC_PCI_IOMAP withut PCI makes absolutely no sense to me. >> >> I can't see why a platform would provide GENERIC_PCI_IOMAP when it >> does not have PCI enabled. > > My understanding for GENERIC_PCI_IOMAP is to enable providing empty > pci_iomap/pci_iounmap. There is also a non-empty implementation that calls to ioremap/iounmap (compiled in when PCI=y). So it's rather an option that every architecture should use if they don't have any special requirements (which seems to be true about SH, too). Maybe a name like "MY_ARCH_DOESNT_NEED_ANYTHING_SPECIAL_FROM_PCI_IOMAP" instead of GENERIC_PCI_IOMAP would be more reflective of its current semantics, but it doesn't look very pretty anyway :) > Digging in git log, arch that used to provide such empty instances > moved to a centralized implementation by using GENERIC_PCI_IOMAP. > For me such empty functions should be available by default when PCI is > not set, but this is not the case today. If we trust the commit message in 97a29d59fc222b36bac3ee3a8ae994f65bf7ffdf it looks like architectures that use their own PCI mapping functions would provide them unconditionally, so we shouldn't unconditionally include stubs in the generic header to avoid double definition. I think the way to fix this most consistently with what other arches do would be to move "select GENERIC_PCI_IOMAP" from arch/sh/Kconfig outside of "config PCI" (towards line 40 or so in the same file). This will ensure that all NOP stubs are applied correctly to avoid such compile-time errors. Happy to provide a 2-line patch to do the above, but I don't have any SH hardware to test. Best, Alexey -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
From: Antonio Borneo <borneo.antonio@gmail.com> Date: Thu, 23 Jul 2015 09:35:51 +0800 > Digging in git log, arch that used to provide such empty instances > moved to a centralized implementation by using GENERIC_PCI_IOMAP. > For me such empty functions should be available by default when PCI is > not set, but this is not the case today. Then that is what we should fix instead of these hackish per-driver patches. Make it so that these fallback implementations always are present when PCI is not set. -- To unsubscribe from this list: send the line "unsubscribe netdev" 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/drivers/net/ethernet/via/Kconfig b/drivers/net/ethernet/via/Kconfig index 9bf3ff1..96c527c 100644 --- a/drivers/net/ethernet/via/Kconfig +++ b/drivers/net/ethernet/via/Kconfig @@ -19,7 +19,7 @@ if NET_VENDOR_VIA config VIA_RHINE tristate "VIA Rhine support" - depends on (PCI || OF_IRQ) + depends on (PCI || (OF_IRQ && GENERIC_PCI_IOMAP)) depends on HAS_DMA select CRC32 select MII @@ -44,7 +44,7 @@ config VIA_RHINE_MMIO config VIA_VELOCITY tristate "VIA Velocity support" - depends on (PCI || (OF_ADDRESS && OF_IRQ)) + depends on (PCI || (OF_ADDRESS && OF_IRQ && GENERIC_PCI_IOMAP)) depends on HAS_DMA select CRC32 select CRC_CCITT
The builds of arch/sh are failing in linux-next with: drivers/net/ethernet/via/via-rhine.c:1098:2: error: implicit declaration of function 'pci_iomap' [-Werror=implicit-function-declaration] drivers/net/ethernet/via/via-rhine.c:1119:2: error: implicit declaration of function 'pci_iounmap' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors make[5]: *** [drivers/net/ethernet/via/via-rhine.o] Error 1 The logic in include/asm-generic/{io,iomap,pci_iomap}.h requires GENERIC_PCI_IOMAP to inline empty pci_iomap/pci_iounmap when PCI is not selected. Add GENERIC_PCI_IOMAP in the dependency list for VIA_RHINE and VIA_VELOCITY. Reported-by: Paul Gortmaker <paul.gortmaker@windriver.com> Signed-off-by: Antonio Borneo <borneo.antonio@gmail.com> --- drivers/net/ethernet/via/Kconfig | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)