EXI: Refactor Windows BBA-TAP interface to a read thread, crash fixes, cleanups #3432
Conversation
stenzek
force-pushed
the
bba-tap-win
branch
2 times, most recently
from
be8aa54
to
ad2ff21
Compare
|
I need to retest this; I first thought I couldn't connect to tap, but then I realized I somehow broke my taps. I'll have to reinstall I guess. |
|
This fixes the constant crashing on my desktop pc. My laptop that had no problems before works too. Tested Kirby Air Ride and everything was working fine. 1080 Avalanche still disconnects when selecting a course, but it always did that. MK:DD is next on my list to test. |
| @@ -27,7 +28,7 @@ void GenerateMacAddress(const MACConsumer type, u8* mac) | |||
| break; | |||
| } | |||
|
|
|||
| srand((unsigned int)time(nullptr)); | |||
| srand(Common::Timer::GetTimeMs()); | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
|
Regarding the casting/buffer issue, I've changed it slightly, it allocates a single struct with an inline buffer for small frames, for larger frames it'll do a second allocation for the buffer. So no casting to/from u8, just to the write class to call the destructor, and not needing a second allocation at all in most cases. Regarding the random MAC change, the only disadvantage to this that I see is if there are any games that still work online (I saw the list on the wiki, but are the servers still online for these games?), and you're bridging to LAN, you'll be creating DHCP leases each time you boot the emulator. I suppose another way around this could be replacing the last byte with the index of the adapter chosen or something, but this would require more refactoring, as the MAC address is set at boot, not reset time, which is when the adapter is actually opened. |
| static const size_t SMALL_FRAME_BUFFER_SIZE = 200; | ||
| struct AsyncIOWritePacket final : public AsyncIOHeader | ||
| { | ||
| u8 small_frame_buffer[SMALL_FRAME_BUFFER_SIZE]; |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
stenzek
changed the title
This also has the added benefit of not crashing under most circumstances. Includes a few other changes, including replacing the atomic<bool> with a Flag, as well as adding a flag for indicating read thread shutdown.
|
Requesting a merge on this: it fixes a ton of issues/crashes and makes BBA actually useable for Phantasy Star Online on Windows. |
EXI: Refactor Windows BBA-TAP interface to a read thread, crash fixes, cleanups
The old interface apparently had performance issues, and was prone to crashes (one of them due to the fact that writes to the tap interface were done asynchronously, using an OVERLAPPED structure on the stack, which obviously does not exist once the method returned)
I don't have many ways of testing this, I've tried with sockettest.dol, and this works okay. Keep in mind, some people have observed that this is slower to get an address on Windows over Linux.
I believe this is correct behavior, because Windows runs spanning tree on its bridge interface, dropping all packets for 20 seconds or so (L2 loop protection). Apparently there's a registry tweak to disable STP, but I had no luck with that as I only got the bridge to work once (seems it's flaky in W10, or my install is hosed). Running NAT on the interface instead brings the time down to around 4-5 seconds (comparable to Linux?)