- 19 Oct, 2021 40 commits
-
-
Pavel Begunkov authored
[ 158.514382] WARNING: CPU: 5 PID: 15251 at fs/io_uring.c:1141 io_free_batch_list+0x269/0x360 [ 158.514426] RIP: 0010:io_free_batch_list+0x269/0x360 [ 158.514437] Call Trace: [ 158.514440] __io_submit_flush_completions+0xde/0x180 [ 158.514444] tctx_task_work+0x14a/0x220 [ 158.514447] task_work_run+0x64/0xa0 [ 158.514448] __do_sys_io_uring_enter+0x7c/0x970 [ 158.514450] __x64_sys_io_uring_enter+0x22/0x30 [ 158.514451] do_syscall_64+0x43/0x90 [ 158.514453] entry_SYSCALL_64_after_hwframe+0x44/0xae We should not touch request internals including req->comp_list.next after putting our ref if it's not final, e.g. we can start freeing requests from the free cache. Fixed: 62ca9cb93e7f8 ("io_uring: optimise io_free_batch_list()") Signed-off-by:Pavel Begunkov <asml.silence@gmail.com> Link: https://lore.kernel.org/r/b1f4df38fbb8f111f52911a02fd418d0283a4e6f.1634047298.git.asml.silence@gmail.comSigned-off-by:
Jens Axboe <axboe@kernel.dk>
-
Pavel Begunkov authored
task_work_add() takes care of waking up the thread, remove useless wake_up_process(). Signed-off-by:
-
Pavel Begunkov authored
Looking at the assembly, the compiler decided to reload req->opcode in io_op_defs[opcode].needs_file instead of one it had in a register, so store it in a temp variable so it can be optimised out. Also move the personality block later, it's better for spilling/etc. as it only depends on @sqe, which we're keeping anyway. By the way, zero req->opcode if it over IORING_OP_LAST, not a problem, at the moment but is safer. Signed-off-by:
-
Pavel Begunkov authored
struct io_submit_state's ->free_list and ->link are hotter and smaller than ->plug, place them first. Signed-off-by:
-
Pavel Begunkov authored
Add extra wq flushing for fallback_work, that's not necessary but safer if invariants of io_fallback_req_func() change. Signed-off-by:
-
Pavel Begunkov authored
Plugging is only needed with requests that also need a file, so hide plugging under a ->needs_file check. Also, place ->needs_file and ->plug bits into the same byte of io_op_defs, it may matter for compilers, e.g. only with the change a tested one decided to optimise two memory testb into a more with two register testb. Signed-off-by:
-
Pavel Begunkov authored
CQE result is a 32-bit integer, so the functions generating CQEs are better to accept not long but ints. Convert io_cqring_fill_event() and other helpers. Signed-off-by:
-
Pavel Begunkov authored
Inline io_poll_complete(), it's simple and doesn't have any particular purpose. Signed-off-by:
-
Pavel Begunkov authored
There is only a single user of io_req_needs_clean() inline it. Signed-off-by:
-
Pavel Begunkov authored
We keep struct io_completion only as a temporal storage of cflags, Place it in io_kiocb, it's cleaner, removes extra bits and even might be used for future optimisations. Signed-off-by:
-
Pavel Begunkov authored
->async_data is a slow path, so it won't matter much if we do the clean up inside io_clean_op(). Moreover, in many cases it's allocated together with setting one or more of IO_REQ_CLEAN_FLAGS flags, so it'd go through io_clean_op() anyway. Control ->async_data allocation with a new flag REQ_F_ASYNC_DATA, so we can do all the maintainence under io_req_needs_clean() fast check. Signed-off-by:
-
Pavel Begunkov authored
Delay reading the next node in io_free_batch_list(), allows the compiler to load the value a bit later improving register spilling in some cases. With gcc 11.1 it helped to move @task_refs variable from the stack to a register and optimises out a couple of per request instructions. Signed-off-by:
-
Pavel Begunkov authored
Attribute cold functions so compilers can optimise them for size. It shrinks the binary by 2.5-3% text data bss dec hex filename 90670 14002 8 104680 198e8 ./fs/io_uring.o 88053 14002 8 102063 18eaf ./fs/io_uring.o Signed-off-by:
-
Pavel Begunkov authored
Currenlty, we allocate one ctx reference per request at submission time and put them at free. It's batched and not so expensive but it still bloats the kernel, adds 2 function calls for rcu and adds some overhead for request counting in io_free_batch_list(). Always keep one reference with a request, even when it's freed and in io_uring request caches. There is extra work at ring exit / quiesce paths, which now need to put all cached requests. io_ring_exit_work() is already looping, so it's not a problem. Add hybrid-busy waiting to io_ctx_quiesce() as well for now. Signed-off-by:
-
Pavel Begunkov authored
->cq_wait and ->poll_wait and waken up in the same manner, use a single waitqueue for both of them. CQ waiters are queued exclusively, so wake up should first go over all pollers and that's what we need. Signed-off-by:
-
Pavel Begunkov authored
io_cqring_ev_posted() doesn't need to wake SQPOLL, it's either done by userspace or with task_work, but no action is required on request completion. Rip off bits waking it up in io_cqring_ev_posted(). Signed-off-by:
-
Pavel Begunkov authored
The invariant of io_wq_work_list is that it's empty IFF ->first is NULL, so no need to initially set ->last. With now having more users of the list it may play a role, i.e. used in each tw iteration and on every completion flushing. Signed-off-by:
-
Pavel Begunkov authored
Even after fully inlining io_alloc_req() my compiler does a NULL check in the path of successful allocation, no hacks like an empty dereference help it. Restructure io_alloc_req() by splitting out refilling part, so the compiler generate a slightly better binary. Signed-off-by:
-
Pavel Begunkov authored
io_req_complete_state() is inlined and used in lots of places, so we want to keep it concise. Move adding a request into a completion batch list from io_req_complete_state() into the consumer, i.e. __io_queue_sqe(). before vs after text data bss dec hex filename 91894 14002 8 105904 19db0 ./fs/io_uring.o 91046 14002 8 105056 19a60 ./fs/io_uring.o Signed-off-by:
-
Pavel Begunkov authored
Add two extra unlikely() in io_submit_sqes() and one around io_req_needs_clean() to help the compiler to avoid extra jumps in hot paths. Signed-off-by:
-
Pavel Begunkov authored
We want ->comp_list in the second cacheline, which is hotter comparing to the 3rd. Swap the field with ->link, which is not as hot and controlled by flags and so not accessed unless there is a link. By the way add a couple of comments for io_kiocb fields. Signed-off-by:
-
Pavel Begunkov authored
For some reason non-off IORING_OP_TIMEOUT always fails links, it's pretty inconvenient and unnecessary limits chaining after it to hard linking, which is far from ideal, e.g. doesn't pair well with timeout cancellation. Add a flag forcing it to not fail links on -ETIME. Signed-off-by:
-
Pavel Begunkov authored
Hiding a pointer to a struct io_buffer in rw.addr is error prone. We have some place in io_kiocb, so keep kbuf's in a separate field without aliasing and risks of it being misused. Signed-off-by:
-
Pavel Begunkov authored
Move io_req_prep() call inside of io_init_req(), it simplifies a bit error handling for callers. Signed-off-by:
-
Pavel Begunkov authored
Never return from io_drain_req() but punt to tw if we've got there but it's a false positive and we shouldn't actually drain. Signed-off-by:
-
Pavel Begunkov authored
Add a helper io_init_req_drain for initialising requests with IOSQE_DRAIN set. Also move bits from preambule of io_drain_req() in there, because we already modify all the bits needed inside the helper. Signed-off-by:
-
Pavel Begunkov authored
Clear ->drain_active in two more cases where we check for a need of draining. It's not a bug, but still may lead to some extra requests being punted to io-wq, and that may be not desirable. Signed-off-by:
-
Pavel Begunkov authored
io_req_complete_state() calls io_clean_op() and it may be not entirely obvious, leave a comment. Signed-off-by:
-
Pavel Begunkov authored
->inflight_entry is not used anymore after converting everything to single linked lists, remove it. Also adjust io_kiocb layout, so all hot bits are in first 3 cachelines. Signed-off-by:
-
Pavel Begunkov authored
Put an explicit check for number of requests to submit. First, we can turn while into do-while and it generates better code, and second that if can be cheaper, e.g. by using CPU flags after sub in io_sqring_entries(). Signed-off-by:
-
Pavel Begunkov authored
If a request completed inline the result should only be zero, it's a grave error otherwise. So, when we see REQ_F_COMPLETE_INLINE it's not even necessary to check the return code, and the flag check can be moved earlier. It's one "if" less for inline completions, and same two checks for it normally completing (ret == 0). Those are two cases we care about the most. Signed-off-by:
-
Pavel Begunkov authored
Extract slow paths from __io_queue_sqe() into a function and inline the hot path. With that we have everything completely inlined on the submission path up until io_issue_sqe(). -> io_submit_sqes() -> io_submit_sqe() (inlined) -> io_queue_sqe() (inlined) -> __io_queue_sqe() (inlined) -> io_issue_sqe() Signed-off-by: -
Pavel Begunkov authored
We don't want the slow path of io_queue_sqe to be inlined, so extract a function from it. text data bss dec hex filename 91950 13986 8 105944 19dd8 ./fs/io_uring.o 91758 13986 8 105752 19d18 ./fs/io_uring.o Signed-off-by:
-
Pavel Begunkov authored
req->ctx->active_drain is a bit too expensive, partially because of two dereferences. Do a trick, if we see it set in io_init_req(), set REQ_F_FORCE_ASYNC and it automatically goes through a slower path where we can catch it. It's nearly free to do in io_init_req() because there is already ->restricted check and it's in the same byte of a bitmask. Signed-off-by:
-
Pavel Begunkov authored
There are two call sites of io_queue_sqe() in io_submit_sqe(), combine them into one, because io_queue_sqe() is inline and we don't want to bloat binary, and will become even bigger text data bss dec hex filename 92126 13986 8 106120 19e88 ./fs/io_uring.o 91966 13986 8 105960 19de8 ./fs/io_uring.o Signed-off-by:
-
Pavel Begunkov authored
Submission state and ctx and coupled together, no need to passs Signed-off-by:
-
Pavel Begunkov authored
io_free_batch_list() iterates all requests in the passed in list, so we don't really need to know the tail but can keep iterating until meet NULL. Just pass the first node into it and it will be enough. Signed-off-by:
-
Pavel Begunkov authored
We now have a single function for batched put of requests, just inline struct req_batch and all related helpers into it. Signed-off-by:
-
Pavel Begunkov authored
First, convert rest of iopoll bits to single linked lists, and also replace per-request list_add_tail() with splicing a part of slist. With that, use io_free_batch_list() to put/free requests. The main advantage of it is that it's now the only user of struct req_batch and friends, and so they can be inlined. The main overhead there was per-request call to not-inlined io_req_free_batch(), which is expensive enough. Signed-off-by:
-
Pavel Begunkov authored
Convert explicit barrier around iopoll_completed to smp_load_acquire() and smp_store_release(). Similar on the callback side, but replaces a single smp_rmb() with per-request smp_load_acquire(), neither imply any extra CPU ordering for x86. Use READ_ONCE as usual where it doesn't matter. Use it to move filling CQEs by iopoll earlier, that will be necessary to avoid traversing the list one extra time in the future. Suggested-by:
Bart Van Assche <bvanassche@acm.org> Signed-off-by:
-
