for these changes. I was mostly just staring in amazement at the screen
while Axel and Thomas were discussing IA32 internals. A particularly
fascinating moment was when Thomas produced the cause of a bug we had
been trying to track down for hours off the top of his head (of course
iret behaves specially when the NT bit is set :-). His slowness must be
excused though, since he hadn't slept for more then 30 hours. ;-)
The code doesn't wholeheartedly deal with multi-processor machines yet.
Axel will certainly do some cleanup...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12239 a95241bf-73f2-0310-859d-f6bbb57e9c96
describe partitions that are out of bounds. We simply ignore those now,
instead of letting the complete scanning fail.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12206 a95241bf-73f2-0310-859d-f6bbb57e9c96
partitioning system module relies on it, otherwise extended
partitions won't be recognized.
* Be a little less lazy and do the priority partition recognition,
i.e. all partitioning systems are asked to identify a partition and
the one that believes it can handle the partition best, wins the pot.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12187 a95241bf-73f2-0310-859d-f6bbb57e9c96
* We apparently cannot use constants like kPartitionTypeIntelExtended
in static array initializers in the kernel.
* Fixed misunderstanding between Axel (boot loader) and me (Intel
partitioning system module) how to deal with the cookie the
identify() hook returns. We switch to a someone cleaner method by
always invoking free_identify_cookie() for it, even if it were passed
to scan(). Since we want to keep the cookie, we add a reference
counter to it.
This should fix a couple of issues with recognizing the boot partition.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12185 a95241bf-73f2-0310-859d-f6bbb57e9c96
Everything compiles here, but I couldn't test it yet, since Haiku
doesn't seem to boot properly at the moment.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12104 a95241bf-73f2-0310-859d-f6bbb57e9c96
after fork()ing and recreate all semaphores we've been using. To do
the latter, we simply maintain a list of all locks.
fork()ing multithreaded teams can still be troublesome.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12097 a95241bf-73f2-0310-859d-f6bbb57e9c96
explicitely when needed.
* Added a ToDo item to _user_resize_area().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12095 a95241bf-73f2-0310-859d-f6bbb57e9c96
objects/ tree, so that it doesn't clash with others (those should be
fixed, too). gdb is (or will be, as soon as I check it in) using this
version, too.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12081 a95241bf-73f2-0310-859d-f6bbb57e9c96
src/libs and build a shared library to be used by all programs that
need it (bash, gdb) instead of letting each have its own copy.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12036 a95241bf-73f2-0310-859d-f6bbb57e9c96
The target debugger won't work though, since quite a bit of code
would need to be moved from the native support files to those for
the target support. But that's not that important at the moment, since
we're mainly want a native debugger.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12027 a95241bf-73f2-0310-859d-f6bbb57e9c96
with the bit number, not the respective mask).
* Added a small hack to allow single stepping to work in qemu.
Apparently the BS bit in DR6 is not set when the debug exception
is handled. So we always assume that a single step event occurred,
when we couldn't recognize any other event, if the hack is enabled.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11998 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added initialization function. Currently only need to init the
single step hack for qemu.
* Fixed a deadlock when the nub thread destroyed the debug info. It was
waiting for itself.
* Moved the filling in the origin info of the debug messages into
thread_hit_debug_event_internal(). No need for code duplication.
* Writing to user memory can now be partial. We also change the area
protection, if it wasn't writable. Necessary for setting software
breakpoints.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11996 a95241bf-73f2-0310-859d-f6bbb57e9c96
the thread count need to be accessed atomically.
* Made debugging through exec*() work.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11993 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Implemented unblocking of pending operations when closing a cookie.
Needed to change the simple one reader semaphore, one writer
semaphore strategy for blocking to a request queue based approach
(otherwise unblocking would have been really hairy).
* Implemented select() support.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11931 a95241bf-73f2-0310-859d-f6bbb57e9c96
* A new select sync pool entry wasn't really added to the pool.
* More debug output.
* The semaphore created in common_{select,poll}() needs to be created
with count 0, otherwise we won't block.
* A serious race condition remains in common_{select,poll)(). When a
selected FD is closed we can't deselect select events anymore, so
that subsequent notify_select_events() issued by the FS/driver might
write into already free()d memory. A solution would be to get_fd()
the involved FDs and operate directly on them instead of using the
indices. I'm just too lazy to do that now.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11929 a95241bf-73f2-0310-859d-f6bbb57e9c96
Entirely implemented on top of the public <debugger.h> API, but more
convenient to use. Lives in a separate library libdebug.so.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11925 a95241bf-73f2-0310-859d-f6bbb57e9c96
and drivers that need to provide special select() support with
managing select notification information.
* Renamed vfs_select.c to *.cpp and fixed the resulting C++ compiler
errors. Added tracing macros.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11886 a95241bf-73f2-0310-859d-f6bbb57e9c96
and free_*_cookie() hooks of the underlying FS were always called
together when the reference count of the FD dropped to zero. When
blocking operations (reading/writing) on the FD were still in progress
this would never happen, though. Now we additionally maintain an open
count and call the close_*() hook when it drops to zero.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@11882 a95241bf-73f2-0310-859d-f6bbb57e9c96
(sourced) before unmounting the image. That solve my problem that mounting
and copying things onto the image after makehdimage is done, damages the
FS integrity for any reason (Axel can't reproduce it, of course :-).
Maybe others have the same problem.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11744 a95241bf-73f2-0310-859d-f6bbb57e9c96
* user_debug_break_or_watchpoint_hit() into separate functions for
break- and watchpoints. The one for breakpoints gets another parameter
which indicates whether the breakpoint was a soft- or hardware
breakpoint. This info is also sent to the debugger.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11714 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Changed the index-based lookup of strings for debug message and
exception codes to a search through entry lists. A bit slower, but more
robust with respect to changes of the underlying enumerations.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11713 a95241bf-73f2-0310-859d-f6bbb57e9c96
Most of the debug events had their own message, but some were grouped
into this `stopped' message with a `why' field to describe what kind
of event it actually was. Hence the user had to check two levels until
they could know what event occurred. So I got rid of this message and
provided separate messages for the concerned events.
* Got rid of B_DEBUG_MESSAGE_GET_WHY_STOPPED and introduced
B_DEBUG_MESSAGE_GET_CPU_STATE which is pretty much the same.
* Put B_DEBUG_MESSAGE_RUN_THREAD and B_DEBUG_MESSAGE_STEP_THREAD into
a single B_DEBUG_MESSAGE_CONTINUE_THREAD message with a boolean
`single_step' field.
* get_debug_why_stopped_string() -> get_debug_message_string().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11712 a95241bf-73f2-0310-859d-f6bbb57e9c96
(there's a separate B_DEBUGGER_MESSAGE_EXCEPTION_OCCURRED now).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11707 a95241bf-73f2-0310-859d-f6bbb57e9c96
condition when the traced team exits, but this seems to be a inherent
problem. The thread that calls _kern_exit_team() apparently still manages
to send the post-syscall message, but when the debugger tries to continue
it, it is already gone. Not really harmful. We'll probably find a remedy
later.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11706 a95241bf-73f2-0310-859d-f6bbb57e9c96
* No longer kill the time when a page fault occurs, but rather send a
SIGSEGV.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11705 a95241bf-73f2-0310-859d-f6bbb57e9c96
of a signal.
* Added sigaction_etc(). It's just a sigaction() for which the thread
in question can be specified.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11704 a95241bf-73f2-0310-859d-f6bbb57e9c96
not to be notified.
* Added debugger commands for setting/getting of thread signal ignore
masks and signal handlers.
* Renamed user_debug_fault_occurred() to a more correct
user_debug_exception_occurred(). It no longer sends a `stopped' message,
but the new one dedicated to exceptions. Additionally the number of the
signal is supplied that will be sent, when the thread continues (without
indicating to ignore the event).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11703 a95241bf-73f2-0310-859d-f6bbb57e9c96
get_debug_exception_string() to get a user-readable string for an
exception type.
* Added more signal support. Now the debugger can set per thread which
signals it wishes to be notified of. It can also just ignore the next
occurrence of a signal.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11701 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added support for debugger handover. The current debugger need to send
a B_DEBUG_MESSAGE_HANDED_OVER message. The rest is completely
transparent to the new debugger. It simply calls install_team_debugger().
* Implemented _user_wait_for_debugger().
* Reworked a few bits to ensure that after a remove_team_debugger() or
after requesting a debugger handover the debugger doesn't get any more
messages.
* When the debugger is removed or dies, the debugged threads should now
cleanup their debug info and continue, instead of waiting at their
debug port forever.
All the new features are not tested. Will happen, when gdb will be
debugged.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11697 a95241bf-73f2-0310-859d-f6bbb57e9c96
(new `flags' parameter, flag B_WAIT_TILL_LOADED). This is the default
behaviour for load_image(), so it behaves like R5's implementation now,
i.e. errors while loading are reported to the caller.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11696 a95241bf-73f2-0310-859d-f6bbb57e9c96
* We no longer notify the debugger about a new image when it is registered,
but when it has been relocated.
* When the application image has been relocated, we notify the one who
initiated its loading.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11694 a95241bf-73f2-0310-859d-f6bbb57e9c96
(maybe this should better go into a private header?). If given the function
waits till the loader is done with loading and relocating the executable.
Otherwise it returns before the loader starts.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11690 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Added _kern_image_relocated() and _kern_loading_app_failed(), which
are invoked by the loader to provide the kernel with more info.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11689 a95241bf-73f2-0310-859d-f6bbb57e9c96
it actually meant.
* Refactored the image initialization process (init_dependencies()) a bit.
topological_sort() is now more flexible.
* The image relocation is now done by relocate_dependencies(). The order
of relocating the images is the same as the initialization order, so
that the app image is the last to be relocated. After having relocated
an image the kernel is notified.
* The FATAL() macro has another parameter, an error code now, and calls
the new syscall _kern_loading_app_failed() instead of _kern_exit_team().
So the error can be reported to the team that called load_image().
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11688 a95241bf-73f2-0310-859d-f6bbb57e9c96
second time for already loaded libraries, when a library/add-on is
loaded.
* image_t features now both, a path and a name. The name is the DT_SONAME,
if present, the leaf name otherwise.
* Modified find_image() to compare both path and name, if necessary.
Added a type mask parameter to specify what image types one is
interested.
* Refactored open_container(). We now also support DT_RPATH, a dependency
search path list that can be specified for the dependencies of a shared
object.
* Extended [un]load_library() to support add-ons.
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11685 a95241bf-73f2-0310-859d-f6bbb57e9c96
Ingo Weinhold