Certain interrupts don't disable interrupts. We were calling
x86_{push,pop}_iframe() without specifically disabling them, thus causing
a race condition with could cause the iframe stack to be invalid. This
could cause all kinds of problems.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21784 a95241bf-73f2-0310-859d-f6bbb57e9c96
After exec() we re-entered the userland without removing the syscall
iframe from the iframe stack, thus leaking one stack slot.
Fixes bug #1304.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21783 a95241bf-73f2-0310-859d-f6bbb57e9c96
released it last, and the counts involved.
Furthermore, release_sem_etc() will now only negate the acquirer thread
ID instead of setting it to -1.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21781 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Simplified the notification framework: removed the updater stuff completely;
it was only there to account for some peculiarities of the node monitor which
we now solved differently.
* NotificationListener no longer includes a doubly linked list link for convenience;
it might want to listen to more than just one service.
* NotificationService cannot have an abstract destructor.
* Changed the _user_stop_watching() syscall to mirror the Be API; ie. it's no
longer possible to just remove some flags separately, just to stop listening
completely.
* Adapted the node monitor implementation to live in the NodeMonitorService class
that uses the new notification framework.
* Removed the public kernel node monitor API - it wasn't useful that way since you
couldn't do a lot with the KMessage in the kernel without using a private API.
Now you will have to use the (private) notification manager to use the node monitor
from inside the kernel. At a later point, we might introduce a public API for that,
too.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21780 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added a Current() method to the same class that returns the current element again.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21779 a95241bf-73f2-0310-859d-f6bbb57e9c96
computed as '.' which resulted in output path ending with '/.', causing mkdir to choke.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21778 a95241bf-73f2-0310-859d-f6bbb57e9c96
This change now actually fixes its logic; thanks for the hint, though :-)
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21777 a95241bf-73f2-0310-859d-f6bbb57e9c96
deliver a line back, but I've changed it such that it only does this for R5.
I've enlarged the timeout back to 60 seconds, as I guess that could have been the culprit.
Tested here, and it at least still works.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21772 a95241bf-73f2-0310-859d-f6bbb57e9c96
Will be used for node monitoring and other stuff, too (like the Registrar or the
VM low memory handler).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21768 a95241bf-73f2-0310-859d-f6bbb57e9c96
private/shared.
* Made AddReference() and CountReferences() inlines.
* The registrar is now using the private Referenceable version in libbe.so.
* Minor cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21767 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Decreased the timeout from 60 to 30 seconds; somehow it seems it waits more
than just once this time in case something fails over here.
* Cleaned up the messy SSL patch.
* Made the whole thing a bit more compliant to our coding style.
* Got rid of the useless pop3_error() macro.
* Fixed error reporting in ReceiveLine().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21765 a95241bf-73f2-0310-859d-f6bbb57e9c96
as a string anymore (it's deprecated). That at least allows mmlr's internet provider to
recognize mails as valid mails rather than spam.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21764 a95241bf-73f2-0310-859d-f6bbb57e9c96
AddDirectoryToHaikuImage with attributes does also work for directory
names with spaces now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21761 a95241bf-73f2-0310-859d-f6bbb57e9c96
the other solution would have been to link against libgcc.a. please comment.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21757 a95241bf-73f2-0310-859d-f6bbb57e9c96
'Printer Name'). In other parts of the print_server we had assumed that
printer spool directory name = printer name. So the change broke some
things.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21753 a95241bf-73f2-0310-859d-f6bbb57e9c96
possible exception thrown from the constructor called by the function
itself, for safety.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21750 a95241bf-73f2-0310-859d-f6bbb57e9c96
Now input works also with a replicanted Terminal, although I have to
click on it twice(!?!?). Don't call exit_thread() anymore in
TermParse::PtyReader(). Hope this is correct. (feel free to beat me on
this).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21747 a95241bf-73f2-0310-859d-f6bbb57e9c96
replicant works (thank you Marc). Return errno instead of the return
value of read() (thank you Jerome).
Terminal now can be instantiated as a replicant, although input doesn't
work :((.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21745 a95241bf-73f2-0310-859d-f6bbb57e9c96
mess with be_app anymore. Renamed TermView private methods to have an
underscore prefix.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21743 a95241bf-73f2-0310-859d-f6bbb57e9c96
with instantiation. Note that if an error occurs in the constructor,
we're pretty much screwed. Made some TermView functions private. Some
small cleanup, and some WIP code.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21742 a95241bf-73f2-0310-859d-f6bbb57e9c96
therefore, we could remove the ugly defines from KMessage again, and compile it
with KMESSAGE_CONTAINER_ONLY.
* Added KMessage::SetDeliveryInfo() to be able to send messages with a correct
header.
* Fixed a bug in KMessage::SendTo() that would not send the senderTeam when passing
a negative value for the parameter, but override it when passing in a valid
value.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21740 a95241bf-73f2-0310-859d-f6bbb57e9c96
single priority 12 thread can make the whole GUI appear frozen.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21736 a95241bf-73f2-0310-859d-f6bbb57e9c96
the new structure tty_settings of which there's only one instance per
master/slave tty pair. Previously the existence of two instances of those
members caused several kinds of problems, e.g. the Terminal setting
the window size on the master and CLI programs readings the unchanged
values from the slave. E.g. less correctly adjusts the display when the
Terminal size changes, now.
* Reorganized writing to a TTY. We do no longer handle writes to master
and slave the same way. Writes to the master are "input" and need to
be processed differently from writes to the slave ("output"). Before,
both were processed first as output then as input, which caused incorrect
behavior. E.g. CRs were not echoed correctly.
* Added canonical ERASE (backspace) and KILL (clear line) processing.
Couldn't really see it work. glibc's fgets() seems to read single
chars, so that we never have anything in the line buffer.
* Added handling for EOF. Works well with Be's Terminal, ours seems to
write an ESC sequence instead of the EOF char (Ctrl-D), though.
* Extended output processing support (ECHOE, ECHOK, ECHONL, OCRNL,
ONLRET, OLCUC).
* Writes use user_memcpy() now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@21735 a95241bf-73f2-0310-859d-f6bbb57e9c96
