* Also defer signals while registering fork hooks.
* While malloc provides fork heap hooks which lock the heaps and unlock/reinit,
malloc_debug provides empty hooks.
* Ideas suggested by Ingo, patch reviewed by him. Thanks a lot!
* Also call fork parent hooks on failure.
* Solve locks-up when combining multithreading and process forking, should help
with #13111.
* There's no need to supply ways to mismatch the buffer duration
and size. Anything should reflect the media_format, this is at least
fixed on API level.
* There's no point actually in providing BTimeSource dependant
functionality. If and when there will be need for something like
that, possibly never, an higher level solution will be integrated.
Team:
- Adjust report generation event to include a final status code for listeners.
CliContext,TeamWindow,ReportUserinterface:
- Use aforementioned status code to indicate whether report generation
succeeded or failed.
DebugReportGenerator:
- Notify listeners if report generation fails. This may have previously
been responsible for some bug reports where it was indicated that the
debugger hung without exiting after being asked to save a report from
a crashed app.
- When dumping disassembly, retrieve it directly if necessary rather than
requesting it via the user interface listener. This also fixes the quirk
that requesting to save a crash report while looking at the source code
of a function would trigger switching it to disassembly visually.
- When walking the list of threads to dump, acquire references to all of
them before starting. Otherwise, it was potentially possible for a running
but not crashed thread to exit while we were generating the report, leaving
us with a pointer to a deleted thread. This was most likely the cause of one
of the crashes reported in #13082.
- When receiving the notification that source code state has changed, clear
the waiting function. Otherwise, it was potentially possible for us to get
other state change notifications, leading to the data semaphore being
released too often. This would then cause later potential waits such as
the stack frame memory dump to not actually wait when they should,
potentially leading them to dereference objects that weren't yet ready.
This fixes another of the crashes in #13802.
- Split MessageCodes.h into a second file that separates out the
application-specific message codes from those used by the core.
Adjust includes accordingly. No functional change.
* The idea is to move the callback based mechanism into a derived
class. The objects can be composed to create working systems.
* The BMediaClient class supply RegisterInput/RegisterOutput
and BeginInput/BeginOutput is moved to BSimpleMediaClient.
* Various minor fixes.
* Capabilities() become Kind()
* Use media_client and media_connection to better model
the different situations of the API. Reduce code duplication
and move media_kit structs/methods into private.
* Various API and beauty improvements.
* Introduce id management for media_connections.
* Fix different issues preventing to correctly begin a connection.
* Those structs are provided as replacement for the classic
media_kit ones. They provide better encapsulation of the data
and the resulting code is heavily simplified. It's planned to
introduce BMediaRoster functions to provide conversion from
media_node and media_input/media_output to media_client
and media_connection.
FunctionInstance:
- Add new state FUNCTION_SOURCE_SUPPRESSED. This signals that the user
explicitly forced disassembly to be loaded despite source code being
available.
LoadSourceCodeJob:
- When forced to disassembly, use the above suppressed state accordingly.
SourceView/TeamWindow/TeamDebugger:
- Adjust to take new state into account as needed.
TeamDebugInfo::GetActiveSourceCode:
- When looking at a function to decide whether to return line information
based on source or disassembly, first examine the source code state. If
the source has never been loaded for that function, but we have it available,
set it on the function at that point. This lazily addresses the fact that
LoadSourceCodeJob is called on behalf of a specific function, and
consequently only sets the source code onto that function, and not all others
present in the same file. This allows us to differentiate between the case
where a function doesn't have source code available at all, versus a function
that has simply been forced to disassembly view at this point in time.
The primary symptom of the above issue was that attempting to set a breakpoint
outside of the currently active function, but within the same file would result
in the breakpoints view indicating that the breakpoint was at line 0 rather
than the appropriate line, and breakpoints would also not be drawn in the
source view for such locations.
Thanks to Humdinger for the heads up!
Chunks may be physically contiguous, but virtually disjoint. Adding
physical addresses may cause ranges to be merged incorrectly.
Signed-off-by: Jessica Hamilton <jessica.l.hamilton@gmail.com>
* The BMediaClient is an higher level API to the media_kit. It
corresponds to what the layout API was for the interface_kit.
The main idea is to allow the developer concentrate only on
higher level details and avoiding handle with the tricky parts
of the media_kit. At the same time the general purpose node that
is implemented inside would allow implementing the best techniques
around thus at the same time reducing code duplication and increasing
efficiency.
* BMediaClient is WIP, this is the initial merge of the branch.
The initial development stone was set long time ago and walked
through various design/implementation phases.
* Ideally we should support this feature by default to allow
future improvements to the plugins management.
* Fixes the major memory corruption that lead to various
crashes on exit in MediaPlayer.
* beaglebone vs beagleboard
* While the boards are almost the same, they have
diff. FDT's now (more memory, different layout etc)
* Make u-boot script more rpi-like
(depend on u-boot for initial addresses)
* Wasn't working, still doesn't.
* Now pulling in latest amdgpu atombios.h headers
* Lots of DisplayPort cleanup and removal of legacy stuff
* Add obsolete atombios header for things that disappear
(amdgpu doesn't support older cards like we do)
* Introduce new code to support later encoder tables
* Drop radeon-specific encoder service from common dp_info
* Tested on HD 5450 hdmi
* Helps prevent mainboard explosions and other bad stuff.
* ...maybe not explosions, but bad things.
* The RX 480 reference design can pull as much as 90W
from its PCIe slot at load. PCIe is rated for 75W.
* This change overloads the PCIe power cables vs the PCIe
slot. The PCIe power cables can handle going over spec.
* Untested since we really can't come close to loading a
RX 480 without hardware OpenGL ^_^
* Should be able to test on real hardware within a week.
- Add new interface TeamFunctionSourceInformation. Currently this
exposes a single function allowing one to query for the currently
active source code given a FunctionDebugInfo instance.
- Implement TeamFunctionSourceInformation on TeamDebugInfo.
- Pass TeamFunctionSourceInformation to Dwarf{Team,Image}DebugInfo.
In turn, make use of it in DwarfImageDebugInfo::GetStatement() in
order to determine whether to return the corresponding assembly
or source statement.
With this piece of information, the debugger is now correctly able to
determine that the user is currently looking at disassembly despite debug
info being available, and consequently adjust its stepping behavior based on
that. Previously, the source code statement was always used, leading to it
not being possible to single step assembly lines in such a circumstance
without manually using run to cursor.
Other related cleanups:
- TeamDebugInfo now inherits BReferenceable directly, rather than relying on
indirectly inheriting it from TeamTypeInformation.
- Remove BReferenceable from TeamTypeInformation. The latter is only an
interface anyways, and inheriting that base class from multiple locations
was causing GCC5 trouble when resolving BReference<TeamDebugInfo>, even
when virtual inheritance was used.
The HashMap constructor was called before the heap is initialized,
ending up calling malloc from the OpenHashMap constructor.
Oddly it was still working on x86 but broke other platforms.
Instead we add a Lookup() static method to Partition,
which by default walks gPartitions for the id,
and recursively calls itself on the children lists.
This means we must add a partition even temporarily to gPartitions
before Scan()ing it though.
Signed-off-by: François Revol <revol@free.fr>
* Only set HAIKU_BOOT_PLATFORM to bios_ia32 if not defined
* Add gnuefi build feature
* Introduce BOOT_LDFLAGS, and move options for passing to linker
into ArchitectureSetup
* x86_64 compile fixes for warnings in boot loader
* loader/elf.cpp: don't include ELF32 support when targeting EFI
* relocation_func.cpp: copy of the relocation code from gnuefi
to make _relocate extern "C", and avoid including <efilib.h>
* boot_loader_efi.ld: copy of gnuefi's elf_x86_64_efi.lds,
modified to include support for C++ constructors, etc. Keep in
sync with the gnuefi package
Signed-off-by: Jessica Hamilton <jessica.l.hamilton@gmail.com>
There was some mixup with the interrupt registers, still:
- The driver uses 16-bit read/write, but on SandyBridge the register is
32 bits
- There is a global interrupt enable bit, which must be set to unmask
everything else
- The bits for vblank interrupt are not the same on SNB and later PCH
based devices, and the code mixed the two.
Move the computation of the interrupt bits to an helper function, and
use it everywhere to make sure we always use the right bits.
Modesetting
===========
My previous hack was setting the transcoder registers, instead of the
display ones. Do that the way it is designed in the driver instead:
- If there is a transcoder, set its registers, but do not set the
display timings. The display will remain set at its native (and only)
resolution, and panel fitting will adjust the output of the transcoder
to match.
- If there is no transcoder, set the display registers directly to the
native resolution, as it was done on previous generation devices.
- fPipeOffset hacks no longer needed
DPMS
====
It seems the panel control register is not readable on PCH? Anyway, the
code would loop forever waiting for the bit to become unset when turning
the display off. Waiting seems to not be needed, so just remove it as
well as the "unlock" bit, which does not work for me and results in a
black screen.
Remaining hacks
===============
I still need to force HEAD_MODE_A_ANALOG to get output on pipe B (LVDS
display) working. I suspect something is common to the two pipes or not
allocated to the right one.
This version will have less side effects on other generations and help
with getting things to work on SandyBridge and possibly later devices.
Please test and report.
