it is not possible to specify rerror, werror and readonly arguments
for drive as drive_init allows/blocks them based on the interface (if=)
specified and none isn't white-listed there.
MORITA Kazutaka [Wed, 26 May 2010 02:35:36 +0000 (11:35 +0900)]
add support for protocol driver create_options
This patch enables protocol drivers to use their create options which
are not supported by the format. For example, protcol drivers can use
a backing_file option with raw format.
Kevin Wolf [Wed, 12 May 2010 14:23:26 +0000 (16:23 +0200)]
qcow2: Fix error handling in l2_allocate
l2_allocate has some intermediate states in which the image is inconsistent.
Change the order to write to the L1 table only after the new L2 table has
successfully been initialized.
Also reset the L2 cache in failure case, it's very likely wrong.
Kevin Wolf [Tue, 13 Apr 2010 09:43:27 +0000 (11:43 +0200)]
qcow2: Clear L2 table cache after write error
If the L2 table was already updated in cache, but writing it to disk has
failed, we must not continue using the changed version in the cache to stay
consistent with what's on the disk.
Alexander Graf [Wed, 26 May 2010 19:04:32 +0000 (21:04 +0200)]
Make cache=unsafe the default for -snapshot
When using -snapshot we don't care about data integrity of the cow file
at all, so let's disable flushing there and squeeze out the last drop
of performance we could possibly get.
- remove unused host state and store pci bus pointer only
- do not map host state access into unused 1fe.10000000 range
- reorder pci region registration
- assign pci i/o region to isa_mem_base
Alexander Graf [Wed, 26 May 2010 15:51:49 +0000 (17:51 +0200)]
Add cache=unsafe parameter to -drive
Usually the guest can tell the host to flush data to disk. In some cases we
don't want to flush though, but try to keep everything in cache.
So let's add a new cache value to -drive that allows us to set the cache
policy to most aggressive, disabling flushes. We call this mode "unsafe",
as guest data is not guaranteed to survive host crashes anymore.
This patch also adds a noop function for aio, so we can do nothing in AIO
fashion.
Gerd Hoffmann [Fri, 21 May 2010 09:54:34 +0000 (11:54 +0200)]
vnc: rich cursor support.
Uses VNC_ENCODING_RICH_CURSOR. Adding XCURSOR support should be
possible without much trouble. Shouldn't be needed though as
RICH_CURSOR is a superset of XCURSOR.
Kevin Wolf [Mon, 17 May 2010 08:36:47 +0000 (10:36 +0200)]
Fix error handling in qemu_read_config_file
We need to close the file even in error case. While at it, make the callers
catch all kind of errors. ENOENT is allowed for default config files, they
are optional.
Jan Kiszka [Sun, 23 May 2010 08:29:34 +0000 (10:29 +0200)]
sdl: Do not disable screensaver by default
Unless we are running in full-screen mode, QEMU's SDL window should not
disable the host's screensaver. The user can still change this behaviour
by setting the environment variable SDL_VIDEO_ALLOW_SCREENSAVER as
desired.
- two pairs of softmmu indexes bind softmmu tlb to cpu tlb in fault handlers
using value of DMMU primary and secondary context registers, so we need to
flush softmmu translations when context registers are changed
sparc64: fix mmu context at trap levels above zero
- cpu_mmu_index return MMU_NUCLEUS_IDX if trap level is not zero
- cpu_get_tb_cpu_state: store trap level and primary context in flags
this allows to restart code translation when address translation is changed
- stop translation block after writing to pstate and tl registers
- stop translation block after writing to alternate space
this can be optimized to stop only if address translation can be changed
by write operation (e.g. by comparing with MMU ASI values)
- refactor code to handle hpstate only if available for current cpu
- conditionally set hypervisor bit in hpstate register
- reorder softmmu indices so user accessable ones go first, translation context
macros supervisor() and hypervisor() adjusted as well
- disable sparcv8 registers for TARGET_SPARC64
- fix cpu_mmu_index to use sparcv9 bits only
sparc64: generate data access exception on RW violation
- separate PRIV and PROT handling
- DPRINTF_MMU macro to clean up debug code
- dump mmu_idx, trap level and mmu context registers
along with address translation values
Artyom Tarasenko [Sat, 22 May 2010 08:38:56 +0000 (10:38 +0200)]
sparc32 protect read-only bits in DMA CSR registers
On a real hardware changing read-only bits has no effect
Use a mask common for SCSI and Ethernet registers. The crucial
bit is DMA_INTR, because setting or clearing it may produce
spurious interrupts.
cirrus_post_load() will be executed twice when loading vm states and then the
wrong physical memory will be registered. This issue may lead to crash qemu.
Stuart Brady [Wed, 12 May 2010 19:42:04 +0000 (20:42 +0100)]
Fix tarbin Makefile rule
The 'tarbin' Makefile rule doesn't include qemu-system-sparc64, but
should do, now that sparc64-softmmu is in the default target list.
The rule attempts to tar up binaries that were not built if a target
list was passed to the configure script -- in which case, it will
either fail, or otherwise include binaries from previous builds.
Fix both problems once and for all by building a list of binaries to
include in the tarball, using the list of targets to be built.
Jan Kiszka [Sat, 15 May 2010 11:03:28 +0000 (13:03 +0200)]
Put dependency files in proper subdir
This seems to resolve subtle breakages of our build system:
Dependency files generated for targets like 'dir/foo.o' were saved as
'foo.d'. Now, if there was also a target 'foo.o', one of the dependency
file was overwritten. Concrete example: libhw*/macio.o vs.
libhw*/ide/macio.o. And this often left a segfaulting build result
behind when changing the "wrong" data structures".
Riku Voipio [Fri, 7 May 2010 12:28:05 +0000 (12:28 +0000)]
linux-user: do not warn for missing pselect6
Libc will fallback gracefully if pselect6 is not available. Thus put
pselect6 to nowarn until the atomicity issues of the original pselect6
patch are dealt with.
(1) The output registers were not marked call-clobbered, even though
they can be modified by called functions.
(2) The thread pointer was not marked reserved.
(3) R4-R6 are call-saved, but not saved by the prologue. Rather than
save them, mark them reserved so that we don't use them.
Now that the prologue is generated after GUEST_BASE is fixed,
we can load it as an immediate, and also avoid reserving the
register if it isn't necessary.
Alpha passes oldset by value in a register, and returns the newset
as the return value; as compared to the standard implementation in
which both are passed by reference. This requires being able to
distinguish negative return values that are not errors. Do this in
the same way as the Alpha Linux kernel, by storing a zero in V0 in
the implementation of the syscall.
At the same time, fix a think-o in the regular sigprocmask path in
which we passed the target, rather than the host, HOW value.
At the same time, tidy the code wrt MIPS and SH4 which have the
same two register return mechanism. Fix confusion between pipe
and pipe2 with an explicit flags=0, when the guest will not be
using the two register return mechanism.
Define OPC_CALL_Jz, generated by tcg_out_calli; use the later
throughout. Unify the calls within qemu_st; adjust the stack
with a single pop if applicable.
Add more OPC values, and tgen_arithr. Use the later throughout.
Note that normal reg/reg arithmetic now uses the Gv,Ev opcode form
instead of the Ev,Gv opcode form used previously. Both forms
disassemble properly, and so there's no visible change when diffing
log files before and after the change. This change makes the operand
ordering within the output routines more natural, and avoids the need
to define an OPC_ARITH_EvGv since a read-modify-write with memory is
not needed within TCG.
Rabin Vincent [Sun, 2 May 2010 09:50:51 +0000 (15:20 +0530)]
arm_timer: reload timer when enabled
Reload the timer when TimerControl is written, if the timer is to be
enabled. Otherwise, if an earlier write to TimerLoad was done while
periodic mode was not set, s->delta may incorrectly still have the value
of the maximum limit instead of the value written to TimerLoad.
This problem is evident on versatileap on current linux-next, which
enables TIMER_CTRL_32BIT before writing to TimerLoad and then enabling
periodic mode and starting the timer. This causes the first periodic
tick to be scheduled to occur after 0xffffffff periods, leading to a
perceived hang while the kernel waits for the first timer tick.
block: Add SG_IO device check in refresh_total_sectors()
This patch adds a special case check for scsi-generic devices in
refresh_total_sectors() to skip the subsequent BlockDriver->bdrv_getlength()
that will be returning -ESPIPE from block/raw-posic.c:raw_getlength() for
BlockDriverState->sg=1 devices.
block: Make find_image_format() return 'raw' BlockDriver for SG_IO devices
This patch adds a special BlockDriverState->sg check in block.c:find_image_format()
after bdrv_file_open() -> block/raw-posix.c:hdev_open() has been called to determine
if we are dealing with a Linux host scsi-generic device.
The patch then returns the BlockDriver * from bdrv_find_format("raw"), skipping the
subsequent bdrv_read() and rest of find_image_format().
block: fix sector comparism in multiwrite_req_compare
The difference between the start sectors of two requests can be larger
than the size of the "int" type, which can lead to a not correctly
sorted multiwrite array and thus spurious I/O errors and filesystem
corruption due to incorrect request merges.
So instead of doing the cute sector arithmetics trick spell out the
exact comparisms.
Spotted by Kevin Wolf based on a testcase from Michael Tokarev.
Before issuing the barrier to the block driver we need to flush our oustanding
queue of write requests, as the flush is supposed to be issued after them.
Stefan Hajnoczi [Fri, 14 May 2010 21:52:30 +0000 (22:52 +0100)]
virtio-blk: Avoid zeroing every request structure
The VirtIOBlockRequest structure is about 40 KB in size. This patch
avoids zeroing every request by only initializing fields that are read.
The other fields are either written to or may not be used at all.
Oprofile shows about 10% of CPU samples in memset called by
virtio_blk_alloc_request(). The workload is
dd if=/dev/vda of=/dev/null iflag=direct bs=8k running concurrently 4
times. This patch makes memset disappear to the bottom of the profile.
target-sparc: Inline some generation of carry for ADDX/SUBX.
Computing carry is trivial for some inputs. By avoiding an
external function call, we generate near-optimal code for
the common cases of add+addx (double-word arithmetic) and
cmp+addx (a setcc pattern).
If the address register overlaps one of the output registers
simply issue the clobbering load last, rather than emitting
an extra move of the address register.
Use int32 types instead of target_ulong when computing ICC. This
simplifies the generated code for 32-bit host and 64-bit guest.
Use the same simplified expressions for ICC as were already used
for XCC in carry flag generation.
Simplify the ADD carry generation to not consider a possible carry-in.
Use the more complex carry computation for ADDX only. Use the same
carry algorithm for the XCC result of ADDX. Similarly for SUB/SUBX.
Use the ADD carry generation functions for TADD/TADDTV. Similarly
for SUB and TSUB/TSUBTV.
The 'quit' Monitor command (implemented by do_quit()) calls
exit() directly, this is problematic under QMP because QEMU
exits before having a chance to send the ok response.
Clients don't know if QEMU exited because of a problem or
because the 'quit' command has been executed.
This commit fixes that by making do_quit() use
qemu_system_shutdown_request(), so that we exit gracefully.
Thanks to Paolo Bonzini <[email protected]> for suggesting
this solution.
Revert "PCI: Convert pci_device_hot_add() to QObject"
Short story: We don't want pci_add in QMP. Long story follows.
pci_add can do two things:
* Hot plug a PCI NIC. device_add is more general.
* Hot plug a PCI disk controller, and a drive connected to it.
The controller is either virtio-blk-pci (if=virtio) or lsi53c895a
(if=scsi). With the latter, the drive is optional. Use drive_add to
hotplug additional SCSI drives. Except drive_add is not available in
QMP.
device_add is more general for controllers and the guest part of
drives. I'm working on a more general alternative for the host part
of drives.
Why am I proposing to remove pci_add from QMP before its replacement is
ready? I want it out sooner rather than later, because it isn't fully
functional (errors and drive_add are missing), and we do not plan to
complete the job. In other words, it's not really usable over QMP now,
and it's not what we want for QMP anyway. Since we don't want it to be
used over QMP, we should take it out, not leave it around as a trap for
the uninitiated.
Dan Berrange confirmed that libvirt has no need for pci_add & friends
over QMP.
projects / qemu.git / log