based protocol as the X vmmouse driver does. It yields proper absolute pointer
positions under both the normal VMWare console and using the builtin VNC server.
Therefore "fixes" the issue seen in #4630.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33298 a95241bf-73f2-0310-859d-f6bbb57e9c96
all the USB drivers working with this single compat header. Usefulness might
be questionable, but since I did the work anyway, why not check it in as well...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33243 a95241bf-73f2-0310-859d-f6bbb57e9c96
the original BeOS input_server will fail to detect the error and never close
the device.
* Remove the empty kernel_cpp header and use the one from kernel util instead.
* Add some missing headers for completeness.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33242 a95241bf-73f2-0310-859d-f6bbb57e9c96
return an error.
* Properly use the name length instead of a hardcoded buffer size when composing
the name of the raw device and ensure proper termination.
* Case new ioctls for Haiku as the target platform. Indeed this driver works
fine on BeOS even though it was written natively for Haiku.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33240 a95241bf-73f2-0310-859d-f6bbb57e9c96
anymore. It could happen that the transfer was already in the process of being
finished, so wasn't in the list of pending transfers anymore. Cancel would then
return even though the callback wasn't called yet. This could lead to a callback
being called after a driver was already unloaded (even after it cleaned up the
pipes it used).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33239 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Also rename it from pad1 to code_area as that's what it is.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33146 a95241bf-73f2-0310-859d-f6bbb57e9c96
transfer to the transfer list and scheduling the transfer descriptors on the
controller by switching the endpoint tail. Since we only check that the endpoint
head is equal to the endpoint tail to conclude that there are no active
transfers, we need to ensure that we don't accidently do that check before the
transfer descriptors are scheduled. Otherwise we could happen to processed a not
yet started transfer and finishing it before an actual transfer has taken place.
This would then lead to 0 byte transfers and toggle mismatches. To fix this we
now protect the transfer addition and tail switching as well as the check in the
finisher with a per-endpoint mutex. Note that we allocate the lock on the heap
and only store the pointer in the endpoint structure as this one is allocated
from the precious physical memory pool. Could fix#4067.
Also switched the Jamfile to UsePrivateKernelHeaders (for the MutexLocker) and
removed BeOS compatibility.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33129 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Draw a pseudo mouse instead of using the bitmap. It's ugly and should be done
properly by someone with more of an artistic talent than me. But it should do
for now.
* Downright removed the little icons. The code is still there, but since the
icons aren't there anymore it will simply fail to load them and not use them.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33078 a95241bf-73f2-0310-859d-f6bbb57e9c96
user and group ids on build platforms that support them. Fixes#4458.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33066 a95241bf-73f2-0310-859d-f6bbb57e9c96
can cause overflows.
* Added a generic IS_VALID_SPOT() macro that checks for overflows and checks if
the area will fit with the given constraints.
* Use the macro to simplify the places where these checks are necessary.
* Use the provided "end" limit instead of the address space end. It currently
doesn't matter but makes more sense.
* Rename newBase variables to alignedBase as that's what they are.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33061 a95241bf-73f2-0310-859d-f6bbb57e9c96
to insert areas so we don't overflow.
* Consequently use the area end (base + size - 1) where appropriate which
prevents overflows in a few places.
* Properly check for reaching the address space end.
* If we've already found a spot we don't need to recheck if we've found one.
* Simplify the B_EXACT_ADDRESS checks down to a simpler single if statement
instead of the four seperate ones.
* Properly calculate the search end for B_EXACT_ADDRESS as well, it's also
base + size - 1.
* Block the full last page now that this actually works without overflowing.
* Some style changes and added spacing.
This should now really fix#2550. Previously the overflow protection didn't
actually work because on allocation we overflowed and completely missed the
protecting area.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33037 a95241bf-73f2-0310-859d-f6bbb57e9c96
and 0xcccccccc (and 64 pages thereafter) in any way will always lead to a crash.
Before it could happen that these ranges were allocated for an area and then
accessing these would not be as evident anymore. Only enabled when the
corresponding paranoid setting is enabled (which it currently is).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33033 a95241bf-73f2-0310-859d-f6bbb57e9c96
of the VM functions handling areas are overflow safe. If an area is created that
spans across the last page many places will run into an integer overflow. This
mostly concerns the area allocation path in find_and_insert_area_slot() and also
vm_create_anonymous_area() where the loop for mapping pages for B_FULL_LOCK
areas overflows and runs more times than it should leading to #2550.
This could be seen as a workaround. The real fix would be to make everything
overflow safe. The thing is that this does also concern the user of the area
which could easily have forgotten to check for overflows as well, so I am a bit
uneasy with handing out areas that could easily lead to such hard to debug
problems. Since this is really an edge case and this single step safes quite a
bit of extra checks I'd actually be OK with keeping it that way.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33032 a95241bf-73f2-0310-859d-f6bbb57e9c96
mapped pages and a non-read and non-write protection to block a certain address
range from being used by anything.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@33030 a95241bf-73f2-0310-859d-f6bbb57e9c96
end up as 0 again in any case. It certainly looks correct without it, removing
so it doesn't confuse the next one reading over it.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32994 a95241bf-73f2-0310-859d-f6bbb57e9c96
break instead of after. Also whitespace after the cursor position wasn't
ignored.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32993 a95241bf-73f2-0310-859d-f6bbb57e9c96
seems to incorrectly set pseconds to 0 for the first track, leading to an LBA of
-150 which messes up session recognition. Works around VMWare issue in #4439.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32990 a95241bf-73f2-0310-859d-f6bbb57e9c96
many pages as are actually missing, not the full count.
* Take into account that free_page_queue_count() can be less than sReservedPages
(when some of the reserved pages have been allocated already) in
vm_page_num_unused_pages(). Before it could return negative and therefore
wrapped numbers.
* Simplify the page scrubber loop by continuing early. Also avoids a needless
interrupt spin lock acquisition when there's nothing to do.
* Some minor coding style cleanup.
* Fix a typo.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32980 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Remove some unused members.
* Avoid initializing the mouse position to non-integer coordinates.
* The frame right/bottom are width/height - 1.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32979 a95241bf-73f2-0310-859d-f6bbb57e9c96
allows for BFS based LiveCDs. Still this whole name matching feels hacky.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32976 a95241bf-73f2-0310-859d-f6bbb57e9c96
as the other node operations.
* Differentiate between data and metadata modification, so that changes to
attributes or stat fields (other than size) dont't prevent direct IO relay.
* Suppress child finish notifications when sending subrequests to the
underlaying filesystem. Otherwise the request can be gone by the time we
try to complete it.
* Fix offsets used for IORequest::CopyData().
* Some refactoring.
* Added locking.
With that a write_overlay can be put on attribute enabled filesystems to either
make them writeable or snapshot them. This allows for example for a working
BFS based LiveCD.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32937 a95241bf-73f2-0310-859d-f6bbb57e9c96
This allows for synchronous uses where subrequests are forked off and waited on.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32936 a95241bf-73f2-0310-859d-f6bbb57e9c96
Otherwise the focus cannot be changed to another window. In one case this could
have a severe sideeffect: When a window is quit it is hidden to remove focus and
remove it from screen. If this didn't work due to the focus lock, the fFocus
member would become a stale pointer after the window was quit and deleted. The
app_server would crash or corrupt memory on the next focus change. Easily seen
when moderately quickly clicking away a few alert windows.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32929 a95241bf-73f2-0310-859d-f6bbb57e9c96
On detaching the views remove themselves from the app local token space. Since
the ServerApp only waits for all ServerWindows to be removed from the window
list and not for their actual destruction, it can happen that the ServerApp is
deleted before the window destruction and hence the view detaching has finished.
The views would then access a stale ServerApp pointer and try to remove their
token from the deleted token space. To avoid that we set the ServerApp pointer
to NULL when the window is removed from the app (as after that the app can be
gone any time) and check for that case when detaching.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32927 a95241bf-73f2-0310-859d-f6bbb57e9c96
only some sub requests worked. Previously we would have simply canceled all
of the subrequests in the loop because we cancel after the first error.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32914 a95241bf-73f2-0310-859d-f6bbb57e9c96
* An off by one error prevented the very last block of a device to be accessed
through IO.
* In case of error the request wasn't notified causing anyone (the page writer
for example) to wait forever for the request to complete.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32913 a95241bf-73f2-0310-859d-f6bbb57e9c96
by writing them into our buffers and directly satisfy read requests by copying
our buffers, filling any gaps by spawning off subrequests to the underlaying
filesystem.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32900 a95241bf-73f2-0310-859d-f6bbb57e9c96
concerned by it as it does not do file data, it simply passes the hook through.
In the write_overlay case we check for write operations and for modified nodes.
In both cases we return B_NOT_SUPPORTED which will cause a fallback to
synchronous IO. The main problem with the fallback is not that it is synchronous
but that the physical buffers of the request will be mapped and filled page
wise, which makes it slow for various reasons. In any case with this setup all
reads to write_overlay that can go through unmodified now do. This should speed
up CD boot as there is no physical to virtual translation overhead and no limit
to page wise reads. In the best case it should now read 256 blocks at the time
instead of always falling back to 2. My tests show no side effects on creating,
writing or partially modifying nodes so far.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32897 a95241bf-73f2-0310-859d-f6bbb57e9c96
capabilities to aid in debugging memory corruption issues.
It does:
* Initialize memory to 0xcc to help turn up use of uninitialized memory
* Set freed memory to 0xdeadbeef to help find accesses of freed memory
* Use the paranoid heap validation to turn up many cases of memory corruption
* Use a simplistic wall check to turn up memory overwrites past allocations
* Take extra steps to validate freed addresses to turn up misaligned frees
It has an interface to en-/disable paranoid validation and to start/stop regular
wall checking. Both are currently just enabled. At a later stage a debug version
of libroot could be used by an application and the checks enabled at will. Note
that due to the paranoid validation and the suboptimal locking this allocator
will perform horribly. Still to find memory corruption issues in the system or
also in your applications it can be helpful to build your installation with it
turned on. To enable it you currently need to edit the Jamfile to sub-include
the malloc_debug instead of the malloc directory.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32894 a95241bf-73f2-0310-859d-f6bbb57e9c96
for the page list but not the area one. Since multiple pages can be allocated
at once, even an area that is not at the top of the list can become empty. In
such a case the area list would previously have lost entries. Also because
we can remove more than one page from any area, not just the top one, we may
need to move forward in the list so that it stays ordered by free pages.
+alphabranch
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32880 a95241bf-73f2-0310-859d-f6bbb57e9c96
such a case as that would try to read info about a non-present medium.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32862 a95241bf-73f2-0310-859d-f6bbb57e9c96
it would overwrite memory and on writing it would crash further down when the
buffer is accessed. Fixes#4383.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32857 a95241bf-73f2-0310-859d-f6bbb57e9c96
we waste space, but we also incorrectly fail to update the size for all writes
that do not go beyond the previous file size.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32825 a95241bf-73f2-0310-859d-f6bbb57e9c96
the parent directory in those cases. Also, when updating the parent directory
we need to update the ".." dirent if we already populated it with a different
value before.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32823 a95241bf-73f2-0310-859d-f6bbb57e9c96
fixes#4358, there changing the Tracker preferences causes some default
directories to be created, like "home" and "config", which were previously
simply killing off the real ones.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32822 a95241bf-73f2-0310-859d-f6bbb57e9c96
ensure that these nodes are never put. As they keep all changes in memory,
putting them and consequently freeing them destroys the modifications without
a way to get them back from the underlaying filesystem. In case a created node
was put but still had its entry in the parent directory this could also lead
to the situation where an invalid node was requested from the underlaying
filesystem, leading to #4347.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32815 a95241bf-73f2-0310-859d-f6bbb57e9c96
to " &" as really the whole rest of the file uses that style.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32813 a95241bf-73f2-0310-859d-f6bbb57e9c96
but this fixes the GCC4 build again, without breaking the function.
+alphabranch
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32772 a95241bf-73f2-0310-859d-f6bbb57e9c96
unflattening if they included any target or reply info. Changing the reader
usage like this makes the first argument a pointer to a buffer. It would
therefore write to random locations, because it would use the literal value of
those header fields as pointers, crashing the using application.
Will check for GCC4 compliance next.
+alphabranch
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32771 a95241bf-73f2-0310-859d-f6bbb57e9c96
Unconditionally release the reader sem when the previous buffer size was 0
instead of testing for fParserWaiting as there is a race condition between
setting it to true in the parser thread and checking for it in the reader
thread so a release_sem() could be missed causing #4343. To reduce the possible
side effect of needlessly looping through acquire_sem() for too many
unconditionally released sems, we eat them in the parser thread once we're sure
we can't miss an event because of that. Also added a TODO about a memcpy()
optimization as pointed out by Rene.
+alphabranch
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32733 a95241bf-73f2-0310-859d-f6bbb57e9c96
media present they will always fail with the no media or media changed errors.
+alphabranch
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32732 a95241bf-73f2-0310-859d-f6bbb57e9c96
don't try to use it for media status polling. In those cases we'll assume a
fixed device with no exchangable medias and therefore always return B_OK.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32731 a95241bf-73f2-0310-859d-f6bbb57e9c96
does calculate the remaining bytes, not the transferred ones, so obviously that
number has to be subtracted from the expected buffer size! Albeit severe, this
wasn't hit that often, because this calculation will only be used when the
transferred size doesn't match the handed in buffer size exactly. So things like
HID or mass storage, that mostly know the exact transfer size weren't affected.
But it could very well explain #4101 where the size isn't previously know.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32680 a95241bf-73f2-0310-859d-f6bbb57e9c96
of load, when using large enough block sizes or when simply having a slow device
this is by far not enough. It is now at 15 seconds, which should reduce timeout
problems to those cases where the device actually get's stuck (because of us
doing something wrong).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32615 a95241bf-73f2-0310-859d-f6bbb57e9c96
into the scope of the lock. Saves an InterruptsSpinLocker.
* Use an InterruptsSpinLocker() as in other places.
* When creating/forking a team fails because the kernel thread cannot be spawned
balance the already sent TEAM_ADDED notification by sending a TEAM_REMOVED one.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32595 a95241bf-73f2-0310-859d-f6bbb57e9c96
send_bit() causes a type conversion from the value handed in to 0 or 1. This
clashed with the usage in send_byte(), that hands over a shifted byte. The
argument was converted to true when it had any value other than 0, whereas
before (where a bool simply was an int) it would have just handed over the value
directly. Therefore the logic in send_bit() that simply masked off the lowest
bit of the value would now not work anymore.
This fixes EDID failing on GCC4 and therefore fixes#2275, the last issue of
#4084 and may also affect #2780.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32593 a95241bf-73f2-0310-859d-f6bbb57e9c96
because all reports schedule transfers on the same endpoint and therefore need
to possibly reschedule. Previously if we got a report that we didn't listen on
all further reports would stop, because noone would schedule a new transfer.
This fixes extra keys not working on my natural keyboard.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32580 a95241bf-73f2-0310-859d-f6bbb57e9c96
been removed. That includes CPU disabling and thread pinning, as that becomes
pointless with only one CPU.
* Return a proper reschedule hint on enqueing a thread, based on the priority
of the current thread vs. the enqueued one.
* Enable dynamic scheduler selection. With one CPU the simple scheduler will
be used, otherwise affine is selected.
* Removed the scheduler type define as we now always auto-select it.
* Some cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32573 a95241bf-73f2-0310-859d-f6bbb57e9c96
testing to determine where we are at sooner than later now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32569 a95241bf-73f2-0310-859d-f6bbb57e9c96
packet that targets the MAC address supplied by argument.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32555 a95241bf-73f2-0310-859d-f6bbb57e9c96
processed. The controller may initialize it to any value when starting to
process the descriptor. If a controller did that it was possible that we thought
a transfer was already done even if it actually just started. Transfers would
then return that weren't processed, returning uninitialized buffers.
Instead of relying on the condition codes we now check that the head and tail
pointers are the same. This guarantees that all transfer descriptors of the
endpoint at hand are completed.
Reverted r32534 again as this one fixes the problem for real. The same things
that were mentioned there could happen here essentially, so in the worst case
the device or controller could stall because of freeing in-use structures.
Fixes#4067.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32551 a95241bf-73f2-0310-859d-f6bbb57e9c96
reused and overwritten by the first descriptor of the new transfer and the first
descriptor will become the new tail. We anticipate this situation in
_AddPendingTransfer() and set the first_descriptor of the transfer data to the
tail already. Since the tail was pretty much cleared to zero, this introduced
a race condition. After adding the pending transfer it can already be found
in the finisher thread. If this happened before actually switching the tail
and first descriptor it would find a descriptor with a condition of 0, meaning
"No Error" and would process the transfer incorrectly. Depending on the count
of descriptors and the timing of the switch taking place this could have
resulted in aborted transfers with actual length 0 or with the correct actual
length but invalid data. In the very worst case it could have freed things still
in use by the controller, resulting in all sorts of device errors. Sadly it
doesn't fix#4067.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32534 a95241bf-73f2-0310-859d-f6bbb57e9c96
transfer descriptors into individual packets. The descriptors are more a
logical thing. As such we do not generate one descriptor per packet but one
for each block of two pages at max. Therefore we only set the initial toggle bit
to the one we stored on the first descriptor and let OHCI use the carry bit for
subsequet descriptors. That error would only be visible on transfers above two
pages in size though.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32532 a95241bf-73f2-0310-859d-f6bbb57e9c96
retired descriptor, it actually contains the value that is to be used next.
Confirmed that by cross-referencing FreeBSD.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32531 a95241bf-73f2-0310-859d-f6bbb57e9c96
toggle carry stays valid, we better make sure we set it correctly. Additionally
even if the toggle carry wasn't clobbered, there are conditions the toggle is
reset without the controller being able to notice (as in clearing a halt state).
Since I don't have any OHCI hardware this is untested, but should fix bug #4067
if correct.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32510 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Keep track of the currently running threads.
* Make use of that info to decide if a thread that becomes ready should preempt
the running thread.
* If we should preempt we send the target CPU a reschedule message.
* This preemption strategy makes keeping track of idle CPUs by means of a bitmap
superflous and it is therefore removed.
* Right now only other CPUs are preempted though, not the current one.
* Add missing initialization of the quantum tracking code.
* Do not extend the quantum of the idle thread based quantum tracking as we want
it to not run longer than necessary. Once the preemption works completely
adding a quantum timer for the idle thread will become unnecessary though.
* Fix thread stealing code, it did missed the last thread in the run queue.
* When stealing, try to steal the highest priority thread that is currently
waiting by taking priorities into account when finding the target run queue.
* Simplify stealing code a bit as well.
* Minor cleanups.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32503 a95241bf-73f2-0310-859d-f6bbb57e9c96
re-insert it at a new place, but by only setting the priority and not the
next_priority field, the thread would actually be enqueued at the same priority
level as before. Didn't cause any real damage, guess it was just an oversight.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32494 a95241bf-73f2-0310-859d-f6bbb57e9c96
not hold the bin lock while setting and only protect the actual freeing. May
reduce bin lock contention a little bit.
* Tiny optimization for force-clearing 0xdeadbeef. Do it after setting to 0xcc,
so that it's less likely we have to do it.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32473 a95241bf-73f2-0310-859d-f6bbb57e9c96
until the data it protects isn't in active use anymore. Previously it would
release the translation map and therefore the page directory reference while
the page dir was still set on the CPU, as only the actual call to
i386_context_switch() will replace the page directory in the control register.
This didn't cause any harm though, as during the context switch interrupts are
disabled and therefore the page directory would only be deferred_delete()ed and
not directly freed/overwritten. Still this is logically more correct.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32216 a95241bf-73f2-0310-859d-f6bbb57e9c96
interrupts when invoking it. The user TLB invalidation function essentially only
reads and writes back control register 3 (cr3) which holds the physical address
of the current page directory. Still a preemption between the read and the write
can cause problems when the last thread of a team dies and therefore the team
is deleted. The context switch on preemption would decrement the refcount of the
object that holds the page directory. Then the team address space is deleted
causing the context switch returning to that thread to not re-acquire a
reference to the object. At that point the page directory as set in cr3 is the
one of the previously run thread (which is fine, as all share the kernel space
mappings we need). Now when the preempted thread continues though, it would
overwrite cr3 with the physical page directory address from before the context
switch still stored in eax, therefore setting the page directory to the one of
the dying thread that now doesn't have the corresponding reference. Further
progressing the thread would release the last reference causing the deletion
of the object and freeing of the, now active again, page directory. The memory
getting overwritten (by deadbeef) now completely corrupts the page directory
causing basically any memory access to fault, in the end resulting in a
triplefault. This should fix bug #3399.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32118 a95241bf-73f2-0310-859d-f6bbb57e9c96
allow for more parallelism. Also introduce seperate locks for the bins and
for page allocation. This greatly reduces lock contention and reduces the
duration the locks are held due to them overall protecting less code. Now only
allocations of the same size hitting the same allocator or allocating larger
chunks of memory should block. Previously, basically any allocation and also
free would be mutually exclusive, making it scale pretty badely.
* Added memalign_nogrow(). As it uses heap_memalign() anyway, there's no real
reason not to allow for an alignment.
* Some cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32074 a95241bf-73f2-0310-859d-f6bbb57e9c96
storage.
* Remove reduntant information from the header (field_count vs. fields_size).
* Remove checksums previously used to validate the message passing by area
mechanism.
* Move variables that are purely used by the instance out of the header and into
the BMessage object.
* Use more sensible types for the different message fields.
* Reduce some field sizes to realistic values.
* Make size_t values into uint32 values so the message format will not change
when later moving to 64 bits.
* Pack the structures used for flat message storage so it doesn't change
because of padding.
* Fix message passing by area. It never worked because the created area was
never actually filled with any data!
* Some more allocation checks with graceful fallbacks (should be all now).
* Some more checks for negative index values (should also be all now).
* Make printing more inline with how the rest of the class works and make some
of the output more consistent.
* Also add the new unsigned types to PrintToStream() output.
* Fix printing of unknown types and invalid BMessages, it would always have
printed only the first entry respectively the same error.
* Added some clarifying comments.
* Cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@32039 a95241bf-73f2-0310-859d-f6bbb57e9c96
cleaner it allows for binary comparison of flattened messages.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31908 a95241bf-73f2-0310-859d-f6bbb57e9c96
for families. This fixes the bug that when a font family was already loaded, but
some of its styles haven't been loaded yet, the app_server would fail to set
the system fonts to those styles. Fixes bug #4148.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31904 a95241bf-73f2-0310-859d-f6bbb57e9c96
Michael Lotz