2019-06-04 16:11:33 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0-only
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
|
* Copyright (C) 2003 Russell King, All Rights Reserved.
|
2006-12-24 03:03:02 +08:00
|
|
|
* Copyright 2006-2007 Pierre Ossman
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
|
|
|
#include <linux/slab.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/module.h>
|
|
|
|
|
#include <linux/blkdev.h>
|
2007-07-17 19:03:35 +08:00
|
|
|
#include <linux/freezer.h>
|
2006-11-14 03:23:52 +08:00
|
|
|
#include <linux/kthread.h>
|
2007-10-23 03:19:53 +08:00
|
|
|
#include <linux/scatterlist.h>
|
2013-07-29 21:20:15 +08:00
|
|
|
#include <linux/dma-mapping.h>
|
2019-07-22 12:03:06 +08:00
|
|
|
#include <linux/backing-dev.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
|
#include <linux/mmc/card.h>
|
|
|
|
|
#include <linux/mmc/host.h>
|
2016-09-20 17:34:38 +08:00
|
|
|
|
2006-12-24 03:03:02 +08:00
|
|
|
#include "queue.h"
|
2016-09-20 17:34:38 +08:00
|
|
|
#include "block.h"
|
2017-01-13 21:14:08 +08:00
|
|
|
#include "core.h"
|
2017-01-13 21:14:14 +08:00
|
|
|
#include "card.h"
|
2017-11-29 21:41:03 +08:00
|
|
|
#include "host.h"
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2019-08-28 20:35:43 +08:00
|
|
|
#define MMC_DMA_MAP_MERGE_SEGMENTS 512
|
|
|
|
|
|
2017-11-29 21:41:04 +08:00
|
|
|
static inline bool mmc_cqe_dcmd_busy(struct mmc_queue *mq)
|
|
|
|
|
{
|
|
|
|
|
/* Allow only 1 DCMD at a time */
|
|
|
|
|
return mq->in_flight[MMC_ISSUE_DCMD];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void mmc_cqe_check_busy(struct mmc_queue *mq)
|
|
|
|
|
{
|
|
|
|
|
if ((mq->cqe_busy & MMC_CQE_DCMD_BUSY) && !mmc_cqe_dcmd_busy(mq))
|
|
|
|
|
mq->cqe_busy &= ~MMC_CQE_DCMD_BUSY;
|
|
|
|
|
|
|
|
|
|
mq->cqe_busy &= ~MMC_CQE_QUEUE_FULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline bool mmc_cqe_can_dcmd(struct mmc_host *host)
|
|
|
|
|
{
|
|
|
|
|
return host->caps2 & MMC_CAP2_CQE_DCMD;
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-30 19:37:38 +08:00
|
|
|
static enum mmc_issue_type mmc_cqe_issue_type(struct mmc_host *host,
|
|
|
|
|
struct request *req)
|
2017-11-29 21:41:04 +08:00
|
|
|
{
|
|
|
|
|
switch (req_op(req)) {
|
|
|
|
|
case REQ_OP_DRV_IN:
|
|
|
|
|
case REQ_OP_DRV_OUT:
|
|
|
|
|
case REQ_OP_DISCARD:
|
|
|
|
|
case REQ_OP_SECURE_ERASE:
|
|
|
|
|
return MMC_ISSUE_SYNC;
|
|
|
|
|
case REQ_OP_FLUSH:
|
|
|
|
|
return mmc_cqe_can_dcmd(host) ? MMC_ISSUE_DCMD : MMC_ISSUE_SYNC;
|
|
|
|
|
default:
|
|
|
|
|
return MMC_ISSUE_ASYNC;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
enum mmc_issue_type mmc_issue_type(struct mmc_queue *mq, struct request *req)
|
|
|
|
|
{
|
2017-11-29 21:41:04 +08:00
|
|
|
struct mmc_host *host = mq->card->host;
|
|
|
|
|
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
if (mq->use_cqe && !host->hsq_enabled)
|
2017-11-29 21:41:04 +08:00
|
|
|
return mmc_cqe_issue_type(host, req);
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
if (req_op(req) == REQ_OP_READ || req_op(req) == REQ_OP_WRITE)
|
|
|
|
|
return MMC_ISSUE_ASYNC;
|
|
|
|
|
|
|
|
|
|
return MMC_ISSUE_SYNC;
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:04 +08:00
|
|
|
static void __mmc_cqe_recovery_notifier(struct mmc_queue *mq)
|
|
|
|
|
{
|
|
|
|
|
if (!mq->recovery_needed) {
|
|
|
|
|
mq->recovery_needed = true;
|
|
|
|
|
schedule_work(&mq->recovery_work);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void mmc_cqe_recovery_notifier(struct mmc_request *mrq)
|
|
|
|
|
{
|
|
|
|
|
struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
|
|
|
|
|
brq.mrq);
|
|
|
|
|
struct request *req = mmc_queue_req_to_req(mqrq);
|
|
|
|
|
struct request_queue *q = req->q;
|
|
|
|
|
struct mmc_queue *mq = q->queuedata;
|
|
|
|
|
unsigned long flags;
|
|
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_lock_irqsave(&mq->lock, flags);
|
2017-11-29 21:41:04 +08:00
|
|
|
__mmc_cqe_recovery_notifier(mq);
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irqrestore(&mq->lock, flags);
|
2017-11-29 21:41:04 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static enum blk_eh_timer_return mmc_cqe_timed_out(struct request *req)
|
|
|
|
|
{
|
|
|
|
|
struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
|
|
|
|
|
struct mmc_request *mrq = &mqrq->brq.mrq;
|
|
|
|
|
struct mmc_queue *mq = req->q->queuedata;
|
|
|
|
|
struct mmc_host *host = mq->card->host;
|
|
|
|
|
enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
|
|
|
|
|
bool recovery_needed = false;
|
|
|
|
|
|
|
|
|
|
switch (issue_type) {
|
|
|
|
|
case MMC_ISSUE_ASYNC:
|
|
|
|
|
case MMC_ISSUE_DCMD:
|
|
|
|
|
if (host->cqe_ops->cqe_timeout(host, mrq, &recovery_needed)) {
|
|
|
|
|
if (recovery_needed)
|
2020-05-08 00:15:33 +08:00
|
|
|
mmc_cqe_recovery_notifier(mrq);
|
2017-11-29 21:41:04 +08:00
|
|
|
return BLK_EH_RESET_TIMER;
|
|
|
|
|
}
|
2020-05-08 14:22:27 +08:00
|
|
|
/* The request has gone already */
|
2018-05-29 21:52:35 +08:00
|
|
|
return BLK_EH_DONE;
|
2017-11-29 21:41:04 +08:00
|
|
|
default:
|
|
|
|
|
/* Timeout is handled by mmc core */
|
|
|
|
|
return BLK_EH_RESET_TIMER;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
static enum blk_eh_timer_return mmc_mq_timed_out(struct request *req,
|
|
|
|
|
bool reserved)
|
|
|
|
|
{
|
2017-11-29 21:41:04 +08:00
|
|
|
struct request_queue *q = req->q;
|
|
|
|
|
struct mmc_queue *mq = q->queuedata;
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
struct mmc_card *card = mq->card;
|
|
|
|
|
struct mmc_host *host = card->host;
|
2017-11-29 21:41:04 +08:00
|
|
|
unsigned long flags;
|
2020-05-08 00:15:33 +08:00
|
|
|
bool ignore_tout;
|
2017-11-29 21:41:04 +08:00
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_lock_irqsave(&mq->lock, flags);
|
2020-05-08 00:15:33 +08:00
|
|
|
ignore_tout = mq->recovery_needed || !mq->use_cqe || host->hsq_enabled;
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irqrestore(&mq->lock, flags);
|
2017-11-29 21:41:04 +08:00
|
|
|
|
2020-05-08 00:15:33 +08:00
|
|
|
return ignore_tout ? BLK_EH_RESET_TIMER : mmc_cqe_timed_out(req);
|
2017-11-29 21:41:04 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void mmc_mq_recovery_handler(struct work_struct *work)
|
|
|
|
|
{
|
|
|
|
|
struct mmc_queue *mq = container_of(work, struct mmc_queue,
|
|
|
|
|
recovery_work);
|
|
|
|
|
struct request_queue *q = mq->queue;
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
struct mmc_host *host = mq->card->host;
|
2017-11-29 21:41:04 +08:00
|
|
|
|
|
|
|
|
mmc_get_card(mq->card, &mq->ctx);
|
|
|
|
|
|
|
|
|
|
mq->in_recovery = true;
|
|
|
|
|
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
if (mq->use_cqe && !host->hsq_enabled)
|
2017-11-29 21:41:07 +08:00
|
|
|
mmc_blk_cqe_recovery(mq);
|
|
|
|
|
else
|
|
|
|
|
mmc_blk_mq_recovery(mq);
|
2017-11-29 21:41:04 +08:00
|
|
|
|
|
|
|
|
mq->in_recovery = false;
|
|
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_lock_irq(&mq->lock);
|
2017-11-29 21:41:04 +08:00
|
|
|
mq->recovery_needed = false;
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irq(&mq->lock);
|
2017-11-29 21:41:04 +08:00
|
|
|
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
if (host->hsq_enabled)
|
|
|
|
|
host->cqe_ops->cqe_recovery_finish(host);
|
|
|
|
|
|
2017-11-29 21:41:04 +08:00
|
|
|
mmc_put_card(mq->card, &mq->ctx);
|
|
|
|
|
|
|
|
|
|
blk_mq_run_hw_queues(q, true);
|
2017-11-29 21:41:03 +08:00
|
|
|
}
|
|
|
|
|
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
static struct scatterlist *mmc_alloc_sg(int sg_len, gfp_t gfp)
|
2011-07-10 05:12:36 +08:00
|
|
|
{
|
|
|
|
|
struct scatterlist *sg;
|
|
|
|
|
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
sg = kmalloc_array(sg_len, sizeof(*sg), gfp);
|
2017-03-13 20:36:36 +08:00
|
|
|
if (sg)
|
2011-07-10 05:12:36 +08:00
|
|
|
sg_init_table(sg, sg_len);
|
|
|
|
|
|
|
|
|
|
return sg;
|
|
|
|
|
}
|
|
|
|
|
|
2011-06-28 22:16:02 +08:00
|
|
|
static void mmc_queue_setup_discard(struct request_queue *q,
|
|
|
|
|
struct mmc_card *card)
|
|
|
|
|
{
|
|
|
|
|
unsigned max_discard;
|
|
|
|
|
|
|
|
|
|
max_discard = mmc_calc_max_discard(card);
|
|
|
|
|
if (!max_discard)
|
|
|
|
|
return;
|
|
|
|
|
|
2018-03-08 09:10:10 +08:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
|
2015-07-14 22:15:12 +08:00
|
|
|
blk_queue_max_discard_sectors(q, max_discard);
|
2011-06-28 22:16:02 +08:00
|
|
|
q->limits.discard_granularity = card->pref_erase << 9;
|
|
|
|
|
/* granularity must not be greater than max. discard */
|
|
|
|
|
if (card->pref_erase > max_discard)
|
|
|
|
|
q->limits.discard_granularity = 0;
|
2013-04-18 20:41:55 +08:00
|
|
|
if (mmc_can_secure_erase_trim(card))
|
2018-03-08 09:10:10 +08:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
|
2011-06-28 22:16:02 +08:00
|
|
|
}
|
|
|
|
|
|
2019-08-28 20:35:43 +08:00
|
|
|
static unsigned int mmc_get_max_segments(struct mmc_host *host)
|
|
|
|
|
{
|
|
|
|
|
return host->can_dma_map_merge ? MMC_DMA_MAP_MERGE_SEGMENTS :
|
|
|
|
|
host->max_segs;
|
|
|
|
|
}
|
|
|
|
|
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
/**
|
|
|
|
|
* mmc_init_request() - initialize the MMC-specific per-request data
|
2020-07-01 20:46:54 +08:00
|
|
|
* @mq: the request queue
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
* @req: the request
|
|
|
|
|
* @gfp: memory allocation policy
|
|
|
|
|
*/
|
2017-11-29 21:41:03 +08:00
|
|
|
static int __mmc_init_request(struct mmc_queue *mq, struct request *req,
|
|
|
|
|
gfp_t gfp)
|
2016-11-29 18:09:13 +08:00
|
|
|
{
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
struct mmc_queue_req *mq_rq = req_to_mmc_queue_req(req);
|
|
|
|
|
struct mmc_card *card = mq->card;
|
|
|
|
|
struct mmc_host *host = card->host;
|
2016-11-29 18:09:13 +08:00
|
|
|
|
2019-08-28 20:35:43 +08:00
|
|
|
mq_rq->sg = mmc_alloc_sg(mmc_get_max_segments(host), gfp);
|
2017-09-20 16:56:14 +08:00
|
|
|
if (!mq_rq->sg)
|
|
|
|
|
return -ENOMEM;
|
2016-11-29 18:09:13 +08:00
|
|
|
|
2016-11-29 18:09:15 +08:00
|
|
|
return 0;
|
|
|
|
|
}
|
2016-11-29 18:09:13 +08:00
|
|
|
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
static void mmc_exit_request(struct request_queue *q, struct request *req)
|
2016-11-29 18:09:15 +08:00
|
|
|
{
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
struct mmc_queue_req *mq_rq = req_to_mmc_queue_req(req);
|
2017-03-13 20:36:36 +08:00
|
|
|
|
mmc: core: Allocate per-request data using the block layer core
The mmc_queue_req is a per-request state container the MMC core uses
to carry bounce buffers, pointers to asynchronous requests and so on.
Currently allocated as a static array of objects, then as a request
comes in, a mmc_queue_req is assigned to it, and used during the
lifetime of the request.
This is backwards compared to how other block layer drivers work:
they usally let the block core provide a per-request struct that get
allocated right beind the struct request, and which can be obtained
using the blk_mq_rq_to_pdu() helper. (The _mq_ infix in this function
name is misleading: it is used by both the old and the MQ block
layer.)
The per-request struct gets allocated to the size stored in the queue
variable .cmd_size initialized using the .init_rq_fn() and
cleaned up using .exit_rq_fn().
The block layer code makes the MMC core rely on this mechanism to
allocate the per-request mmc_queue_req state container.
Doing this make a lot of complicated queue handling go away. We only
need to keep the .qnct that keeps count of how many request are
currently being processed by the MMC layer. The MQ block layer will
replace also this once we transition to it.
Doing this refactoring is necessary to move the ioctl() operations
into custom block layer requests tagged with REQ_OP_DRV_[IN|OUT]
instead of the custom code using the BigMMCHostLock that we have
today: those require that per-request data be obtainable easily from
a request after creating a custom request with e.g.:
struct request *rq = blk_get_request(q, REQ_OP_DRV_IN, __GFP_RECLAIM);
struct mmc_queue_req *mq_rq = req_to_mq_rq(rq);
And this is not possible with the current construction, as the request
is not immediately assigned the per-request state container, but
instead it gets assigned when the request finally enters the MMC
queue, which is way too late for custom requests.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
[Ulf: Folded in the fix to drop a call to blk_cleanup_queue()]
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Tested-by: Heiner Kallweit <hkallweit1@gmail.com>
2017-05-18 17:29:32 +08:00
|
|
|
kfree(mq_rq->sg);
|
|
|
|
|
mq_rq->sg = NULL;
|
2016-11-29 18:09:14 +08:00
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
static int mmc_mq_init_request(struct blk_mq_tag_set *set, struct request *req,
|
|
|
|
|
unsigned int hctx_idx, unsigned int numa_node)
|
|
|
|
|
{
|
|
|
|
|
return __mmc_init_request(set->driver_data, req, GFP_KERNEL);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void mmc_mq_exit_request(struct blk_mq_tag_set *set, struct request *req,
|
|
|
|
|
unsigned int hctx_idx)
|
|
|
|
|
{
|
|
|
|
|
struct mmc_queue *mq = set->driver_data;
|
|
|
|
|
|
|
|
|
|
mmc_exit_request(mq->queue, req);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static blk_status_t mmc_mq_queue_rq(struct blk_mq_hw_ctx *hctx,
|
|
|
|
|
const struct blk_mq_queue_data *bd)
|
|
|
|
|
{
|
|
|
|
|
struct request *req = bd->rq;
|
|
|
|
|
struct request_queue *q = req->q;
|
|
|
|
|
struct mmc_queue *mq = q->queuedata;
|
|
|
|
|
struct mmc_card *card = mq->card;
|
2017-11-29 21:41:04 +08:00
|
|
|
struct mmc_host *host = card->host;
|
2017-11-29 21:41:03 +08:00
|
|
|
enum mmc_issue_type issue_type;
|
|
|
|
|
enum mmc_issued issued;
|
2017-11-29 21:41:04 +08:00
|
|
|
bool get_card, cqe_retune_ok;
|
2017-11-29 21:41:03 +08:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
|
|
if (mmc_card_removed(mq->card)) {
|
|
|
|
|
req->rq_flags |= RQF_QUIET;
|
|
|
|
|
return BLK_STS_IOERR;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
issue_type = mmc_issue_type(mq, req);
|
|
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_lock_irq(&mq->lock);
|
2017-11-29 21:41:03 +08:00
|
|
|
|
2018-08-21 20:05:55 +08:00
|
|
|
if (mq->recovery_needed || mq->busy) {
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irq(&mq->lock);
|
2017-11-29 21:41:04 +08:00
|
|
|
return BLK_STS_RESOURCE;
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
switch (issue_type) {
|
2017-11-29 21:41:04 +08:00
|
|
|
case MMC_ISSUE_DCMD:
|
|
|
|
|
if (mmc_cqe_dcmd_busy(mq)) {
|
|
|
|
|
mq->cqe_busy |= MMC_CQE_DCMD_BUSY;
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irq(&mq->lock);
|
2017-11-29 21:41:04 +08:00
|
|
|
return BLK_STS_RESOURCE;
|
|
|
|
|
}
|
|
|
|
|
break;
|
2017-11-29 21:41:03 +08:00
|
|
|
case MMC_ISSUE_ASYNC:
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
/*
|
|
|
|
|
* For MMC host software queue, we only allow 2 requests in
|
|
|
|
|
* flight to avoid a long latency.
|
|
|
|
|
*/
|
|
|
|
|
if (host->hsq_enabled && mq->in_flight[issue_type] > 2) {
|
|
|
|
|
spin_unlock_irq(&mq->lock);
|
|
|
|
|
return BLK_STS_RESOURCE;
|
|
|
|
|
}
|
2017-11-29 21:41:03 +08:00
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
/*
|
|
|
|
|
* Timeouts are handled by mmc core, and we don't have a host
|
|
|
|
|
* API to abort requests, so we can't handle the timeout anyway.
|
|
|
|
|
* However, when the timeout happens, blk_mq_complete_request()
|
|
|
|
|
* no longer works (to stop the request disappearing under us).
|
|
|
|
|
* To avoid racing with that, set a large timeout.
|
|
|
|
|
*/
|
|
|
|
|
req->timeout = 600 * HZ;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
2018-08-21 20:05:55 +08:00
|
|
|
/* Parallel dispatch of requests is not supported at the moment */
|
|
|
|
|
mq->busy = true;
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
mq->in_flight[issue_type] += 1;
|
|
|
|
|
get_card = (mmc_tot_in_flight(mq) == 1);
|
2017-11-29 21:41:04 +08:00
|
|
|
cqe_retune_ok = (mmc_cqe_qcnt(mq) == 1);
|
2017-11-29 21:41:03 +08:00
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irq(&mq->lock);
|
2017-11-29 21:41:03 +08:00
|
|
|
|
|
|
|
|
if (!(req->rq_flags & RQF_DONTPREP)) {
|
|
|
|
|
req_to_mmc_queue_req(req)->retries = 0;
|
|
|
|
|
req->rq_flags |= RQF_DONTPREP;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (get_card)
|
|
|
|
|
mmc_get_card(card, &mq->ctx);
|
|
|
|
|
|
2017-11-29 21:41:04 +08:00
|
|
|
if (mq->use_cqe) {
|
|
|
|
|
host->retune_now = host->need_retune && cqe_retune_ok &&
|
|
|
|
|
!host->hold_retune;
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
blk_mq_start_request(req);
|
|
|
|
|
|
|
|
|
|
issued = mmc_blk_mq_issue_rq(mq, req);
|
|
|
|
|
|
|
|
|
|
switch (issued) {
|
|
|
|
|
case MMC_REQ_BUSY:
|
|
|
|
|
ret = BLK_STS_RESOURCE;
|
|
|
|
|
break;
|
|
|
|
|
case MMC_REQ_FAILED_TO_START:
|
|
|
|
|
ret = BLK_STS_IOERR;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
ret = BLK_STS_OK;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (issued != MMC_REQ_STARTED) {
|
|
|
|
|
bool put_card = false;
|
|
|
|
|
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_lock_irq(&mq->lock);
|
2017-11-29 21:41:03 +08:00
|
|
|
mq->in_flight[issue_type] -= 1;
|
|
|
|
|
if (mmc_tot_in_flight(mq) == 0)
|
|
|
|
|
put_card = true;
|
2018-08-21 20:05:55 +08:00
|
|
|
mq->busy = false;
|
2018-11-16 16:10:06 +08:00
|
|
|
spin_unlock_irq(&mq->lock);
|
2017-11-29 21:41:03 +08:00
|
|
|
if (put_card)
|
|
|
|
|
mmc_put_card(card, &mq->ctx);
|
2018-08-21 20:05:55 +08:00
|
|
|
} else {
|
|
|
|
|
WRITE_ONCE(mq->busy, false);
|
2017-11-29 21:41:03 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static const struct blk_mq_ops mmc_mq_ops = {
|
|
|
|
|
.queue_rq = mmc_mq_queue_rq,
|
|
|
|
|
.init_request = mmc_mq_init_request,
|
|
|
|
|
.exit_request = mmc_mq_exit_request,
|
|
|
|
|
.complete = mmc_blk_mq_complete,
|
|
|
|
|
.timeout = mmc_mq_timed_out,
|
|
|
|
|
};
|
|
|
|
|
|
2017-09-22 20:36:57 +08:00
|
|
|
static void mmc_setup_queue(struct mmc_queue *mq, struct mmc_card *card)
|
|
|
|
|
{
|
|
|
|
|
struct mmc_host *host = card->host;
|
2019-02-28 00:02:11 +08:00
|
|
|
unsigned block_size = 512;
|
2017-09-22 20:36:57 +08:00
|
|
|
|
2018-03-08 09:10:10 +08:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_NONROT, mq->queue);
|
|
|
|
|
blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, mq->queue);
|
2017-09-22 20:36:57 +08:00
|
|
|
if (mmc_can_erase(card))
|
|
|
|
|
mmc_queue_setup_discard(mq->queue, card);
|
|
|
|
|
|
2019-06-25 17:20:41 +08:00
|
|
|
if (!mmc_dev(host)->dma_mask || !*mmc_dev(host)->dma_mask)
|
|
|
|
|
blk_queue_bounce_limit(mq->queue, BLK_BOUNCE_HIGH);
|
2017-09-22 20:36:57 +08:00
|
|
|
blk_queue_max_hw_sectors(mq->queue,
|
|
|
|
|
min(host->max_blk_count, host->max_req_size / 512));
|
2019-08-28 20:35:43 +08:00
|
|
|
if (host->can_dma_map_merge)
|
|
|
|
|
WARN(!blk_queue_can_use_dma_map_merging(mq->queue,
|
|
|
|
|
mmc_dev(host)),
|
|
|
|
|
"merging was advertised but not possible");
|
|
|
|
|
blk_queue_max_segments(mq->queue, mmc_get_max_segments(host));
|
2019-02-28 00:02:11 +08:00
|
|
|
|
|
|
|
|
if (mmc_card_mmc(card))
|
|
|
|
|
block_size = card->ext_csd.data_sector_size;
|
|
|
|
|
|
|
|
|
|
blk_queue_logical_block_size(mq->queue, block_size);
|
2019-08-28 20:35:43 +08:00
|
|
|
/*
|
|
|
|
|
* After blk_queue_can_use_dma_map_merging() was called with succeed,
|
|
|
|
|
* since it calls blk_queue_virt_boundary(), the mmc should not call
|
|
|
|
|
* both blk_queue_max_segment_size().
|
|
|
|
|
*/
|
|
|
|
|
if (!host->can_dma_map_merge)
|
|
|
|
|
blk_queue_max_segment_size(mq->queue,
|
2019-02-28 00:02:11 +08:00
|
|
|
round_down(host->max_seg_size, block_size));
|
2017-09-22 20:36:57 +08:00
|
|
|
|
2019-06-06 03:08:27 +08:00
|
|
|
dma_set_max_seg_size(mmc_dev(host), queue_max_segment_size(mq->queue));
|
|
|
|
|
|
2017-11-29 21:41:04 +08:00
|
|
|
INIT_WORK(&mq->recovery_work, mmc_mq_recovery_handler);
|
2017-11-29 21:41:03 +08:00
|
|
|
INIT_WORK(&mq->complete_work, mmc_blk_mq_complete_work);
|
|
|
|
|
|
|
|
|
|
mutex_init(&mq->complete_lock);
|
|
|
|
|
|
|
|
|
|
init_waitqueue_head(&mq->wait);
|
|
|
|
|
}
|
|
|
|
|
|
2019-09-12 12:13:55 +08:00
|
|
|
static inline bool mmc_merge_capable(struct mmc_host *host)
|
|
|
|
|
{
|
|
|
|
|
return host->caps2 & MMC_CAP2_MERGE_CAPABLE;
|
|
|
|
|
}
|
|
|
|
|
|
2018-11-15 00:02:16 +08:00
|
|
|
/* Set queue depth to get a reasonable value for q->nr_requests */
|
|
|
|
|
#define MMC_QUEUE_DEPTH 64
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* mmc_init_queue - initialise a queue structure.
|
|
|
|
|
* @mq: mmc queue
|
|
|
|
|
* @card: mmc card to attach this queue
|
|
|
|
|
*
|
|
|
|
|
* Initialise a MMC card request queue.
|
|
|
|
|
*/
|
2018-11-16 16:10:06 +08:00
|
|
|
int mmc_init_queue(struct mmc_queue *mq, struct mmc_card *card)
|
2017-11-29 21:41:03 +08:00
|
|
|
{
|
2018-11-15 00:02:16 +08:00
|
|
|
struct mmc_host *host = card->host;
|
2017-11-29 21:41:03 +08:00
|
|
|
int ret;
|
|
|
|
|
|
2018-11-15 00:02:16 +08:00
|
|
|
mq->card = card;
|
|
|
|
|
mq->use_cqe = host->cqe_enabled;
|
2018-11-16 16:10:06 +08:00
|
|
|
|
|
|
|
|
spin_lock_init(&mq->lock);
|
2018-11-15 00:02:16 +08:00
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
memset(&mq->tag_set, 0, sizeof(mq->tag_set));
|
2018-11-15 00:02:16 +08:00
|
|
|
mq->tag_set.ops = &mmc_mq_ops;
|
|
|
|
|
/*
|
|
|
|
|
* The queue depth for CQE must match the hardware because the request
|
|
|
|
|
* tag is used to index the hardware queue.
|
|
|
|
|
*/
|
mmc: Add MMC host software queue support
Now the MMC read/write stack will always wait for previous request is
completed by mmc_blk_rw_wait(), before sending a new request to hardware,
or queue a work to complete request, that will bring context switching
overhead and spend some extra time to poll the card for busy completion
for I/O writes via sending CMD13, especially for high I/O per second
rates, to affect the IO performance.
Thus this patch introduces MMC software queue interface based on the
hardware command queue engine's interfaces, which is similar with the
hardware command queue engine's idea, that can remove the context
switching. Moreover we set the default queue depth as 64 for software
queue, which allows more requests to be prepared, merged and inserted
into IO scheduler to improve performance, but we only allow 2 requests
in flight, that is enough to let the irq handler always trigger the
next request without a context switch, as well as avoiding a long latency.
Moreover the host controller should support HW busy detection for I/O
operations when enabling the host software queue. That means, the host
controller must not complete a data transfer request, until after the
card stops signals busy.
From the fio testing data in cover letter, we can see the software
queue can improve some performance with 4K block size, increasing
about 16% for random read, increasing about 90% for random write,
though no obvious improvement for sequential read and write.
Moreover we can expand the software queue interface to support MMC
packed request or packed command in future.
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Baolin Wang <baolin.wang7@gmail.com>
Link: https://lore.kernel.org/r/4409c1586a9b3ed20d57ad2faf6c262fc3ccb6e2.1581478568.git.baolin.wang7@gmail.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
2020-02-12 12:12:56 +08:00
|
|
|
if (mq->use_cqe && !host->hsq_enabled)
|
2018-11-15 00:02:16 +08:00
|
|
|
mq->tag_set.queue_depth =
|
|
|
|
|
min_t(int, card->ext_csd.cmdq_depth, host->cqe_qdepth);
|
|
|
|
|
else
|
|
|
|
|
mq->tag_set.queue_depth = MMC_QUEUE_DEPTH;
|
2017-11-29 21:41:03 +08:00
|
|
|
mq->tag_set.numa_node = NUMA_NO_NODE;
|
2019-02-15 19:13:24 +08:00
|
|
|
mq->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_BLOCKING;
|
2017-11-29 21:41:03 +08:00
|
|
|
mq->tag_set.nr_hw_queues = 1;
|
|
|
|
|
mq->tag_set.cmd_size = sizeof(struct mmc_queue_req);
|
|
|
|
|
mq->tag_set.driver_data = mq;
|
|
|
|
|
|
2019-08-28 20:35:43 +08:00
|
|
|
/*
|
|
|
|
|
* Since blk_mq_alloc_tag_set() calls .init_request() of mmc_mq_ops,
|
|
|
|
|
* the host->can_dma_map_merge should be set before to get max_segs
|
|
|
|
|
* from mmc_get_max_segments().
|
|
|
|
|
*/
|
2019-09-12 12:13:55 +08:00
|
|
|
if (mmc_merge_capable(host) &&
|
|
|
|
|
host->max_segs < MMC_DMA_MAP_MERGE_SEGMENTS &&
|
2019-08-28 20:35:43 +08:00
|
|
|
dma_get_merge_boundary(mmc_dev(host)))
|
|
|
|
|
host->can_dma_map_merge = 1;
|
|
|
|
|
else
|
|
|
|
|
host->can_dma_map_merge = 0;
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
ret = blk_mq_alloc_tag_set(&mq->tag_set);
|
|
|
|
|
if (ret)
|
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
|
|
mq->queue = blk_mq_init_queue(&mq->tag_set);
|
|
|
|
|
if (IS_ERR(mq->queue)) {
|
|
|
|
|
ret = PTR_ERR(mq->queue);
|
|
|
|
|
goto free_tag_set;
|
|
|
|
|
}
|
|
|
|
|
|
2019-07-22 12:03:06 +08:00
|
|
|
if (mmc_host_is_spi(host) && host->use_spi_crc)
|
|
|
|
|
mq->queue->backing_dev_info->capabilities |=
|
|
|
|
|
BDI_CAP_STABLE_WRITES;
|
|
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
mq->queue->queuedata = mq;
|
2018-11-15 00:02:16 +08:00
|
|
|
blk_queue_rq_timeout(mq->queue, 60 * HZ);
|
2017-11-29 21:41:03 +08:00
|
|
|
|
2018-11-15 00:02:16 +08:00
|
|
|
mmc_setup_queue(mq, card);
|
2017-11-29 21:41:03 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
|
|
free_tag_set:
|
|
|
|
|
blk_mq_free_tag_set(&mq->tag_set);
|
|
|
|
|
return ret;
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:18 +08:00
|
|
|
void mmc_queue_suspend(struct mmc_queue *mq)
|
2017-11-29 21:41:03 +08:00
|
|
|
{
|
|
|
|
|
blk_mq_quiesce_queue(mq->queue);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The host remains claimed while there are outstanding requests, so
|
|
|
|
|
* simply claiming and releasing here ensures there are none.
|
|
|
|
|
*/
|
|
|
|
|
mmc_claim_host(mq->card->host);
|
|
|
|
|
mmc_release_host(mq->card->host);
|
|
|
|
|
}
|
|
|
|
|
|
2017-11-29 21:41:18 +08:00
|
|
|
void mmc_queue_resume(struct mmc_queue *mq)
|
2017-11-29 21:41:03 +08:00
|
|
|
{
|
|
|
|
|
blk_mq_unquiesce_queue(mq->queue);
|
|
|
|
|
}
|
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
void mmc_cleanup_queue(struct mmc_queue *mq)
|
|
|
|
|
{
|
2007-07-24 15:28:11 +08:00
|
|
|
struct request_queue *q = mq->queue;
|
2006-11-15 05:08:16 +08:00
|
|
|
|
2017-11-29 21:41:18 +08:00
|
|
|
/*
|
|
|
|
|
* The legacy code handled the possibility of being suspended,
|
|
|
|
|
* so do that here too.
|
|
|
|
|
*/
|
|
|
|
|
if (blk_queue_quiesced(q))
|
|
|
|
|
blk_mq_unquiesce_queue(q);
|
2010-01-09 06:43:00 +08:00
|
|
|
|
2017-11-29 21:40:59 +08:00
|
|
|
blk_cleanup_queue(q);
|
2019-05-03 03:07:14 +08:00
|
|
|
blk_mq_free_tag_set(&mq->tag_set);
|
2017-11-29 21:40:59 +08:00
|
|
|
|
2017-11-29 21:41:03 +08:00
|
|
|
/*
|
|
|
|
|
* A request can be completed before the next request, potentially
|
|
|
|
|
* leaving a complete_work with nothing to do. Such a work item might
|
|
|
|
|
* still be queued at this point. Flush it.
|
|
|
|
|
*/
|
|
|
|
|
flush_work(&mq->complete_work);
|
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
mq->card = NULL;
|
|
|
|
|
}
|
|
|
|
|
|
2008-07-22 20:35:42 +08:00
|
|
|
/*
|
|
|
|
|
* Prepare the sg list(s) to be handed of to the host driver
|
|
|
|
|
*/
|
2011-07-10 05:12:36 +08:00
|
|
|
unsigned int mmc_queue_map_sg(struct mmc_queue *mq, struct mmc_queue_req *mqrq)
|
2007-05-12 06:26:16 +08:00
|
|
|
{
|
2017-05-19 21:37:27 +08:00
|
|
|
struct request *req = mmc_queue_req_to_req(mqrq);
|
2007-05-12 06:26:16 +08:00
|
|
|
|
2017-09-20 16:56:14 +08:00
|
|
|
return blk_rq_map_sg(mq->queue, req, mqrq->sg);
|
2007-05-12 06:26:16 +08:00
|
|
|
}
|