source: libcfa/src/concurrency/io/types.hfa @ cf34e82

//
// Cforall Version 1.0.0 Copyright (C) 2020 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// io/types.hfa -- PRIVATE
// Types used by the I/O subsystem
//
// Author           : Thierry Delisle
// Created On       : Fri Jul 31 16:22:47 2020
// Last Modified By :
// Last Modified On :
// Update Count     :
//

#pragma once

#include <limits.h>

extern "C" {
        #include <linux/types.h>
}

#include "bits/locks.hfa"
#include "iofwd.hfa"
#include "kernel/fwd.hfa"

#if defined(CFA_HAVE_LINUX_IO_URING_H)
        #include "monitor.hfa"

        struct processor;
        monitor io_arbiter$;

        //-----------------------------------------------------------------------
        // Ring Data structure
        // represent the io_uring submission ring which contains operations that will be sent to io_uring for processing
        struct __sub_ring_t {
                // lock needed because remote processors might need to flush the instance 
                __spinlock_t lock;

                struct {
                        // Head and tail of the ring (associated with array)
                        volatile __u32 * head;   // one passed last index consumed by the kernel
                        volatile __u32 * tail;   // one passed last index visible to the kernel
                        volatile __u32 released; // one passed last index released back to the free list

                        // The actual kernel ring which uses head/tail
                        // indexes into the sqes arrays
                        __u32 * array;
                } kring;

                struct {
                        volatile __u32 head;
                        volatile __u32 tail;
                        // The ring which contains free allocations
                        // indexes into the sqes arrays
                        __u32 * array;
                } free_ring;

                // number of sqes to submit on next system call.
                volatile __u32 to_submit;

                // number of entries and mask to go with it
                const __u32 * num;
                const __u32 * mask;

                // Submission flags, currently only IORING_SETUP_SQPOLL
                __u32 * flags;

                // number of sqes not submitted
                // From documentation : [dropped] is incremented for each invalid submission queue entry encountered in the ring buffer.
                __u32 * dropped;

                // A buffer of sqes (not the actual ring)
                struct io_uring_sqe * sqes;

                // The location and size of the mmaped area
                void * ring_ptr;
                size_t ring_sz;

                // for debug purposes, whether or not the last flush was due to a arbiter flush
                bool last_external;
        };

        // represent the io_uring completion ring which contains operations that have completed
        struct __cmp_ring_t {
                // needed because remote processors can help drain the buffer
                volatile bool try_lock;

                // id of the ring, used for the helping/topology algorithms
                unsigned id;

                // timestamp from last time it was drained
                unsigned long long ts;

                // Head and tail of the ring
                volatile __u32 * head;
                volatile __u32 * tail;

                // number of entries and mask to go with it
                const __u32 * mask;
                const __u32 * num;

                // I don't know what this value is for
                __u32 * overflow;

                // the kernel ring
                volatile struct io_uring_cqe * cqes;

                // The location and size of the mmaped area
                void * ring_ptr;
                size_t ring_sz;
        };

        // struct representing an io operation that still needs processing
        // actual operations are expected to inherit from this
        struct __outstanding_io {
                // intrusive link fields
                inline dlink(__outstanding_io);

                // primitive on which to block until the io is processed
                oneshot waitctx;
        };
    P9_EMBEDDED( __outstanding_io, dlink(__outstanding_io) )

        // queue of operations that are outstanding
        struct __outstanding_io_queue {
                // spinlock for protection
                // TODO: changing to a lock that blocks, I haven't examined whether it should be a kernel or user lock
                __spinlock_t lock;

                // the actual queue
                dlist(__outstanding_io) queue;

                // volatile used to avoid the need for taking the lock if it's empty
                volatile bool empty;
        };

        // struct representing an operation that was submitted 
        struct __external_io {
                // inherits from outstanding io
                inline __outstanding_io;

                // pointer and count to an array of ids to be submitted
                __u32 * idxs;
                __u32 have;

                // whether or not these can be accumulated before flushing the buffer
                bool lazy;
        };

        // complete io_context, contains all the data for io submission and completion
        struct __attribute__((aligned(64))) io_context$ {
                // arbiter, used in cases where threads for migrated at unfortunate moments
                io_arbiter$ * arbiter;

                // which prcessor the context is tied to
                struct processor * proc;

                // queue of io submissions that haven't beeen processed.
                __outstanding_io_queue ext_sq;

                // io_uring ring data structures
                struct __sub_ring_t sq;
                struct __cmp_ring_t cq;

                // flag the io_uring rings where created with
                __u32 ring_flags;

                // file descriptor that identifies the io_uring instance
                int fd;
        };

        // short hand to check when the io_context was last processed (io drained)
        static inline unsigned long long ts(io_context$ *& this) {
                const __u32 head = *this->cq.head;
                const __u32 tail = *this->cq.tail;

                // if there is no pending completions, just pretend it's infinetely recent
                if(head == tail) return ULLONG_MAX;

                return this->cq.ts;
        }

        // structure represeting allocations that couldn't succeed locally
        struct __pending_alloc {
                // inherit from outstanding io
                inline __outstanding_io;

                // array and size of the desired allocation
                __u32 * idxs;
                __u32 want;

                // output param, the context the io was allocated from 
                io_context$ * ctx;
        };

        // arbiter that handles cases where the context tied to the local processor is unable to satisfy the io
        monitor __attribute__((aligned(64))) io_arbiter$ {
                // contains a queue of io for pending allocations
                __outstanding_io_queue pending;
        };

        //-----------------------------------------------------------------------
        // Misc
        // Weirdly, some systems that do support io_uring don't actually define these
        #ifdef __alpha__
                /*
                * alpha is the only exception, all other architectures
                * have common numbers for new system calls.
                */
                #ifndef __NR_io_uring_setup
                        #define __NR_io_uring_setup           535
                #endif
                #ifndef __NR_io_uring_enter
                        #define __NR_io_uring_enter           536
                #endif
                #ifndef __NR_io_uring_register
                        #define __NR_io_uring_register        537
                #endif
        #else /* !__alpha__ */
                #ifndef __NR_io_uring_setup
                        #define __NR_io_uring_setup           425
                #endif
                #ifndef __NR_io_uring_enter
                        #define __NR_io_uring_enter           426
                #endif
                #ifndef __NR_io_uring_register
                        #define __NR_io_uring_register        427
                #endif
        #endif

        // void __ioctx_prepare_block(io_context$ & ctx);
#endif