Re: [PATCH] 2.6.16-rc6 calls check_acpi_pci() on x86 with ACPI disabled

Re: [PATCH] 2.6.16-rc6 calls check_acpi_pci() on x86 with ACPI disabled

[Andrew Morton]

Herbert Poetzl <herbert@13thfloor.at> wrote:
>
> 
> Hi Andrew! Folks!
> 
> check_acpi_pci() is called form arch/i386/kernel/setup.c
> even if CONFIG_ACPI is not defined, but the code in
> include/asm/acpi.h doesn't provide it in this case, 

Well that's a shame.

> so either we need to move the declaration outside the 
> CONFIG_ACPI check, or alternatively move the call in
> setup.c inside the CONFIG_ACPI one
> 
> attached two patches which would do this

Prefer the first version.  But it'll break if CONFIG_X86_IO_APIC &&
!CONFIG_ACPI

So how's about this?



From: Herbert Poetzl <herbert@13thfloor.at>

check_acpi_pci() is called from arch/i386/kernel/setup.c even if
CONFIG_ACPI is not defined, but the code in include/asm/acpi.h doesn't
provide it in this case.

Signed-off-by: Andrew Morton <akpm@osdl.org>
---

 include/asm/acpi.h |   10 ++++++----
 1 files changed, 6 insertions(+), 4 deletions(-)

diff -puN include/asm/acpi.h~dont-check_acpi_pci-on-x86-with-acpi-disabled include/asm/acpi.h
--- devel/include/asm/acpi.h~dont-check_acpi_pci-on-x86-with-acpi-disabled	2006-03-14 17:42:11.000000000 -0800
+++ devel-akpm/include/asm/acpi.h	2006-03-14 17:44:50.000000000 -0800
@@ -103,6 +103,12 @@ __acpi_release_global_lock (unsigned int
         :"=r"(n_hi), "=r"(n_lo)     \
         :"0"(n_hi), "1"(n_lo))
 
+#if defined(CONFIG_ACPI) && defined(CONFIG_X86_IO_APIC)
+extern void check_acpi_pci(void);
+#else
+static inline void check_acpi_pci(void) { }
+#endif
+
 #ifdef CONFIG_ACPI 
 extern int acpi_lapic;
 extern int acpi_ioapic;
@@ -128,8 +134,6 @@ extern int acpi_gsi_to_irq(u32 gsi, unsi
 extern int skip_ioapic_setup;
 extern int acpi_skip_timer_override;
 
-extern void check_acpi_pci(void);
-
 static inline void disable_ioapic_setup(void)
 {
 	skip_ioapic_setup = 1;
@@ -142,8 +146,6 @@ static inline int ioapic_setup_disabled(
 
 #else
 static inline void disable_ioapic_setup(void) { }
-static inline void check_acpi_pci(void) { }
-
 #endif
 
 static inline void acpi_noirq_set(void) { acpi_noirq = 1; }
_

Re: [PATCH] 2.6.16-rc6 calls check_acpi_pci() on x86 with ACPI disabled

[Herbert Poetzl]

On Tue, Mar 14, 2006 at 05:45:40PM -0800, Andrew Morton wrote:
> Herbert Poetzl <herbert@13thfloor.at> wrote:
> >
> > 
> > Hi Andrew! Folks!
> > 
> > check_acpi_pci() is called form arch/i386/kernel/setup.c
> > even if CONFIG_ACPI is not defined, but the code in
> > include/asm/acpi.h doesn't provide it in this case, 
> 
> Well that's a shame.
> 
> > so either we need to move the declaration outside the 
> > CONFIG_ACPI check, or alternatively move the call in
> > setup.c inside the CONFIG_ACPI one
> > 
> > attached two patches which would do this
> 
> Prefer the first version.  But it'll break if CONFIG_X86_IO_APIC &&
> !CONFIG_ACPI
> 
> So how's about this?

hmm, well, the comment around the check_acpi_pci() call
says: "Checks more than just ACPI actually", so I didn't
want to make it depend on ACPI in the 'first' version,
which now would change semantics, but if it is fine to
make it depend on ACPI, the second version might be the
simpler solution (which should have the same semantic as
your version ... I think

maybe the ACPI folks should clarify if this stuff has to
be run if ACPI is off, in which case renaming the thing
might be a good idea ...

best,
Herbert

> From: Herbert Poetzl <herbert@13thfloor.at>
> 
> check_acpi_pci() is called from arch/i386/kernel/setup.c even if
> CONFIG_ACPI is not defined, but the code in include/asm/acpi.h doesn't
> provide it in this case.
> 
> Signed-off-by: Andrew Morton <akpm@osdl.org>
> ---
> 
>  include/asm/acpi.h |   10 ++++++----
>  1 files changed, 6 insertions(+), 4 deletions(-)
> 
> diff -puN include/asm/acpi.h~dont-check_acpi_pci-on-x86-with-acpi-disabled include/asm/acpi.h
> --- devel/include/asm/acpi.h~dont-check_acpi_pci-on-x86-with-acpi-disabled	2006-03-14 17:42:11.000000000 -0800
> +++ devel-akpm/include/asm/acpi.h	2006-03-14 17:44:50.000000000 -0800
> @@ -103,6 +103,12 @@ __acpi_release_global_lock (unsigned int
>          :"=r"(n_hi), "=r"(n_lo)     \
>          :"0"(n_hi), "1"(n_lo))
>  
> +#if defined(CONFIG_ACPI) && defined(CONFIG_X86_IO_APIC)
> +extern void check_acpi_pci(void);
> +#else
> +static inline void check_acpi_pci(void) { }
> +#endif
> +
>  #ifdef CONFIG_ACPI 
>  extern int acpi_lapic;
>  extern int acpi_ioapic;
> @@ -128,8 +134,6 @@ extern int acpi_gsi_to_irq(u32 gsi, unsi
>  extern int skip_ioapic_setup;
>  extern int acpi_skip_timer_override;
>  
> -extern void check_acpi_pci(void);
> -
>  static inline void disable_ioapic_setup(void)
>  {
>  	skip_ioapic_setup = 1;
> @@ -142,8 +146,6 @@ static inline int ioapic_setup_disabled(
>  
>  #else
>  static inline void disable_ioapic_setup(void) { }
> -static inline void check_acpi_pci(void) { }
> -
>  #endif
>  
>  static inline void acpi_noirq_set(void) { acpi_noirq = 1; }
> _

Re: [PATCH] 2.6.16-rc6 calls check_acpi_pci() on x86 with ACPI disabled

[Andrew Morton]

Herbert Poetzl <herbert@13thfloor.at> wrote:
>
> On Tue, Mar 14, 2006 at 05:45:40PM -0800, Andrew Morton wrote:
> > Herbert Poetzl <herbert@13thfloor.at> wrote:
> > >
> > > 
> > > Hi Andrew! Folks!
> > > 
> > > check_acpi_pci() is called form arch/i386/kernel/setup.c
> > > even if CONFIG_ACPI is not defined, but the code in
> > > include/asm/acpi.h doesn't provide it in this case, 
> > 
> > Well that's a shame.
> > 
> > > so either we need to move the declaration outside the 
> > > CONFIG_ACPI check, or alternatively move the call in
> > > setup.c inside the CONFIG_ACPI one
> > > 
> > > attached two patches which would do this
> > 
> > Prefer the first version.  But it'll break if CONFIG_X86_IO_APIC &&
> > !CONFIG_ACPI
> > 
> > So how's about this?
> 
> hmm, well, the comment around the check_acpi_pci() call
> says: "Checks more than just ACPI actually", so I didn't
> want to make it depend on ACPI in the 'first' version,
> which now would change semantics, but if it is fine to
> make it depend on ACPI, the second version might be the
> simpler solution (which should have the same semantic as
> your version ... I think
> 
> maybe the ACPI folks should clarify if this stuff has to
> be run if ACPI is off, in which case renaming the thing
> might be a good idea ...

Yes, actually I didn't check closely enough - arch/i386/kernel/acpi/* gets
built even if CONFIG_ACPI=n (!)

So the code will actualy compile and link OK if we do:

#ifdef CONFIG_X86_IO_APIC
extern void check_acpi_pci(void);
#else
static inline void check_acpi_pci(void) { }
#endif

But we'd need the acpi guys to tell us what's actually intended here,
please.  Does it make sense to be calling this function in a non-ACPI
kernel?



erk, your patch was against include/asm/...  - please don't do that - it
doesn't work very well if the patch receiver isn't using a setup-for-i386
tree.

Re: [PATCH] 2.6.16-rc6 calls check_acpi_pci() on x86 with ACPI disabled

[Herbert Poetzl]

On Tue, Mar 14, 2006 at 06:06:51PM -0800, Andrew Morton wrote:
> Herbert Poetzl <herbert@13thfloor.at> wrote:
> >
> > On Tue, Mar 14, 2006 at 05:45:40PM -0800, Andrew Morton wrote:
> > > Herbert Poetzl <herbert@13thfloor.at> wrote:
> > > >
> > > > 
> > > > Hi Andrew! Folks!
> > > > 
> > > > check_acpi_pci() is called form arch/i386/kernel/setup.c
> > > > even if CONFIG_ACPI is not defined, but the code in
> > > > include/asm/acpi.h doesn't provide it in this case, 
> > > 
> > > Well that's a shame.
> > > 
> > > > so either we need to move the declaration outside the 
> > > > CONFIG_ACPI check, or alternatively move the call in
> > > > setup.c inside the CONFIG_ACPI one
> > > > 
> > > > attached two patches which would do this
> > > 
> > > Prefer the first version.  But it'll break if CONFIG_X86_IO_APIC &&
> > > !CONFIG_ACPI
> > > 
> > > So how's about this?
> > 
> > hmm, well, the comment around the check_acpi_pci() call
> > says: "Checks more than just ACPI actually", so I didn't
> > want to make it depend on ACPI in the 'first' version,
> > which now would change semantics, but if it is fine to
> > make it depend on ACPI, the second version might be the
> > simpler solution (which should have the same semantic as
> > your version ... I think
> > 
> > maybe the ACPI folks should clarify if this stuff has to
> > be run if ACPI is off, in which case renaming the thing
> > might be a good idea ...
> 
> Yes, actually I didn't check closely enough - arch/i386/kernel/acpi/* gets
> built even if CONFIG_ACPI=n (!)
> 
> So the code will actualy compile and link OK if we do:
> 
> #ifdef CONFIG_X86_IO_APIC
> extern void check_acpi_pci(void);
> #else
> static inline void check_acpi_pci(void) { }
> #endif
> 
> But we'd need the acpi guys to tell us what's actually intended here,
> please.  Does it make sense to be calling this function in a non-ACPI
> kernel?

yes, maybe it needs to be split into two parts, one
generic and the other acpi related ...

> erk, your patch was against include/asm/... 
> - please don't do that - 
> it doesn't work very well if the patch receiver isn't using a
> setup-for-i386 tree.

oops, sorry, I should know that, but I just forgot
about that ... will try to remember next time ...

best,
Herbert