The Device Manager code barfs when sector_t is 64bit wide (i.e. an u64) and
CONFIG_LBD is off. This happens on printk(), resulting in wrong memory
accesses, but also on sscanf(), resulting in overflows (because it uses %lu
for a long long in this case). And region_t, chunk_t are typedefs for
sector_t, so we have warnings for these, too.
Andrew Morton suggested simply using "%llu" and casting sector_t to unsigned
long long; but he missed the sscanf()'s, which cannot be fixed this way.
I've used %llu instead of %Lu for standards conformance, as suggested by
The problem is this code in drivers/md/dm.h:
/*
* FIXME: I think this should be with the definition of sector_t
* in types.h.
*/
#ifdef CONFIG_LBD
#define SECTOR_FORMAT "%Lu"
#else
#define SECTOR_FORMAT "%lu"
#endif
So we must fix the FIXME. However, having sector_t defined by the arch is
wrong in all current cases, and we fix these: we can simply decide this in
linux/types.h following CONFIG_LBD.
So, I have also removed HAVE_SECTOR_T; you can readd it, but it has no users.
All 64-bit arch use a 64-bit wide long for sector_t; almost all 32-bit archs
use a long long only if CONFIG_LBD is on. The only exception is h8300: for
that case, we add CONFIG_LBD = y and we are again in the general case.
And x86_64 does not need to define sector_t on its own.
Sample warnings (from both 2.6.8.1 and 2.6.9-rc2):
drivers/md/dm-raid1.c: In function `get_mirror':
drivers/md/dm-raid1.c:930: warning: long unsigned int format, sector_t arg (arg 3)
drivers/md/dm-raid1.c: In function `mirror_status':
drivers/md/dm-raid1.c:1200: warning: long unsigned int format, region_t arg (arg 4)
drivers/md/dm-raid1.c:1200: warning: long unsigned int format, region_t arg (arg 5)
drivers/md/dm-raid1.c:1206: warning: long unsigned int format, sector_t arg (arg 5)
drivers/md/dm-raid1.c:1212: warning: long unsigned int format, sector_t arg (arg 5)
Signed-off-by: Paolo 'Blaisorblade' Giarrusso <[email protected]>
---
linux-2.6.9-current-paolo/arch/h8300/Kconfig | 4 ++++
linux-2.6.9-current-paolo/drivers/md/dm.h | 10 ----------
linux-2.6.9-current-paolo/include/asm-h8300/types.h | 3 ---
linux-2.6.9-current-paolo/include/asm-i386/types.h | 5 -----
linux-2.6.9-current-paolo/include/asm-mips/types.h | 5 -----
linux-2.6.9-current-paolo/include/asm-ppc/types.h | 5 -----
linux-2.6.9-current-paolo/include/asm-s390/types.h | 5 -----
linux-2.6.9-current-paolo/include/asm-sh/types.h | 5 -----
linux-2.6.9-current-paolo/include/asm-x86_64/types.h | 3 ---
linux-2.6.9-current-paolo/include/linux/types.h | 17 ++++++++++++++---
10 files changed, 18 insertions(+), 44 deletions(-)
diff -puN drivers/md/dm.h~fix-dm-warnings drivers/md/dm.h
--- linux-2.6.9-current/drivers/md/dm.h~fix-dm-warnings 2004-10-16 21:58:29.912802872 +0200
+++ linux-2.6.9-current-paolo/drivers/md/dm.h 2004-10-16 21:58:29.954796488 +0200
@@ -22,16 +22,6 @@
#define DMEMIT(x...) sz += ((sz >= maxlen) ? \
0 : scnprintf(result + sz, maxlen - sz, x))
-/*
- * FIXME: I think this should be with the definition of sector_t
- * in types.h.
- */
-#ifdef CONFIG_LBD
-#define SECTOR_FORMAT "%Lu"
-#else
-#define SECTOR_FORMAT "%lu"
-#endif
-
#define SECTOR_SHIFT 9
/*
diff -puN include/asm-h8300/types.h~fix-dm-warnings include/asm-h8300/types.h
--- linux-2.6.9-current/include/asm-h8300/types.h~fix-dm-warnings 2004-10-16 21:58:29.913802720 +0200
+++ linux-2.6.9-current-paolo/include/asm-h8300/types.h 2004-10-16 21:58:29.954796488 +0200
@@ -55,9 +55,6 @@ typedef unsigned long long u64;
typedef u32 dma_addr_t;
-#define HAVE_SECTOR_T
-typedef u64 sector_t;
-
typedef unsigned int kmem_bufctl_t;
#endif /* __KERNEL__ */
diff -puN include/asm-i386/types.h~fix-dm-warnings include/asm-i386/types.h
--- linux-2.6.9-current/include/asm-i386/types.h~fix-dm-warnings 2004-10-16 21:58:29.914802568 +0200
+++ linux-2.6.9-current-paolo/include/asm-i386/types.h 2004-10-16 21:58:29.955796336 +0200
@@ -58,11 +58,6 @@ typedef u32 dma_addr_t;
#endif
typedef u64 dma64_addr_t;
-#ifdef CONFIG_LBD
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-#endif
-
typedef unsigned short kmem_bufctl_t;
#endif /* __ASSEMBLY__ */
diff -puN include/asm-mips/types.h~fix-dm-warnings include/asm-mips/types.h
--- linux-2.6.9-current/include/asm-mips/types.h~fix-dm-warnings 2004-10-16 21:58:29.915802416 +0200
+++ linux-2.6.9-current-paolo/include/asm-mips/types.h 2004-10-16 21:58:29.955796336 +0200
@@ -94,11 +94,6 @@ typedef unsigned long long phys_t;
typedef unsigned long phys_t;
#endif
-#ifdef CONFIG_LBD
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-#endif
-
typedef unsigned short kmem_bufctl_t;
#endif /* __ASSEMBLY__ */
diff -puN include/asm-ppc/types.h~fix-dm-warnings include/asm-ppc/types.h
--- linux-2.6.9-current/include/asm-ppc/types.h~fix-dm-warnings 2004-10-16 21:58:29.916802264 +0200
+++ linux-2.6.9-current-paolo/include/asm-ppc/types.h 2004-10-16 21:58:29.955796336 +0200
@@ -57,11 +57,6 @@ typedef __vector128 vector128;
typedef u32 dma_addr_t;
typedef u64 dma64_addr_t;
-#ifdef CONFIG_LBD
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-#endif
-
typedef unsigned int kmem_bufctl_t;
#endif /* __ASSEMBLY__ */
diff -puN include/asm-s390/types.h~fix-dm-warnings include/asm-s390/types.h
--- linux-2.6.9-current/include/asm-s390/types.h~fix-dm-warnings 2004-10-16 21:58:29.917802112 +0200
+++ linux-2.6.9-current-paolo/include/asm-s390/types.h 2004-10-16 21:58:29.955796336 +0200
@@ -90,11 +90,6 @@ typedef union {
} subreg;
} register_pair;
-#ifdef CONFIG_LBD
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-#endif
-
#endif /* ! __s390x__ */
#endif /* __ASSEMBLY__ */
#endif /* __KERNEL__ */
diff -puN include/asm-sh/types.h~fix-dm-warnings include/asm-sh/types.h
--- linux-2.6.9-current/include/asm-sh/types.h~fix-dm-warnings 2004-10-16 21:58:29.918801960 +0200
+++ linux-2.6.9-current-paolo/include/asm-sh/types.h 2004-10-16 21:58:29.956796184 +0200
@@ -53,11 +53,6 @@ typedef unsigned long long u64;
typedef u32 dma_addr_t;
-#ifdef CONFIG_LBD
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-#endif
-
typedef unsigned int kmem_bufctl_t;
#endif /* __ASSEMBLY__ */
diff -puN include/asm-x86_64/types.h~fix-dm-warnings include/asm-x86_64/types.h
--- linux-2.6.9-current/include/asm-x86_64/types.h~fix-dm-warnings 2004-10-16 21:58:29.919801808 +0200
+++ linux-2.6.9-current-paolo/include/asm-x86_64/types.h 2004-10-16 21:58:29.956796184 +0200
@@ -48,9 +48,6 @@ typedef unsigned long long u64;
typedef u64 dma64_addr_t;
typedef u64 dma_addr_t;
-typedef u64 sector_t;
-#define HAVE_SECTOR_T
-
typedef unsigned short kmem_bufctl_t;
#endif /* __ASSEMBLY__ */
diff -puN include/linux/types.h~fix-dm-warnings include/linux/types.h
--- linux-2.6.9-current/include/linux/types.h~fix-dm-warnings 2004-10-16 21:58:29.951796944 +0200
+++ linux-2.6.9-current-paolo/include/linux/types.h 2004-10-16 21:58:29.956796184 +0200
@@ -125,12 +125,23 @@ typedef __s64 int64_t;
/*
* The type used for indexing onto a disc or disc partition.
- * If required, asm/types.h can override it and define
- * HAVE_SECTOR_T
+ * You cannot override it any more in asm- includes; define CONFIG_LBD
+ * to turn it inside a long long (only on 32-bit archs).
+ * The DM code must also scanf sector_t's, so here we define SECTOR_FORMAT
+ * for them.
*/
-#ifndef HAVE_SECTOR_T
+
+#ifndef CONFIG_LBD
typedef unsigned long sector_t;
+#define SECTOR_FORMAT "%lu"
+#else /* CONFIG_LBD */
+#if BITS_PER_LONG == 64
+#error Cannot define CONFIG_LBD on 64-bit archs.
+#else
+typedef unsigned long long sector_t;
+#define SECTOR_FORMAT "%llu"
#endif
+#endif /* CONFIG_LBD */
/*
* The type of an index into the pagecache. Use a #define so asm/types.h
diff -puN arch/h8300/Kconfig~fix-dm-warnings arch/h8300/Kconfig
--- linux-2.6.9-current/arch/h8300/Kconfig~fix-dm-warnings 2004-10-16 21:58:29.952796792 +0200
+++ linux-2.6.9-current-paolo/arch/h8300/Kconfig 2004-10-16 21:58:29.956796184 +0200
@@ -25,6 +25,10 @@ config UID16
bool
default y
+config LBD
+ bool
+ default y
+
config RWSEM_GENERIC_SPINLOCK
bool
default y
_
[email protected] wrote:
>
> The Device Manager code barfs when sector_t is 64bit wide (i.e. an u64) and
> CONFIG_LBD is off. This happens on printk(), resulting in wrong memory
> accesses, but also on sscanf(), resulting in overflows (because it uses %lu
> for a long long in this case). And region_t, chunk_t are typedefs for
> sector_t, so we have warnings for these, too.
This patch caused the x86_64 build to puke: "Not allowed to define
CONFIG_LBD on this architecture" or some such.
I'd much prefer that you simply remove SECTOR_FORMAT completely.
On Friday 22 October 2004 11:53, Andrew Morton wrote:
> [email protected] wrote:
> > The Device Manager code barfs when sector_t is 64bit wide (i.e. an u64)
> > and CONFIG_LBD is off. This happens on printk(), resulting in wrong
> > memory accesses, but also on sscanf(), resulting in overflows (because it
> > uses %lu for a long long in this case). And region_t, chunk_t are
> > typedefs for sector_t, so we have warnings for these, too.
> This patch caused the x86_64 build to puke: "Not allowed to define
> CONFIG_LBD on this architecture" or some such.
Yes, there is an #error directive for this.
Who would ever need to use CONFIG_LBD when sector_t is already 64-bit? So this
is flagged as an error (feel free to remove that, if you know a good reason).
In fact, this is your fault, given both the 2.6.9 kernel and the 2.6.10-rc1
one:
config LBD
bool "Support for Large Block Devices"
depends on X86 || MIPS32 || PPC32 || ARCH_S390_31 || SUPERH
help
Say Y here if you want to attach large (bigger than 2TB) discs to
your machine, or if you want to have a raid or loopback device
bigger than 2TB. Otherwise say N.
So if you define LBD on a x86_64 kernel, you get to keep the pieces.
A make oldconfig would probably help - don't you think so?
> I'd much prefer that you simply remove SECTOR_FORMAT completely.
Impossible - doing a scanf and then copying the value is much uglier. I just
add 5 lines of code to avoid the problem, while removing the silly
HAVE_SECTOR_T trick and moving the CONFIG_LBD handling to arch-independent
code.
--
Paolo Giarrusso, aka Blaisorblade
Linux registered user n. 292729
BlaisorBlade <[email protected]> wrote:
>
> > I'd much prefer that you simply remove SECTOR_FORMAT completely.
>
> Impossible - doing a scanf and then copying the value is much uglier.
Put this in ll_rw_blk.c:
/*
* Parse an ascii decimal number into a sector_t. Return 1 on success
*/
int str_to_sector_t(const char *str, sector_t *sector)
{
unsigned long long val;
int ret;
ret = sscanf(str, "%llu", &val);
if (ret == 1)
*sector = val;
return ret;
}
EXPORT_SYMBOL(str_to_sector_t);