* [PATCH] Fix hwcap handling in ld.so
@ 2004-03-18 14:53 Jakub Jelinek
0 siblings, 0 replies; only message in thread
From: Jakub Jelinek @ 2004-03-18 14:53 UTC (permalink / raw)
To: Ulrich Drepper; +Cc: Glibc hackers
Hi!
Wonder why this has been broken for so long unnoticed.
unsigned long long x, y;
...
(x & y) > x
is 0 for any x, y.
Guess I should add a GCC warning for this kind of things.
I believe the following patch is correct and seems to work for me
on both PIII (where it doesn't decide to use a /usr/lib/sse2/ library
neither without nor with LD_LIBRARY_PATH=/usr/lib, without this
patch it happily uses /usr/lib/sse2/ library without
LD_LIBRARY_PATH=/usr/lib, but doesn't use it with the env var)
and on x86-64 (where it uses /usr/lib/sse2/).
When GLRO(dl_osversion) is referenced in dl-cache.c, I don't see how
GLRO(dl_hwcap) could be undefined, since it is for statically linked
apps defined in dl-support.c together with _dl_osversion.
2004-03-18 Jakub Jelinek <jakub@redhat.com>
* sysdeps/generic/dl-cache.c (_dl_load_cache_lookup): Remove hwcap
variable and weak_extern for _dl_hwcap.
(_DL_HWCAP_TLS_MASK): Define.
(HWCAP_CHECK): Fix checking of non-platform hwcap bits. Use
lib->osversion instead of cache_new->libs[middle].osversion.
--- libc/sysdeps/generic/dl-cache.c.jj 2004-03-08 12:01:51.000000000 +0100
+++ libc/sysdeps/generic/dl-cache.c 2004-03-18 14:55:34.774381649 +0100
@@ -243,12 +243,7 @@ _dl_load_cache_lookup (const char *name)
if (cache_new != (void *) -1)
{
- /* This file ends in static libraries where we don't have a hwcap. */
- unsigned long int *hwcap;
uint64_t platform;
-#ifndef SHARED
- weak_extern (_dl_hwcap);
-#endif
/* This is where the strings start. */
cache_data = (const char *) cache_new;
@@ -256,22 +251,25 @@ _dl_load_cache_lookup (const char *name)
/* Now we can compute how large the string table is. */
cache_data_size = (const char *) cache + cachesize - cache_data;
- hwcap = &GLRO(dl_hwcap);
platform = _dl_string_platform (GLRO(dl_platform));
if (platform != (uint64_t) -1)
platform = 1ULL << platform;
/* Only accept hwcap if it's for the right platform. */
+#ifdef USE_TLS
+# define _DL_HWCAP_TLS_MASK (1LL << 63)
+#else
+# define _DL_HWCAP_TLS_MASK 0
+#endif
#define HWCAP_CHECK \
- if (GLRO(dl_osversion) \
- && cache_new->libs[middle].osversion > GLRO(dl_osversion)) \
+ if (GLRO(dl_osversion) && lib->osversion > GLRO(dl_osversion)) \
continue; \
if (_DL_PLATFORMS_COUNT && platform != -1 \
&& (lib->hwcap & _DL_HWCAP_PLATFORM) != 0 \
&& (lib->hwcap & _DL_HWCAP_PLATFORM) != platform) \
continue; \
- if (hwcap \
- && ((lib->hwcap & *hwcap & ~_DL_HWCAP_PLATFORM) > *hwcap)) \
+ if (lib->hwcap \
+ & ~(GLRO(dl_hwcap) | _DL_HWCAP_PLATFORM | _DL_HWCAP_TLS_MASK)) \
continue
SEARCH_CACHE (cache_new);
}
Jakub
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2004-03-18 14:53 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-03-18 14:53 [PATCH] Fix hwcap handling in ld.so Jakub Jelinek
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).