The character backend refactoring introduced an undesirable busy wait.
The busy wait happens if can_read returns zero and there is data available
on the character device's file descriptor. Then, the I/O watch will
fire continuously and, with TCG, the CPU thread will never run.
1) Char backend asks front end if it can write
2) Front end says no
3) poll() finds the char backend's descriptor is available
4) Goto (1)
What we really want is this (note that step 3 avoids the busy wait):
1) Char backend asks front end if it can write
2) Front end says no
3) poll() goes on without char backend's descriptor
4) Goto (1) until qemu_chr_accept_input() called
5) Char backend asks front end if it can write
6) Front end says yes
7) poll() finds the char backend's descriptor is available
8) Backend handler called
After this patch, the IOWatchPoll source and the watch source are
separated. The IOWatchPoll is simply a hook that runs during the prepare
phase on each main loop iteration. The hook adds/removes the actual
source depending on the return value from can_read.
A simple reproducer is
qemu-system-i386 -serial mon:stdio
... followed by banging on the terminal as much as you can. :) Without
this patch, emulation will hang.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1365177573-11817-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
# By Peter Crosthwaite (2) and others
# Via Stefan Hajnoczi
* stefanha/trivial-patches:
xilinx_zynq: Cleanup ssi_create_slave
petalogix_ml605_mmu: Cleanup ssi_create_slave()
target-s390: Fix SRNMT
linux-user: Don't omit comma for strace of rt_sigaction()
test-visitor-serialization: Fix some memory leaks
# By Alex Bligh (2) and Felipe Franciosi (2)
# Via Stefano Stabellini
* sstabellini/xen-2013-04-05:
Allow xen guests to plug disks of 1 TiB or more
Introduce 64 bit integer write interface to xenstore
Xen PV backend: Disable use of O_DIRECT by default as it results in crashes.
Xen PV backend: Move call to bdrv_new from blk_init to blk_connect
# By Stefan Hajnoczi (4) and Kevin Wolf (3)
# Via Kevin Wolf
* kwolf/for-anthony:
qcow2: Fix L1 write error handling in qcow2_update_snapshot_refcount
qcow2: Return real error in qcow2_update_snapshot_refcount
block: clean up I/O throttling wait_time code
block: drop duplicated slice extension code
block: keep I/O throttling slice time constant
block: fix I/O throttling accounting blind spot
usb-storage: Forward serial number to scsi-disk
It ignored the error code, and at least the 'goto fail' is obvious
nonsense as it creates an endless loop (if the next attempt doesn't
magically succeed) and leaves the in-memory L1 table in big-endian
instead of converting it back.
In error cases, there's no point in writing an updated L1 table, so
skip this part for them.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The wait_time variable is in seconds. Reflect this in a comment and use
NANOSECONDS_PER_SECOND instead of BLOCK_IO_SLICE_TIME * 10 (which
happens to have the right value).
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The current slice is extended when an I/O request exceeds the limit.
There is no need to extend the slice every time we check a request.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It is not necessary to adjust the slice time at runtime. We already
extend the current slice in order to carry over accounting into the next
slice. Changing the actual slice time value introduces oscillations.
The guest may experience large changes in throughput or IOPS from one
moment to the next when slice times are adjusted.
Reported-by: Benoît Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I/O throttling relies on bdrv_acct_done() which is called when a request
completes. This leaves a blind spot since we only charge for completed
requests, not submitted requests.
For example, if there is 1 operation remaining in this time slice the
guest could submit 3 operations and they will all be submitted
successfully since they don't actually get accounted for until they
complete.
Originally we probably thought this is okay since the requests will be
accounted when the time slice is extended. In practice it causes
fluctuations since the guest can exceed its I/O limit and it will be
punished for this later on.
Account for I/O upon submission so that I/O limits are enforced
properly.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
usb-storage takes care to fetch the USB serial number from -drive
options, but it neglected to pass its own 'serial' property to the
scsi-disk it creates. With this patch, the 'serial' qdev property and
the 'serial' option in -drive behave the same and correctly apply the
serial number on both USB and SCSI level.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
The current xen backend driver implementation uses int64_t variables
to store the size of the corresponding backend disk/file. It also uses
an int64_t variable to store the block size of that image. When writing
the number of sectors (file_size/block_size) to xenstore, however, it
passes these values as 32 bit signed integers. This will cause an
overflow for any disk of 1 TiB or more.
This patch changes the xen backend driver to use a 64 bit integer write
xenstore function.
Signed-off-by: Felipe Franciosi <felipe@paradoxo.org>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
The current implementation of xen_backend only provides 32 bit integer
functions to write to xenstore. This patch adds two functions that
allow writing 64 bit integers (one generic function and another for
the backend only).
This patch also fixes the size of the char arrays used to represent
these integers as strings (originally 32 bytes, however no more than
12 bytes are needed for 32 bit integers and no more than 21 bytes are
needed for 64 bit integers).
Signed-off-by: Felipe Franciosi <felipe@paradoxo.org>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Due to what is almost certainly a kernel bug, writes with O_DIRECT may
continue to reference the page after the write has been marked as
completed, particularly in the case of TCP retransmit. In other
scenarios, this "merely" risks data corruption on the write, but with
Xen pages from domU are only transiently mapped into dom0's memory,
resulting in kernel panics when they are subsequently accessed.
This brings PV devices in line with emulated devices. Removing
O_DIRECT is safe as barrier operations are now correctly passed
through.
See:
http://lists.xen.org/archives/html/xen-devel/2012-12/msg01154.html
for more details.
Signed-off-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
This commit delays the point at which bdrv_new (and hence blk_open
on the underlying device) is called from blk_init to blk_connect.
This ensures that in an inbound live migrate, the block device is
not opened until it has been closed at the other end. This is in
preparation for supporting devices with open/close consistency
without using O_DIRECT. This commit does NOT itself change O_DIRECT
semantics.
Signed-off-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
# By Luiz Capitulino
# Via Luiz Capitulino
* luiz/queue/qmp:
chardev: drop the Memory chardev driver
hmp: human-monitor-command: stop using the Memory chardev driver
Monitor: Make output buffer dynamic
qstring: add qstring_get_length()
With the recent m25p80 cleanup there is no need to use
ssi_create_slave_no_init() anymore. Just use ssi_create_slave().
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
With the recent m25p80 cleanup there is no need to use
ssi_create_slave_no_init() anymore. Just use ssi_create_slave().
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Fallthough into abort = oops.
Cc: qemu-trivial@nongnu.org
Cc: Alexander Graf <agraf@suse.de>
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
It's not used anymore since the last commit.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
The Memory chardev driver was added because, as the Monitor's output
buffer was static, we needed a way to accumulate the output of an
HMP commmand when ran by human-monitor-command.
However, the Monitor's output buffer is now dynamic, so it's possible
for the human-monitor-command to use it instead of the Memory chardev
driver.
This commit does that change, but there are two important
observations about it:
1. We need a way to signal to the Monitor that it shouldn't call
chardev functions when flushing its output. This is done
by adding a new flag to the Monitor object called skip_flush
(which is set to true by qmp_human_monitor_command())
2. The current code has buffered semantics: QMP clients will
only see a command's output if it flushes its output with
a new-line character. This commit changes this to unbuffered,
which means that QMP clients will see a command's output
whenever the command prints anything.
I don't think this will matter in practice though, as I believe
all HMP commands print the new-line character anyway.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
Commit f628926bb4 changed monitor_flush()
to retry on qemu_chr_fe_write() errors. However, the Monitor's output
buffer can keep growing while the retry is not issued and this can
cause the buffer to overflow.
To reproduce this issue, just start qemu and type on the Monitor:
(qemu) ?
This will cause an assertion to trig.
To fix this problem this commit makes the Monitor buffer dynamic,
which means that it can grow as much as needed.
Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Gerd Hoffmann <kraxel@redhat.com>
Pass the 'last' parameter of print_signal() through to
print_raw_param(); this fixes a problem where we weren't printing
the comma separator for strace of rt_sigaction() when the signal
was an unnamed (ie realtime) one:
6856 rt_sigaction(230xf6fff870,0xf6fff8fc) = 0
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch fixes some of the memory leaks in test-visitor-serialization but not all of them.
Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Allow the clock_gettime() code using monotonic clock to be utilized on
more POSIX compliannt OS's. This started as a fix for OpenBSD which was
listed in one function as part of the previous hard coded list of OS's
for the functions to support but not in the other.
Signed-off-by: Brad Smith <brad@comstyle.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 20130405003748.GH884@rox.home.comstyle.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
A common dependency of the constant's current users:
- hw/apic_common.c
- hw/i386/kvmvapic.c
- target-i386/cpu.c
is "target-i386/cpu.h".
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1363821803-3380-9-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The new function acpi_table_install() installs any blob the caller passes
in. In the next patches this function will be promoted from helper role to
extern.
Reimplementing the logic should make it easier to understand. It also
removes a buffer overflow when
has_header &&
cumulative_file_size < ACPI_TABLE_HDR_SIZE - ACPI_TABLE_PFX_SIZE
(In that case the g_realloc() call in the read() loop used to shrink the
"acpi_tables" array, causing an out-of-bounds read access when copying the
header out of "acpi_tables".)
The new code isn't more daring alignment-wise than its predecessor:
"acpi_table_header" is packed, and the uint32_t fields are at offsets 6,
26, and 34.
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1363821803-3380-7-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
As one consequence, strtok() -- which modifies its argument -- is replaced
with g_strsplit().
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1363821803-3380-6-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The upcoming changes will need a cleanup section at the end of the
function, plus OptsVisitor reports errors via Error. For now keep
channeling any Errors to stderr.
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1363821803-3380-4-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The data is binary, not textual.
Also, acpi_table_add() abuses the "char *f" pointer -- which normally
points to file names to load -- to poke into the table. Introduce "char
unsigned *table_start" for that purpose.
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1363821803-3380-3-git-send-email-lersek@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Alberto Garcia <agarcia@igalia.com>
Message-id: 1364412581-3672-4-git-send-email-hdegoede@redhat.com
Cc: Alberto Garcia <agarcia@igalia.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
chardev-frontends need to explictly check, increase and decrement the
avail_connections "property" of the chardev when they are not using a
qdev-chardev-property for the chardev.
This fixes things like:
qemu-kvm -chardev stdio,id=foo -device isa-serial,chardev=foo \
-mon chardev=foo
Working, where they should fail. Most of the changes here are due to
old hardware emulation code which is using serial_hds directly rather then
a qdev-chardev-property.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Message-id: 1364412581-3672-3-git-send-email-hdegoede@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Add qemu_chr_fe_claim / _release helper functions for properly dealing with
avail_connections.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Message-id: 1364412581-3672-2-git-send-email-hdegoede@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
When the conditions blocking receiving are cleared, check for buffered rx
packets.
Signed-off-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
Signed-off-by: Edgar E. Iglesias <edgar.iglesias@gmail.com>
Otherwise make will refuse to build updated .po files.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
While investigating why a 32 bit Windows 2003 guest wasn't able to
successfully perform a shutdown /h, it was discovered that commit
afafe4bbe0 inadvertently dropped the
initialization of the s4_val used to handle s4 shutdown.
Initialize the value as before.
Signed-off-by: Bruce Rogers <brogers@suse.com>
Message-id: 1364928100-487-1-git-send-email-brogers@suse.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
commit 5ec01c2e96 broke "-cpu ..,enforce",
as it has moved kvm_check_features_against_host() after the
filter_features_for_kvm() call. filter_features_for_kvm() removes all
features not supported by the host, so this effectively made
kvm_check_features_against_host() impossible to fail.
This patch changes the call so we check for host feature support before
filtering the feature bits.
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Message-id: 1364935692-24004-1-git-send-email-ehabkost@redhat.com
Cc: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
# By Hanweidong (1) and Stefano Stabellini (1)
# Via Stefano Stabellini
* sstabellini/xen-2013-04-03:
xen-mapcache: pass the right size argument to test_bits
xen-mapcache: replace last_address_index with a last_entry pointer
Compute the correct size for test_bits().
qemu_get_ram_ptr() and qemu_safe_ram_ptr() will call xen_map_cache()
with size is 0 if the requested address is in the RAM. Then
xen_map_cache() will pass the size 0 to test_bits() for checking if the
corresponding pfn was mapped in cache. But test_bits() will always
return 1 when size is 0 without any bit testing. Actually, for this
case, test_bits should check one bit. So this patch introduced a
__test_bit_size which is greater than 0 and a multiple of XC_PAGE_SIZE,
then test_bits can work correctly with __test_bit_size
>> XC_PAGE_SHIFT as its size.
Signed-off-by: Zhenguo Wang <wangzhenguo@huawei.com>
Signed-off-by: Weidong Han <hanweidong@huawei.com>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Replace last_address_index and last_address_vaddr with a single pointer
to the last MapCacheEntry used.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
