On Fri, Dec 06, 2013 at 02:53:00PM +0000, Zheng, Lv wrote:
> Hi,
>
> > From: Zheng, Lv
> > Sent: Friday, December 06, 2013 4:52 PM
> > To: Wysocki, Rafael J; Brown, Len
> >
> > In Linux kernel, ACPICA is wrapped and safely exported by CONFIG_ACPI. So
> > all external modules should depend on CONFIG_ACPI rather than using ACPICA
> > header directly for stubbing. But if we moves <acpi/acpi.h> inclusions
> > into "#ifdef CONFIG_ACPI", build breakge can help to detect wrong ACPICA
> > dependent modules.
> >
> > One of the build breakage is:
> > arch/x86/pci/mmconfig-shared.c:389:1: error: unknown type name 'acpi_status'
> > arch/x86/pci/mmconfig-shared.c:389:47: warning: 'struct acpi_resource' declared inside parameter list [enabled by default]
> > arch/x86/pci/mmconfig-shared.c:389:47: warning: its scope is only this definition or declaration, which is probably not what you want
> > [enabled by default]
> > arch/x86/pci/mmconfig-shared.c: In function 'check_mcfg_resource':
> > arch/x86/pci/mmconfig-shared.c:392:33: error: storage size of 'address' isn't known
> > arch/x86/pci/mmconfig-shared.c:393:2: error: unknown type name 'acpi_status'
> > arch/x86/pci/mmconfig-shared.c:395:9: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:395:19: error: 'ACPI_RESOURCE_TYPE_FIXED_MEMORY32' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:395:19: note: each undeclared identifier is reported only once for each function it appears in
> > arch/x86/pci/mmconfig-shared.c:397:8: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:399:11: error: 'AE_OK' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:400:35: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:401:33: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:402:19: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:404:11: error: 'AE_CTRL_TERMINATE' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:407:10: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:407:20: error: 'ACPI_RESOURCE_TYPE_ADDRESS32' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:408:10: error: dereferencing pointer to incomplete type
> > arch/x86/pci/mmconfig-shared.c:408:20: error: 'ACPI_RESOURCE_TYPE_ADDRESS64' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:411:2: error: implicit declaration of function 'acpi_resource_to_address64' [-Werror=implicit-
> > function-declaration]
> > arch/x86/pci/mmconfig-shared.c:412:2: error: implicit declaration of function 'ACPI_FAILURE' [-Werror=implicit-function-declaration]
> > arch/x86/pci/mmconfig-shared.c:414:31: error: 'ACPI_MEMORY_RANGE' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c:392:33: warning: unused variable 'address' [-Wunused-variable]
> > arch/x86/pci/mmconfig-shared.c: At top level:
> > arch/x86/pci/mmconfig-shared.c:425:1: error: unknown type name 'acpi_status'
> > arch/x86/pci/mmconfig-shared.c:425:41: error: unknown type name 'acpi_handle'
> > arch/x86/pci/mmconfig-shared.c: In function 'is_acpi_reserved':
> > arch/x86/pci/mmconfig-shared.c:447:2: error: implicit declaration of function 'acpi_get_devices' [-Werror=implicit-function-
> > declaration]
> > arch/x86/pci/mmconfig-shared.c:447:30: error: 'find_mboard_resource' undeclared (first use in this function)
> > arch/x86/pci/mmconfig-shared.c: At top level:
> > arch/x86/pci/mmconfig-shared.c:389:20: warning: 'check_mcfg_resource' defined but not used [-Wunused-function]
>
> Let me say something about this error messages. Someone may face difficulties to reproduce it.
> Currently, this message cannot be seen in any kernel build.
> This is because this patch series doesn't include a fix to remove <acpi/acpi.h> inclusion from linux/sfi_acpi.h.
> To achieve this, I have offered 2 options:
> 1. Redefining ACPICA table stuff for CONFIG_ACPI=n environment, or
> 2. Implementing stubs for all ACPICA functions/globals/macros and introducing <linux/acpica.h> (where acpi/acpi.h is included) as a top level header.
> 3. Other solutions...
>
> If we choose solution 1, then this issue is exposed by a build test where CONFIG_ACPI=n and CONFIG_SFI=y.
> Since we haven't decided what to do for the table structures and it is really not a good approach to achieve the build safety for mmconfig-share.c by relying on link-stage optimization rather than compile-stage referencing. IMO, we need to sort it out, thus I make it a part of this patchset.
>
> Well, if we choose solution 2, this patch can only help to reduce the size of the kernel image.
OK, I assume you will make sure one of these fixes gets merged before the
change that causes the breakage, so bisection continues to work.
Therefore, I think it's best if both changes are merged through the same
tree, and I won't apply this via the PCI tree. Let me know if you need me
to do something.
Bjorn
> >
> > The root cause of this breakage is:
> > 1. The following commit doesn't protect ACPICA functions like
> > acpi_get_devices()/acpi_walk_resources()/acpi_resource_to_address64() in
> > CONFIG_SFI=y and CONFIG_ACPI=n environment:
> > Commit: 5f0db7a2fb78895a197f64e548333b3bbd433996
> > Author: Feng Tang <feng.tang@xxxxxxxxx>
> > Subject: SFI: Hook PCI MMCONFIG
> > First check ACPI, and if that fails, ask SFI to find the MCFG.
> > So it actually depends on the logic that in !CONFIG_ACPI builds, code
> > blocks under "if (!acpi_disabled)" will not be linked in.
> >
> > This patch fixes this issue by introducing a stub for MCFG entry checker
> > is_acpi_reserved() in !CONFIG_ACPI builds.
> >
> > Signed-off-by: Lv Zheng <lv.zheng@xxxxxxxxx>
> > Cc: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> > Cc: Len Brown <lenb@xxxxxxxxxx>
> > Cc: Feng Tang <feng.tang@xxxxxxxxx>
> > Cc: sfi-devel@xxxxxxxxxxxxxxxxxx
> > Cc: linux-pci@xxxxxxxxxxxxxxx
> > ---
> > arch/x86/pci/mmconfig-shared.c | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
> > index 248642f..2c04b22 100644
> > --- a/arch/x86/pci/mmconfig-shared.c
> > +++ b/arch/x86/pci/mmconfig-shared.c
> > @@ -369,6 +369,7 @@ static int __init pci_mmcfg_check_hostbridge(void)
> > return !list_empty(&pci_mmcfg_list);
> > }
> >
> > +#ifdef CONFIG_ACPI
> > static acpi_status check_mcfg_resource(struct acpi_resource *res, void *data)
> > {
> > struct resource *mcfg_res = data;
> > @@ -435,6 +436,12 @@ static int is_acpi_reserved(u64 start, u64 end, unsigned not_used)
> >
> > return mcfg_res.flags;
> > }
> > +#else
> > +static inline int is_acpi_reserved(u64 start, u64 end, unsigned not_used)
> > +{
> > + return 0;
> > +}
> > +#endif
> >
> > typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
> >
> > --
> > 1.7.10
>
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
