Skip to contentSkip to conversation

klibc

PATCH In review Bob Picco 5 participants 7 messages Jun 1, 2006 – Jun 4, 2006 Kbuild

lore
Share & tools
lore

Bob Picco

author
#0 lore raw

PATCH 1/1 klibc

This change enabled me to boot on OpenPower 710. Otherwise klibc's statfs is built for 32 bit and not for 64 bit. The failure was in kinit when calling run_init. run_init would attempt to validate ramfs and failed calling statfs of "/".

 usr/include/sys/vfs.h |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
usr/include/sys/vfs.h +2 -1
Index: linux-2.6.17-rc5-mm2/usr/include/sys/vfs.h
===================================================================
@@ -13,7 +13,8 @@
 /* struct statfs64 -- there seems to be two standards -
    one for 32 and one for 64 bits, and they're incompatible... */
 
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) && \
+	!defined(__powerpc64__)
 
 struct statfs {
 	uint32_t f_type;
#1 lore raw

usr/include/sys/vfs.h line 17

Quoted 4 lines · Bob Picco
usr/include/sys/vfs.h +2 -1
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) && \
+	!defined(__powerpc64__)

Why not just !defined(__LP64__) ?

#2 lore raw
Quoted 7 lines · Brian F. G. Bidulock

usr/include/sys/vfs.h line 17

Quoted 4 lines · Bob Picco
usr/include/sys/vfs.h +2 -1
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) && \
+	!defined(__powerpc64__)

Why not just !defined(__LP64__) ?

_BITSIZE == 64 is really the right formula... if I remember correctly, there were some 64-bit platforms (Alpha?) which didn't conform, though. I will look at this later today.

-hpa

#3 lore raw
Quoted 9 lines · H. Peter Anvin
Quoted 7 lines · Brian F. G. Bidulock

usr/include/sys/vfs.h line 17

Quoted 4 lines · Bob Picco
usr/include/sys/vfs.h +2 -1
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) && \
+	!defined(__powerpc64__)

Why not just !defined(__LP64__) ?

Quoted 4 lines

_BITSIZE == 64 is really the right formula... if I remember correctly, there were some 64-bit platforms (Alpha?) which didn't conform, though. I will look at this later today.

-hpa

indeed aboves line contains an mistake by an earlier patch of mine.

-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) +#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__arch64__)

is atm needed to get statfs right on sparc.

Signature · 1 line
maks
#4 lore raw
Quoted 17 lines · maximilian attems
Quoted 10 lines · H. Peter Anvin
Quoted 6 lines · Brian F. G. Bidulock

usr/include/sys/vfs.h line 17

Quoted 4 lines · Bob Picco
usr/include/sys/vfs.h +2 -1
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__) && \
+	!defined(__powerpc64__)

Why not just !defined(__LP64__) ?

_BITSIZE == 64 is really the right formula... if I remember correctly, there were some 64-bit platforms (Alpha?) which didn't conform, though. I will look at this later today.

-hpa

indeed aboves line contains an mistake by an earlier patch of mine.

(quoted diff) +1 -1
-#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__sparc_v9__)
+#if !defined(__x86_64__) && !defined(__ia64__) && !defined(__arch64__)
Quoted 2 lines

is atm needed to get statfs right on sparc.

__arch64__ is ugly; it doesn't say it's a sparc thing. I have added -D__sparc_v9__ to the sparc64 MCONFIG file, so I think that's fine.

Perhaps the right thing to do is to make this an archconfig.h configurable.

-hpa

#5 lore raw

From: "H. Peter Anvin" <[email protected]>
Date: Sat, 03 Jun 2006 15:26:05 -0700

Quoted 4 lines

__arch64__ is ugly; it doesn't say it's a sparc thing. I have added -D__sparc_v9__ to the sparc64 MCONFIG file, so I think that's fine.

Perhaps the right thing to do is to make this an archconfig.h configurable.

Please don't do this, I'll explain why.

Using __sparc_v9__ is incorrect, here is the lowdown on these defines:

1) __sparc_v9__ means "-mcpu={ultrasparc*,niagara}" or "-mcpu=v9". Although this is implied by "-m64" it can be used for 32-bit code too.

__sparc_v9__ means "using v9 instructions" which does not necessarily mean 64-bit

2) "__sparc__ && __arch64__" means 64-bit sparc

People often get this wrong, and it certainly is confusing.

Jakub Jelinek gives a good explanation on the gcc mailing list here:

	http://gcc.gnu.org/ml/gcc/2002-12/msg00108.html

Thanks.

#6 lore raw
Quoted 14 lines · David Miller
Quoted 5 lines

__arch64__ is ugly; it doesn't say it's a sparc thing. I have added -D__sparc_v9__ to the sparc64 MCONFIG file, so I think that's fine.

Perhaps the right thing to do is to make this an archconfig.h configurable.

Please don't do this, I'll explain why.

Using __sparc_v9__ is incorrect, here is the lowdown on these defines:

1) __sparc_v9__ means "-mcpu={ultrasparc*,niagara}" or "-mcpu=v9". Although this is implied by "-m64" it can be used for 32-bit code too.

Quoted 7 lines

__sparc_v9__ means "using v9 instructions" which does not necessarily mean 64-bit

2) "__sparc__ && __arch64__" means 64-bit sparc

People often get this wrong, and it certainly is confusing.

Okay, how about __sparc64__ then (via -D in the MCONFIG file.)

I did change this particular instance to be a archconfig variable, so it doesn't exist (I try very much to avoid architecture-specific #ifdefs in bulk code.)

-hpa