AS scheduler alternates between issuing read and write batches. It does
the batch switch only after all requests from the previous batch are
completed.
When switching to a write batch, if there is an on-going read request,
it waits for its completion and indicates its intention of switching by
setting ad->changed_batch and the new direction but does not update the
batch_expire_time for the new write batch which it does in the case of
no previous pending requests.
On completion of the read request, it sees that we were waiting for the
switch and schedules work for kblockd right away and resets the
ad->changed_data flag.
Now when kblockd enters dispatch_request where it is expected to pick
up a write request, it in turn ends the write batch because the
batch_expire_timer was not updated and shows the expire timestamp for
the previous batch.
This results in the write starvation for all the cases where there is
the intention for switching to a write batch, but there is a previous
in-flight read request and the batch gets reverted to a read_batch
right away.
This also holds true in the reverse case (switching from a write batch
to a read batch with an in-flight write request).
I've checked that this bug exists on 2.6.11, 2.6.18, 2.6.24 and
linux-2.6-block git HEAD. I've tested the fix on x86 platforms with
SCSI drives where the driver asks for the next request while a current
request is in-flight.
This patch is based off linux-2.6-block git HEAD.
Bug reproduction:
A simple scenario which reproduces this bug is:
- dd if=/dev/hda3 of=/dev/null &
- lilo
The lilo takes forever to complete.
This can also be reproduced fairly easily with the earlier dd and
another test
program doing msync().
The example test program below should print out a message after every
iteration
but it simply hangs forever. With this bugfix it makes forward progress.
====
Example test program using msync() (thanks to suleiman AT google DOT
com)
inline uint64_t
rdtsc(void)
{
int64_t tsc;
__asm __volatile("rdtsc" : "=A" (tsc));
return (tsc);
}
int
main(int argc, char **argv)
{
struct stat st;
uint64_t e, s, t;
char *p, q;
long i;
int fd;
if (argc < 2) {
printf("Usage: %s <file>\n", argv[0]);
return (1);
}
if ((fd = open(argv[1], O_RDWR | O_NOATIME)) < 0)
err(1, "open");
if (fstat(fd, &st) < 0)
err(1, "fstat");
p = mmap(NULL, st.st_size, PROT_READ | PROT_WRITE,
MAP_SHARED, fd, 0);
t = 0;
for (i = 0; i < 1000; i++) {
*p = 0;
msync(p, 4096, MS_SYNC);
s = rdtsc();
*p = 0;
__asm __volatile(""::: "memory");
e = rdtsc();
if (argc > 2)
printf("%d: %lld cycles %jd %jd\n",
i, e - s, (intmax_t)s, (intmax_t)e);
t += e - s;
}
printf("average time: %lld cycles\n", t / 1000);
return (0);
}
Cc: <stable@kernel.org>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
It blindly copies everything in the io_context, including the lock.
That doesn't work so well for either lock ordering or lockdep.
There seems zero point in swapping io contexts on a request to request
merge, so the best point of action is to just remove it.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Since it's acquired from irq context, all locking must be of the
irq safe variant. Most are already inside the queue lock (which
already disables interrupts), but the io scheduler rmmod path
always has irqs enabled and the put_io_context() path may legally
be called with irqs enabled (even if it isn't usually). So fixup
those two.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
If the two requests belong to the same io context, we will attempt
to lock the same lock twice. But swapping contexts is pointless in
that case, so just check for rioc == nioc before doing the double
lock and copy.
Tested-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
elv_register() always returns 0, and there isn't anything it does where
it should return an error (the only error condition is so grave that
it's handled with a BUG_ON).
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
New write batches currently start from where the last one completed.
We have no idea where the head is after switching batches, so this
makes little sense. Instead, start the next batch from the request
with the earliest deadline in the hope that we avoid a deadline
expiry later on.
Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Two comments refer to deadlines applying to reads only. This is
not the case.
Signed-off-by: Aaron Carroll <aaronc@gelato.unsw.edu.au>
Acked-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Some of the code has been gradually transitioned to using the proper
struct request_queue, but there's lots left. So do a full sweet of
the kernel and get rid of this typedef and replace its uses with
the proper type.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
kmalloc_node() and kmem_cache_alloc_node() were not available in a zeroing
variant in the past. But with __GFP_ZERO it is possible now to do zeroing
while allocating.
Use __GFP_ZERO to remove the explicit clearing of memory via memset whereever
we can.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Switch the kblockd flushing from a global flush to a more specific
flush_work().
(akpm: bypassed maintainers, sorry. There are other patches which depend on
this)
Cc: "Maciej W. Rozycki" <macro@linux-mips.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Jens Axboe <axboe@suse.de>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix units mismatch (jiffies vs msecs) in as-iosched.c, spotted by Xiaoning
Ding <dingxn@cse.ohio-state.edu>.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We implemented the missing bits to allow this some time ago, and
they are integrated in AS. So remove the __module_get() to allow
the module to be unloaded.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Conflicts:
drivers/infiniband/core/iwcm.c
drivers/net/chelsio/cxgb2.c
drivers/net/wireless/bcm43xx/bcm43xx_main.c
drivers/net/wireless/prism54/islpci_eth.c
drivers/usb/core/hub.h
drivers/usb/input/hid-core.c
net/core/netpoll.c
Fix up merge failures with Linus's head and fix new compilation failures.
Signed-Off-By: David Howells <dhowells@redhat.com>
- ->init_queue() does not need the elevator passed in
- ->put_request() is a hot path and need not have the queue passed in
- cfq_update_io_seektime() does not need cfqd passed in
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Pass the work_struct pointer to the work function rather than context data.
The work function can use container_of() to work out the data.
For the cases where the container of the work_struct may go away the moment the
pending bit is cleared, it is made possible to defer the release of the
structure by deferring the clearing of the pending bit.
To make this work, an extra flag is introduced into the management side of the
work_struct. This governs auto-release of the structure upon execution.
Ordinarily, the work queue executor would release the work_struct for further
scheduling or deallocation by clearing the pending bit prior to jumping to the
work function. This means that, unless the driver makes some guarantee itself
that the work_struct won't go away, the work function may not access anything
else in the work_struct or its container lest they be deallocated.. This is a
problem if the auxiliary data is taken away (as done by the last patch).
However, if the pending bit is *not* cleared before jumping to the work
function, then the work function *may* access the work_struct and its container
with no problems. But then the work function must itself release the
work_struct by calling work_release().
In most cases, automatic release is fine, so this is the default. Special
initiators exist for the non-auto-release case (ending in _NAR).
Signed-Off-By: David Howells <dhowells@redhat.com>
All on stack DECLARE_COMPLETIONs should be replaced by:
DECLARE_COMPLETION_ONSTACK
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
CFQ implements this on its own now, but it's really block layer
knowledge. Tells a device queue to start dispatching requests to
the driver, taking care to unplug if needed. Also fixes the issue
where as/cfq will invoke a stopped queue, which we really don't
want.
Signed-off-by: Jens Axboe <axboe@suse.de>
It's ok if the read path is a lot more costly, as long as inc/dec is
really cheap. The inc/dec will happen for each created/freed io context,
while the reading only happens when a disk queue exits.
Signed-off-by: Jens Axboe <axboe@suse.de>
Get rid of the as_rq request type. With the added elevator_private2, we
have enough room in struct request to get rid of any arq allocation/free
for each request.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Right now, every IO scheduler implements its own backmerging (except for
noop, which does no merging). That results in duplicated code for
essentially the same operation, which is never a good thing. This patch
moves the backmerging out of the io schedulers and into the elevator
core. We save 1.6kb of text and as a bonus get backmerging for noop as
well. Win-win!
Signed-off-by: Jens Axboe <axboe@suse.de>
Right now ->flags is a bit of a mess: some are request types, and
others are just modifiers. Clean this up by splitting it into
->cmd_type and ->cmd_flags. This allows introduction of generic
Linux block message types, useful for sending generic Linux commands
to block devices.
Signed-off-by: Jens Axboe <axboe@suse.de>
They all duplicate macros to check for empty root and/or node, and
clearing a node. So put those in rbtree.h.
Signed-off-by: Jens Axboe <axboe@suse.de>
A process flag to indicate whether we are doing sync io is incredibly
ugly. It also causes performance problems when one does a lot of async
io and then proceeds to sync it. Part of the io will go out as async,
and the other part as sync. This causes a disconnect between the
previously submitted io and the synced io. For io schedulers such as CFQ,
this will cause us lost merges and suboptimal behaviour in scheduling.
Remove PF_SYNCWRITE completely from the fsync/msync paths, and let
the O_DIRECT path just directly indicate that the writes are sync
by using WRITE_SYNC instead.
Signed-off-by: Jens Axboe <axboe@suse.de>
Use hlist instead of list_head for request hashtable in deadline-iosched
and as-iosched. It also can remove the flag to know hashed or unhashed.
Signed-off-by: Akinobu Mita <mita@miraclelinux.com>
Signed-off-by: Jens Axboe <axboe@suse.de>
block/as-iosched.c | 45 +++++++++++++++++++--------------------------
block/deadline-iosched.c | 39 ++++++++++++++++-----------------------
2 files changed, 35 insertions(+), 49 deletions(-)
* git://git.infradead.org/~dwmw2/rbtree-2.6:
[RBTREE] Switch rb_colour() et al to en_US spelling of 'color' for consistency
Update UML kernel/physmem.c to use rb_parent() accessor macro
[RBTREE] Update hrtimers to use rb_parent() accessor macro.
[RBTREE] Add explicit alignment to sizeof(long) for struct rb_node.
[RBTREE] Merge colour and parent fields of struct rb_node.
[RBTREE] Remove dead code in rb_erase()
[RBTREE] Update JFFS2 to use rb_parent() accessor macro.
[RBTREE] Update eventpoll.c to use rb_parent() accessor macro.
[RBTREE] Update key.c to use rb_parent() accessor macro.
[RBTREE] Update ext3 to use rb_parent() accessor macro.
[RBTREE] Change rbtree off-tree marking in I/O schedulers.
[RBTREE] Add accessor macros for colour and parent fields of rb_node
There's a race between shutting down one io scheduler and firing up the
next, in which a new io could enter and cause the io scheduler to be
invoked with bad or NULL data.
To fix this, we need to maintain the queue lock for a bit longer.
Unfortunately we cannot do that, since the elevator init requires to be
run without the lock held. This isn't easily fixable, without also
changing the mempool API. So split the initialization into two parts,
and alloc-init operation and an attach operation. Then we can
preallocate the io scheduler and related structures, and run the attach
inside the lock after we detach the old one.
This patch has survived 30 minutes of 1 second io scheduler switching
with a very busy io load.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
They were abusing the rb_color field to mark nodes which weren't currently
on the tree. Fix that to use the same method as eventpoll did -- setting
the parent pointer to point back to itself. And use the appropriate
accessor macros for setting and reading the parent.
Signed-off-by: David Woodhouse <dwmw2@infradead.org>
On rmmod path, cfq/as waits to make sure all io-contexts was
freed. However, it's using complete(), not wait_for_completion().
I think barrier() is not enough in here. To avoid the following case,
this patch replaces barrier() with smb_wmb().
cpu0 visibility cpu1
[ioc_gnone=NULL,ioc_count=1]
ioc_gnone = &all_gone NULL,ioc_count=1
atomic_read(&ioc_count) NULL,ioc_count=1
wait_for_completion() NULL,ioc_count=0 atomic_sub_and_test()
NULL,ioc_count=0 if ( && ioc_gone)
[ioc_gone==NULL,
so doesn't call complete()]
&all_gone,ioc_count=0
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Jens Axboe <axboe@suse.de>
Unlike other ioscheds, as-iosched handles alias by chaing them using
rq->queuelist. As aliased requests are very rare in the first place,
this complicates merge/dispatch handling without meaningful
performance improvement. This patch updates as-iosched to dump
aliased requests into dispatch queue as other ioscheds do.
Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jens Axboe <axboe@suse.de>
Kill the arq->state poison statement in as_add_request(), it can trigger
for perfectly valid code that just reuses a request after io completion
instead of freeing it and allocating a new one. We probably should
introduce a blk_init_request() to start from scratch, but for now just
kill it as we will be removing the as specific poisoning soon.
Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some leftover comments referring to drivers/block that are now block/.
They don't add any information we don't already have, so kill them.
Signed-off-by: Coywolf Qi Hunt <qiyong@fc-cn.com>
Signed-off-by: Jens Axboe <axboe@suse.de>
drivers/block/ is right now a mix of core and driver parts. Lets move
the core parts to a new top level directory. Al will move the fs/
related block parts to block/ next.
Signed-off-by: Jens Axboe <axboe@suse.de>
