Index: libcfa/src/concurrency/io.cfa =================================================================== --- libcfa/src/concurrency/io.cfa (revision 4c925cd31e0a0d5ba86584befde68164c77404b8) +++ libcfa/src/concurrency/io.cfa (revision 49981556d70010c36f072dc3534ff01415259a87) @@ -93,5 +93,5 @@ //============================================================================================= static unsigned __collect_submitions( struct __io_data & ring ); - static uint32_t __release_consumed_submission( struct __io_data & ring ); + static __u32 __release_consumed_submission( struct __io_data & ring ); static inline void process(struct io_uring_cqe & cqe ) { @@ -100,5 +100,5 @@ data->result = cqe.res; - unpark( data->thrd __cfaabi_dbg_ctx2 ); + post( data->sem ); } @@ -136,5 +136,5 @@ unsigned head = *ring.completion_q.head; unsigned tail = *ring.completion_q.tail; - const uint32_t mask = *ring.completion_q.mask; + const __u32 mask = *ring.completion_q.mask; // Nothing was new return 0 @@ -143,5 +143,5 @@ } - uint32_t count = tail - head; + __u32 count = tail - head; /* paranoid */ verify( count != 0 ); for(i; count) { @@ -224,5 +224,5 @@ // - [* struct io_uring_sqe, uint32_t] __submit_alloc( struct __io_data & ring, uint64_t data ) { + [* struct io_uring_sqe, __u32] __submit_alloc( struct __io_data & ring, __u64 data ) { /* paranoid */ verify( data != 0 ); @@ -230,9 +230,9 @@ __attribute((unused)) int len = 0; __attribute((unused)) int block = 0; - uint32_t cnt = *ring.submit_q.num; - uint32_t mask = *ring.submit_q.mask; + __u32 cnt = *ring.submit_q.num; + __u32 mask = *ring.submit_q.mask; disable_interrupts(); - uint32_t off = __tls_rand(); + __u32 off = __tls_rand(); enable_interrupts( __cfaabi_dbg_ctx ); @@ -241,8 +241,8 @@ // Look through the list starting at some offset for(i; cnt) { - uint64_t expected = 0; - uint32_t idx = (i + off) & mask; + __u64 expected = 0; + __u32 idx = (i + off) & mask; struct io_uring_sqe * sqe = &ring.submit_q.sqes[idx]; - volatile uint64_t * udata = (volatile uint64_t *)&sqe->user_data; + volatile __u64 * udata = &sqe->user_data; if( *udata == expected && @@ -270,5 +270,5 @@ } - static inline uint32_t __submit_to_ready_array( struct __io_data & ring, uint32_t idx, const uint32_t mask ) { + static inline __u32 __submit_to_ready_array( struct __io_data & ring, __u32 idx, const __u32 mask ) { /* paranoid */ verify( idx <= mask ); /* paranoid */ verify( idx != -1ul32 ); @@ -277,15 +277,15 @@ __attribute((unused)) int len = 0; __attribute((unused)) int block = 0; - uint32_t ready_mask = ring.submit_q.ready_cnt - 1; + __u32 ready_mask = ring.submit_q.ready_cnt - 1; disable_interrupts(); - uint32_t off = __tls_rand(); + __u32 off = __tls_rand(); enable_interrupts( __cfaabi_dbg_ctx ); - uint32_t picked; + __u32 picked; LOOKING: for() { for(i; ring.submit_q.ready_cnt) { picked = (i + off) & ready_mask; - uint32_t expected = -1ul32; + __u32 expected = -1ul32; if( __atomic_compare_exchange_n( &ring.submit_q.ready[picked], &expected, idx, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) ) { break LOOKING; @@ -316,9 +316,9 @@ } - void __submit( struct io_context * ctx, uint32_t idx ) __attribute__((nonnull (1))) { + void __submit( struct io_context * ctx, __u32 idx ) __attribute__((nonnull (1))) { __io_data & ring = *ctx->thrd.ring; // Get now the data we definetely need - volatile uint32_t * const tail = ring.submit_q.tail; - const uint32_t mask = *ring.submit_q.mask; + volatile __u32 * const tail = ring.submit_q.tail; + const __u32 mask = *ring.submit_q.mask; // There are 2 submission schemes, check which one we are using @@ -332,5 +332,5 @@ } else if( ring.eager_submits ) { - uint32_t picked = __submit_to_ready_array( ring, idx, mask ); + __u32 picked = __submit_to_ready_array( ring, idx, mask ); for() { @@ -429,11 +429,11 @@ unsigned to_submit = 0; - uint32_t tail = *ring.submit_q.tail; - const uint32_t mask = *ring.submit_q.mask; + __u32 tail = *ring.submit_q.tail; + const __u32 mask = *ring.submit_q.mask; // Go through the list of ready submissions for( i; ring.submit_q.ready_cnt ) { // replace any submission with the sentinel, to consume it. - uint32_t idx = __atomic_exchange_n( &ring.submit_q.ready[i], -1ul32, __ATOMIC_RELAXED); + __u32 idx = __atomic_exchange_n( &ring.submit_q.ready[i], -1ul32, __ATOMIC_RELAXED); // If it was already the sentinel, then we are done @@ -451,16 +451,16 @@ } - static uint32_t __release_consumed_submission( struct __io_data & ring ) { - const uint32_t smask = *ring.submit_q.mask; + static __u32 __release_consumed_submission( struct __io_data & ring ) { + const __u32 smask = *ring.submit_q.mask; if( !try_lock(ring.submit_q.release_lock __cfaabi_dbg_ctx2) ) return 0; - uint32_t chead = *ring.submit_q.head; - uint32_t phead = ring.submit_q.prev_head; + __u32 chead = *ring.submit_q.head; + __u32 phead = ring.submit_q.prev_head; ring.submit_q.prev_head = chead; unlock(ring.submit_q.release_lock); - uint32_t count = chead - phead; + __u32 count = chead - phead; for( i; count ) { - uint32_t idx = ring.submit_q.array[ (phead + i) & smask ]; + __u32 idx = ring.submit_q.array[ (phead + i) & smask ]; ring.submit_q.sqes[ idx ].user_data = 0; } Index: libcfa/src/concurrency/io/setup.cfa =================================================================== --- libcfa/src/concurrency/io/setup.cfa (revision 4c925cd31e0a0d5ba86584befde68164c77404b8) +++ libcfa/src/concurrency/io/setup.cfa (revision 49981556d70010c36f072dc3534ff01415259a87) @@ -298,5 +298,5 @@ if( params_in.poll_complete ) params.flags |= IORING_SETUP_IOPOLL; - uint32_t nentries = params_in.num_entries != 0 ? params_in.num_entries : 256; + __u32 nentries = params_in.num_entries != 0 ? params_in.num_entries : 256; if( !is_pow2(nentries) ) { abort("ERROR: I/O setup 'num_entries' must be a power of 2\n"); @@ -362,15 +362,15 @@ // Get the pointers from the kernel to fill the structure // submit queue - sq.head = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.head); - sq.tail = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail); - sq.mask = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask); - sq.num = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries); - sq.flags = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags); - sq.dropped = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped); - sq.array = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.array); + sq.head = (volatile __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.head); + sq.tail = (volatile __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail); + sq.mask = ( const __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask); + sq.num = ( const __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries); + sq.flags = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags); + sq.dropped = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped); + sq.array = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.array); sq.prev_head = *sq.head; { - const uint32_t num = *sq.num; + const __u32 num = *sq.num; for( i; num ) { sq.sqes[i].user_data = 0ul64; @@ -395,10 +395,10 @@ // completion queue - cq.head = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.head); - cq.tail = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail); - cq.mask = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask); - cq.num = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries); - cq.overflow = ( uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow); - cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes); + cq.head = (volatile __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.head); + cq.tail = (volatile __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail); + cq.mask = ( const __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask); + cq.num = ( const __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries); + cq.overflow = ( __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow); + cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes); // some paranoid checks @@ -448,5 +448,5 @@ void __ioctx_register($io_ctx_thread & ctx, struct epoll_event & ev) { ev.events = EPOLLIN | EPOLLONESHOT; - ev.data.u64 = (uint64_t)&ctx; + ev.data.u64 = (__u64)&ctx; int ret = epoll_ctl(iopoll.epollfd, EPOLL_CTL_ADD, ctx.ring->fd, &ev); if (ret < 0) { Index: libcfa/src/concurrency/io/types.hfa =================================================================== --- libcfa/src/concurrency/io/types.hfa (revision 4c925cd31e0a0d5ba86584befde68164c77404b8) +++ libcfa/src/concurrency/io/types.hfa (revision 49981556d70010c36f072dc3534ff01415259a87) @@ -17,4 +17,8 @@ #if defined(CFA_HAVE_LINUX_IO_URING_H) + extern "C" { + #include + } + #include "bits/locks.hfa" @@ -23,25 +27,25 @@ struct __submition_data { // Head and tail of the ring (associated with array) - volatile uint32_t * head; - volatile uint32_t * tail; - volatile uint32_t prev_head; + volatile __u32 * head; + volatile __u32 * tail; + volatile __u32 prev_head; // The actual kernel ring which uses head/tail // indexes into the sqes arrays - uint32_t * array; + __u32 * array; // number of entries and mask to go with it - const uint32_t * num; - const uint32_t * mask; + const __u32 * num; + const __u32 * mask; // Submission flags (Not sure what for) - uint32_t * flags; + __u32 * flags; // number of sqes not submitted (whatever that means) - uint32_t * dropped; + __u32 * dropped; // Like head/tail but not seen by the kernel - volatile uint32_t * ready; - uint32_t ready_cnt; + volatile __u32 * ready; + __u32 ready_cnt; __spinlock_t lock; @@ -58,13 +62,13 @@ struct __completion_data { // Head and tail of the ring - volatile uint32_t * head; - volatile uint32_t * tail; + volatile __u32 * head; + volatile __u32 * tail; // number of entries and mask to go with it - const uint32_t * mask; - const uint32_t * num; + const __u32 * mask; + const __u32 * num; // number of cqes not submitted (whatever that means) - uint32_t * overflow; + __u32 * overflow; // the kernel ring @@ -79,5 +83,5 @@ struct __submition_data submit_q; struct __completion_data completion_q; - uint32_t ring_flags; + __u32 ring_flags; int fd; bool eager_submits:1; @@ -89,6 +93,6 @@ // IO user data struct __io_user_data_t { - int32_t result; - $thread * thrd; + __s32 result; + oneshot sem; }; Index: libcfa/src/concurrency/iocall.cfa =================================================================== --- libcfa/src/concurrency/iocall.cfa (revision 4c925cd31e0a0d5ba86584befde68164c77404b8) +++ libcfa/src/concurrency/iocall.cfa (revision 49981556d70010c36f072dc3534ff01415259a87) @@ -32,8 +32,8 @@ #include "io/types.hfa" - extern [* struct io_uring_sqe, uint32_t] __submit_alloc( struct __io_data & ring, uint64_t data ); - extern void __submit( struct io_context * ctx, uint32_t idx ) __attribute__((nonnull (1))); - - static inline void ?{}(struct io_uring_sqe & this, uint8_t opcode, int fd) { + extern [* struct io_uring_sqe, __u32] __submit_alloc( struct __io_data & ring, __u64 data ); + extern void __submit( struct io_context * ctx, __u32 idx ) __attribute__((nonnull (1))); + + static inline void ?{}(struct io_uring_sqe & this, __u8 opcode, int fd) { this.opcode = opcode; #if !defined(IOSQE_ASYNC) @@ -51,8 +51,8 @@ } - static inline void ?{}(struct io_uring_sqe & this, uint8_t opcode, int fd, void * addr, uint32_t len, uint64_t off ) { + static inline void ?{}(struct io_uring_sqe & this, __u8 opcode, int fd, void * addr, __u32 len, __u64 off ) { (this){ opcode, fd }; this.off = off; - this.addr = (uint64_t)(uintptr_t)addr; + this.addr = (__u64)(uintptr_t)addr; this.len = len; } @@ -105,17 +105,17 @@ (void)timeout; (void)cancellation; \ if( !context ) context = __get_io_context(); \ - __io_user_data_t data = { 0, active_thread() }; \ + __io_user_data_t data = { 0 }; \ struct __io_data & ring = *context->thrd.ring; \ struct io_uring_sqe * sqe; \ - uint32_t idx; \ - uint8_t sflags = REGULAR_FLAGS & submit_flags; \ - [sqe, idx] = __submit_alloc( ring, (uint64_t)(uintptr_t)&data ); \ + __u32 idx; \ + __u8 sflags = REGULAR_FLAGS & submit_flags; \ + [sqe, idx] = __submit_alloc( ring, (__u64)(uintptr_t)&data ); \ sqe->flags = sflags; #define __submit_wait \ /*__cfaabi_bits_print_safe( STDERR_FILENO, "Preparing user data %p for %p\n", &data, data.thrd );*/ \ - verify( sqe->user_data == (uint64_t)(uintptr_t)&data ); \ + verify( sqe->user_data == (__u64)(uintptr_t)&data ); \ __submit( context, idx ); \ - park( __cfaabi_dbg_ctx ); \ + wait( data.sem ); \ if( data.result < 0 ) { \ errno = -data.result; \ @@ -149,5 +149,8 @@ extern int fsync(int fd); - extern int sync_file_range(int fd, int64_t offset, int64_t nbytes, unsigned int flags); + + typedef __off64_t off_t; + typedef __off64_t off64_t; + extern int sync_file_range(int fd, off64_t offset, off64_t nbytes, unsigned int flags); struct msghdr; @@ -160,6 +163,6 @@ extern int connect(int sockfd, const struct sockaddr *addr, socklen_t addrlen); - extern int fallocate(int fd, int mode, uint64_t offset, uint64_t len); - extern int posix_fadvise(int fd, uint64_t offset, uint64_t len, int advice); + extern int fallocate(int fd, int mode, off_t offset, off_t len); + extern int posix_fadvise(int fd, off_t offset, off_t len, int advice); extern int madvise(void *addr, size_t length, int advice); @@ -190,5 +193,5 @@ sqe->fd = fd; sqe->off = offset; - sqe->addr = (uint64_t)(uintptr_t)iov; + sqe->addr = (__u64)iov; sqe->len = iovcnt; sqe->rw_flags = 0; @@ -207,5 +210,12 @@ __submit_prelude - (*sqe){ IORING_OP_WRITEV, fd, iov, iovcnt, offset }; + sqe->opcode = IORING_OP_WRITEV; + sqe->ioprio = 0; + sqe->fd = fd; + sqe->off = offset; + sqe->addr = (__u64)iov; + sqe->len = iovcnt; + sqe->rw_flags = 0; + sqe->__pad2[0] = sqe->__pad2[1] = sqe->__pad2[2] = 0; __submit_wait @@ -220,11 +230,18 @@ __submit_prelude - (*sqe){ IORING_OP_FSYNC, fd }; - - __submit_wait - #endif -} - -int cfa_sync_file_range(int fd, int64_t offset, int64_t nbytes, unsigned int flags, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { + sqe->opcode = IORING_OP_FSYNC; + sqe->ioprio = 0; + sqe->fd = fd; + sqe->off = 0; + sqe->addr = 0; + sqe->len = 0; + sqe->rw_flags = 0; + sqe->__pad2[0] = sqe->__pad2[1] = sqe->__pad2[2] = 0; + + __submit_wait + #endif +} + +int cfa_sync_file_range(int fd, off64_t offset, off64_t nbytes, unsigned int flags, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { #if !defined(CFA_HAVE_LINUX_IO_URING_H) || !defined(CFA_HAVE_IORING_OP_SYNC_FILE_RANGE) return sync_file_range(fd, offset, nbytes, flags); @@ -275,5 +292,5 @@ (*sqe){ IORING_OP_SEND, sockfd }; - sqe->addr = (uint64_t)buf; + sqe->addr = (__u64)buf; sqe->len = len; sqe->msg_flags = flags; @@ -290,5 +307,5 @@ (*sqe){ IORING_OP_RECV, sockfd }; - sqe->addr = (uint64_t)buf; + sqe->addr = (__u64)buf; sqe->len = len; sqe->msg_flags = flags; @@ -305,6 +322,6 @@ (*sqe){ IORING_OP_ACCEPT, sockfd }; - sqe->addr = (uint64_t)(uintptr_t)addr; - sqe->addr2 = (uint64_t)(uintptr_t)addrlen; + sqe->addr = (__u64)addr; + sqe->addr2 = (__u64)addrlen; sqe->accept_flags = flags; @@ -320,12 +337,12 @@ (*sqe){ IORING_OP_CONNECT, sockfd }; - sqe->addr = (uint64_t)(uintptr_t)addr; - sqe->off = (uint64_t)(uintptr_t)addrlen; - - __submit_wait - #endif -} - -int cfa_fallocate(int fd, int mode, uint64_t offset, uint64_t len, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { + sqe->addr = (__u64)addr; + sqe->off = (__u64)addrlen; + + __submit_wait + #endif +} + +int cfa_fallocate(int fd, int mode, off_t offset, off_t len, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { #if !defined(CFA_HAVE_LINUX_IO_URING_H) || !defined(CFA_HAVE_IORING_OP_FALLOCATE) return fallocate( fd, mode, offset, len ); @@ -344,5 +361,5 @@ } -int cfa_fadvise(int fd, uint64_t offset, uint64_t len, int advice, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { +int cfa_fadvise(int fd, off_t offset, off_t len, int advice, int submit_flags, Duration timeout, io_cancellation * cancellation, io_context * context) { #if !defined(CFA_HAVE_LINUX_IO_URING_H) || !defined(CFA_HAVE_IORING_OP_FADVISE) return posix_fadvise( fd, offset, len, advice ); @@ -351,5 +368,5 @@ (*sqe){ IORING_OP_FADVISE, fd }; - sqe->off = (uint64_t)offset; + sqe->off = (__u64)offset; sqe->len = len; sqe->fadvise_advice = advice; @@ -366,5 +383,5 @@ (*sqe){ IORING_OP_MADVISE, 0 }; - sqe->addr = (uint64_t)addr; + sqe->addr = (__u64)addr; sqe->len = length; sqe->fadvise_advice = advice; @@ -381,5 +398,5 @@ (*sqe){ IORING_OP_OPENAT, dirfd }; - sqe->addr = (uint64_t)pathname; + sqe->addr = (__u64)pathname; sqe->open_flags = flags; sqe->len = mode; @@ -414,5 +431,5 @@ __submit_prelude - (*sqe){ IORING_OP_STATX, dirfd, pathname, mask, (uint64_t)statxbuf }; + (*sqe){ IORING_OP_STATX, dirfd, pathname, mask, (__u64)statxbuf }; sqe->statx_flags = flags; @@ -456,5 +473,5 @@ } else { - sqe->off = (uint64_t)-1; + sqe->off = (__u64)-1; } sqe->len = len; @@ -464,5 +481,5 @@ } else { - sqe->splice_off_in = (uint64_t)-1; + sqe->splice_off_in = (__u64)-1; } sqe->splice_flags = flags | (SPLICE_FLAGS & submit_flags);