IPC_HLE is actually IOS HLE. The actual IPC emulation is not in
IPC_HLE, but in HW/WII_IPC.cpp. So calling IPC_HLE IOS is more
accurate. (If IOS LLE gets ever implemented, it'll likely be at
a lower level -- Starlet LLE.)
This also totally gets rid of the IPC_HLE prefix in file names, and
moves some source files to their own subdirectories to make the file
hierarchy cleaner.
We're going to get ~14 additional source files with the USB PR,
and this is really needed to keep things from becoming a total pain.
Hiding and not implementing the copy constructor is a pre-C++11 thing.
It should also be noted that a copy constructor, as defined by the
language, contains a const qualifier on its parameter, so this wouldn't
have prevented copies from being performed.
Now that everything has been changed to use the new structs, the old
methods and structs can be removed.
And while I was changing the base device class, I also moved the
"unsupported command" code to a separate function. It was pretty silly
to copy the same 3 lines for ~5 commands.
This adds well-defined structs that are responsible for parsing
resource requests, instead of duplicating the logic and offsets all
over IOS HLE. Command handler functions are now passed parsed requests
instead of a command address.
This may not seem like a very important change, but it removes the
need to remember all of the struct offsets or copy/paste existing
struct request variables. It also prevents nasty bugs which have
occurred in the past, such as parsing an ioctl as if it were an ioctlv
(that's way too easy to do if you pass command addresses directly);
or writing something to 0x0, which can easily happen by mistake with
a close handler that can be called with invalid command addresses.
Bonus changes:
- The return code is not an obscure Memory::Write_U32 anymore, but an
explicit, more obvious SetReturnValue() call. (Which correctly takes
a s32 instead of a u32, since return codes are signed.)
- Open handlers are now only responsible for returning an IOS ret code,
and Close handlers don't return anything and don't have to worry
about checking whether the request is a real one anymore.
- DumpAsync was moved to the ioctlv request struct, because it did
not really make sense to make it part of the IOS device and it only
works for ioctlvs.
All current usages have been removed; they will be readded in a
later commit.
As of this commit, nothing uses the structs yet. Usages will be
migrated progressively.
Hiding and not implementing the copy constructor is a pre-C++11 thing.
It should also be noted that a copy constructor, as defined by the language,
contains a const qualifier on its parameter, so this wouldn't have
prevented copies from being performed.
It also follows that if the copy constructor is deleted, then copy
assignment should also be forbidden.
wxChoice controls don't display any titles.
By the way, why is the file called DebuggerPanel.cpp
even though it implements the Video debug panel specifically?
The current implementations do many things wrong but work well enough to
run the Dragon Quest X installer until the very end. The game itself
crashes when being launched from its System Menu channel unfortunately
so it is hard to verify whether the install properly worked or not.
There are plenty of "TODO(wfs)" sprinkled around this codebase with
things that are knowingly done wrong. The most important one right now
is that content extraction is done by buffering everything into memory
instead of properly streaming the data to disk (and processing
asynchronously), which causes freezes. It is likely to not cause any
practical issues since only the installer and the updater should use
this anyway.
Without this, attempts to savestate std::set will fail with an error
about dropping the const qualifier.
<Lioncash> leoetlino: I'll try to break it down: So, when you do a
ranged-for on a container, it's essentially syntactic sugar over begin
and end iterators. std::set is an associative container where the key
type is the same as the value type, and so it's required that all
iterator functions return constant iterators. If this wasn't a
requirement, it would allow changing the ordering of elements from
outside of the set's API (this is bad).
This attempts to make some bit arithmetic more self-documenting and also
make it easier during review to identify potential off-by-one errors by
making it possible to just specify which bits are being extracted.
Functions both support the case where bits being extracted can vary and
fixed bit extraction. In the case the bits are fixed, compile-time asserts
are present to prevent accidental API usage at compile-time.
e.g. Instead of shifting and masking to get bits 10 to 15,
Common::ExtractBits<10, 15>(value) can just be done instead.
There are several things wrong with this implementation. The first being
that since we still don't have a proper ticket/tmd handling library, we
hardcode offsets once again to fetch TMD fields. The second being that
we don't stream data to the disk and we buffer everything in memory. The
third being that we don't properly fetch the content index for
decryption, which is prone to breaking.
But hey, it works well enough to install the DQX channel!
This library implements basic parsing support for some of the IOS ES
formats we need to extract data from. Currently only implements TMD
functions, but some ticket handling functions from DiscIO should likely
be moved here in the future.
These two functions load either a signed ticket or a raw ticket from the
emulated NAND.
The ticket signature skip is refactored out of the ticket writing in
order to be usable by the raw ticket reading function.
Refactor the existing DiscIO::AddTicket to not require the caller to
pass the requested title ID. We do not have the title ID in the ES case,
and it needs to be extracted from the ticket. Since this is always a
safe operation to do (title ID is always in the ticket), the
implementation is made default.
This constant isn't particularly helpful, mainly because it's not
applicable to all DSP instructions. Some instructions don't have encoding
space for registers, and not all instructions that do encode registers
have one at the first five bits.
This change also has the benefit of removing all includes to the
interpreter within the JIT code, which keeps them fully separate from one
another. Changes to the interpreter header won't require some of the JIT
code to be rebuilt.
Dolphin is able to generate one with all correct default settings, so
we don't need to ship with a pre-generated SYSCONF and worry about
syncing default settings.
Additionally, this commit changes SysConf to work with session SYSCONFs
so that Dolphin is able to generate a default one even for Movie/TAS.
Which SYSCONF needs to be touched is explicitly specified to avoid
confusion about which file SysConf is managing.
(Another notable change is that the Wii root functions are moved into
Core to prevent Common from depending on Core.)
Reset():
We only need to close IOS devices which were opened, and we can do that
simply by iterating over s_fdmap and closing any opened device.
With this change, s_device_map can be cleared at once.
SetDefaultContentFile():
We can just use s_es_handles which is guaranteed to contain three valid
ES devices. Gets rid of a downcast.
Since the Open command won't ever return with the stub, there's no way
we will get a Close/IOCtl/IOCtlV for it, so we don't have to
implement it at all.
is_hardware is an obscure name (what does hardware mean?) and it forces
us to assume that anything that !is_hardware is a FileIO device. This
assumption prevents properly restoring OH0 child devices (which will be
implemented in the USB PR), so this commit replaces the is_hardware
bool with a device type.
Confirmed by a hardware test and a quick diassembly of /dev/es.
I'm not aware of anything that opens several ES handles, but
technically, this fixes a small inaccuracy in IOS HLE.
For IOCTL_STM_EVENTHOOK, IOS checks if there is already an event hook
to prevent overriding an existing event hook message with a new one,
without first releasing it.
Currently, the country is always overridden depending on the Wii
language setting. This makes it impossible to change the country
independently from the language, unlike on a Wii, since a language
will always be associated to an unique country (which is hardcoded
in Dolphin's codebase).
This behaviour was added in c881262 and changed in PR 4319 to happen
every time emulation is started. It doesn't seem to be needed anymore,
as a quick testing shows that a Japanese system menu is able to load
in Japanese even with the region set to France, or anything other than
Japan in fact. So what this commit changes is drop this code.
Fixes issue 9884.
Removes #defines which have been unused for years and cleans up
naming.
This also changes IPC_REP_ASYNC to simply IPC_REPLY because it turns
out it's actually not specific to async replies, but used for all
command replies.
- Use an enum instead of defines.
- Only use the FS_ prefix for return codes which are actually related
to FS stuff, not for everything.
- Remove duplicated error codes and clean up the names.
Size is internally stored as a size_t, so having an int parameter
would cause implicit sign-conversion from a signed value to an
unsigned value to occur.
This removes Open() and Close() functions from devices whenever they
did nothing more than the base class (setting m_Active, returning a
default reply).
Also, since IOS close commands practically always return FS_SUCCESS,
writing the return code is moved to HandleCommand() in WII_IPC_HLE,
which has two benefits: it's not duplicated all over the place
(so people will not forget it) and it gets rid of having to check
the force parameter, since HandleCommand() is always called for
real IOS commands, so command_address is guaranteed to be valid.
It looks like at some point Dolphin device IDs coincided with IOS file
descriptors, but this is not the case anymore, so keeping those
WriteReturnCode(GetDeviceID()) in every single IOS HLE device,
as if the device ID was used as IOS fd, is both unnecessary
and confusing.
Jan 04 22:55:01 <leoetlino> fwiw, it looks like there are new warnings in the RegCache code
Jan 04 22:55:04 <leoetlino> Source/Core/Core/PowerPC/Jit64/FPURegCache.cpp:13:33: warning: declaration shadows a variable in the global namespace [-Wshadow]
Jan 04 22:56:19 <@Lioncash> yeah, the jit global should have a g_ prefix.
This fixes shadowing warnings and adds the g_ prefix to a global.
Other settings options are nouns rather than verbs so this change makes the configuration option consistent with others. Also makes the menu option label the same as the windows title.
The constructor shouldn't be used as a dumping ground for all UI-related
initialization. Doing so makes it somewhat more difficult to reason about
how certain UI elements get created. It also puts unrelated identifiers in
the same scope.
This separates the UI creation out so code relevant to each component
is self-contained.
We don't really have to keep track of device opens/closes manually,
since we can already check that by calling IsOpened() on the device.
This also replaces some loops with for range loops.
Certain parts of the standard library try to determine whether or not a
transfer operation should either be a copy or a move. The prevalent notion
of move constructors/assignment operators is that they should not throw,
they simply move an already existing resource somewhere else.
This is typically done with 'std::move_if_noexcept'. Like the name says,
if a type's move constructor is noexcept, then the functions retrieves an
r-value reference (for move semantics), or an l-value (for copy semantics)
if it is not noexcept.
As IOFile deletes the copy constructor and copy assignment operators,
using IOFile with certain parts of the standard library can fail in
unexcepted ways (especially when used with various container
implementations). This prevents that.
'operator void*' is basically a pre-C++11-ism that was used, as C++03
only had the notion of implicit type-conversion operators, but not explicit type
conversion operators (allowing implicit conversion of a file handle to
bool can go downhill pretty quickly).
ExecuteCommand was becoming pretty confusing with unused variables
for some commands, confusing names (device ID != IOS file descriptor),
duplicated checks, not keeping the indentation level low, and having
tons of things into a single function.
This commit gives more correct names to variables, deduplicates the
device checking code, and splits ExecuteCommand so that it's
easier to read.
It's worth noting that some device checks have been forgotten in the
past, which has caused a bug (which was recently fixed in 288e75f6).
GCMemcard is only used outside of the core and has
no real reason to use the core's RTC. GetEmulatedTime
isn't designed to be called when the core isn't running.
Should fix https://bugs.dolphin-emu.org/issues/9871
From what I can tell, the emulated GPU places (0,0) at the lower left of
the image, and we were generating texture coordinates so that (0,0) was
at the upper-left in the expansion geometry shader, causing textures
used by point sprites to be flipped vertically.
Fixes the upside-down A button in Mario Golf.
Since in this case we're setting it based on the state at record start
time, not when a register is loaded, UseMemory would not be called, so
this could potentially wipe out texture memory that was valid.
This should ensure that when playing with loop enabled, the first frame is
in the same state each time. There is potentially still issues when the
start frame is set to something other than zero, but I'm not sure how we
could work around this without capturing the entire state on each frame.
Doing it from the add dialogs instead would prevent the call to these dialogs outside of a breakpointWindow which would be necessary for hotkeys binding.
Instead of needing different switch cases for
converting countries to regions in multiple places,
we now only need a single country-to-region switch case
(in DiscIO/Enums.cpp), and we get a nice Region type.
This also makes it a strongly-typed enum.
Considering that the flushing mode is a trait/behavior for the register
cache, it doesn't really make sense to have the enum separate from it.
This also has the benefit of removing constants from global scope.
This is done to not have the device combo box be too small in width when making the main sizer fit into the window. Not fitting the sizer would alternatively break Hidpi so it was best to just add an empty sizer to workaround this problem.
I think it's best to remove these if we are going to be adding new hotkeys since these would work no matter what so I can simply make these the default one instead.
This was kind of a pointless function, considering the parameter wasn't
used at all, so the other Flush() function could have been just directly
used instead.
Windows-1252 was sometimes being referred to as ASCII or ANSI
in Dolphin, which is incorrect. ASCII is only a subset of
Windows-1252, and ANSI is (rather improperly) used in Windows
to refer to the current code page (which often is 1252 on
Western systems, but can also be something entirely different).
The commit also replaces "SJIS" with "Shift JIS". "SJIS"
isn't misleading, but "Shift JIS" is more commonly used.
GetName() creates a new evdev device which calls tons of ioctls. But the
main culprit is close() which for input devices appears to be a slow
path in the kernel.
This commit reduces PopulateDevices() by 50% on my laptop, but ~730 ms
is still ridiculously slow for something that isn't needed right away.
We can't rely on the OS returning files and directories
in a deterministic order, so we should sort them on our own
if we want VolumeDirectory to work for movies and netplay.
This fixes a bug which caused Movie (input recording or playback) or
netplay not to be stopped. DolphinWX previously triggered a STM power
event, and then the STM directly stopped the emulation; the code
which stops Movie/Netplay was completely skipped.
This is fixed by moving it /before/ sending the shutdown event.
This allows removing DSPCore and DSPTables includes from the header file.
Doing allows resolving quite a bit of indirect includes that were present
throughout the DSP source files.
Another plus with this is that changes to the DSPEmitter don't require an
almost total rebuild of all DSP source files. The underlying reason for
most of the files being rebuilt it because DSPMemoryMap is used quite
extensively, however its header includes DSPTables.h. DSPTables.h includes
DSPEmitter.h as it uses the DSPEmitter type in a typedef. So any change to
the emitter would propagate through the DSPMemoryMap header. This will no
longer happen.
This is actually used as the DSP JIT, so this should be with the other JIT
source files.
This commit also makes it so changes to the JIT emitter don't require
recompiling all of the DSP core (i.e. changing the JIT won't require the
interpreter to be rebuilt).
Jit64 inherits from Jitx86Base which inherits from JitBase. JitBase
contains jo and js, which are instances of the JitOptions and JitState
structs. Because of the inheritance, there's no actual need to access the
jit global in order to get to these instances. They're already accessible
via the class hierarchy.
This happened when the geometry shader was disabled, and the uniform
buffer was grown to a larger size. The update would be skipped, leaving
the old buffer to be included in the descriptor set.
OnBootDrive used the "drives" member std::vector for drive paths, but
since PR #4363, this vector is not populated anymore, so we were
accessing it out of bounds.
Actually, drives was not needed in the first place, since we can
get the wxMenu from the event, and from there, get the label directly.
Much of Jit64Util consists of essentials, not utilities. Breaking these
out into their own files also prevents unrelated includes from being
present near other classes.
This also makes it easier to find and change certain components of the
x86-64 JIT, should it be necessary.
These provide the same semantics, however aggregate initialization
doesn't force the structs to be trivially copyable. memset, on the other
hand, does.
JUTWarningConsole_f calls vprintf, but in a way we currently don't
handle (which messes up the printed message). However, it is a standard
debug print function, so we can directly hook it instead of waiting for
the vprintf call.
This is necessary to fix debug output in a few games now that vprintf
is properly detected in more games.
This is more logical as the mic is plugged into an EXI slot so it should be configured via the GameCube config dialog. This also allows to pass the right port number for the new dialog.
This also moves the pipeline and descriptor set layouts used for texture
conversion (texel buffers) to ObjectCache, and shares a binding location
with the SSBO set.
Makes the information panel self-contained.
This was done first, as opposed to isolating the GameConfig panel--the
first panel in the group--as this panel had code all over the place in
ISOProperties, so I figured it'd be best to fix this one up first.
These disc images are encrypted and signed using a different set of keys.
We only care about the master key, so we check the signature issuer. If
it matches the debug issuer, then we'll use the RVT-R key. Otherwise,
the previous set of common and Korean keys are used.
The SDL backend crashes when you close a joystick after SDL_Quit has
been called. Some backends don't need to be shutdown and
re-initialized everytime, we can just ask to enumerate devices again.
It's questionable whether ES_LAUNCH should write *anything* to the
command structure at all, ever, given that it never actually returns
it back through the mailbox. But it *definitely* shouldn't write
anything to it if it has just launched a DOL, because otherwise it might
clobber code/data from the just-loaded application.
When booting "cooked" executables, BATs should already be set up and
enabled. They should only really be disabled when booting NAND contents
in real mode.
Should we ever introduce anything else that has to be done when a command
buffer is executed (e.g. invalidating constants from previous commit), we
don't have to update all the callers.
This could happen because the vertex memory was already committed, if a
uniform buffer allocation failed and caused a command buffer to be
executed, it would be associated with the previous command buffer rather
than the buffer containing the draw that consumed these vertices.
This was happening when a fence wait happened mid-frame. The data written
between the fence being queued and the allocation occuring was incorrectly
assumed to be consumed by the GPU.
Léo Lam