From d90a42c43003827d296deebbb08ef091d257398b Mon Sep 17 00:00:00 2001 From: Joe Maples Date: Wed, 28 Feb 2018 17:43:20 +0100 Subject: [PATCH] block: Add Maple I/O Scheduler Signed-off-by: Joe Maples --- block/Kconfig.iosched | 8 + block/Makefile | 1 + block/maple-iosched.c | 426 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 435 insertions(+) create mode 100644 block/maple-iosched.c diff --git a/block/Kconfig.iosched b/block/Kconfig.iosched index c98ed9f5a6ed..ceba3485cbe2 100644 --- a/block/Kconfig.iosched +++ b/block/Kconfig.iosched @@ -115,6 +115,10 @@ config IOSCHED_BFQ a low latency to interactive and soft real-time applications. Details in Documentation/block/bfq-iosched.txt +config IOSCHED_MAPLE + tristate "Maple I/O scheduler" + default n + config BFQ_GROUP_IOSCHED bool "BFQ hierarchical scheduling support" depends on IOSCHED_BFQ && BLK_CGROUP @@ -181,6 +185,9 @@ choice any workload. It also tries to guarantee low latency to interactive and soft real-time applications. + config DEFAULT_MAPLE + bool "Maple" if IOSCHED_MAPLE=y + endchoice config DEFAULT_IOSCHED @@ -197,6 +204,7 @@ config DEFAULT_IOSCHED default "zen" if DEFAULT_ZEN default "row" if DEFAULT_ROW default "bfq" if DEFAULT_BFQ + default "maple" if DEFAULT_MAPLE endmenu diff --git a/block/Makefile b/block/Makefile index d802bb187112..341c777baabd 100644 --- a/block/Makefile +++ b/block/Makefile @@ -27,6 +27,7 @@ obj-$(CONFIG_IOSCHED_VR) += vr-iosched.o obj-$(CONFIG_IOSCHED_ZEN) += zen-iosched.o obj-$(CONFIG_IOSCHED_ROW) += row-iosched.o obj-$(CONFIG_IOSCHED_BFQ) += bfq-iosched.o +obj-$(CONFIG_IOSCHED_MAPLE) += maple-iosched.o obj-$(CONFIG_BLOCK_COMPAT) += compat_ioctl.o obj-$(CONFIG_BLK_CMDLINE_PARSER) += cmdline-parser.o diff --git a/block/maple-iosched.c b/block/maple-iosched.c new file mode 100644 index 000000000000..89b32833a321 --- /dev/null +++ b/block/maple-iosched.c @@ -0,0 +1,426 @@ +/* + * Maple I/O Scheduler + * Based on Zen and SIO. + * + * Copyright (C) 2016 Joe Maples + * (C) 2012 Brandon Berhent + * + * Maple uses a first come first serve style algorithm with seperated read/write + * handling to allow for read biases. By prioritizing reads, simple tasks should improve + * in performance. Maple also uses hooks for the powersuspend driver to increase + * expirations when power is suspended to decrease workload. + */ +#include +#include +#include +#include +#include +#include +#include + +#define MAPLE_IOSCHED_PATCHLEVEL (8) + +enum { ASYNC, SYNC }; + +/* Tunables */ +static const int sync_read_expire = 100; /* max time before a read sync is submitted. */ +static const int sync_write_expire = 350; /* max time before a write sync is submitted. */ +static const int async_read_expire = 200; /* ditto for read async, these limits are SOFT! */ +static const int async_write_expire = 500; /* ditto for write async, these limits are SOFT! */ +static const int fifo_batch = 16; /* # of sequential requests treated as one by the above parameters. */ +static const int writes_starved = 3; /* max times reads can starve a write */ +static const int sleep_latency_multiple = 5; /* multple for expire time when device is asleep */ + +/* Elevator data */ +struct maple_data { + /* Request queues */ + struct list_head fifo_list[2][2]; + + /* Attributes */ + unsigned int batched; + unsigned int starved; + + /* Settings */ + int fifo_expire[2][2]; + int fifo_batch; + int writes_starved; + int sleep_latency_multiple; +}; + +static inline struct maple_data * +maple_get_data(struct request_queue *q) { + return q->elevator->elevator_data; +} + +static void +maple_merged_requests(struct request_queue *q, struct request *rq, + struct request *next) +{ + /* + * If next expires before rq, assign its expire time to rq + * and move into next position (next will be deleted) in fifo. + */ + if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist)) { + if (time_before(next->fifo_time, rq->fifo_time)) { + list_move(&rq->queuelist, &next->queuelist); + rq->fifo_time = next->fifo_time; + } + } + + /* Delete next request */ + rq_fifo_clear(next); +} + +static void +maple_add_request(struct request_queue *q, struct request *rq) +{ + struct maple_data *mdata = maple_get_data(q); + const int sync = rq_is_sync(rq); + const int dir = rq_data_dir(rq); + const bool display_on = is_display_on(); + + /* + * Add request to the proper fifo list and set its + * expire time. + */ + + /* inrease expiration when device is asleep */ + unsigned int fifo_expire_suspended = mdata->fifo_expire[sync][dir] * sleep_latency_multiple; + if (display_on && mdata->fifo_expire[sync][dir]) { + rq->fifo_time = jiffies + mdata->fifo_expire[sync][dir]; + list_add_tail(&rq->queuelist, &mdata->fifo_list[sync][dir]); + } else if (!display_on && fifo_expire_suspended) { + rq->fifo_time = jiffies + mdata->fifo_expire[sync][dir]; + list_add_tail(&rq->queuelist, &mdata->fifo_list[sync][dir]); + } +} + +static struct request * +maple_expired_request(struct maple_data *mdata, int sync, int data_dir) +{ + struct list_head *list = &mdata->fifo_list[sync][data_dir]; + struct request *rq; + + if (list_empty(list)) + return NULL; + + /* Retrieve request */ + rq = rq_entry_fifo(list->next); + + /* Request has expired */ + if (time_after(jiffies, rq->fifo_time)) + return rq; + + return NULL; +} + +static struct request * +maple_choose_expired_request(struct maple_data *mdata) +{ + struct request *rq_sync_read = maple_expired_request(mdata, SYNC, READ); + struct request *rq_sync_write = maple_expired_request(mdata, SYNC, WRITE); + struct request *rq_async_read = maple_expired_request(mdata, ASYNC, READ); + struct request *rq_async_write = maple_expired_request(mdata, ASYNC, WRITE); + + /* Reset (non-expired-)batch-counter */ + mdata->batched = 0; + + /* + * Check expired requests. + * Asynchronous requests have priority over synchronous. + * Read requests have priority over write. + */ + + if (rq_async_read && rq_sync_read) { + if (time_after(rq_sync_read->fifo_time, rq_async_read->fifo_time)) + return rq_async_read; + } else if (rq_async_read) { + return rq_async_read; + } else if (rq_sync_read) { + return rq_sync_read; + } + + if (rq_async_write && rq_sync_write) { + if (time_after(rq_sync_write->fifo_time, rq_async_write->fifo_time)) + return rq_async_write; + } else if (rq_async_write) { + return rq_async_write; + } else if (rq_sync_write) { + return rq_sync_write; + } + + return NULL; +} + +static struct request * +maple_choose_request(struct maple_data *mdata, int data_dir) +{ + struct list_head *sync = mdata->fifo_list[SYNC]; + struct list_head *async = mdata->fifo_list[ASYNC]; + + /* Increase (non-expired-)batch-counter */ + mdata->batched++; + + + /* + * Retrieve request from available fifo list. + * Asynchronous requests have priority over synchronous. + * Read requests have priority over write. + */ + if (!list_empty(&async[data_dir])) + return rq_entry_fifo(async[data_dir].next); + if (!list_empty(&sync[data_dir])) + return rq_entry_fifo(sync[data_dir].next); + + if (!list_empty(&async[!data_dir])) + return rq_entry_fifo(async[!data_dir].next); + if (!list_empty(&sync[!data_dir])) + return rq_entry_fifo(sync[!data_dir].next); + + return NULL; +} + +static inline void +maple_dispatch_request(struct maple_data *mdata, struct request *rq) +{ + /* + * Remove the request from the fifo list + * and dispatch it. + */ + rq_fifo_clear(rq); + elv_dispatch_add_tail(rq->q, rq); + + if (rq_data_dir(rq)) { + mdata->starved = 0; + } else { + if (!list_empty(&mdata->fifo_list[SYNC][WRITE]) || + !list_empty(&mdata->fifo_list[ASYNC][WRITE])) + mdata->starved++; + } +} + +static int +maple_dispatch_requests(struct request_queue *q, int force) +{ + struct maple_data *mdata = maple_get_data(q); + struct request *rq = NULL; + int data_dir = READ; + const bool display_on = is_display_on(); + + /* + * Retrieve any expired request after a batch of + * sequential requests. + */ + if (mdata->batched >= mdata->fifo_batch) + rq = maple_choose_expired_request(mdata); + + /* Retrieve request */ + if (!rq) { + /* Treat writes fairly while suspended, otherwise allow them to be starved */ + if (display_on && mdata->starved >= mdata->writes_starved) + data_dir = WRITE; + else if (!display_on && mdata->starved >= 1) + data_dir = WRITE; + + rq = maple_choose_request(mdata, data_dir); + if (!rq) + return 0; + } + + /* Dispatch request */ + maple_dispatch_request(mdata, rq); + + return 1; +} + +static struct request * +maple_former_request(struct request_queue *q, struct request *rq) +{ + struct maple_data *mdata = maple_get_data(q); + const int sync = rq_is_sync(rq); + const int data_dir = rq_data_dir(rq); + + if (rq->queuelist.prev == &mdata->fifo_list[sync][data_dir]) + return NULL; + + /* Return former request */ + return list_entry(rq->queuelist.prev, struct request, queuelist); +} + +static struct request * +maple_latter_request(struct request_queue *q, struct request *rq) +{ + struct maple_data *mdata = maple_get_data(q); + const int sync = rq_is_sync(rq); + const int data_dir = rq_data_dir(rq); + + if (rq->queuelist.next == &mdata->fifo_list[sync][data_dir]) + return NULL; + + /* Return latter request */ + return list_entry(rq->queuelist.next, struct request, queuelist); +} + +static int maple_init_queue(struct request_queue *q, struct elevator_type *e) +{ + struct maple_data *mdata; + struct elevator_queue *eq; + + eq = elevator_alloc(q, e); + if (!eq) + return -ENOMEM; + + /* Allocate structure */ + mdata = kmalloc_node(sizeof(*mdata), GFP_KERNEL, q->node); + if (!mdata) { + kobject_put(&eq->kobj); + return -ENOMEM; + } + eq->elevator_data = mdata; + + /* Initialize fifo lists */ + INIT_LIST_HEAD(&mdata->fifo_list[SYNC][READ]); + INIT_LIST_HEAD(&mdata->fifo_list[SYNC][WRITE]); + INIT_LIST_HEAD(&mdata->fifo_list[ASYNC][READ]); + INIT_LIST_HEAD(&mdata->fifo_list[ASYNC][WRITE]); + + /* Initialize data */ + mdata->batched = 0; + mdata->fifo_expire[SYNC][READ] = sync_read_expire; + mdata->fifo_expire[SYNC][WRITE] = sync_write_expire; + mdata->fifo_expire[ASYNC][READ] = async_read_expire; + mdata->fifo_expire[ASYNC][WRITE] = async_write_expire; + mdata->fifo_batch = fifo_batch; + mdata->writes_starved = writes_starved; + mdata->sleep_latency_multiple = sleep_latency_multiple; + + spin_lock_irq(q->queue_lock); + q->elevator = eq; + spin_unlock_irq(q->queue_lock); + return 0; +} + +static void +maple_exit_queue(struct elevator_queue *e) +{ + struct maple_data *mdata = e->elevator_data; + + /* Free structure */ + kfree(mdata); +} + +/* + * sysfs code + */ + +static ssize_t +maple_var_show(int var, char *page) +{ + return sprintf(page, "%d\n", var); +} + +static ssize_t +maple_var_store(int *var, const char *page, size_t count) +{ + char *p = (char *) page; + + *var = simple_strtol(p, &p, 10); + return count; +} + +#define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ +static ssize_t __FUNC(struct elevator_queue *e, char *page) \ +{ \ + struct maple_data *mdata = e->elevator_data; \ + int __data = __VAR; \ + if (__CONV) \ + __data = jiffies_to_msecs(__data); \ + return maple_var_show(__data, (page)); \ +} +SHOW_FUNCTION(maple_sync_read_expire_show, mdata->fifo_expire[SYNC][READ], 1); +SHOW_FUNCTION(maple_sync_write_expire_show, mdata->fifo_expire[SYNC][WRITE], 1); +SHOW_FUNCTION(maple_async_read_expire_show, mdata->fifo_expire[ASYNC][READ], 1); +SHOW_FUNCTION(maple_async_write_expire_show, mdata->fifo_expire[ASYNC][WRITE], 1); +SHOW_FUNCTION(maple_fifo_batch_show, mdata->fifo_batch, 0); +SHOW_FUNCTION(maple_writes_starved_show, mdata->writes_starved, 0); +SHOW_FUNCTION(maple_sleep_latency_multiple_show, mdata->sleep_latency_multiple, 0); +#undef SHOW_FUNCTION + +#define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ +static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ +{ \ + struct maple_data *mdata = e->elevator_data; \ + int __data; \ + int ret = maple_var_store(&__data, (page), count); \ + if (__data < (MIN)) \ + __data = (MIN); \ + else if (__data > (MAX)) \ + __data = (MAX); \ + if (__CONV) \ + *(__PTR) = msecs_to_jiffies(__data); \ + else \ + *(__PTR) = __data; \ + return ret; \ +} +STORE_FUNCTION(maple_sync_read_expire_store, &mdata->fifo_expire[SYNC][READ], 0, INT_MAX, 1); +STORE_FUNCTION(maple_sync_write_expire_store, &mdata->fifo_expire[SYNC][WRITE], 0, INT_MAX, 1); +STORE_FUNCTION(maple_async_read_expire_store, &mdata->fifo_expire[ASYNC][READ], 0, INT_MAX, 1); +STORE_FUNCTION(maple_async_write_expire_store, &mdata->fifo_expire[ASYNC][WRITE], 0, INT_MAX, 1); +STORE_FUNCTION(maple_fifo_batch_store, &mdata->fifo_batch, 1, INT_MAX, 0); +STORE_FUNCTION(maple_writes_starved_store, &mdata->writes_starved, 1, INT_MAX, 0); +STORE_FUNCTION(maple_sleep_latency_multiple_store, &mdata->sleep_latency_multiple, 1, INT_MAX, 0); +#undef STORE_FUNCTION + +#define DD_ATTR(name) \ + __ATTR(name, S_IRUGO|S_IWUSR, maple_##name##_show, \ + maple_##name##_store) + +static struct elv_fs_entry maple_attrs[] = { + DD_ATTR(sync_read_expire), + DD_ATTR(sync_write_expire), + DD_ATTR(async_read_expire), + DD_ATTR(async_write_expire), + DD_ATTR(fifo_batch), + DD_ATTR(writes_starved), + DD_ATTR(sleep_latency_multiple), + __ATTR_NULL +}; + +static struct elevator_type iosched_maple = { + .ops = { + .elevator_merge_req_fn = maple_merged_requests, + .elevator_dispatch_fn = maple_dispatch_requests, + .elevator_add_req_fn = maple_add_request, + .elevator_former_req_fn = maple_former_request, + .elevator_latter_req_fn = maple_latter_request, + .elevator_init_fn = maple_init_queue, + .elevator_exit_fn = maple_exit_queue, + }, + + .elevator_attrs = maple_attrs, + .elevator_name = "maple", + .elevator_owner = THIS_MODULE, +}; + +static int __init maple_init(void) +{ + /* Register elevator */ + elv_register(&iosched_maple); + + return 0; +} + +static void __exit maple_exit(void) +{ + /* Unregister elevator */ + elv_unregister(&iosched_maple); +} + +module_init(maple_init); +module_exit(maple_exit); + +MODULE_AUTHOR("Joe Maples"); +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("Maple I/O Scheduler"); +MODULE_VERSION("1.0");