linux_old1/block/blk-mq-sysfs.c

425 lines
10 KiB
C
Raw Normal View History

blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/backing-dev.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/workqueue.h>
#include <linux/smp.h>
#include <linux/blk-mq.h>
#include "blk-mq.h"
#include "blk-mq-tag.h"
static void blk_mq_sysfs_release(struct kobject *kobj)
{
}
struct blk_mq_ctx_sysfs_entry {
struct attribute attr;
ssize_t (*show)(struct blk_mq_ctx *, char *);
ssize_t (*store)(struct blk_mq_ctx *, const char *, size_t);
};
struct blk_mq_hw_ctx_sysfs_entry {
struct attribute attr;
ssize_t (*show)(struct blk_mq_hw_ctx *, char *);
ssize_t (*store)(struct blk_mq_hw_ctx *, const char *, size_t);
};
static ssize_t blk_mq_sysfs_show(struct kobject *kobj, struct attribute *attr,
char *page)
{
struct blk_mq_ctx_sysfs_entry *entry;
struct blk_mq_ctx *ctx;
struct request_queue *q;
ssize_t res;
entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
ctx = container_of(kobj, struct blk_mq_ctx, kobj);
q = ctx->queue;
if (!entry->show)
return -EIO;
res = -ENOENT;
mutex_lock(&q->sysfs_lock);
if (!blk_queue_dying(q))
res = entry->show(ctx, page);
mutex_unlock(&q->sysfs_lock);
return res;
}
static ssize_t blk_mq_sysfs_store(struct kobject *kobj, struct attribute *attr,
const char *page, size_t length)
{
struct blk_mq_ctx_sysfs_entry *entry;
struct blk_mq_ctx *ctx;
struct request_queue *q;
ssize_t res;
entry = container_of(attr, struct blk_mq_ctx_sysfs_entry, attr);
ctx = container_of(kobj, struct blk_mq_ctx, kobj);
q = ctx->queue;
if (!entry->store)
return -EIO;
res = -ENOENT;
mutex_lock(&q->sysfs_lock);
if (!blk_queue_dying(q))
res = entry->store(ctx, page, length);
mutex_unlock(&q->sysfs_lock);
return res;
}
static ssize_t blk_mq_hw_sysfs_show(struct kobject *kobj,
struct attribute *attr, char *page)
{
struct blk_mq_hw_ctx_sysfs_entry *entry;
struct blk_mq_hw_ctx *hctx;
struct request_queue *q;
ssize_t res;
entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
q = hctx->queue;
if (!entry->show)
return -EIO;
res = -ENOENT;
mutex_lock(&q->sysfs_lock);
if (!blk_queue_dying(q))
res = entry->show(hctx, page);
mutex_unlock(&q->sysfs_lock);
return res;
}
static ssize_t blk_mq_hw_sysfs_store(struct kobject *kobj,
struct attribute *attr, const char *page,
size_t length)
{
struct blk_mq_hw_ctx_sysfs_entry *entry;
struct blk_mq_hw_ctx *hctx;
struct request_queue *q;
ssize_t res;
entry = container_of(attr, struct blk_mq_hw_ctx_sysfs_entry, attr);
hctx = container_of(kobj, struct blk_mq_hw_ctx, kobj);
q = hctx->queue;
if (!entry->store)
return -EIO;
res = -ENOENT;
mutex_lock(&q->sysfs_lock);
if (!blk_queue_dying(q))
res = entry->store(hctx, page, length);
mutex_unlock(&q->sysfs_lock);
return res;
}
static ssize_t blk_mq_sysfs_dispatched_show(struct blk_mq_ctx *ctx, char *page)
{
return sprintf(page, "%lu %lu\n", ctx->rq_dispatched[1],
ctx->rq_dispatched[0]);
}
static ssize_t blk_mq_sysfs_merged_show(struct blk_mq_ctx *ctx, char *page)
{
return sprintf(page, "%lu\n", ctx->rq_merged);
}
static ssize_t blk_mq_sysfs_completed_show(struct blk_mq_ctx *ctx, char *page)
{
return sprintf(page, "%lu %lu\n", ctx->rq_completed[1],
ctx->rq_completed[0]);
}
static ssize_t sysfs_list_show(char *page, struct list_head *list, char *msg)
{
char *start_page = page;
struct request *rq;
page += sprintf(page, "%s:\n", msg);
list_for_each_entry(rq, list, queuelist)
page += sprintf(page, "\t%p\n", rq);
return page - start_page;
}
static ssize_t blk_mq_sysfs_rq_list_show(struct blk_mq_ctx *ctx, char *page)
{
ssize_t ret;
spin_lock(&ctx->lock);
ret = sysfs_list_show(page, &ctx->rq_list, "CTX pending");
spin_unlock(&ctx->lock);
return ret;
}
static ssize_t blk_mq_hw_sysfs_queued_show(struct blk_mq_hw_ctx *hctx,
char *page)
{
return sprintf(page, "%lu\n", hctx->queued);
}
static ssize_t blk_mq_hw_sysfs_run_show(struct blk_mq_hw_ctx *hctx, char *page)
{
return sprintf(page, "%lu\n", hctx->run);
}
static ssize_t blk_mq_hw_sysfs_dispatched_show(struct blk_mq_hw_ctx *hctx,
char *page)
{
char *start_page = page;
int i;
page += sprintf(page, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER; i++) {
unsigned long d = 1U << (i - 1);
page += sprintf(page, "%8lu\t%lu\n", d, hctx->dispatched[i]);
}
return page - start_page;
}
static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
char *page)
{
ssize_t ret;
spin_lock(&hctx->lock);
ret = sysfs_list_show(page, &hctx->dispatch, "HCTX pending");
spin_unlock(&hctx->lock);
return ret;
}
static ssize_t blk_mq_hw_sysfs_ipi_show(struct blk_mq_hw_ctx *hctx, char *page)
{
ssize_t ret;
spin_lock(&hctx->lock);
ret = sprintf(page, "%u\n", !!(hctx->flags & BLK_MQ_F_SHOULD_IPI));
spin_unlock(&hctx->lock);
return ret;
}
static ssize_t blk_mq_hw_sysfs_ipi_store(struct blk_mq_hw_ctx *hctx,
const char *page, size_t len)
{
struct blk_mq_ctx *ctx;
unsigned long ret;
unsigned int i;
if (kstrtoul(page, 10, &ret)) {
pr_err("blk-mq-sysfs: invalid input '%s'\n", page);
return -EINVAL;
}
spin_lock(&hctx->lock);
if (ret)
hctx->flags |= BLK_MQ_F_SHOULD_IPI;
else
hctx->flags &= ~BLK_MQ_F_SHOULD_IPI;
spin_unlock(&hctx->lock);
hctx_for_each_ctx(hctx, ctx, i)
ctx->ipi_redirect = !!ret;
return len;
}
static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
{
return blk_mq_tag_sysfs_show(hctx->tags, page);
}
static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
{
unsigned int i, first = 1;
ssize_t ret = 0;
blk_mq_disable_hotplug();
for_each_cpu(i, hctx->cpumask) {
if (first)
ret += sprintf(ret + page, "%u", i);
else
ret += sprintf(ret + page, ", %u", i);
first = 0;
}
blk_mq_enable_hotplug();
ret += sprintf(ret + page, "\n");
return ret;
}
blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_dispatched = {
.attr = {.name = "dispatched", .mode = S_IRUGO },
.show = blk_mq_sysfs_dispatched_show,
};
static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_merged = {
.attr = {.name = "merged", .mode = S_IRUGO },
.show = blk_mq_sysfs_merged_show,
};
static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_completed = {
.attr = {.name = "completed", .mode = S_IRUGO },
.show = blk_mq_sysfs_completed_show,
};
static struct blk_mq_ctx_sysfs_entry blk_mq_sysfs_rq_list = {
.attr = {.name = "rq_list", .mode = S_IRUGO },
.show = blk_mq_sysfs_rq_list_show,
};
static struct attribute *default_ctx_attrs[] = {
&blk_mq_sysfs_dispatched.attr,
&blk_mq_sysfs_merged.attr,
&blk_mq_sysfs_completed.attr,
&blk_mq_sysfs_rq_list.attr,
NULL,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_queued = {
.attr = {.name = "queued", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_queued_show,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_run = {
.attr = {.name = "run", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_run_show,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
.attr = {.name = "dispatched", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_dispatched_show,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
.attr = {.name = "pending", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_rq_list_show,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_ipi = {
.attr = {.name = "ipi_redirect", .mode = S_IRUGO | S_IWUSR},
.show = blk_mq_hw_sysfs_ipi_show,
.store = blk_mq_hw_sysfs_ipi_store,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
.attr = {.name = "tags", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_tags_show,
};
static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_cpus = {
.attr = {.name = "cpu_list", .mode = S_IRUGO },
.show = blk_mq_hw_sysfs_cpus_show,
};
blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
static struct attribute *default_hw_ctx_attrs[] = {
&blk_mq_hw_sysfs_queued.attr,
&blk_mq_hw_sysfs_run.attr,
&blk_mq_hw_sysfs_dispatched.attr,
&blk_mq_hw_sysfs_pending.attr,
&blk_mq_hw_sysfs_ipi.attr,
&blk_mq_hw_sysfs_tags.attr,
&blk_mq_hw_sysfs_cpus.attr,
blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
NULL,
};
static const struct sysfs_ops blk_mq_sysfs_ops = {
.show = blk_mq_sysfs_show,
.store = blk_mq_sysfs_store,
};
static const struct sysfs_ops blk_mq_hw_sysfs_ops = {
.show = blk_mq_hw_sysfs_show,
.store = blk_mq_hw_sysfs_store,
};
static struct kobj_type blk_mq_ktype = {
.sysfs_ops = &blk_mq_sysfs_ops,
.release = blk_mq_sysfs_release,
};
static struct kobj_type blk_mq_ctx_ktype = {
.sysfs_ops = &blk_mq_sysfs_ops,
.default_attrs = default_ctx_attrs,
.release = blk_mq_sysfs_release,
};
static struct kobj_type blk_mq_hw_ktype = {
.sysfs_ops = &blk_mq_hw_sysfs_ops,
.default_attrs = default_hw_ctx_attrs,
.release = blk_mq_sysfs_release,
};
void blk_mq_unregister_disk(struct gendisk *disk)
{
struct request_queue *q = disk->queue;
block: fix memory leaks on unplugging block device All objects, which are allocated in blk_mq_register_disk, must be released in blk_mq_unregister_disk. I use a KVM virtual machine and virtio disk to reproduce this issue. kmemleak: 18 new suspected memory leaks (see /sys/kernel/debug/kmemleak) $ cat /sys/kernel/debug/kmemleak | head -n 30 unreferenced object 0xffff8800b6636150 (size 8): comm "kworker/0:2", pid 65, jiffies 4294809903 (age 86.358s) hex dump (first 8 bytes): 76 69 72 74 69 6f 34 00 virtio4. backtrace: [<ffffffff8165d41e>] kmemleak_alloc+0x4e/0xb0 [<ffffffff8118cfc5>] __kmalloc_track_caller+0xf5/0x260 [<ffffffff81155b11>] kstrdup+0x31/0x60 [<ffffffff812242be>] sysfs_new_dirent+0x2e/0x140 [<ffffffff81224678>] create_dir+0x38/0xe0 [<ffffffff812249e3>] sysfs_create_dir_ns+0x73/0xc0 [<ffffffff8130dfa9>] kobject_add_internal+0xc9/0x340 [<ffffffff8130e535>] kobject_add+0x65/0xb0 [<ffffffff813f34f8>] device_add+0x128/0x660 [<ffffffff813f3a4a>] device_register+0x1a/0x20 [<ffffffff813ae6f8>] register_virtio_device+0x98/0xe0 [<ffffffff813b0cce>] virtio_pci_probe+0x12e/0x1c0 [<ffffffff81340675>] local_pci_probe+0x45/0xa0 [<ffffffff81341a51>] pci_device_probe+0x121/0x130 [<ffffffff813f67f7>] driver_probe_device+0x87/0x390 [<ffffffff813f6b3b>] __device_attach+0x3b/0x40 unreferenced object 0xffff8800b65aa1d8 (size 144): Fixes: 320ae51feed5 (blk-mq: new multi-queue block IO queueing mechanism) Cc: Jens Axboe <axboe@kernel.dk> Signed-off-by: Andrey Vagin <avagin@openvz.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-12-06 13:06:41 +08:00
struct blk_mq_hw_ctx *hctx;
struct blk_mq_ctx *ctx;
int i, j;
queue_for_each_hw_ctx(q, hctx, i) {
hctx_for_each_ctx(hctx, ctx, j) {
kobject_del(&ctx->kobj);
kobject_put(&ctx->kobj);
}
kobject_del(&hctx->kobj);
kobject_put(&hctx->kobj);
}
blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
kobject_uevent(&q->mq_kobj, KOBJ_REMOVE);
kobject_del(&q->mq_kobj);
block: fix memory leaks on unplugging block device All objects, which are allocated in blk_mq_register_disk, must be released in blk_mq_unregister_disk. I use a KVM virtual machine and virtio disk to reproduce this issue. kmemleak: 18 new suspected memory leaks (see /sys/kernel/debug/kmemleak) $ cat /sys/kernel/debug/kmemleak | head -n 30 unreferenced object 0xffff8800b6636150 (size 8): comm "kworker/0:2", pid 65, jiffies 4294809903 (age 86.358s) hex dump (first 8 bytes): 76 69 72 74 69 6f 34 00 virtio4. backtrace: [<ffffffff8165d41e>] kmemleak_alloc+0x4e/0xb0 [<ffffffff8118cfc5>] __kmalloc_track_caller+0xf5/0x260 [<ffffffff81155b11>] kstrdup+0x31/0x60 [<ffffffff812242be>] sysfs_new_dirent+0x2e/0x140 [<ffffffff81224678>] create_dir+0x38/0xe0 [<ffffffff812249e3>] sysfs_create_dir_ns+0x73/0xc0 [<ffffffff8130dfa9>] kobject_add_internal+0xc9/0x340 [<ffffffff8130e535>] kobject_add+0x65/0xb0 [<ffffffff813f34f8>] device_add+0x128/0x660 [<ffffffff813f3a4a>] device_register+0x1a/0x20 [<ffffffff813ae6f8>] register_virtio_device+0x98/0xe0 [<ffffffff813b0cce>] virtio_pci_probe+0x12e/0x1c0 [<ffffffff81340675>] local_pci_probe+0x45/0xa0 [<ffffffff81341a51>] pci_device_probe+0x121/0x130 [<ffffffff813f67f7>] driver_probe_device+0x87/0x390 [<ffffffff813f6b3b>] __device_attach+0x3b/0x40 unreferenced object 0xffff8800b65aa1d8 (size 144): Fixes: 320ae51feed5 (blk-mq: new multi-queue block IO queueing mechanism) Cc: Jens Axboe <axboe@kernel.dk> Signed-off-by: Andrey Vagin <avagin@openvz.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-12-06 13:06:41 +08:00
kobject_put(&q->mq_kobj);
blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li <shli@fusionio.com> Alexander Gordeev <agordeev@redhat.com> Christoph Hellwig <hch@infradead.org> Mike Christie <michaelc@cs.wisc.edu> Matias Bjorling <m@bjorling.me> Jeff Moyer <jmoyer@redhat.com> Acked-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-10-24 16:20:05 +08:00
kobject_put(&disk_to_dev(disk)->kobj);
}
int blk_mq_register_disk(struct gendisk *disk)
{
struct device *dev = disk_to_dev(disk);
struct request_queue *q = disk->queue;
struct blk_mq_hw_ctx *hctx;
struct blk_mq_ctx *ctx;
int ret, i, j;
kobject_init(&q->mq_kobj, &blk_mq_ktype);
ret = kobject_add(&q->mq_kobj, kobject_get(&dev->kobj), "%s", "mq");
if (ret < 0)
return ret;
kobject_uevent(&q->mq_kobj, KOBJ_ADD);
queue_for_each_hw_ctx(q, hctx, i) {
kobject_init(&hctx->kobj, &blk_mq_hw_ktype);
ret = kobject_add(&hctx->kobj, &q->mq_kobj, "%u", i);
if (ret)
break;
if (!hctx->nr_ctx)
continue;
hctx_for_each_ctx(hctx, ctx, j) {
kobject_init(&ctx->kobj, &blk_mq_ctx_ktype);
ret = kobject_add(&ctx->kobj, &hctx->kobj, "cpu%u", ctx->cpu);
if (ret)
break;
}
}
if (ret) {
blk_mq_unregister_disk(disk);
return ret;
}
return 0;
}