diff options
author | Nicolás Reynolds <apoyosis@correo.inta.gob.ar> | 2012-04-17 20:37:01 -0300 |
---|---|---|
committer | Nicolás Reynolds <apoyosis@correo.inta.gob.ar> | 2012-04-17 20:37:01 -0300 |
commit | 76480b6b18868754fc95ff41d02885898c54367f (patch) | |
tree | b3c9945f331f2cadd2be8a7dba5b2097c519998f /cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch | |
parent | 7a97ab610eea972771b91f51cb180f0f195b4da3 (diff) |
Updating the cross toolchain for mips64el
Added patches from upstream and diff'ed with them. Binutils and
linux-libre-api-headers are already built. GCC Core builts but
`mips64el-unknown-linux-gnu-gcc -v dummy.c` fails because of missing -EL on
i686-pc-linux-gnu-as. Exporting mips64el-...-binutils to the PATH solves the
problem but mips64el-...-ld fails with missing crt*.
No idea why target gcc decides to use the host as instead of the target one,
since the PKGBUILD hasn't fundamentally changed.
Diffstat (limited to 'cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch')
-rw-r--r-- | cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch | 29 |
1 files changed, 29 insertions, 0 deletions
diff --git a/cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch b/cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch new file mode 100644 index 000000000..46f52e98d --- /dev/null +++ b/cross/cross-mips64el-unknown-linux-gnu-glibc-headers/glibc-2.15-arena.patch @@ -0,0 +1,29 @@ +diff --git a/malloc/arena.c b/malloc/arena.c +index d3cf4b9..b1c9469 100644 +--- a/malloc/arena.c ++++ b/malloc/arena.c +@@ -828,7 +828,7 @@ arena_get2(mstate a_tsd, size_t size) + { + if (mp_.arena_max != 0) + narenas_limit = mp_.arena_max; +- else ++ else if (narenas > mp_.arena_test) + { + int n = __get_nprocs (); + +@@ -842,7 +842,14 @@ arena_get2(mstate a_tsd, size_t size) + } + repeat:; + size_t n = narenas; +- if (__builtin_expect (n <= mp_.arena_test || n < narenas_limit, 0)) ++ /* NB: the following depends on the fact that (size_t)0 - 1 is a ++ very large number and that the underflow is OK. If arena_max ++ is set the value of arena_test is irrelevant. If arena_test ++ is set but narenas is not yet larger or equal to arena_test ++ narenas_limit is 0. There is no possibility for narenas to ++ be too big for the test to always fail since there is not ++ enough address space to create that many arenas. */ ++ if (__builtin_expect (n <= narenas_limit - 1, 0)) + { + if (catomic_compare_and_exchange_bool_acq (&narenas, n + 1, n)) + goto repeat; |