| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
situations
|
|
When I imported INLINE_SYSCALL, etc. from glibc, I just copied whole
macros for each ABI. Actually, except for argument type (long
vs. long long), internal_syscall[0-4] are identical for all ABI, and
internal_syscall[5-6] for N32/N64 are identical for N32/N64.
And I just disabled FORCE_FRAME_POINTER since it had caused build
failure, but including <alloca.h> can solve it.
Here is a patch to handle these issues, plus asm/__asm__ and
volatile/__volatile__ conversion.
Thanks for Peter S. Mazinger who did basic work of this cleanup.
|
|
clean up E500 math support somewhat.
From Steve Papacharalambous with mods by me.
|
|
protect its use by #ifdef INTERNAL_SYSCALL since not all
arch's has INTERNAL_SYSCALL.
All arch's not having INTERNAL_SYSCALL in
libc/sysdeps/linux/<arch>/bits/syscalls.h should get an update
from glibc.
|
|
|
|
in the powerpc arch code and make more code common with
classic PowerPC. From Steve Papacharalambous
Lets hope I didn't break something.
|
|
Current ucontext_t does not match with kernel's definition. Also, on
N32 ABI, we can not use kernel's struct sigcontext since kernel's
sizeof(long) and user's sizeof(long) is differ.
For both data structure, just importing header files from recent glibc
fixes the mismatch. So here is a patch importing sys/ucontext.h and
bits/sigcontext.t from glibc-2.5.
|
|
|
|
gcc barf otherwise.
|
|
This patch imports INLINE_SYSCALL, etc. from glibc-2.5 ports and
implements _syscall[0-7] by using them, like arm and powerpc.
I think this patch does not change functionality, but gives better
__SYSCALL_CLOBBERS for N32/N64.
|
|
|
|
st_[amc]time_nsec to each kernel_stat, and define STAT_HAVE_NSEC to convert them
|
|
|
|
|
|
Hi. I have found a bug in realpath that did not allow me to use it with
symlinks. Although I am using 0.9.28.1, the bug is present in 0.9.28 and
also in the trunk.
I attach a suggested patch. Please let me know if the solution is correct.
|
|
|
|
I'm seeing a hang in applications that open and close files with a
non-threaded uClibc. There is code in fclose() to remove a file handle from
_stdio_openlist however it's commented out, allegedly because it is dead.
_stdio_openlist is used in several places, including _stdio_term when a
program exits. fclose() frees the struct, so I don't see how the code can
possibly be redundant.
In the __UCLIBC_HAS_THREADS__ && __STDIO_BUFFERS case the file is removed from
_stdio_openlist by __STDIO_OPENLIST_DEC_USE.
The attached patch fixes this bug be re-enabling the code.
|
|
|
|
from Carmelo Amoroso <carmelo73@gmail.com>.
|
|
Math support to be done.
From Steve Papacharalambous, stevep@freescale.com
|
|
|
|
disable strict-aliasing until someone gets the balls to clean this stuff up
|
|
|
|
|
|
__USE_FILE_OFFSET64 is defined. This was the case for all
architectures except those that use packed structures by default
(i.e., CRIS).
|
|
I have successfully made gethostbyname_r(), res_init(), and gethostid() fully
reentrant. In addition, I have added a NULL check to inet_aton(). This is
where SEG FAULTs were coming from when gethostbyname_r() was called.
|
|
|
|
|
|
Hi Erik,
Thanks for all your great work. I found a set of bugs in
resolv.c . Basically, there is code that looks like:
BIGLOCK;
__nameserversXX=__nameservers;
__nameserverXX=__nameserver;
BIGUNLOCK;
i = __dns_lookup(dname, type, __nameserversXX, __nameserverXX, &packet, &a);
which is a problem because the declarations are
int __nameservers;
char * __nameserver[MAX_SERVERS];
int __searchdomains;
char * __searchdomain[MAX_SEARCH];
so you can see that __nameserver is a pointer. Copying the
pointer to __nameserverXX doesn't protect the global variable
space. I have attached a patch and the new file. I hope you
will incorporate these bug fixes. I spent quite a bit of time
tracking them down.
Many thanks,
Ron
|
|
gethostbyname2_r tries to resolve an ipv6 address from /etc/hosts using
get_hosts_byname_r, but with AF_INET instead of the supplied address family.
This returns ipv4 addresses marked as ipv6 ones. Fix from nbd.
|
|
identical cc1 input for LFS enabled (compared to before r17652) but makes sure that we pickup the non-LFS if LFS is turned off (like r17652 did in this case).
|
|
Sounds like a typo, really
|
|
However, retesting on m68k showed up a problem that had appeared in
uClibc since the last time I tried. Specifically, revision 15785 did:
-#define HEAP_GRANULARITY (sizeof (HEAP_GRANULARITY_TYPE))
+#define HEAP_GRANULARITY (__alignof__ (HEAP_GRANULARITY_TYPE))
-#define MALLOC_ALIGNMENT (sizeof (double))
+#define MALLOC_ALIGNMENT (__alignof__ (double))
The problem is that
(a) MALLOC_HEADER_SIZE == MALLOC_ALIGNMENT
(b) the header contains a size value of type size_t
(c) sizeof (size_t) is 4 on m68k, but...
(d) __alignof__ (double) is only 2 (the largest alignment used on m68k)
So we only allocate 2 bytes for the 4-byte header, and the least
significant 2 bytes of the size are in the user's area rather than
the header. The patch below fixes that problem by redefining
MALLOC_HEADER_SIZE to:
MAX (MALLOC_ALIGNMENT, sizeof (size_t))
(but without the help of the MAX macro ;)). However, we really would
like to have word alignment on Coldfire. It makes a big performance
difference, and because we have to allocate a 4-byte header anyway,
what wastage there is will be confined to the end of the allocated block.
Any wastage will also be limited to 2 bytes per allocation compared to
the current alignment.
I've therefore used the __aligned__ type attribute to create a double
type that has at least sizeof (size_t) bytes of alignment. I've
introduced a new __attribute_aligned__ macro for this. It might seem
silly protecting against old or non-GNU compilers here, but the extra
alignment is only an optimisation, and having the macro is more in the
spirit of the other attribute code.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
screw us up
|
|
The patch below tweaks the ARM nommu startup code so that doubleword stack
alignment is preserved. This is required on EABI targets.
|
