According to the POSIX specification:
> The sem_timedwait() function shall fail if:
> [EINVAL]
> The process or thread would have blocked, and the abstime
> parameter specified a nanoseconds field value less than zero
> or greater than or equal to 1000 million.
Fixes #16693, #16694.
Haiku's Be-style semaphores have a global limit, but the
POSIX-style semaphores do not. Functionally there will be a limit
as to how many can be waited on at a time, but one will probably
hit thread limits before that happens.
This matches what FreeBSD does: specifies 256 for _POSIX_SEM_NSEMS_MAX
but then returns -1 for the sysconf value.
Fixes #16692.
If a multibyte Unicode character is specified to MultibyteToWchar
one byte at a time, then resetting the state will lose the data
for the second character. We thus shouldn't re-read but instead
keep the existing character and just read the second one.
Fixes #19035.
Otherwise, when the current locale timezone is one that has
an offset to GMT of 0, we would wind up with its name in tm_zone
and not "GMT".
Follows up to #19047.
We can be called with a timezone that isn't the same as the one
set in the current locale, in particular from gmtime(). We thus need
to handle this case properly by checking that the gmtoff is really
the timezone in the locale, or if it's GMT, to just use a static
string for the tm_zone instead.
Fixes #19047.
-1 was our previous value for AT_FDCWD, so we should retain support
for it here for the moment. All other negative values will now result
in an error.
Part of #19048.
The local transferEndOffset should be relative to the parent (us),
not to the underlying filesystem. We thus need to subtract our
own offset from the operation's offset.
This fixes a regression from be8080575a5ae9e8ecba3494afdd081cb4fd94a2.
Errors transferring data should now be properly reported in more cases.
i.e. not on Deskbar team menu, menu field menus or the clock
options menu but all regular menus at the top of the window
and other context-click menus.
Change-Id: If551aecd040500973a92fda2138b6ffe03cda279
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8154
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Reviewed-by: John Scipione <jscipione@gmail.com>
Some things that call these macros put expressions in them; to avoid
operator precedence problems we should put them in parens.
Change-Id: I57335ffb190c484778c4bc8075d5ca3597f1e7e5
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8184
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Lookups are more frequent than cache deletions.
Signed-off-by: Augustin Cavalier <waddlesplash@gmail.com>
A compile benchmark seems to have little if any performance improvement.
However, this codepath will be hit when allocating pages that are
cached, so it may provide a more significant improvement once file
caches are fully utilized.
Implemented using __builtin_clz where available, otherwise using
an algorithm derived from "Bit Twiddling Hacks" which is similar
to the one ramfs uses. GCC and Clang seem to unroll the loop on
x86 at least (but it doesn't matter there as the builtin exists,
implemented using the "bsr" instruction.)
After settings Backgrounds to None for all worskapces, I was getting a 100%
reproducible segment violation on "fShowingBitmap->fBitmap->Bounds()" with every
resolution change.
Change-Id: I2a3640a3c33e571d772c1b38b21087346215dca8
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8176
Reviewed-by: Adrien Destugues <pulkomandy@pulkomandy.tk>
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
... whatever motivated this to only work on disabled text views
has been resolved.
Change-Id: I3ae9d056870f1163d3fa074ab9a1dad96c0192f8
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8168
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
* Fully implement break-before-make PTE updating, fixing the
race condition that necessitates it.
* Incorporate align-down logic into the normal flow of the table
traversal.
* Drop support for huge pages and for table freeing for now. They
can be reimplemented once we have a fully functional baseline.
* Further adoption of the new ProcessRange coming in subsequent
patches.
Change-Id: I852c63aaeab48b847b4bb99b328dfe071d27bb29
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8139
Reviewed-by: Fredrik Holmqvist <fredrik.holmqvist@gmail.com>
* clang, unlike GCC, does not allow a weak declaration to override
a strong definition. We need to make herror's definition in
addition to declaring the weak alias above.
Change-Id: Iac1c00e66bdd73d6e214e2302e18a0d51b18c9d6
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8148
Reviewed-by: Fredrik Holmqvist <fredrik.holmqvist@gmail.com>
Does the same thing but _inherited is used so we can change the class.
Change-Id: I6c0bf7b35917da55a3bd4aa15b5bb80063f2b02a
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8150
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This allows both guard pages and freed allocations, when memory reuse is
disabled, to be returned to the system, massively reducing the pages that
are actually needed.
The guarded heap obviously still wastes a lot of memory due to using a
dedicated page for every tiny allocation. But with this change it
becomes realistic to run the full userland under the guarded heap for
more than a couple of minutes and with more complex applications than
before.
Change-Id: I479e822f9dbeea3968d41b66fc55b0d285a1f57e
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8166
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This shifts another big chunk of packagefs memory usage into dedicated
object caches. It saves a small amount of memory since now we use
caches sized exactly to the object size plus alignment rather than
rounding up to the block allocator size, and also makes clearer in memory
usage statistics how much memory packagefs is using.
We don't need to set the top here, it is already set by CalcRect()
a couple lines up. Our calculation was slightly off causing bug.
Fixes #12558
Change-Id: I022cc661cf55a9d016fd722c701f05fc1f9fea4f
Reviewed-on: https://review.haiku-os.org/c/haiku/+/8165
Haiku-Format: Haiku-format Bot <no-reply+haikuformatbot@haiku-os.org>
Reviewed-by: waddlesplash <waddlesplash@gmail.com>
This is the only case the POSIX specification indicates is portable.
We may be able to implement some of the others that Linux or the BSDs
support if it turns out there are applications that want them, though.
Fixes #19032.
