Skip to contentSkip to conversation

ia64 allmodconfig bustage in 2.6.30-rc2

DISCUSSION Andrew Morton 3 participants 3 messages Apr 17, 2009 – Apr 18, 2009 PARAVIRT_OPS interface

lore
Share & tools
lore
#0 lore raw

`make allmodconfig'
`make vmlinux'

CC arch/ia64/kernel/asm-offsets.s
In file included from include/linux/bitops.h:17,

                 from include/linux/kernel.h:15,
                 from include/linux/sched.h:52,
                 from arch/ia64/kernel/asm-offsets.c:9:

/usr/src/devel/arch/ia64/include/asm/bitops.h: In function `set_bit':
/usr/src/devel/arch/ia64/include/asm/bitops.h:47: error: implicit declaration of function `BUILD_BUG_ON'

I fiddled around a bit but failed to fix it then I got bored.

#1 lore raw
Quoted 10 lines

`make allmodconfig'
`make vmlinux'

CC arch/ia64/kernel/asm-offsets.s
In file included from include/linux/bitops.h:17,

                from include/linux/kernel.h:15,
                from include/linux/sched.h:52,
                from arch/ia64/kernel/asm-offsets.c:9:

/usr/src/devel/arch/ia64/include/asm/bitops.h: In function `set_bit':
/usr/src/devel/arch/ia64/include/asm/bitops.h:47: error: implicit declaration of function `BUILD_BUG_ON'

Line 47 of bitops.h is a CMPXCHG_BUGCHECK() and allmodconfig has CONFIG_IA64_DEBUG_CMPXCHG=y, so we pick the debug version that uses ia64_getreg(_IA64_REG_IP) to printk a message about where we are stuck.

Now allmodconfig picks CONFIG_PARAVIRT=y ... so this ends up taking us to __paravirt_getreg(), which is where the BUILD_BUG_ON() gets introduced. So we need to include <linux/kernel.h> somewhere (or somewhere earlier than we currently do) to make this alright.

But our probem is that kernel.h goes off and includes bitops.h long before it defines BUILD_BUG_ON. Just moving the definition of BUILD_BUG_ON up before we #include anything else (ugly) fixes this immediate problem ... but then there are lots of other errors from asm/bitops.h :-(

Deleting the two BUILD_BUG_ON instances from paravirt_privop.h does get the build up on its feet and running. But that loses us some compile time checking for a coding/usage error.

Yamahata-san: How important do you think the BUILD_BUG_ON() checks are here? What would the failure look like if we dropped them, and at some later time someone coded a call to ia64_getreg with a non-constant register number as argument?

-Tony

#2 lore raw
In reply to Andrew Morton
Quoted 31 lines · Tony
Quoted 10 lines

`make allmodconfig'
`make vmlinux'

CC arch/ia64/kernel/asm-offsets.s
In file included from include/linux/bitops.h:17,

                from include/linux/kernel.h:15,
                from include/linux/sched.h:52,
                from arch/ia64/kernel/asm-offsets.c:9:

/usr/src/devel/arch/ia64/include/asm/bitops.h: In function `set_bit':
/usr/src/devel/arch/ia64/include/asm/bitops.h:47: error: implicit declaration of function `BUILD_BUG_ON'

Line 47 of bitops.h is a CMPXCHG_BUGCHECK() and allmodconfig has CONFIG_IA64_DEBUG_CMPXCHG=y, so we pick the debug version that uses ia64_getreg(_IA64_REG_IP) to printk a message about where we are stuck.

Now allmodconfig picks CONFIG_PARAVIRT=y ... so this ends up taking us to __paravirt_getreg(), which is where the BUILD_BUG_ON() gets introduced. So we need to include <linux/kernel.h> somewhere (or somewhere earlier than we currently do) to make this alright.

But our probem is that kernel.h goes off and includes bitops.h long before it defines BUILD_BUG_ON. Just moving the definition of BUILD_BUG_ON up before we #include anything else (ugly) fixes this immediate problem ... but then there are lots of other errors from asm/bitops.h :-(

Deleting the two BUILD_BUG_ON instances from paravirt_privop.h does get the build up on its feet and running. But that loses us some compile time checking for a coding/usage error.

Quoted 7 lines

Yamahata-san: How important do you think the BUILD_BUG_ON() checks are here? What would the failure look like if we dropped them, and at some later time someone coded a call to ia64_getreg with a non-constant register number as argument?

-Tony

Oh, sorry. I think removing it doesn't spoil compile time check because the error the BUILD_BUG_ON() catches is also caught when CONFIG_PARAVIRT=n. Here is the patch.

From f8f6a92b7433debdc8e70157ae2d70cbadcdd90b Mon Sep 17 00:00:00 2001
From: Isaku Yamahata <[email protected]>
Date: Sat, 18 Apr 2009 12:15:23 +0900
Subject: [PATCH] ia64/pv_ops: fix allmodconfig compilation breakage.

This patch fixes the following compilation error caused by recursive inclusion of kernel.h which defines BUILD_BUG_ON().
In this case, the case it catches will be caught by the case CONFIG_PARAVIRT=n, so removing it would not hurt compile time check very much. So fix the breakage by removing it.

CC arch/ia64/kernel/asm-offsets.s
In file included from include/linux/bitops.h:17,

                 from include/linux/kernel.h:15,
                 from include/linux/sched.h:52,
                 from arch/ia64/kernel/asm-offsets.c:9:

arch/ia64/include/asm/bitops.h: In function 'set_bit':
arch/ia64/include/asm/bitops.h:47: error: implicit declaration of function 'BUILD_BUG_ON'

---
 arch/ia64/include/asm/paravirt_privop.h |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
arch/ia64/include/asm/paravirt_privop.h +0 -2
index 3d29511..8f6cb11 100644
@@ -445,7 +445,6 @@ paravirt_set_rr0_to_rr4(unsigned long val0, unsigned long val1,
 		register unsigned long ia64_intri_res asm ("r8");	\
 		register unsigned long __reg asm ("r8") = (reg);	\
 									\
-		BUILD_BUG_ON(!__builtin_constant_p(reg));		\
 		asm volatile (paravirt_alt_bundle(__PARAVIRT_BR,	\
 						  PARAVIRT_TYPE(GETREG) \
 						  + (reg))		\
@@ -464,7 +463,6 @@ paravirt_set_rr0_to_rr4(unsigned long val0, unsigned long val1,
 		register unsigned long ia64_clobber1 asm ("r8");	\
 		register unsigned long ia64_clobber2 asm ("r9");	\
 									\
-		BUILD_BUG_ON(!__builtin_constant_p(reg));		\
 		asm volatile (paravirt_alt_bundle(__PARAVIRT_BR,	\
 						  PARAVIRT_TYPE(SETREG) \
 						  + (reg))		\
Signature · 6 lines
1.6.0.2



-- 
yamahata