the code is written to pre-init the thread pointer in static linked
programs that pull in __stack_chk_fail or dynamic-linked programs that
lookup the symbol. no explicit canary is set; the canary will be
whatever happens to be in the thread structure at the offset gcc
hard-coded. this can be improved later.
ditch the priority inheritance locks; use malloc's version of lock
i did some testing trying to switch malloc to use the new internal
lock with priority inheritance, and my malloc contention test got
20-100 times slower. if priority inheritance futexes are this slow,
it's simply too high a price to pay for avoiding priority inversion.
maybe we can consider them somewhere down the road once the kernel
folks get their act together on this (and perferably don't link it to
glibc's inefficient lock API)...
as such, i've switch __lock to use malloc's implementation of
lightweight locks, and updated all the users of the code to use an
array with a waiter count for their locks. this should give optimal
performance in the vast majority of cases, and it's simple.
malloc is still using its own internal copy of the lock code because
it seems to yield measurably better performance with -O3 when it's
inlined (20% or more difference in the contention stress test).
internal locks: new owner of contended lock must set waiters flag
this bug probably would have gone unnoticed since it's only used in
the fallback code for systems where priority-inheritance locking
fails. unfortunately this approach results in one spurious wake
syscall on the final unlock, when there are no waiters remaining. the
alternative (possibly better) would be to use broadcast wakes instead
of reflagging the waiter unconditionally, and let each waiter reflag
itself; this saves one syscall at the expense of invoking the
"thundering herd" effect (worse performance degredation) when there
are many waiters.
ideally we would be able to update all of our locks to use an array of
two ints rather than a single int, and use a separate counter system
like proper mutexes use; then we could avoid all spurious wake calls
without resorting to broadcasts. however, it's not clear to me that
priority inheritance futexes support this usage. the kernel sets the
waiters flag for them (just like we're doing now) and i can't tell if
it's safe to bypass the kernel when unlocking just because we know
(from private data, the waiter count) that there are no waiters. this
is something that could be explored in the future.
we use priority inheritance futexes if possible so that the library
cannot hit internal priority inversion deadlocks in the presence of
realtime priority scheduling (full support to be added later).
i tried to go with improving the old binary-search-based algorithm,
but between growth in the number of ranges, bad performance, and lack
of confidence in the binary search code's stability under changes in
the table, i decided it was worth the extra 1.8k to have something
clean and maintainable.
also note that, like the alpha and punct tables, there's definitely
room to optimize the nonspacing/wide tables by overlapping subtables.
this is not a high priority, but i've begun looking into how to do it,
and i suspect the table sizes can be roughly halved. if that turns out
to be true, the new, fast, table-based implementation will be roughly
the same size as if i had just extended the old binary search one.
also special-case ß (U+00DF) as lowercase even though it does not have
a mapping to uppercase. unicode added an uppercase version of this
character but does not map it, presumably because the uppercase
version is not actually used except for some obscure purpose...
destubify iswalpha and update iswpunct to unicode 6.1
alpha is defined as unicode property "Alphabetic" plus category Nd
minus ASCII digits minus 2 special-cased Thai punctuation marks
supposedly misclassified by Unicode as letters.
punct is defined as all of unicode except control, alphanumeric, and
space characters.
the tables were generated by a simple tool based on the code posted
previously to the mailing list. in the future, this and other code
used for maintaining locale/iconv/i18n data will be published either
in the main source repository or in a separate locale data generation
repository.
note that dlerror is specified to be non-thread-safe, so no locking is
performed on the error flag or message aside from the rwlock already
held by dlopen or dlsym. if 2 invocations of dlsym are generating
errors at the same time, they could clobber each other's results, but
the resulting string, albeit corrupt, will still be null-terminated.
any use of dlerror in such a situation could not be expected to give
meaningful results anyway.
the _concept_ of this wrapper has been tested extensively, but the
integration with the build/install system, and using a persistent
specfile rather than one generated at build-time, have not been
heavily tested and may need minor tweaks.
this approach should be a lot more robust (and easier to improve) than
writing a shell script that's responsible for trying to mimic gcc's
logic about whether it's compiling or linking, building shared libs or
executable files, etc. it's also lighter weight and should result in
mildly faster builds when using the wrapper.
make floatscan correctly set errno for overflow/underflow
care is taken that the setting of errno correctly reflects underflow
condition. scanning exact denormal values does not result in ERANGE,
nor does scanning values (such as the usual string definition of
FLT_MIN) which are actually less than the smallest normal number but
which round to a normal result.
only the decimal case is handled so far; hex float require a separate
fix to come later.
skip leading zeros even after decimal point in floatscan
in principle this should just be an optimization, but it happens to
also fix a nasty bug where values like 0.00000000001 were getting
caught by the early zero detection path and wrongly scanned as zero.
- add the rest of the junk traditionally in sys/param.h
- add prototypes for some nonstandard functions
- add _GNU_SOURCE to their source files so the compiler can check proto
this code worked in strtod, but not in scanf. more evidence that i
should design a better interface for discarding multiple tail
characters than just calling unget repeatedly...
introduce new wide scanf code and remove the last remnants of old scanf
at this point, strto* and all scanf family functions are using the new
unified integer and floating point parser/converter code.
the wide scanf is largely a wrapper for ordinary byte-based scanf;
since numbers can only contain ascii characters, only strings need to
be handled specially.
fix buffer overflow in vfprintf on long writes to unbuffered files
vfprintf temporarily swaps in a local buffer (for the duration of the
operation) when the target stream is unbuffered; this both simplifies
the implementation of functions like dprintf (they don't need their
own buffers) and eliminates the pathologically bad performance of
writing the formatted output with one or more write syscalls per
formatting field.
in cases like dprintf where we are dealing with a virgin FILE
structure, everything worked correctly. however for long-lived files
(like stderr), it's possible that the buffer bounds were already set
for the internal zero-size buffer. on the next write, __stdio_write
would pick up and use the new buffer provided by vfprintf, but the
bound (wend) field was still pointing at the internal zero-size
buffer's end. this in turn allowed unbounded writes to the temporary
buffer.
the l prefix is redundant/no-op with printf, since default promotions
always promote floats to double; however, it is valid, and printf was
wrongly rejecting it.
shunget cannot unget eof status, causing wcstol to leave endptr
pointing to the wrong place when scanning, for example, L"0x". cheap
fix is to make the read function provide an infinite stream of bogus
characters rather than eof. really this is something of a design flaw
in how the shgetc system is used for strto* and wcsto*; in the long
term, I believe multi-character unget should be scrapped and replaced
with a function that can subtract from the f->shcnt counter.
new scanf implementation and corresponding integer parser/converter
advantages over the old code:
- correct results for floating point (old code was bogus)
- wide/regular scanf separated so scanf does not pull in wide code
- well-defined behavior on integers that overflow dest type
- support for %[a-b] ranges with %[ (impl-defined by widely used)
- no intermediate conversion of fmt string to wide string
- cleaner, easier to share code with strto* functions
- better standards conformance for corner cases
the old code remains in the source tree, as the wide versions of the
scanf-family functions are still using it. it will be removed when no
longer needed.
fix crash in wordfree if we_offs is not initialized by the caller
I'm not sure if it's legal for wordexp to modify this field, but this
is the only easy/straightforward fix, and applications should not
care. if it's an issue, i can work out a different (but more complex)
solution later.
floatscan: fix incorrect count of leading nonzero digits
this off-by-one error was causing values with just one digit past the
decimal point to be treated by the integer case. in many cases it
would yield the correct result, but if expressions are evaluated in
excess precision, double rounding may occur.
fcntl values 1024 and up are universal, arch-independent. later I'll
add some of the other linux-specific ones for notify, leases, pipe
size, etc. here too.
fix signedness error handling invalid multibyte sequences in regexec
the "< 0" test was always false due to use of an unsigned type. this
resulted in infinite loops on 32-bit machines (adding -1U to a pointer
is the same as adding -1) and crashes on 64-bit machines (offsetting
the string pointer by 4gb-1b when an illegal sequence was hit).
rename __sa_restorer to sa_restorer in struct sigaction
this is legal since sa_* is in the reserved namespace for signal.h,
per posix. note that the sa_restorer field is not used anywhere, so
programs that are trying to use it may still break, but at least
they'll compile. if it turns out such programs actually need to be
able to set their own sa_restorer to function properly, i'll add the
necessary code to sigaction.c later.
TRE wants to treat + and ? after a +, ?, or * as special; ? means
ungreedy and + is reserved for future use. however, this is
non-conformant. although redundant, these redundant characters have
well-defined (no-op) meaning for POSIX ERE, and are actually _literal_
characters (which TRE is wrongly ignoring) in POSIX BRE mode.
the simplest fix is to simply remove the unneeded nonstandard
functionality. as a plus, this shaves off a small amount of bloat.
use macros instead of inline functions in shgetc.h
at -Os optimization level, gcc refuses to inline these functions even
though the inlined code would roughly the same size as the function
call, and much faster. the easy solution is to make them into macros.
fix spurious overflows in strtoull with small bases
whenever the base was small enough that more than one digit could
still fit after UINTMAX_MAX/36-1 was reached, only the first would be
allowed; subsequent digits would trigger spurious overflow, making it
impossible to read the largest values in low bases.
optimize floatscan downscaler to skip results that won't be needed
when upscaling, even the very last digit is needed in cases where the
input is exact; no digits can be discarded. but when downscaling, any
digits less significant than the mantissa bits are destined for the
great bitbucket; the only influence they can have is their presence
(being nonzero). thus, we simply throw them away early. the result is
nearly a 4x performance improvement for processing huge values.
the particular threshold LD_B1B_DIG+3 is not chosen sharply; it's
simply a "safe" distance past the significant bits. it would be nice
to replace it with a sharp bound, but i suspect performance will be
comparable (within a few percent) anyway.
simplify/debloat radix point alignment code in floatscan
now that this is the first operation, it can rely on the circular
buffer contents not being wrapped when it begins. we limit the number
of digits read slightly in the initial parsing loops too so that this
code does not have to consider the case where it might cause the
circular buffer to wrap; this is perfectly fine because KMAX is chosen
as a power of two for circular-buffer purposes and is much larger than
it otherwise needs to be, anyway.
these changes should not affect performance at all.
upscaling by even one step too much creates 3-29 extra iterations for
the next loop. this is still suboptimal since it always goes by 2^29
rather than using a smaller upscale factor when nearing the target,
but performance on common, small-magnitude, few-digit values has
already more than doubled with this change.
fix float scanning of certain values ending in zeros
for example, "1000000000" was being read as "1" due to this loop
exiting early. it's necessary to actually update z and zero the
entries so that the subsequent rounding code does not get confused;
before i did that, spurious inexact exceptions were being raised.
note that there's no need for a precise cutoff, because exponents this
large will always result in overflow or underflow (it's impossible to
read enough digits to compensate for the exponent magnitude; even at a
few nanoseconds per digit it would take hundreds of years).
add "scan helper getc" and rework strtod, etc. to use it
the immediate benefit is a significant debloating of the float parsing
code by moving the responsibility for keeping track of the number of
characters read to a different module.
by linking shgetc with the stdio buffer logic, counting logic is
defered to buffer refill time, keeping the calls to shgetc fast and
light.
in the future, shgetc will also be useful for integrating the new
float code with scanf, which needs to not only count the characters
consumed, but also limit the number of characters read based on field
width specifiers.
shgetc may also become a useful tool for simplifying the integer
parsing code.
this version is intended to be fully conformant to the ISO C, POSIX,
and IEEE standards for conversion of decimal/hex floating point
strings to float, double, and long double (ld64 or ld80 only at
present) values. in particular, all results are intended to be rounded
correctly according to the current rounding mode. further, this
implementation aims to set the floating point underflow, overflow, and
inexact flags to reflect the conversion performed.
a moderate amount of testing has been performed (by nsz and myself)
prior to integration of the code in musl, but it still may have bugs.
so far, only strto(d|ld|f) use the new code. scanf integration will be
done as a separate commit, and i will add implementations of the wide
character functions later.
alloca cannot be a function. #define it to the gcc builtin if possible
gcc makes this mapping by default anyway, but it will be disabled by
-fno-builtin (and presumably by -std=c99 or similar). for the main
program the error will be reported by the linker, and the issue can
easily be fixed, but for dynamic-loaded so files, the error cannot be
detected until dlopen time, at which point it has become very obscure.
when the "r" (register) constraint is used to let gcc choose a
register, gcc will sometimes assign the same register that was used
for one of the other fixed-register operands, if it knows the values
are the same. one common case is multiple zero arguments to a syscall.
this horribly breaks the intended usage, which is swapping the GOT
pointer from ebx into the temp register and back to perform the
syscall.
presumably there is a way to fix this with advanced usage of register
constaints on the inline asm, but having bad memories about hellish
compatibility issues with different gcc versions, for the time being
i'm just going to hard-code specific registers to be used. this may
hurt the compiler's ability to optimize, but it will fix serious
miscompilation issues.
so far the only function i know what compiled incorrectly is
getrlimit.c, and naturally the bug only applies to shared (PIC)
builds, but it may be more extensive and may have gone undetected..
the buffer in getaddrinfo really only matters when /etc/hosts is huge,
but in that case, the huge number of syscalls resulting from a tiny
buffer would seriously impact the performance of every name lookup.
the buffer in __dns.c has also been enlarged a bit so that typical
resolv.conf files will fit fully in the buffer. there's no need to
make it so large as to dominate the syscall overhead for large files,
because resolv.conf should never be large.
nsz [Thu, 29 Mar 2012 12:05:16 +0000 (14:05 +0200)]
math: remove x86 modf asm
the int part was wrong when -1 < x <= -0 (+0.0 instead of -0.0)
and the size and performace gain of the asm version was negligible
nsz [Tue, 27 Mar 2012 20:49:37 +0000 (22:49 +0200)]
math: fix a regression in powl and do some cleanups
previously a division was accidentally turned into integer div
(w = -i/NXT;) instead of long double div (w = -i; w /= NXT;)