- Call ABI_AlignStack even on x86-64.
- Have ABI_AlignStack respect the difference in current alignment
between the root JIT function, which has a prolog, and
ProtectFunction thunks, which do not. This was causing many games
to crash on start on OS X. Since this might otherwise mean changing
the stack pointer before every call...
- Have one prolog/epilog function rather than two (one of which
definitely did not do what it was thought to do), and make it
actually work like a normal one, so that the stack frame shows up
properly in the debugger. There should be no performance impact.
ES_DECRYPT is usually called with the in IV parameter equal to the out
IV parameter. Dolphin was preemptively zeroing out the out buffer,
causing it to read zeroes for the IV. (Also be more correct when those
parameters *aren't* equal.)
Dolphin now has the dubious honor of having an exploit (smashstack) work
in it. It crashes in Project M, though.
When servicing a write-register request, it wrote the contents to the
register offset plus 0x8d070000, which corresponds to the actual
hardware registers, presumably in case the application wanted to read
them directly rather than with a read-register request. WriteToHardware
doesn't handle cached writes to registers, so it decided the address was
RAM, applied RAM_MASK, and happily wrote the register contents to
0x81070000, causing random corruption. Since the PPC does not normally
have access to those registers, there is no reason to be doing this in
the first place. Use a member to store these values instead.
(Also add a proper DoState.)
On OS X, openL2CAPChannelSync registers events on the current
thread's run loop, so Connect needs to be called on a thread that's
going to do CFRunLoopRun; this was causing all Wiimote input to be
ignored. Easiest way to do that is to use the Wiimote thread, and
have Read call CFRunLoopRun to block on events, bringing OS X's
Wiimote event loop in line with every other platform's. This also
means that the thread can't be stopped and recreated by Prepare,
so make Prepare notify it instead, which has the side effect of not
making the GUI block on Prepare. (It would be nice if the GUI also
did not block on searching for devices, because blocking the GUI
is gross, but for now...)
IOdarwin.mm was assuming that scanning was complete when the run loop
was stopped (which the scan callback does), but somebody else was
stopping the run loop first, causing the scan to be aborted. Wait until
the scan is actually complete.
Not only do debuggers catch the expected SIGSEGV by default, I'm not
sure there's a working way to configure either gdb or lldb not to. This
commit uses an OS X-specific mechanism to add an exception handler that
runs before the default one.
It isn't easily accessible with sigaction or Mach exceptions (well,
requires an additional system call in the latter), and isn't necessary.
(and get rid of the enum, because it's only used once, and the comments
are more expressive than enum names)
It tried to cast incoming pointers to native ones. The values are
required to be null anyway, so it doesn't really matter, but it fixes a
warning at least.
(This is currently pointless, as the code in question is not used on OS
X anyway, but I'd like to see that option come back. In any case, fixes
the warning)
MSVC insisted on using a copy assignment where a move was intended and
ought to be used. This would have been caught, because the class in
question inherits from NonCopyable, which declares a move assignment
operator, which is supposed to delete the implicitly declared copy
assignment operator, but of course MSVC didn't do that either, causing a
class that should have been safe to be unsafe.
(Intertwined enough that's it's easier to do in one patch.)
(1) /dev/es did not support state save, which could cause crashes and
incorrect behavior after loading.
(2) NANDContentLoader tried to read all of a title's contents into
memory when it was first opened. Two issues:
- If any contents were missing, it bailed out. However, with DLC,
only some of the contents may be downloaded, as determined by the
permission bits in the ticket. Instead, return an appropriate error
when a content is accessed that doesn't exist on the filesystem
(don't bother checking the permission bits though).
- Everything was loaded into memory - even if it consisted of 3 GB of
songs, which caused Dolphin to lag out for quite a while (and would
fail on 32-bit). Instead, open content on demand.
This is required to be able to move objects that inherit from it.
(Note that this patch also #ifs out the class for the externals that
include it yet are compiled in pre-C++11 mode. It shouldn't matter,
since those externals don't use it.)
- Make "copy data into bundle" depend on the files actually being
changed, rather than being run on every build.
- postprocess_bundle depends on system files and checking the Dolphin
binary and all that, and would be iffy to try to avoid rerunning;
but it's only needed to produce a redistributable bundle, so add
SKIP_POSTPROCESS_BUNDLE to skip it for development.
comex