If one runs a simple fio request with random read/write with a
20%/80% ratio, the numbers are incredibly bad when using the CFQ scheduler.
IOmeter | | | |
64K, randrw | NOOP | CFQ | deadline |
randrwmix=80 | | | |
--------------+-------+------+----------+
blkback |103/27 |32/10 | 102/27 |
--------------+-------+------+----------+
QEMU qdisk |103/27 |102/27| 102/27 |
The problem as explained by Vivek Goyal was:
".. that difference is that sync vs async requests. In the case of
a kernel thread submitting IO, [..] all the WRITES might be being
considered as async and will go in a different queue. If you mix those
with some READS, they are always sync and will go in differnet queue.
In presence of sync queue, CFQ will idle and choke up WRITES in
an attempt to improve latencies of READs.
In case of AIO [note: this is what QEMU qdisk is doing] , [..]
it is direct IO and both READS and WRITES will be considered SYNC
and will go in a single queue and no choking of WRITES will take place."
The solution is quite simple, tack on REQ_SYNC (which is
what the WRITE_ODIRECT macro points to) and the numbers go
back up.
Suggested-by: Vivek Goyal <vgoyal@redhat.com
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We used to the plug/unplug on the submit_bio. But that means
if within a stream of WRITE, WRITE, WRITE,...,WRITE we have
one READ, it could stall the pipeline (as the 'submio_bio'
could trigger the unplug_fnc to be called and stall/sync
when doing the READ). Instead we want to move the unplugging
when the whole (or as a much as possible) ring buffer has been
processed. This also eliminates us doing plug/unplug for
each request.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
And also shorten the name if it has blkback to blkbk.
This results in the symbol table (if compiled in the kernel)
to be much shorter, prettier, and also easier to search for.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
There is no need for it, as the address is updated constatly
in the root of the Linux kernel.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Daniel Stodden suggested to eliminate vbd.c and interface.c, inlining the
critical bits where they belong, respectively.
Leaving only blkback.c for the data- and xenbus.c for the control path.
Suggested-by: Daniel Stodden <daniel.stodden@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
They were used to check if the queue does not have QUEUE_FLAG_DEAD
set. That is not necessary anymore as the 'submit_io' call
ends up doing that for us.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
After the commit 0faa8cca88
(" xen/blkback: remove per-queue plugging") we forgot
to retrieve the 'struct request_queue' from the block device.
This puts the functionality back in and fixes a NULL pointer
bug.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The commit 976222e05e
xen/blkback: Move the check for misaligned I/O higher.
moved it a bit to high. The preq->vbdev was not set, so the
check for misaligned I/O would cause a NULL pointer derefence.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The previous name ('fast_flush_area') had nothing to do with what it
does right now. Changing the names so that the code dealing with
mapping pages in and out of the guest is called xen_blkbk_[map|unmap].
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We move it up higher to be in same loop that actually computes
the sector number.
This way, all of the code that deals with verifying that the
request is correct is all done before we do any of the
page mapping, I/O submission, etc.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We take out the chunk of code dealing with mapping to the guest
of pages into the xen_blk_map_buf code. And we also move the
vbd_translate to be done much earlier.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Moving it so that the code that 'fast_flush_area' code is
close to the code that deals with it so that the reader
won't lose focus.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We seperate the bio allocation (bio_alloc) from the bio submission so
that the error paths are much easier, and also so that the bio
submission can be done in one tight loop. It also makes the
plug/unplug calls much much easier.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
commit 7eaceaccab ("block: remove per-queue plugging")
added two new interfaces to plug and unplug: blk_start_plug
and blk_finish_plug. Lets use those.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
+ sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring_area->addr;
WARNING: line over 80 characters
+ BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
as breaking them up really does not help that much.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
checkpatch.pl suggested that we don't use the typdef in common.h
and this triggered this avalanche of patches.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The WRITE_BARRIER was missing the REQ_WRITE option. This
was causing the blktap to die.
Signed-off-by: Tom Goetz <tom.goetz@virtualcomputer.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
The patch titled:"xen/blkback: Use 'vzalloc' for page arrays and pre-allocate pages."
allocates the structures and its member variables using the 'vzalloc'.
Daniel Stodden pointed out that vzalloc is good when we use
big number of pages - while these are at the max two pages.
We can do this using kzalloc. Also the GFP_HIGHMEM does not
work properly with Xen, so take that out.
We will have to revisit this when a "get_empty_pages_and_pagevec"
type API shows up to leverage that.
BugLink: http://mid.gmane.org/1299898639.11681.227.camel@agari.van.xensource.com
CC: Daniel Stodden <daniel.stodden@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Instead of doing copy grants lets do mapping grants using
the M2P(and P2M) override mechanism.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Conflicts:
drivers/xen/blkback/blkback.c
Previously we would allocate the array for page using 'kmalloc' which
we can as easily do with 'vzalloc'. The pre-allocation of pages
was done a bit differently in the past - it used to be that
the balloon driver would export "alloc_empty_pages_and_pagevec"
which would have in one function created an array, allocated
the pages, balloned the pages out (so the memory behind those
pages would be non-present), and provide us those pages.
This was OK as those pages were shared between other guest and
the only thing we needed was to "swizzel" the MFN of those pages
to point to the other guest MFN. We can still "swizzel" the MFNs
using the M2P (and P2M) override API calls, but for the sake of
simplicity we are dropping the balloon API calls. We can return
to those later on.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Following in the steps of patch:
"xen: Union the blkif_request request specific fields" this patch
changes the blkback. Per the original patch:
"Prepare for extending the block device ring to allow request
specific fields, by moving the request specific fields for
reads, writes and barrier requests to a union member."
Cc: Owen Smith <owen.smith@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Bundle the lot of discrete variables into a single structure.
This is based on what was done in the xen-netback driver:
xen: netback: Move global/static variables into struct xen_netbk.
(094944631cc5a9d6e623302c987f78117c0bf7ac)
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cherry-pick and modified from 69d64727c42eecd47fdf82c15a54474d21a4012a
("blkback/blktap2: simplify address translations"):
"There are quite a number of places where e.g. page->va->page
translations happen.
Besides yielding smaller code (source and binary), a second goal is to
make it easier to determine where virtual addresses of pages allocated
through alloc_empty_pages_and_pagevec() are really used (in turn in
order to determine whether using highmem pages would be possible
there)."
The second goal is not the purpose of this patch - it is just to
make it easier to read the code.
linux-2.6-pvops:
* Stripped drivers/xen/gntdev/*
* Stripped drivers/xen/netback/*
[v2: Stripped blktap off]
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Daniel Stodden <daniel.stodden@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
A guest can cause the backend driver to leak a kernel thread. Such
leaked threads hold references to the device, whichmakes the device
impossible to tear down. If shut down, the guest remains a zombie
domain, the xenwatch process hangs, and most xm commands will stop
working.
This patch tries to do the following for blkback:
- identify/extract idempotent teardown operations,
- add/move the invocation of said teardown operation
right before we're about to allocate new resources in the
Connected states.
[ linux-2.6.18-xen.hg 59f097ef181b ]
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Keir Fraser <keir@xen.org>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
First cut at flushing blkback data when first connecting
blkback. This should avoid the pygrub issues we are experiencing
in (RedHat bugzilla) 466681.
[ 2.6.18-xen.hg commit 63b4d7f56688 ]
Signed-off-by: Chris Lalancette <clalance@redhat.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Support dynamic resizing of virtual block devices. This patch supports
both file backed block devices as well as physical devices that can be
dynamically resized on the host side.
Signed-off-by: K. Y. Srinivasan <ksrinivasan@novell.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
blkbk is rather generic for a modular distro style kernel.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
We neglect to check the return value of xenbus_register_backend
and take actions when that fails. This patch fixes that and adds
code to deal with those type of failures.
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Impact: build fix
I'm not sure if blkback should be using these functions, but in the
meantime export them to allow blkback to be a module.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Conflicts:
drivers/xen/Makefile
* 'for-linus' of git://oss.sgi.com/xfs/xfs:
xfs: use proper interfaces for on-stack plugging
xfs: fix xfs_debug warnings
xfs: fix variable set but not used warnings
xfs: convert log tail checking to a warning
xfs: catch bad block numbers freeing extents.
xfs: push the AIL from memory reclaim and periodic sync
xfs: clean up code layout in xfs_trans_ail.c
xfs: convert the xfsaild threads to a workqueue
xfs: introduce background inode reclaim work
xfs: convert ENOSPC inode flushing to use new syncd workqueue
xfs: introduce a xfssyncd workqueue
xfs: fix extent format buffer allocation size
xfs: fix unreferenced var error in xfs_buf.c
Also, applied patch from Tony Luck that fixes ia64:
xfs_destroy_workqueues() should not be tagged with__exit
in the branch before merging.