Use the session data to store information about the HTTP session state.
This is based on reverse engineering of the HTTP module.
The AddRequestHeader function is still mostly incorrect, this will be fixed in follow up PRs
This allows LLE library applets (like swkbd) to properly close and return to the application instead of hanging.
There is still a bug in our rasterizer cache that may cause crashes some time after an applet is closed, but that is tangential to this change and should be tackled separately.
Right now only MappedBuffers that only span a single page and are not aligned are implemented.
MappedBuffers are unmapped during the reply part of ReplyAndReceive. Only unmapping of ReadOnly buffers is currently implemented.
Most applications call AcquireRight before calling RegisterInterruptRelayQueue so we can't assign the thread id there.
This fixes the bug with LLE applets not launching properly.
This is true for all interrupts except PDC0 and PDC1, which should be triggered for all registered threads.
TODO: The real GSP module seems to only trigger PDC0 after updating the screens (both top and bottom). PDC1 doesn't seem to be triggered at all.
The registered interrupt event is unique to each session that calls RegisterInterruptRelayQueue, and only that event should be reset when UnregisterInterruptRelayQueue is called.
The only functional change is the error handling of GSP_GPU::ReadHWRegs function. We previously didn't return error codes (not even for success). The new returns were found by reverse engineering the GSP module.
Terminating processes with ready threads is not currently implemented and will assert. It is currently unknown how the 3DS kernel stops ready threads or threads running in another core.
Some services can have multiple clients at the same time, and they identify the different clients using the server session as a key.
This parameter (if present) should be a structure that contains the per-session data for each service.
The data can be retrieved using ServiceFramework::GetSessionData(session)
The File class now holds a list of connected sessions along with data unique to each session.
A subfile is a window into an existing file. They have a few limitations compared to normal files:
* They can't be written to.
* They can't be flushed.
* Their size can not be changed.
* New subfiles can't be created from another subfile.
We don't yet implement the SEEDDB store.
This number is supposed to be modified by the AddSeed function (0x087A0180) but there's too little info about it on 3dbrew.
The FS module overrides whatever value was in the saveid high dword with 0x48000 when trying to open the archive.
This fixes the problem where the Home Menu would create a few SharedExtData archives with 0x48000 as the saveid high, but then try to open them with 0 as the high value and fail.
In a real 3DS, threads that call svcSyncRequest are put to sleep until the server responds via svcReplyAndReceive. Our HLE services don't implement this mechanism and are effectively immediate from the 3DS's point of view. This commit makes it so that we at least simulate the IPC delay.
Specific HLE handlers might need to put their callers to sleep for a longer period of time to simulate IO timings. This is their responsibility but doing so is currently not implemented.
See https://gist.github.com/ds84182/4a7690c5376e045cab9129ca4185bbeb for a test that was not passing before this commit.
An HLE service function that wants to perform an async operation should put the caller guest thread to sleep using SleepClientThread, passing in a callback to execute when the thread is resumed.
SleepClientThread returns a Kernel::Event that should be signaled to resume the guest thread when the host async operation completes.
This could happen if the guest application performs a request with static buffer id X, and the service module responds with another static buffer with the same id X.
This was incorrect behavior that somehow found its way to 3dbrew. The correct behavior is to sleep until the port becomes available again and then return a session to it.
This is currently unimplemented due to the inability to put a guest thread to sleep during HLE requests.
The correct behavior was reverse engineered by TuxSH a while ago but we never corrected the code in citra.
Kernel/Arbiters: When doing ArbitrateAddress(Signal), always pick the highest priority thread, using the first one that was put to sleep if more than one thread with the same highest priority exists.
The real kernel requires services to set up their static buffer targets ahead of time. This implementation does not require that and will simply create the storage for the buffers as they are processed in the incoming IPC request.
Static buffers are kept in an unordered_map keyed by their buffer id, and are written into the already-setup area of the request thread when responding an IPC request.
This fixes a regression (crash) introduced in #2992.
This PR introduces more warnings due to the [[deprecated]] attribute being added to void PushStaticBuffer(VAddr buffer_vaddr, size_t size, u8 buffer_id); and VAddr PopStaticBuffer(size_t* data_size);
The error code 0xC920181A will be returned by svcReplyAndReceive when the wakeup callback runs.
This lets LLE services be properly notified of clients closing the connection so they can end their handler threads instead of letting them linger indefinitely, taking up connection slots in their parent port.
With this commit, you can run native applets if they are in the correct folder of your virtual NAND.
Trying to exit the applet will currently cause an invalid read loop due to svcExitProcess not being implemented.
The table was taken from the real APT service, but is incomplete due to the sheer amount of data it contains. There's 29 applets with 7 possible titleids. This table should be filled as needed.
This descriptor requires the target process to set up a "receive buffer" beforehand, where the input data will be written to when the descriptor is processed.
Subv