- Moves all test code from DSPTool into UnitTests/Core/DSPAssemblyTest.
- Converts test files (which could only be loaded if they were in the
shell's working directory, so basically never) into C++ values.
- Enables most of the commented-out tests.
- Removes non-deterministic random code test.
Disassemble code without the additional text for humans, like the
current PC and opcode hex values, so that it can be reassembled.
I'm not updating any commented-out tests here.
You may want to read the PR #2047 comments before reading this.
Dolphin attempts to support an unencrypted type of Wii discs
that apparently is identified by a 4-byte integer at 0x60
being non-zero. I don't know what discs (if any) would be
using that format, so I haven't been able to test Dolphin's
support for it, but it has probably been broken for a while.
The old implementation is very short but also strange.
In CreateVolumeFromFilename, we read a 4-byte integer from
0x60, and if it's non-zero, we create a CVolumeGC object
instead of a CVolumeWiiCrypted object. This might seem like
it makes no sense, but it presumably worked in the past
because IsVolumeWiiDisc used to check the volume type by
reading the magic word for Wii straight from the disc,
meaning that CVolumeGC objects representing unencrypted Wii
discs would be treated as Wii discs by pretty much all of
Dolphin's code except for the volume implementation code.
(It wasn't possible to simply use CVolumeWiiCrypted, because
that class only handled encrypted discs, like the name says.)
However, that stopped working as intended because of ace0607.
And furthermore, bb93336 made it even more broken by making
parts of Dolphin expect that data read from Wii discs needed
to be decrypted (rather than the volume implementation
implicitly deciding whether to decrypt when Read was called).
Disclaimer: Like I said before, I haven't been able to test
any of this because I don't have any discs that use this
unencrypted Wii disc format, so this is all theoretical.
Later, PR #2047 tried to remove Dolphin's support for
the unencrypted Wii disc format because seemingly no
discs used it, but the PR got closed without being merged.
At the end of that PR, I said that I would make a new PR
with a better implementation for the format after PR #2353
was merged. Now that PR #2353 is merged (two years later...)
and PR #5521 is merged, the new implementation was easy to
make, and here it is!
Untested.
Normal users don't care about it. In fact, people care so
little about it that the Wii implementation of it was broken
starting from when it was implemented (eb65601) to 7 years
later (e0a47c1), apparently without anyone reporting it.
- coef: Explicitly set 23 different values that are used by GBA UCode,
and tweaked overall parameters to more closely match those 23 values.
- irom: Moved a few functions to their proper places, updated BootUCode
to configure DMA transfers using AX registers as well as IX registers
(the GBA UCode uses this to do two sequential transfers in one call),
and added partial functions used by GBA UCode.
All functions were reverse-engineered solely based off of observed
effects on the virtual machine: register states before-and-after, dmem
interactions, and DMA transfers. The specific coefficients were observed
being read from dmem, and must be exactly those values to function
properly. I have no knowledge of how the official ROM implements these
functions, or how it is implemented overall.
Tested with The Legend of Zelda: Four Swords Adventures, Final Fantasy
Crystal Chronicles, and Billy Hatcher and the Giant Egg (to download
ChuChu Rocket!).
`P_REG1C` had the same value as `P_ACCL`, so was causing spurious errors
when used with ACCM registers. Gcdsptool (which calls this `P_ACCLM`)
gives it the value `P_REG | 0x1c10` instead, and handles errors in the
same block as other REG## enums.
This makes the interface slightly cleaner and a bit more consistent
with the other getters. Still not fully the same, since the others
don't really handle failures with std::optional; but at least the
value is returned by value now, as opposed to having the function
take a pointer to a u64.
This gets rid of some assumptions that non-DiscIO code was making about
volume types. It's better to encapsulate as many of the volume type
differences as possible in DiscIO.
Made possible by PR #2353.
This is more reliable, as this guarantees subsystems will be
shut down in the same order they were initialised (if they were
initialised). It also allows us to stop keeping track of what needs to
be shut down manually and just return in case of errors.
This should prevent the emulator from getting totally stuck when
the boot process does fail.
This makes it hard to support different boot params for different boot
types. We should not be making the assumption that Dolphin will
always be booting directly from a file (and in particular, only
using a string).
It's incompatible with future changes that will allow Dolphin to boot
a NAND title properly from well, the NAND, as opposed to booting from
WADs. (And no, treating the title TMD as a "bootable" path doesn't
count. Especially when that approach won't work with NAND images
or IOS LLE.)
And it's confusing to expose this functionality from the UI. It's
pretty bad for UX to change the play button's behaviour depending on
whether the user has launched something before, configured a default
file to boot, added a directory to their game paths.
This commit moves the write function to where it should be (IOS),
especially when ES::ImportTicket() is the only place to use it.
Prevents misusing the ticket import function, and removes one unsafe
direct write to the NAND that does not go through IOS.
This also fixes the destination path: the session root is the one which
should be used for determining the ticket path, not the configured one.
The whole NANDContentLoader stuff is truly awful and will be removed
as soon as possible.
For now, this fixes a bug that was exposed by std::optional::operator*.
std::optional makes a few things a bit neater and less error prone.
However, we still cannot use C++17 (unfortunately), so this commit
adds an implementation of std::optional that we can use right now.
Based on https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/lib/gtl/optional.h
which seems to be fairly similar to C++17's <optional> and standards
compliant. It's one of the few implementations that handle propagating
type traits like copy constructibility, just like libc++/libstdc++.
`MainWindow` initializes a number of input interfaces but never shuts
them down. This was causing a crash-after-exit on macOS where the
ControllerInterface backend stores a `std::thread` object in a static
variable and only stops it when ControllerInterface::Shutdown is called.
FormLayout is not intended for this sort of dialog:
> QFormLayout is a convenience layout class that lays out its children
> in a two-column form. The left column consists of labels and the right
> column consists of "field" widgets (line editors, spin boxes, etc.).
pSysMem is of the type const void* -- because of this, it makes the
original delete[] call undefined behavior, as deleting a void pointer is
undefined behavior.
Also punning types into existence, like what was done for the stereo
image header is undefined behavior as well. The proper way to do this is
to either manually add all individual bytes manually, or memcpy the
struct into memory.
As we want to deallocate the memory before returning, and because
pSysMem is a const void*, we keep a unique_ptr to the data and just pass
pSysMem a raw pointer to the data.
Use @Orphis's FindFFmpeg module from ppsspp:
2149d3db7f
From that commit:
> This new module should be able to handle both libraries in the regular
> paths and fallback to pkg-config.
> It is also able to find dynamic libraries, not just static libraries.
> It will generate imported targets with the name FFmpeg::<lib> that you
> can use in your scripts.
The current prefixing makes it harder to build test executables directly
from the command line, since the target name breaks CMake convention and
doesn't match the name passed to `add_dolphin_test`. They all have "Test"
somewhere in the name anyways.
gcc complains that the printf %x formatting instruction expects an
'unsigned int' but we pass a 'size_t'. We add the 'z' length formatting
specifier used for 'size_t'
This adds the WARNPC directive from xkas/asar to complement the existing ORG
directive. A common useful idiom is "WARNPC 0xXXXX\nORG 0xXXXX," which only
seeks forward and raises an error if you've already written to that part
of the file.
Fixes https://bugs.dolphin-emu.org/issues/10265 (Star Wars: The Clone
Wars hangs on loading screen with DSP-HLE and JIT Recompiler).
The Clone Wars hangs upon initial boot if this interrupt happens too
quickly after submitting a command list. When played in DSP-LLE, the
interrupt lags by about 160,000 cycles, though any value greater than or
equal to 814 will work. In other games, the lag can be as small as 50,000
cycles (in Metroid Prime) and as large as 718,092 cycles (in Tales of
Symphonia!).
All credit to @hthh, who put in a heroic(!) amount of detective work and
discovered that The Clone Wars tracks a "AXCommandListCycles" variable
which matches the aforementioned 160,000 cycles. It's initialized to ~2500
cycles for a minimal, empty command list, so that should be a safe number
for pretty much anything a game does (*crosses fingers*).
These settings are already loaded and saved to the SYSCONF. The INI
load/saves are redundant and do not work anyway because they are
overwritten by SYSCONF.
This file is pretty small now that it doesn't handle Wii
partitions anymore, so let's move its contents to Volume.cpp.
This is also more consistent with how blob creation works.
This happened to work without any problems because the only way for a
file system to be invalid was to not have the right GC/Wii magic word
in the unencrypted area, and a volume could not be created without
having the right GC/Wii magic word there. Now that file systems read
the magic word from a partition instead, a fix is needed.
I replaced m_OffsetShift with m_Wii in bb93336 to support
the decrypt parameter for read functions. Doing that is no
longer necessary, so m_offset_shift is now used like before.
By removing mutable state in VolumeWiiCrypted, this change makes
partition-related code simpler. It also gets rid of other ugly things,
like ISOProperties's "over 9000" loop that creates a list of
partitions by trying possible combinations, and DiscScrubber's
volume swapping that recreates the entire volume when it needs to
change partition.
- Makes DSP-LLE code checksums the same as those from DSP-HLE. I'm
assuming DSP-HLE was doing it correctly, since there are numerous
references to these pre-endian-swapped checksums (including in
DSPHost.cpp itself).
- Fixes disassembly when dumping code from DSP-LLE, which was using the
wrong endianness and giving totally bogus output.
- Reveals error messages of the format, "Bah! ReadAnnotatedAssembly
couldn't find the file ../../docs/DSP/DSP_UC_AX_07F88145.txt," which
seems to be intended behavior that was previously hidden.
This change centralizes all of the path handling and file writing logic
in DumpDSPCode. DSP-HLE also gains the feature of DSP-LLE to
automatically disassemble dumped code and write it to an accompanying
text file.
With the relocation of DumpDSPCode to DSPCodeUtils, the only remaining
function in DSPLLETools is DumpCWCode. This function 1) is not used
anywhere (not even in DSPTool), 2) doesn't seem to really do anything,
and 3) has a single comment saying "TODO make this useful :p"
This is something that should be the responsibility of the frontend
booting the game. Making this part of the host 'interface' inherently
requires frontends to leak internal details (much like the other
UI-related functions in the interface).
This also decouples more behavior from the debugger and the
initialization process in the wx frontend. This also eliminates several
usages of the parent menubar in the debugger code window.
VolumeDirectory doesn't support necessities like TMDs,
so thanks to 5.0-2172 (18968ab), EmulatedBS2_Wii crashes
when the inserted disc is a VolumeDirectory.
This commit fixes that.
This commit makes our DOL booting code very similar to our
ELF booting code. One exception is that the DOL booting
code still always calls SetupBAT. (Note that EmulatedBS2_GC
calls SetupBAT even if no disc is inserted.) I'm not sure
if there's a point to the difference, but I thought I'd
better avoid changing it so that I don't break anything.
For thread safety reasons, the currently inserted volume must
only be accessed by the DVD thread (or by the CPU thread if it
calls DVDThread::WaitUntilIdle() first). After this commit,
only DVDThread.cpp can access the volume, which prevents code in
other files from accessing the volume in a non-threadsafe way.
Direct access to the WAD bytes is required to read contents with proper
padding data (since they can sometimes end up being outside of the
data app section). Allowing the whole buffer to be accessed directly
would be error prone, so this commit adds GetContent() to WiiWAD
for getting raw content data by index.
These cannot be booted, so it is bad UX to show them in the UI as if
they were regular titles, and yet have different behaviour for them.
And technically, there is no reason to allow them to be used to boot
in the first place.
Another reason they should not be shown is that Dolphin fails
spectacularly with WADs that have a valid boot content index, but are
not PPC titles (e.g. IOS WADs). The only reliable way to avoid this
is to check for the title type and only show channels, just like
the Wii System Menu.
Mistakenly used the wrong TMD to clean up the import.
The original TMD is the one that is supposed to be used when
cancelling an import, but I forgot it's in the /import directory after
starting an import.
This exposes all ES title management ioctlvs to avoid duplicating IOS
code everywhere and to make it easier to reuse (since this way it's
not unnecessarily tied to the PPC IPC mechanism anymore) and unit test.
Some functions were also renamed for consistency with the other names,
*and* with official names.
It's better to just let the calling code provide a volume
object instead of needing one SetVolume for each way of
creating a volume. This simplifies InsertDiscCallback and
is needed for the following commits.
Just like DeleteTitle, Using CNANDContentManager is overkill,
inefficient and useless. And it results in a few failures in
situations where a delete should just always work.
But here it gets bonus points, because it manages to actually use
the TMD for deleting contents, when IOS does none of that and just
deletes files ending with .app in the title content directory. :)
This enables constructing an IOS instance that is not tied to emulation
and that can be simply used for internal purposes (ES, FS).
NAND root initialisation was moved to IOS since we cannot rely on HW
doing that for us anymore, and technically the NAND is entirely managed
by IOS anyway.
Default column width was being used for math related to automatic column sizing, in the case of COLUMN_SIZE the default width was -1 which resulted in an issue where an errant scrollbar would appear.
The MemoryWindow.cpp strings should be the same as the
MemoryCheckDlg.cpp strings so that translators don't
have to translate very similar strings twice.
Also adding i18n comments. These strings are *very* easy
to misinterpret in my opinion.
This is unsafe, because the NAND should not be accessed and messed with
while it is being used. In fact, this kind of inappropriate behaviour
will not be possible when we get NAND image support.
And even if there were no safety issue, there is still no reason
a *getter* function should *do* something that has lasting effects
on user data. GetWiiFSPath() should *just* return a path!
This adds code to update the PPC's UID and GID, which fixes a
regression with the system menu. Ideally, we would simply rely on IOS's
ES_Launch here and not duplicate the logic here, but we cannot do that
properly until the direct WAD launch hack is dropped, *and* until the
IOS changes that would enable internal calls are ready.
Since this fixes a semi-important regression with managing saves from
the SM, and the duplication is not too terrible, I believe it is a good
idea to get this fixed right now to avoid affecting users too much.
I do plan to fix this properly in the future.
* CNANDContentManager does things that are absolutely useless. In
particular, it parses the ticket, the TMD, reads contents, etc.
when we only need to remove the title directory.
* This means it will fail if the ticket cannot be found, when that
should not be the case.
* This also obviously caused DeleteTitle to be incredibly inefficient.
* We are already removing the title directory later in the function,
as CNANDContentManager does not even delete titles correctly.
DeleteTitle != DeleteTitleContents.
Since these button names are printed on all real controllers,
we should show them in the same way as they are printed on
the controllers, regardless of the user's language. It seems
like this was intended all along (except for "Start"), but the
_ markers in TASInputDlg.cpp (accidentally?) led to the button
names in the controller configs also becoming translatable.
I'm making exceptions for "L" and "R" because translators
may want to mark them in some way (for instance "L-Digital")
to clarify the difference from "L-Analog" and "R-Analog".
I'm also making an exception for START/PAUSE because it's
referred to as スタート in Japanese games.
I'm changing "Home" and "Start" to uppercase for consistency
with how Nintendo refers to those buttons, and because someone
who isn't familiar with the Latin script might not know the
connection between the lowercase and uppercase letters (most
users likely do know the connection, but we shouldn't assume it),
and because leaving "Start" as "Start" makes it "collide" with
unrelated strings, such as the string for the button that starts
a netplay session.
To rename "Start" and "Home" without breaking INI
compatibility, I added a ui_name variable like in f5c82ad.
* It should take a ticket view, not a title ID.
* It's missing a lot of checks.
* It's not deleting tickets properly.
* It's not deleting only the ticket it needs to delete.
* It should not return -1017 when the ticket doesn't exist.
* It's not returning the proper error code when a read/write fails.
* It's not cleaning up the ticket directory if there is nothing left.
This commit fixes its implementation.
* Supporting other ticket types makes the logic slightly more complex.
* There have been no such non-RSA2048 tickets seen during the Wii's
lifetime.
* The Wii's IOS doesn't even have support for them.
Stops CMake from saying "BlueZ NOT found, disabling bluetooth support"
on other OSes. Windows, macOS, and Android support Bluetooth using other
libraries. I'm not sure if non-Linux, non-Android Unices (like FreeBSD)
need another message?
Whenever the EmulationActivity crashes and the app gets back to the
TvMainActivity, dolphin will crash tring to restor the mBrowseFragment
since we don't save the adapter data in the bundle.
This is quick hack to avoid the crash. The proper fix would be to save the
adapter data in the bundle and restore it before restoring the fragment
when the activity gets recreated.
Crash Stacktrace:
Process: org.dolphinemu.dolphinemu.debug, PID: 30353
java.lang.RuntimeException: Unable to start activity ComponentInfo{org.dolphinemu.dolphinemu.debug/org.dolphinemu.dolphinemu.ui.main.TvMainActivity}: java.lang.IllegalArgumentException: Invalid position 1 requested
at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:2691)
at android.app.ActivityThread.handleLaunchActivity(ActivityThread.java:2752)
at android.app.ActivityThread.-wrap12(ActivityThread.java)
at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1461)
at android.os.Handler.dispatchMessage(Handler.java:102)
at android.os.Looper.loop(Looper.java:154)
at android.app.ActivityThread.main(ActivityThread.java:6120)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:865)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:755)
Caused by: java.lang.IllegalArgumentException: Invalid position 1 requested
at android.support.v17.leanback.app.BrowseFragment.createMainFragment(BrowseFragment.java:509)
at android.support.v17.leanback.app.BrowseFragment.replaceMainFragment(BrowseFragment.java:1454)
at android.support.v17.leanback.app.BrowseFragment.setAdapter(BrowseFragment.java:764)
at org.dolphinemu.dolphinemu.ui.main.TvMainActivity.buildRowsAdapter(TvMainActivity.java:183)
at org.dolphinemu.dolphinemu.ui.main.TvMainActivity.onCreate(TvMainActivity.java:59)
at android.app.Activity.performCreate(Activity.java:6664)
at android.app.Instrumentation.callActivityOnCreate(Instrumentation.java:1118)
at android.app.ActivityThread.performLaunchActivity(ActivityThread.java:2644)
at android.app.ActivityThread.handleLaunchActivity(ActivityThread.java:2752)
at android.app.ActivityThread.-wrap12(ActivityThread.java)
at android.app.ActivityThread$H.handleMessage(ActivityThread.java:1461)
at android.os.Handler.dispatchMessage(Handler.java:102)
at android.os.Looper.loop(Looper.java:154)
at android.app.ActivityThread.main(ActivityThread.java:6120)
at java.lang.reflect.Method.invoke(Native Method)
at com.android.internal.os.ZygoteInit$MethodAndArgsCaller.run(ZygoteInit.java:865)
at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:755)
The accuracy doesn't match ppc, and worse, it doesn't set the error flags if the input is zero.
Lets stop to ship broken instructions, so right now, the interpreter is the closest one.
Fixes warning:
```
Source/Core/DiscIO/NANDImporter.cpp:55:17: warning: format specifies type 'unsigned long' but the argument has type 'u64' (aka 'unsigned long long') [-Wformat]
file.GetSize(), NAND_BIN_SIZE);
^~~~~~~~~~~~~~
1 warning generated.
```
This changes some parts of IOS (actually just ES) to reuse more crypto
code from IOSC or Common::AES.
TicketReader still returns the title key directly as opposed to having
ES use IOSC directly to avoid duplicating the title key IV stuff.
Side effects:
* A nasty unbounded array access bug is now fixed.
* ES_Decrypt/ES_Encrypt now returns sane results for keys other than
the SD key.
* Titles with a Korean ticket can now be decrypted properly.
And in the future, we can look into implementing ioctlv 0x3c and 0x3d
now that we have the proper "infra" for IOSC calls.
This prevents the IOS crypto code and keys from being spread over
the codebase. Things only have to be implemented once, and can be
used everywhere from the IOS code.
Additionally, since ES exposes some IOSC calls directly (DeleteObject
and Encrypt/Decrypt), we need this for proper emulation.
Currently, this only supports AES key objects.
Netplay uses a blank NAND, which means that homebrew launchers like
Gecko will force users to install IOSes.
Expecting netplay users to have a proper NAND setup is unrealistic,
and we don't actually give them a good way of syncing NANDs, so
let's extend the hack to netplay/TAS until we have a better way
of dealing with the issue.
This changes the main IOS code (roughly the equivalent of the kernel)
to a class instead of being a set of free functions + tons of static
variables.
The reason for this change is that keeping tons of static variables
like that prevents us from making an IOS instance and reusing IOS
code easily.
Converting the IOS code to a class also allows us to mostly decouple
IOS from the PPC emulation.
The more interesting changes are in Core/IOS/IOS. Everything else is
mostly just boring stuff required by this change...
* Because the devices themselves call back to the main IOS code
for various things (getting the current version, replying to a
request, and other syscall-like functions), just like processes in
IOS call kernel syscalls, we have to pass a reference to the kernel
to anything that uses IOS syscalls.
* Change DoState to save device names instead of device IDs to simplify
AddDevice() and get rid of an ugly static count.
* Change ES_Launch's ack to be sent at IOS boot, now that we can do
this properly.
It only marks a string for translation. It doesn't actually do anything
at runtime, so the string will always be displayed in English. Even if
we would've had a way to make the translation work, we shouldn't
translate this, because OSD doesn't support non-ASCII characters.
Some strings were marked with _trans in some places but not
others. This commit adds extra _trans markers so that the
usage of _trans is consistent.
This shouldn't have any effect on which strings actually get
translated. (Note that _trans doesn't do anything at runtime.)
I also added a few new i18n comments.
std::string's operator+ will handle this. Also move std::string to where
they're actually needed. There's no need to construct an unnecessary
string if the first failure case occurs.
This removes the need for multiple texture files to store the mipmap
chain for a texture. As many mipmaps will be loaded as are present in
the DDS file, and any remaining mipmaps will fall back to the old
behavior.
This changes the IOS code to handle ES contexts inside of ES, instead
of leaking out implementation details into the IPC request dispatcher.
The intent is to clarify what's shared between every single ES context,
and what is specific to an ES context. (Not much.) This should reduce
the number of static members in the ES class.
The other changes are there just because we now keep track of the
IPC FD inside of ES.
Future plans:
* After the WAD direct launch hack is dropped, the title context
will be made a class member.
* Have proper function prototypes, instead of having every single one
of them take ioctlv requests. This will allow reusing IOS code in
other parts of the Dolphin codebase without having to construct
ioctlv requests.
Use Bitmap.setPixels() instead of Bitmap.copyPixelsFromBuffer() the former
use non pre-multiplied values of the colors which is what we expect to
come from the native code.
GTK2 is a dependency on Linux whenever USE_X11 is true, but we were
not linking or adding the include directory for GTK for DolphinWX.
Fixes a regression introduced by 6197d9622.
Simple quality-of-life addition that allows "uninstalling" WADs
(removing the corresponding installed title) from the NAND.
The option is only enabled when the WAD can be uninstalled
The motivation for this is actually to encourage proper usage of the
WAD launch feature (installing it to the NAND first), so we can
drop the "direct WAD title launch" hack.
Looking more carefully at the IOS ticket view generation code reveals
that the first field in the TicketView struct is copied over from
the ticket version, extended to 4 bytes.
This implements ES_SetUid, which is used by the system menu to change
its own permissions. This is required for implementing permission
checks and proper NAND metadata support in the future.
This fixes an error condition on macOS when HIDAPI calls
IOHIDManagerCreate and IOHIDManagerClose on different threads. The
error behavior is non-deterministic, but can cause EXC_BAD_ACCES and
kill the program.
This will be required for permission checks in the future.
Note that this is only for the PPC as we do not have actual processes.
Keeping track of other modules' UIDs/GIDs is virtually useless anyway.
UID/GID changes are implemented in the following functions:
* ES_Launch
* ES_DIVerify
ES_SetUid is not implemented yet because it'd need further changes.
Fixes an error with the CoreAudio backend, which apparently doesn't
allow you to set the volume before starting the stream:
```
59:31:087 AudioCommon/CoreAudioSoundStream.cpp:97 E[Audio]: error setting volume
```
This shouldn't cause any problems with other backends, since the mixer
starts with silence anyways.
Shawn Hoffman