The vector was not constructed with the proper size, which results in a
buffer overflow as we were using memcpy.
This commit fixes that mistake and also uses a safer way of copying the
ticket view data (std::vector::insert instead of memcpy).
This commit fixes ES_Launch to work mostly the same as the real IOS
(except temporary, internal files such as /sys/launch.sys and title
handling; the latter will be handled in a future PR).
First of all, this adds two IOS functions, which correspond to two
IOS syscalls: 0x41 (boot_ppc) and 0x42 (boot_ios).
boot_ios() writes the new version to 0x3140, loads the new kernel,
which then proceeds to reinit IPC and load modules as part of its
boot process. Note that this doesn't include writing to any of the
other constants in the 0x3100 region.
In Dolphin, this is implemented by changing the active IOS
version variable, writing to 0x3140 and resetting all devices. This
has exactly the same effect as the real syscall.
The other syscall, boot_ppc(), writes code to the EXI boot buffer,
pokes all constants to memory before bootstrapping the PPC with a
binary from the NAND.
We skip the low level stuff and just load the DOL to memory (and set
the PPC's PC to 0x3400), which is essentially what IOS does.
The other change is mostly related to how ES_Launch is handled.
With a real IOS, if the launched title type is 00000001 (system) and
the title is not 1-2 (System Menu), ES calls boot_ios().
Otherwise, ES handles the launch as a PPC title. It reads the TMD
to determine the required IOS version. If it is the same, boot_ppc()
is called directly. If not, ES saves the title to launch to the NAND
before launching the new IOS. After the new IOS has finished booting,
it will notice the flag and then launch the requested title.
What this commit does is really just implement this logic into IOS HLE.
The result is a fix for a regression introduced by SetupMemory,
where reloading an IOS would have overwritten some OS constants.
This fixes booting games from the disc channel.
No idea why this wasn't implemented whereas ES_DeleteTicket and
ES_DeleteTitleContent were.
This probably fixes title deletion in old System Menus, and maybe
the new ones as well in some cases; I've seen 4.3 use this ioctlv.
On startup, wxWidgets pops up an assertion error:
> ./src/osx/menu_osx.cpp(648): assert ""IsAttached()"" failed in
> Refresh(): can't refresh unatteched menubar
Starting in #4916, upon startup wxWidgets pops up an assertion error:
> ./src/common/cmdline.cpp(527): assert ""Assert failure"" failed in
> FindOptionByAnyName(): Unknown option verbose
Fix this by overriding wxApp::OnCmdLineParsed to disable the default
handling (since we also disable the default options in
DolphinApp::OnInitCmdLine).
We already have a TMDReader, so let's actually use it.
And move ESFormats to IOS::ES, since it's definitely part of IOS.
This adds a DiscIO dependency on Core which will be fixed in a
follow-up PR.
The /tmp directory is cleared every time IOS boots up (when the FS
driver is initialized), *not* when /dev/fs is opened.
Although this should have no effect, it fixes the case where files
could be left in /tmp and seen before opening /dev/fs.
Gets some constants out of the ControllerEmu namespace, and modifies
ControlGroup so that it uses the enum type itself to represent the
underlying type, rather than a u32 value.
As this is a base class with virtuals, there needs to be an out-of-line
function definition to prevent the vtable of the class being placed within
every translation unit it's used in (i.e. every JIT implementation).
Getting and setting configuration from the base config layer are common
and repetitive tasks. This commit adds some simpler to use functions to
make the new system easier to work with.
Config::Get and Config::Set are intended to make switching from
SConfig a bit less painful. They always operate on the main system.
Example usage:
// before
auto base_layer = Config::GetLayer(Config::LayerType::Base);
auto core = base_layer->GetOrCreateSection(Config::System::Main, "Core");
u8 language;
core->Get("Language", &language, 0);
SetData("IPL.LNG", language);
// now
auto base_layer = Config::GetLayer(Config::LayerType::Base);
auto core = base_layer->GetOrCreateSection(Config::System::Main, "Core");
SetData("IPL.LNG", core->Get<u8>("Language", 0));
// or simply
SetData("IPL.LNG", Config::Get<u8>("Core", "Language", 0));
It is kind of silly to connect all of the configured Wii remotes (from
the user config; NOT netplay assigned remotes), then connect/disconnect
additional Wii remotes *after* the core has booted.
(The bWii check has been removed, because it's actually unneeded;
m_wiimote_map is always usable regardless of bWii. And we can't get
info about the currently running game without booting the core with our
current config system…)
This should fix Netplay trying to connect all configured Wii remotes.
Fixes a logic bug I introduced as part of #4942. We were not
handling the "read past EOF" case correctly, which caused
requested_read_length to underflow in some cases.
Also fixes a comparison (though this is unlikely to change anything).
Instead, the JitInterface namespace functions should be used instead. This
gets rid of all usages of the JIT global from the wxWidgets UI code.
The null check isn't needed as the JIT core would already need to be
initialized in order to be within a paused state. The null check is just a
remnant from 2011 that existed before the check for a paused state was
added.
This changes the read request handler to work just like IOS:
* To make things clearer, we now return early from error conditions,
instead of having nested ifs.
* IOS does an additional check on the requested read length, and
substracts the current seek position from it, if the read would
cause IOS to read past the EOF (not sure what the purpose of this
check is, but IOS does it, so we should too).
* The most significant one: IOS does *not* return the requested read
length, or update the file seek position with it. Instead, it uses
the *actual* read length.
As a result of simply doing what IOS does, this fixes _Mushroom Men_.
The game creates a save file, reads 2560 bytes from it, then
immediately writes 16384 bytes to it. With IOS, the first read does not
change the seek position at all, so the save data is written at
offset 0, not 2560. With Dolphin, the read erroneously set the
seek position to 2560, which caused the data to be written at
the wrong location.
Behavior confirmed by comparing IPC replies with IOS LLE and by looking
at the FS module in IOS.
What we actually care about is whether it's a GCZ file,
not whether it's compressed. (This commit doesn't change
the behavior, since the beginning of CompressSelection
discards items that aren't BlobType::GCZ or BlobType::PLAIN.)
- There's no clear definition of what it means for a GC/Wii game
to be compressed. GC games in GCZ are obviously compressed,
but what about formats like WBFS and CISO that just discard data?
- Hardcoded colors might have bad contrast with the used theme.
- It feels Windows XP to me.
YYCJ is one of the last titles to be completely broken in Dolphin.
It would hang right after the Wii remote screen. Looking at the
game's debug messages reveals that it was failing to find some of
its files.
IOS LLE booted the game just fine, which confirmed that it was an issue
with IOS HLE.
By comparing the ioctlv requests and responses with IOS, it turns out
that one of the very first ES replies was different between IOS HLE and
IOS: there was a mismatch for the content fd returned by ES.
Changing the initial content FD to what IOS returns fixes the issue.
IOS
000000: 00 00 00 08 00 00 00 00 00 00 00 07 00 00 00 09 ................
000010: 00 00 00 01 00 00 00 00 01 38 66 f0 00 00 00 20 .........8f....
000020: 00 00 00 00 00 00 00 00 00 00 00 00 81 36 d3 18 .............6..
000030: 81 36 d3 18 00 00 ff ff ff ff ff ff ff ff ff ff .6..............
Dolphin
000000: 00 00 00 08 06 00 00 00 00 00 00 07 00 00 00 09 ................
000010: 00 00 00 01 00 00 00 00 01 38 66 f0 00 00 00 20 .........8f....
000020: 00 00 00 00 00 00 00 00 00 00 00 00 81 36 d3 18 .............6..
000030: 81 36 d3 18 00 00 ff ff ff ff ff ff ff ff ff ff .6..............
So where did 0x6000000 come from?
4bd5674 changed "Wiimote" to "Wii Remote" in the GUI
(intentionally) but also did the same change for two INI
keys (seemingly unintentional, breaks backwards compatibility,
and is inconsistent with the INI's filename). This commit
reverts the INI keys but not the GUI strings.
This commit uses the same approach as cbd539e used for GameCube
sticks (but I made sure to avoid the bug that 56531a0 fixed).
Given a std::map can't have duplicate keys, iterating over the map
explicitly isn't necessary, and find() can just be used instead.
Also, instead of manually calling push_back() for every entry to
be added, the range constructor of std::vector can be used instead to add
the whole range all at once.
We (the Microsoft C++ team) use the dolphin project as part of our "Real world code" tests.
I noticed a few issues in windows specific code when building dolphin with the MSVC compiler
in its conformance mode (/permissive-). For more information on /permissive- see our blog
https://blogs.msdn.microsoft.com/vcblog/2016/11/16/permissive-switch/.
These changes are to address 3 different types of issues:
1) Use of qualified names in member declarations
struct A {
void A::f() { } // error C4596: illegal qualified name in member declaration
// remove redundant 'A::' to fix
};
2) Binding a non-const reference to a temporary
struct S{};
// If arg is in 'in' parameter, then it should be made const.
void func(S& arg){}
int main() {
//error C2664: 'void func(S &)': cannot convert argument 1 from 'S' to 'S &'
//note: A non-const reference may only be bound to an lvalue
func( S() );
//Work around this by creating a local, and using it to call the function
S s;
func( s );
}
3) Add missing #include <intrin.h>
Because of the workaround you are using in the code you will need to include
this. This is because of changes in the libraries and not /permissive-
This adds memory values for IOS11, 20, 30, 50, 51, 52, 60 and 70.
Unfortunately, IOS40 (in its working version) is not present on NUS, so
constants for that one are still missing.
This is something I removed by mistake. It didn't break anything in
most titles, but the Mii Channel *requires* write requests to
/dev/usb/kbd to succeed before exiting, so this commit readds the stub.
The latest version has tons of security fixes (which is expected for a
library such as mbedtls).
Updating also allows getting rid of a few deprecation warnings.
Turns out it is completely unneeded and it actually works better
*without* it.
Just try launching the system menu from the HBC; in current master, it
will disconnect the remote and not connect it automatically again. With
this change, it will.
The recent IOS initialization changes caused the Bluetooth device to
no longer exist before "starting" IOS (as it should be…), which meant
that Core could not activate Wii remotes during the boot process
anymore.
But that is actually completely useless, because we can just have the
emulated Bluetooth code itself activate Wii remotes as appropriate,
at the right moment.
wxWidgets headers don't play well with some of the macros defined in
Windows headers and perform their own magic to fix things, as long as
they're included entirely either before or after any Windows headers.
This file can cause a conflict in other DolphinWX files because NetPlay
headers directly include ENet headers, which leak Windows header macros.
To fix this, explicitly tell wxWidgets here that it needs to re-clean
macros.
We can return early from invalid conditions, which allows getting rid
of quite a few levels of indentation.
And let's not duplicate the new_position > file_size check.
ControllerEmu::Control instances have a unique_ptr<ControlReference>
member, which is passed either an InputReference or OutputReference.
Without this virtual destructor, deleting a derived class through a
pointer to the base class is undefined behavior.
This prevents Dolphin from writing to /sys/uid.sys (on the host; root
partition) when installing a WAD before starting emulation, because
the session root is not initialized at that moment.
Incidentally, this also gets rid of a singleton.
instruction tables
Previously, all of the internals that handled how the instruction tables
are initialized were exposed externally. However, this can all be made
private to each CPU backend.
If each backend has an Init() function, then this is where the instruction
tables should be initialized, it shouldn't be the responsibility of
external code to ensure internal validity.
This allows for getting rid of all the table initialization shenanigans
within JitInterface and PPCTables.
ControllerEmu, the class, is essentially acting like a namespace for
ControlGroup. This makes it impossible to forward declare any of the
internals. It also globs a bunch of classes together which is kind of a
pain to manage.
This splits ControlGroup and the classes it contains into their own source
files and situates them all within a namespace, which gets them out of
global scope.
Since this allows forward declarations for the once-internal classes, it
now requires significantly less files to be rebuilt if anything is changed
in the ControllerEmu portion of code.
It does not split out the settings classes yet, however, as it
would be preferable to make a settings base class that all settings derive
from, but this would be a functional change -- this commit only intends to
move around existing code. Extracting the settings class will be done in
another commit.
Several of the things done while performing a scan are logically their own
behavior (e.g. loading a titles file, checking if an entry should be added, etc).
The three parameter AnalogStick constructor takes an internal name, a
display name, and a default radius argument. The delegated constructor is
the one that calls the ControlGroup constructor, setting the group type,
so passing the group type here is a logic bug.
The only reason this appeared to work despite this bug is because
GROUP_TYPE_STICK has a value of 1, and the default radius value used for
attachment sticks is 1.0.
This implements MIOS's PPC bootstrapping functionality, which enables
users to start a GameCube game from the Wii System Menu.
Because we aren't doing Starlet LLE (and don't have a boot1), we can
just jump to MIOS when the emulated software does an ES_LAUNCH or uses
ioctlv 0x25 to launch BC.
Note that the process is more complex on a real Wii and goes through
several more steps before getting to MIOS:
* The System Menu detects a GameCube disc and launches BC (1-100)
instead of the game. [Dolphin does this too.]
* BC, which is reportedly very similar to boot1, lowers the Hollywood
clock speed to the Flipper's and then launches boot2.
* boot2 sees the lowered clock speed and launches MIOS (1-101) instead
of the System Menu.
MIOS runs instead of IOS in GC mode and has an embedded GC IPL (which
is the code actually responsible for loading the disc game) and a PPC
bootstrap code. To get things working properly, we simply need to load
both to memory, then jump to the bootstrap code at 0x3400.
Obviously, because of the way this works, a real MIOS is required.
It held a raw pointer to a IOS::HLE::Device::BluetoothEmu that is not
guaranteed to exist (and of course, nothing checked that it wasn't
nullptr), but what is more, it's totally unnecessary because we have
IOS::HLE::GetDeviceByName().
Since we cannot always inform the host that Wii remotes are
disconnected from ES, that is now done in BluetoothEmu's destructor.
Unless IOS failed at ES_Launch, it doesn't appear to write anything
back to the request after a launch, because the request is never
actually replied to in the normal way.
So let's just drop the writes to make things less confusing.
This ioctlv is used to launch BC. Not sure if that's useful,
since only the system menu is known to launch BC and it does that
through a regular ES_LAUNCH; but let's implement it anyway.
(Implementation based on IOS59.)
Some minor changes to make things slightly less confusing:
* Reinit doesn't actually init anything. It just adds static devices to
the map, so let's give it an actually descriptive name. And let's not
expose it in the header when it should not be.
* Reset's parameter name was changed from "force" -- which totally does
not describe what it does -- to "clear_devices".
* Add a reload function which handles the reload process properly
(reset all devices, set up memory values, re-add devices) and
without publicly exposing implementation details.
Shouldn't make any difference in practice because
both IsRecordingInput and IsNetPlayRunning should
be false if a temporary NAND isn't being set up,
but doing it this way is cleaner regardless.
Splits DVD reads up into smaller chunks so that data is available
before the final interrupt is triggered. This better simulates the DMA
that happens on a real device, which some games will take advantage of -
by either playing back data as it is loading or by using data that is
going to be overwritten shortly by an outstanding read.
Better separation of concerns. Relegates `ControllerInterface` to
enumerating input controls, and the new `ControlReference` deals with
combining inputs and configuration expression parsing.
ControllerEmu is a massive class with a lot of nested public classes.
The only reason these are nested is because the outer class acts as a
namespace. There's no reason to keep these classes nested just for that.
Keeping these classes nested makes it impossible to forward declare them, which leads to quite a few includes in other headers, making compilation take
longer.
This moves the source files to their own directory so classes can be
separated as necessary to their own source files, and be namespaced under the
ControllerEmu namespace.
To use it, with a modern LLVM (3.9+), set your CMAKE_PREFIX_PATH
to point to the LLVM install folder or to a LLVM build folder.
We're linking ALL of LLVM libs since I don't really know which ones we need.
LTO will take care of sliming the binary size...
The second output vector should not be written to for
IOCTLV_NCD_READCONFIG. If it is, the system menu will never attempt
to open /dev/net/wd/command and request a Wi-Fi scan.
Symbols map may not only end with a \n, but they may also end with \r\n and only the \n would get removed. This is the case with the Super Mario Sunshine map file which resulted in a weird looking symbols list and thus made it harder to scroll through it. This removes the \r after the \n has been removed if it's present.
Using the AVCodecContext contained in AVStream for muxing is officially
discouraged[1] and AVStream::codec was deprecated in favor of
AVStream::codecpar in libavformat 57.33.100 / 57.5.0.
1: [FFmpeg-cvslog] lavf: replace AVStream.codec with AVStream.codecpar: https://ffmpeg.org/pipermail/ffmpeg-cvslog/2016-April/099152.html
Any functions left exposed are used elsewhere through the main_window
global. May as well prevent any more functions from being used in that
manner where possible.
Utilizes the event system (which is what should have been done here
initially), in order to prevent coupling between two different window frames.
This also makes booting games more versatile using the UI event system,
as the event can just act as a carrier for the filename, making directly
calling boot functions unnecessary. All that's needed is for the event to
propagate to the frame.
Since files from Data/Sys are collected and added to a built macOS .app
bundle using GLOB, any new files won't get picked up until the next time
CMake is run. Tell CMake it should re-run itself every time the directory
is touched.
This prevents panic alerts from showing up three times when starting
Wii emulation whenever libusb could not be initialized. The user has
already seen a warning at startup -- no need to warn them 3 more times.
The built-in `configure_file` command correctly handles the case where
none of the variables change and scmrev.h doesn't need to be rebuilt.
This saves a full re-link of Dolphin any time CMake is re-run.
This reimplements the USB HID v4 IOS device using the new common
USB code (to reuse more code and allow emulated HIDs to be added
more easily in the future).
The main difference is that HIDs now have to be whitelisted, like
every other USB device for OH0 and VEN.
libusb on Windows is limited to only a single context. Trying to open
more than one can cause device enumerations to fail randomly.
libusb is thread-safe and we don't use the manual polling support (with
`poll()`) so this should be safe.
The Host class will be used by the OH0, VEN and HID implementations
as the base class for the IOS HLE device. It handles scanning devices,
detecting device changes and everything that will be needed for OH0,
VEN and HID to be implemented, while mostly abstracting libusb away.
The Device class is for actual USB devices. This commit adds a
LibusbDevice which interacts with a real USB device and enables
USB passthrough.
The NullAudio backend is guaranteed to be compiled in, so no reason
to check it.
In addition to that, if it wasn't valid, it wouldn't work as a fallback
in InitSoundStream as there are uses to g_sound_stream later.
Keeps associated data together. It also eliminates the possibility of out
parameters not being initialized properly. For example, consider the
following example:
-- some FramebufferManager implementation --
void FBMgrImpl::GetTargetSize(u32* width, u32* height) override
{
// Do nothing
}
-- somewhere else where the function is used --
u32 width, height;
framebuffer_manager_instance->GetTargetSize(&width, &height);
if (texture_width != width) <-- Uninitialized variable usage
{
...
}
It makes it much more obvious to spot any initialization issues, because
it requires something to be returned, as opposed to allowing an
implementation to just not do anything.
Movie basically just wants to get the title ID of
the initally booted game, so let's set the title ID in
ConfigManager at boot like we do with the regular game ID.
Aside from being cleaner, this should make the approach to
title IDs compatible with booting non-disc software (WADs).
This is unnecessary now that IOS::HLE is responsible for writing the
values to memory; removing the writes also prevents the IOS minor
version from being mangled (by the write to 0x3142).
Instead of using install() commands, we use the MACOSX_PACKAGE_LOCATION
property, which will allow the files to be identified and updated individually
by the build system without having to remove the entire folder and copy it
each time.
deploy-mac.py is now idempotent and should be working properly, so we'll
call it all the time from now on.
On macOS, we want them copied in the bundle directly, otherwise we will
install them later in the system folder.
Obviously not working for Windows, but that's not any different from before!
Forward declaring functions from a completely different header inside a cpp
file can lead to linker errors. Forward declaring also doesn't really
provide any benefit within cpp files unless it's to bring an internally
linked function within the same file into scope.
Hopefully will fix the crash in vkDestroyInstance on the NV Shield TV,
and likely reduce boot times slightly for drivers that take a while
to create instances.
Makes it more self-documenting which stack is being loaded or stored to,
as C, D, and magic numbers are extremely vague. It also enforces a
strongly-typed API instead of accepting arbitrary integral values.
It also adds the two other missing stack register names -- loop address
and loop counter.
This fixes a crash when trying to open the advanced input config dialog on the wiimote extensions. The device_cbox wasn't initialised and it should have been with the wiimote one.
These quite clearly have a dependency on the emitter itself, so these
should be a part of the emitter itself.
The template function can be modified to just simply take functions as a
parameter.
This function is exceptionally large. Everything within a switch like this
also makes it quite error prone. Separating the functions out makes it
easier to change a certain request implementation as well as improving
code locality.
This fixes savestates when using Bluetooth passthrough by keeping track
of pending transfer commands and discarding them on state load, so that
the emulated software receives a reply to IOS requests as expected.
With this change, savestates in BT passthrough should work as long as
no Wiimote is connected when creating the savestate and when
restoring it. Yes, I know this is an important limitation -- but
that is probably the best we can do, and it's still better than
completely broken savestates.
This fixes a regression from 5.0-1556, but I don't know why
the regression occurred or why this fixes it. (Many games
failed to fully boot - I tried Metroid Prime and Twilight
Princess (both GC), and they never got to the title screen.)
