- 04 May, 2023 2 commits
-
-
Breno Leitao authored
Currently uring CMD operation relies on having large SQEs, but future operations might want to use normal SQE. The io_uring_cmd currently only saves the payload (cmd) part of the SQE, but, for commands that use normal SQE size, it might be necessary to access the initial SQE fields outside of the payload/cmd block. So, saves the whole SQE other than just the pdu. This changes slightly how the io_uring_cmd works, since the cmd structures and callbacks are not opaque to io_uring anymore. I.e, the callbacks can look at the SQE entries, not only, in the cmd structure. The main advantage is that we don't need to create custom structures for simple commands. Creates io_uring_sqe_cmd() that returns the cmd private data as a null pointer and avoids casting in the callee side. Also, make most of ublk_drv's sqe->cmd priv structure into const, and use io_uring_sqe_cmd() to get the private structure, removing the unwanted cast. (There is one case where the cast is still needed since the header->{len,addr} is updated in the private structure) Suggested-by:Pavel Begunkov <asml.silence@gmail.com> Signed-off-by:
Breno Leitao <leitao@debian.org> Reviewed-by:
Keith Busch <kbusch@kernel.org> Reviewed-by:
Christoph Hellwig <hch@lst.de> Reviewed-by:
Jens Axboe <axboe@kernel.dk>
-
Breno Leitao authored
Create a simple helper that returns the size of the SQE. The SQE could have two size, depending of the flags. If IO_URING_SETUP_SQE128 flag is set, then return a double SQE, otherwise returns the sizeof of io_uring_sqe (64 bytes). Signed-off-by:
-
- 03 May, 2023 1 commit
-
-
Tobias Holl authored
Pages that are from the same folio do not necessarily need to be consecutive. In that case, we cannot consolidate them into a single bvec entry. Before applying the huge page optimization from commit 57bebf80 ("io_uring/rsrc: optimise registered huge pages"), check that the memory is actually consecutive. Cc: stable@vger.kernel.org Fixes: 57bebf80 ("io_uring/rsrc: optimise registered huge pages") Signed-off-by:
Tobias Holl <tobias@tholl.xyz> [axboe: formatting] Signed-off-by:
-
- 20 Apr, 2023 1 commit
-
-
Jens Axboe authored
This reverts commit edd47826. There's really no specific need to disallow multiple sources of buffers, and io_uring really should not be mandating this by itself. We should be able to solely rely on GUP making these decisions. As this also stands in the way of a cleanup where io_uring is the odd one out, kill it. Link: https://lore.kernel.org/all/61ded378-51a8-1dcb-b631-fda1903248a9@gmail.com/Signed-off-by:
-
- 19 Apr, 2023 9 commits
-
-
David Wei authored
A multishot timeout submission will repeatedly generate completions with the IORING_CQE_F_MORE cflag set. Depending on the value of the `off' field in the submission, these timeouts can either repeat indefinitely until cancelled (`off' = 0) or for a fixed number of times (`off' > 0). Only noseq timeouts (i.e. not dependent on the number of I/O completions) are supported. An indefinite timer will be cancelled if the CQ ever overflows. Signed-off-by:
David Wei <davidhwei@meta.com> Link: https://lore.kernel.org/r/20230418225817.1905027-1-davidhwei@meta.comSigned-off-by:
-
Pavel Begunkov authored
Make rsrc nodes independent from rsrd_data, for that we keep ctx and rsrc type in nodes. Signed-off-by:
-
Pavel Begunkov authored
We only have two rsrc types, buffers and files, replace virtual callbacks for putting resources down with a switch..case. Signed-off-by:
-
Pavel Begunkov authored
Instead of passing rsrc_data and a resource to io_rsrc_put_work() just forward node, that's all the function needs. Signed-off-by:
-
Pavel Begunkov authored
io_rsrc_put_work() is simple enough to be open coded into its only caller. Signed-off-by:
-
Pavel Begunkov authored
Unless a node was flushed by io_rsrc_ref_quiesce(), it'll carry a resource. Replace ->inline_items with an empty flag, which is initialised to false and only raised in io_rsrc_ref_quiesce(). Signed-off-by:
-
Pavel Begunkov authored
struct io_rsrc_node carries a number of resources represented by struct io_rsrc_put. That was handy before for sync overhead ammortisation, but all complexity is gone and nodes are simple and lightweight. Let's allocate a separate node for each resource. Nodes and io_rsrc_put and not much different in size, and former are cached, so node allocation should work better. That also removes some overhead for nested iteration in io_rsrc_node_ref_zero() / __io_rsrc_put_work(). Another reason for the patch is that it greatly reduces complexity by moving io_rsrc_node_switch[_start]() inside io_queue_rsrc_removal(), so users don't have to care about it. Signed-off-by:
-
Pavel Begunkov authored
For io_queue_rsrc_removal() we should always use the current active rsrc node, don't pass it directly but let the function grab it from the context. Signed-off-by:
-
Pavel Begunkov authored
->llist was needed for rsrc node destruction offload, which is removed now. Get rid of the unused field. Signed-off-by:
-
- 15 Apr, 2023 11 commits
-
-
Pavel Begunkov authored
We can queue up a rsrc into a list in io_queue_rsrc_removal() while allocating io_rsrc_put and so simplify the function. Signed-off-by:
-
Pavel Begunkov authored
At maximum io_install_fixed_file() removes only one file, so no need to keep needs_switch state and we can call io_rsrc_node_switch() right after removal. Signed-off-by:
-
Pavel Begunkov authored
Inline offset variable, so we don't use it without subjecting it to array_index_nospec() first. Signed-off-by:
-
Pavel Begunkov authored
Inline the part of io_rsrc_node_switch_start() that checks whether the cache is empty or not, as most of the times it will have some number of entries in there. Signed-off-by:
-
Pavel Begunkov authored
Instead of waiting for rsrc_data->refs to be downed to zero, check whether there are rsrc nodes queued for completion, that's easier then maintaining references. Signed-off-by:
-
Pavel Begunkov authored
For io_rsrc_ref_quiesce() to progress it should execute all task_work items, including deferred ones. However, currently nobody would wake us, and so let's set ctx->cq_wait_nr, so io_req_local_work_add() would wake us up. Fixes: c0e0d6ba ("io_uring: add IORING_SETUP_DEFER_TASKRUN") Signed-off-by:
-
Pavel Begunkov authored
Replace completions with waitqueues for rsrc data quiesce, the main wakeup condition is when data refs hit zero. Note that data refs are only changes under ->uring_lock, so we prepare before mutex_unlock() reacquire it after taking the lock back. This change will be needed in the next patch. Signed-off-by:
-
Pavel Begunkov authored
Refactor io_rsrc_ref_quiesce() by moving the first mutex_unlock(), so we don't have to have a second mutex_unlock() further in the loop. It prepares us to the next patch. Signed-off-by:
-
Pavel Begunkov authored
Kill io_rsrc_node::node and check refs instead, it's set when the nodes refcount hits zero, and it won't change afterwards. Signed-off-by:
-
Pavel Begunkov authored
We use array_index_nospec() for registered buffer indexes, but don't use it while poking into rsrc tags, fix that. Fixes: 634d00df ("io_uring: add full-fledged dynamic buffers support") Signed-off-by:
-
Pavel Begunkov authored
Add a constant IO_NOTIF_UBUF_FLAGS for struct ubuf_info flags that notifications use. That should minimise merge conflicts for planned changes touching both io_uring and net at the same time. Signed-off-by:
-
- 12 Apr, 2023 9 commits
-
-
Pavel Begunkov authored
SCM file accounting is a slow path and is only used for UNIX files. Extract a helper out of io_rsrc_file_put() that does the SCM unaccounting. Signed-off-by:
-
Pavel Begunkov authored
We use io_rsrc_node_switch() coupled with io_rsrc_node_switch_start() for a bunch of cases including initialising ctx->rsrc_node, i.e. by passing NULL instead of rsrc_data. Leave it to only deal with actual node changing. For that, first remove it from io_uring_create() and add a function allocating the first node. Then also remove all calls to io_rsrc_node_switch() from files/buffers register as we already have a node installed and it does essentially nothing. Signed-off-by:
-
Pavel Begunkov authored
struct io_rsrc_node::rsrc_data field is initialised on rsrc removal and shouldn't be used before that, still let's play safe and zero the field on alloc. Signed-off-by:
-
Pavel Begunkov authored
We store one pre-allocated rsrc node in ->rsrc_backup_node, merge it with ->rsrc_node_cache. Signed-off-by:
-
Pavel Begunkov authored
Add a lockdep chek to make sure that file and buffer updates hold ->uring_lock. Signed-off-by:
-
Pavel Begunkov authored
We don't post CQEs from the IRQ context, add a check catching that. Signed-off-by:
-
Pavel Begunkov authored
The kernel test robot complains about __io_remove_buffers(). io_uring/kbuf.c:221 __io_remove_buffers() warn: variable dereferenced before check 'bl->buf_ring' (see line 219) That check is not needed as ->buf_ring will always be set, so we can remove it and so silence the warning. Signed-off-by:
-
Pavel Begunkov authored
io_uring/io_uring.c:432 io_prep_async_work() error: we previously assumed 'req->file' could be null (see line 425). Even though it's a false positive as there will not be REQ_F_ISREG set without a file, let's add a simple check to make the kernel test robot happy. We don't care about performance here, but assumingly it'll be optimised out by the compiler. Signed-off-by:
-
Jens Axboe authored
We know now what the completion context is for the uring_cmd completion handling, so use that to have io_req_task_complete() decide what the best way to complete the request is. This allows batching of the posted completions if we have multiple pending, rather than always doing them one-by-one. Signed-off-by:
-
- 06 Apr, 2023 7 commits
-
-
Pavel Begunkov authored
Chains of memory accesses are never good for performance. The req->task->io_uring->in_cancel in io_req_local_work_add() is there so that when a task is exiting via io_uring_try_cancel_requests() and starts waiting for completions, it gets woken up by every new task_work item queued. Do a little trick by announcing waiting in io_uring_try_cancel_requests(), making io_req_local_work_add() wake us up. We also need to check for deferred tw items after prepare_to_wait(TASK_INTERRUPTIBLE); Signed-off-by:
-
Pavel Begunkov authored
Separate ->task_complete path in __io_cq_unlock_post_flush(). Signed-off-by:
-
Pavel Begunkov authored
Every task_work will try to wake the task to be executed, which causes excessive scheduling and additional overhead. For some tw it's justified, but others won't do much but post a single CQE. When a task waits for multiple cqes, every such task_work will wake it up. Instead, the task may give a hint about how many cqes it waits for, io_req_local_work_add() will compare against it and skip wake ups if #cqes + #tw is not enough to satisfy the waiting condition. Task_work that uses the optimisation should be simple enough and never post more than one CQE. It's also ignored for non DEFER_TASKRUN rings. Signed-off-by:
-
Pavel Begunkov authored
We'll need to grab some information from the previous request in the tw list, inline llist_add(), it'll be used in the following patch. Signed-off-by:
-
Pavel Begunkov authored
We pass 'allow_local' into io_req_task_work_add() but will need more flags. Replace it with a flags bit field and name this allow_local flag. Signed-off-by:
-
Pavel Begunkov authored
Instead of smp_mb() + __io_cqring_wake() in __io_cq_unlock_post_flush() use equivalent io_cqring_wake(). With that we can clean it up further and remove __io_cqring_wake(). Signed-off-by:
-
Pavel Begunkov authored
We currently pin the ctx for io_req_local_work_add() with percpu_ref_get/put, which implies two rcu_read_lock/unlock pairs and some extra overhead on top in the fast path. Replace it with a pure rcu read and let io_ring_exit_work() synchronise against it. Signed-off-by:
-
