Instead of clearing a hardcoded 16 bytes, adjust for the actual number
of instructions modified. The implementation will still only clear a
single cacheline so it doesn't change behaviour.
This used to exist in the FEXCore header since the unaligned handler was
done in the frontend. Once it got moved in to FEXCore it had stayed
there. Move it over now.
In the case of a visibility tear when one thread is backpatching while
another is executing. The executing thread can /potentially/ see the
writing of instructions depending on coherency rules or filling of
cachelines.
By ensuring the DMB instructions are backpatched over the NOP
instructions first, this ensures correct atomic visibility even on tear.
When code buffers are shared between threads, FEX needs to be careful
around backpatching its code buffers, since one thread might have
backpatched the code that another thread was also planning on
backpatching.
To handle this case, when the handler fails to find a backpatchable
instruction, check if it was already backpatched. This can be determined
by atomically reading the instructions back and seeing if they have
turned in to the non-atomic variants.
In most cases we can just return saying that it has been handled, in the
case of a store we need to back the PC up 4 bytes to ensure the DMB is
executed before the non-atomic store.
These handlers don't do any code backpatching so locking the spinlock
futex isn't necessary. Move them before the lock to make them a bit more
efficient once code buffers get shared.
A feature of FEX's JIT is that when an unaligned atomic load/store
operation occurs, the instructions will be backpatched in to a barrier
plus a non-atomic memory instruction. This is the half-barrier technique
that still ensures correct visibility of loadstores in an unaligned
context.
The problem with this approach is that the dmb instructions are HEAVY,
because they effectively stop the world until all memory operations in
flight are visible. But it is a necessary evil since unaligned atomics
aren't a thing on ARM processors. FEAT_LSE only gives you unaligned
atomics inside of a 16-byte granularity, which doesn't match x86
behaviour of cacheline size (effectively always 64B).
This adds a new TSO option to disable the half-barrier on unaligned
atomic and instead only convert it to a regular loadstore instruction,
ommiting the half-barrier. This gives more insight in to how well a
CPU's LRCPC implementation is by not stalling on DMB instructions when
possible.
Originally implemented as a test to see if this makes Sonic Adventure 2
run full speed with TSO enabled (but all available TSO options disabled)
on NVIDIA Orin. Unfortunately this basically makes the code no longer
stall on dmb instructions and instead just showing how bad the LRCPC
implementation is, since the stalls show up on `ldapur` instructions
instead.
Tested Sonic Adventure 2 on X13s and it ran at 60FPS there without the
hack anyway.
This is no longer necessary to be part of the public API. Moves the
header internally.
Needed to pass through `IsAddressInCodeBuffer` from CPUBackend through
the Context object, but otherwise no functional change.
Currently no functional change but public API breaks should come early.
The thread state object will be used for looking up thread specific
codebuffers in the future when we support MDWE with code mirrors.
It is not an external component, and it makes paths needlessly long.
Ryan seemed amenable to this when we discussed on IRC earlier.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Ryan Houdek