Peter Maydell [Tue, 4 Oct 2016 10:01:39 +0000 (11:01 +0100)]
Merge remote-tracking branch 'remotes/famz/tags/for-upstream' into staging
# gpg: Signature made Sun 02 Oct 2016 02:49:58 BST
# gpg: using RSA key 0xCA35624C6A9171C6
# gpg: Good signature from "Fam Zheng <[email protected]>"
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: 5003 7CB7 9706 0F76 F021 AD56 CA35 624C 6A91 71C6
* remotes/famz/tags/for-upstream:
docker: Build in a clean directory
smbios: fix uuid copy
xenpv: Fix qemu_uuid compiling error
Currently we configure and build under "$QEMU_SRC/tests/docker" which is
dubious. Create a fixed directory (to be friendly to ccache) and change
to there before calling build_qemu.
Peter Maydell [Fri, 30 Sep 2016 22:45:56 +0000 (23:45 +0100)]
Merge remote-tracking branch 'remotes/borntraeger/tags/s390x-20160927' into staging
Couple of s390x patches:
- some PCI cleanups
- fix build error due to uuid rework
- fix potential deadlock in sigp handling
- enable ccw devices in BIOS and enforce checking in QEMU
# gpg: Signature made Wed 28 Sep 2016 12:27:03 BST
# gpg: using RSA key 0x117BBC80B5A61C7C
# gpg: Good signature from "Christian Borntraeger (IBM) <[email protected]>"
# Primary key fingerprint: F922 9381 A334 08F9 DBAB FBCA 117B BC80 B5A6 1C7C
* remotes/borntraeger/tags/s390x-20160927:
s390x/kvm: fix build against qemu_uuid
s390x/css: {c,h,t,r,x}sch: require enable AND device number valid
pc-bios/s390-ccw.img: rebuild image
pc-bios/s390-ccw: enable subchannel for IPL I/O devices
s390x/kvm: Fix potential deadlock in sigp handling
s390x/pci: code cleanup
s390x/pci: assign msix io region for each pci device
s390x/pci: re-arrange variable declarations
# gpg: Signature made Thu 29 Sep 2016 13:09:09 BST
# gpg: using RSA key 0x2238EB86D5F797C2
# gpg: Good signature from "Yongbok Kim <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 8600 4CF5 3415 A5D9 4CFA 2B5C 2238 EB86 D5F7 97C2
* remotes/jnsnow/tags/ide-pull-request:
ide: Fix memory leak in ide_register_restart_cb()
MAINTAINERS: Add some more headers to the IDE section
ahci: clear aiocb in ncq_cb
ide: fix DMA register transitions
Fix a memory leak in ide_register_restart_cb() in hw/ide/core.c and add
idebus_unrealize() in hw/ide/qdev.c to have calls to
qemu_del_vm_change_state_handler() to deal with the dangling change
state handler during hot-unplugging ide devices which might lead to a
crash.
John Snow [Mon, 26 Sep 2016 18:33:37 +0000 (14:33 -0400)]
ahci: clear aiocb in ncq_cb
Similar to existing fixes for IDE (87ac25fd) and ATAPI (7f951b2d), the
AIOCB must be cleared in the callback. Otherwise, we may accidentally
try to reset a dangling pointer in bdrv_aio_cancel() from a port reset.
John Snow [Thu, 29 Sep 2016 18:46:15 +0000 (14:46 -0400)]
ide: fix DMA register transitions
ATA8-APT defines the state transitions for both a host controller and
for the hardware device during the lifecycle of a DMA transfer, in
section 9.7 "DMA command protocol."
One of the interesting tidbits here is that when a device transitions
from DDMA0 ("Prepare state") to DDMA1 ("Data_Transfer State"), it can
choose to set either BSY or DRQ to signal this transition, but not both.
as ide_sector_dma_start is the last point in our preparation process
before we begin the real data transfer process (for either AHCI or BMDMA),
this is the correct transition point for DDMA0 to DDMA1.
I have chosen !BSY && DRQ for QEMU to make the transition from DDMA0 the
most obvious.
Since 9c5ce8db, the uuid is wrongly copied, as QemuUUID 'in' argument is
already a pointer.
Fixes ASAN complaining:
hw/smbios/smbios.c:489:5: runtime error: load of address 0x7fffcdb91b00
with insufficient space for an object of type '__int128 unsigned'
* remotes/bonzini/tags/for-upstream: (28 commits)
replay: allow replay stopping and restarting
replay: vmstate for replay module
replay: move internal data to the structure
cpus-common: lock-free fast path for cpu_exec_start/end
tcg: Make tb_flush() thread safe
cpus-common: Introduce async_safe_run_on_cpu()
cpus-common: simplify locking for start_exclusive/end_exclusive
cpus-common: remove redundant call to exclusive_idle()
cpus-common: always defer async_run_on_cpu work items
docs: include formal model for TCG exclusive sections
cpus-common: move exclusive work infrastructure from linux-user
cpus-common: fix uninitialized variable use in run_on_cpu
cpus-common: move CPU work item management to common code
cpus-common: move CPU list management to common code
linux-user: Add qemu_cpu_is_self() and qemu_cpu_kick()
linux-user: Use QemuMutex and QemuCond
cpus: Rename flush_queued_work()
cpus: Move common code out of {async_, }run_on_cpu()
cpus: pass CPUState to run_on_cpu helpers
build-sys: put glib_cflags in QEMU_CFLAGS
...
trace: Properly initialize dynamic event states in hot-plugged vCPUs
Every time a vCPU is hot-plugged, it will "inherit" its tracing state
from the global state array. That is, if *any* existing vCPU has an
event enabled, new vCPUs will have too.
* remotes/kraxel/tags/pull-input-20160928-1:
ps2: do not generate invalid key codes for unknown keys
ps2: use QEMU qcodes instead of scancodes
ps2: allow keycode translation for all scancode sets
ps2: correctly handle 'get/set scancode' command
ps2: reject unknown commands, instead of blindly accepting them
* remotes/kraxel/tags/pull-ui-20160928-1:
ui/vnc-enc-tight: remove switch and have single return
spice/gl: render DisplaySurface via opengl
console: track gl_block state in QemuConsole
console: skip same-size resize
Stefan Hajnoczi [Tue, 27 Sep 2016 15:18:36 +0000 (16:18 +0100)]
linux-aio: fix re-entrant completion processing
Commit 0ed93d84edabc7656f5c998ae1a346fe8b94ca54 ("linux-aio: process
completions from ioq_submit()") added an optimization that processes
completions each time ioq_submit() returns with requests in flight.
This commit introduces a "Co-routine re-entered recursively" error which
can be triggered with -drive format=qcow2,aio=native.
(gdb) bt
#0 0x00007ffff0a046f5 in raise () at /lib64/libc.so.6
#1 0x00007ffff0a062fa in abort () at /lib64/libc.so.6
#2 0x0000555555ac0013 in qemu_coroutine_enter (co=0x5555583464d0) at util/qemu-coroutine.c:113
#3 0x0000555555a4b663 in qemu_laio_process_completions (s=s@entry=0x555557e2f7f0) at block/linux-aio.c:218
#4 0x0000555555a4b874 in ioq_submit (s=s@entry=0x555557e2f7f0) at block/linux-aio.c:331
#5 0x0000555555a4ba12 in laio_do_submit (fd=fd@entry=13, laiocb=laiocb@entry=0x555559d38ae0, offset=offset@entry=2932727808, type=type@entry=1) at block/linux-aio.c:383
#6 0x0000555555a4bbd3 in laio_co_submit (bs=<optimized out>, s=0x555557e2f7f0, fd=13, offset=2932727808, qiov=0x555559d38e20, type=1) at block/linux-aio.c:402
#7 0x0000555555a4fd23 in bdrv_driver_preadv (bs=bs@entry=0x55555663bcb0, offset=offset@entry=2932727808, bytes=bytes@entry=8192, qiov=qiov@entry=0x555559d38e20, flags=0) at block/io.c:804
#8 0x0000555555a52b34 in bdrv_aligned_preadv (bs=bs@entry=0x55555663bcb0, req=req@entry=0x555559d38d20, offset=offset@entry=2932727808, bytes=bytes@entry=8192, align=align@entry=512, qiov=qiov@entry=0x555559d38e20, flags=0) at block/io.c:1041
#9 0x0000555555a52db8 in bdrv_co_preadv (child=<optimized out>, offset=2932727808, bytes=8192, qiov=qiov@entry=0x555559d38e20, flags=flags@entry=0) at block/io.c:1133
#10 0x0000555555a29629 in qcow2_co_preadv (bs=0x555556635890, offset=6178725888, bytes=8192, qiov=0x555557527840, flags=<optimized out>) at block/qcow2.c:1509
#11 0x0000555555a4fd23 in bdrv_driver_preadv (bs=bs@entry=0x555556635890, offset=offset@entry=6178725888, bytes=bytes@entry=8192, qiov=qiov@entry=0x555557527840, flags=0) at block/io.c:804
#12 0x0000555555a52b34 in bdrv_aligned_preadv (bs=bs@entry=0x555556635890, req=req@entry=0x555559d39000, offset=offset@entry=6178725888, bytes=bytes@entry=8192, align=align@entry=1, qiov=qiov@entry=0x555557527840, flags=0) at block/io.c:1041
#13 0x0000555555a52db8 in bdrv_co_preadv (child=<optimized out>, offset=offset@entry=6178725888, bytes=bytes@entry=8192, qiov=qiov@entry=0x555557527840, flags=flags@entry=0) at block/io.c:1133
#14 0x0000555555a4515a in blk_co_preadv (blk=0x5555566356d0, offset=6178725888, bytes=8192, qiov=0x555557527840, flags=0) at block/block-backend.c:783
#15 0x0000555555a45266 in blk_aio_read_entry (opaque=0x5555577025e0) at block/block-backend.c:991
#16 0x0000555555ac0cfa in coroutine_trampoline (i0=<optimized out>, i1=<optimized out>) at util/coroutine-ucontext.c:78
It turned out that re-entrant ioq_submit() and completion processing
between three requests caused this error. The following check is not
sufficient to prevent recursively entering coroutines:
if (laiocb->co != qemu_coroutine_self()) {
qemu_coroutine_enter(laiocb->co);
}
As the following coroutine backtrace shows, not just the current
coroutine (self) can be entered. There might also be other coroutines
that are currently entered and transferred control due to the qcow2 lock
(CoMutex):
(gdb) qemu coroutine 0x5555583464d0
#0 0x0000555555ac0c90 in qemu_coroutine_switch (from_=from_@entry=0x5555583464d0, to_=to_@entry=0x5555572f9890, action=action@entry=COROUTINE_ENTER) at util/coroutine-ucontext.c:175
#1 0x0000555555abfe54 in qemu_coroutine_enter (co=0x5555572f9890) at util/qemu-coroutine.c:117
#2 0x0000555555ac031c in qemu_co_queue_run_restart (co=co@entry=0x5555583462c0) at util/qemu-coroutine-lock.c:60
#3 0x0000555555abfe5e in qemu_coroutine_enter (co=0x5555583462c0) at util/qemu-coroutine.c:119
#4 0x0000555555a4b663 in qemu_laio_process_completions (s=s@entry=0x555557e2f7f0) at block/linux-aio.c:218
#5 0x0000555555a4b874 in ioq_submit (s=s@entry=0x555557e2f7f0) at block/linux-aio.c:331
#6 0x0000555555a4ba12 in laio_do_submit (fd=fd@entry=13, laiocb=laiocb@entry=0x55555a338b40, offset=offset@entry=2911477760, type=type@entry=1) at block/linux-aio.c:383
#7 0x0000555555a4bbd3 in laio_co_submit (bs=<optimized out>, s=0x555557e2f7f0, fd=13, offset=2911477760, qiov=0x55555a338e80, type=1) at block/linux-aio.c:402
#8 0x0000555555a4fd23 in bdrv_driver_preadv (bs=bs@entry=0x55555663bcb0, offset=offset@entry=2911477760, bytes=bytes@entry=8192, qiov=qiov@entry=0x55555a338e80, flags=0) at block/io.c:804
#9 0x0000555555a52b34 in bdrv_aligned_preadv (bs=bs@entry=0x55555663bcb0, req=req@entry=0x55555a338d80, offset=offset@entry=2911477760, bytes=bytes@entry=8192, align=align@entry=512, qiov=qiov@entry=0x55555a338e80, flags=0) at block/io.c:1041
#10 0x0000555555a52db8 in bdrv_co_preadv (child=<optimized out>, offset=2911477760, bytes=8192, qiov=qiov@entry=0x55555a338e80, flags=flags@entry=0) at block/io.c:1133
#11 0x0000555555a29629 in qcow2_co_preadv (bs=0x555556635890, offset=6157475840, bytes=8192, qiov=0x5555575df720, flags=<optimized out>) at block/qcow2.c:1509
#12 0x0000555555a4fd23 in bdrv_driver_preadv (bs=bs@entry=0x555556635890, offset=offset@entry=6157475840, bytes=bytes@entry=8192, qiov=qiov@entry=0x5555575df720, flags=0) at block/io.c:804
#13 0x0000555555a52b34 in bdrv_aligned_preadv (bs=bs@entry=0x555556635890, req=req@entry=0x55555a339060, offset=offset@entry=6157475840, bytes=bytes@entry=8192, align=align@entry=1, qiov=qiov@entry=0x5555575df720, flags=0) at block/io.c:1041
#14 0x0000555555a52db8 in bdrv_co_preadv (child=<optimized out>, offset=offset@entry=6157475840, bytes=bytes@entry=8192, qiov=qiov@entry=0x5555575df720, flags=flags@entry=0) at block/io.c:1133
#15 0x0000555555a4515a in blk_co_preadv (blk=0x5555566356d0, offset=6157475840, bytes=8192, qiov=0x5555575df720, flags=0) at block/block-backend.c:783
#16 0x0000555555a45266 in blk_aio_read_entry (opaque=0x555557231aa0) at block/block-backend.c:991
#17 0x0000555555ac0cfa in coroutine_trampoline (i0=<optimized out>, i1=<optimized out>) at util/coroutine-ucontext.c:78
Use the new qemu_coroutine_entered() function instead of comparing
against qemu_coroutine_self(). This is correct because:
1. If a coroutine is not entered then it must have yielded to wait for
I/O completion. It is therefore safe to enter.
2. If a coroutine is entered then it must be in
ioq_submit()/qemu_laio_process_completions() because otherwise it
would be yielded while waiting for I/O completion. Therefore it will
check laio->ret and return from ioq_submit() instead of yielding,
i.e. it's guaranteed not to hang.
This fixes problems with translated set 1, where most make code were wrong.
This fixes problems with set 3 for extended keys (like arrows) and lot of other keys.
Added a FIXME for set 3, where most keys must not (by default) deliver a break code.
Detailed list of changes on untranslated set 2:
- change of ALTGR break code from 0xe4 to 0xf0 0x08
- change of ALTGR_R break code from 0xe0 0xe4 to 0xe0 0xf0 0x08
- change of F7 make code from 0x02 to 0x83
- change of F7 break code from 0xf0 0x02 to 0xf0 0x83
- change of PRINT make code from 0xe0 0x7c to 0xe0 0x12 0xe0 0x7c
- change of PRINT break code from 0xe0 0xf0 0x7c to 0xe0 0xf0 0x7c 0xe0 0xf0 0x12
- change of PAUSE key: new make code = old make code + old break code, no more break code
- change on RO break code from 0xf3 to 0xf0 0x51
- change on KP_COMMA break code from 0xfe to 0xf0 0x6d
Detailed list of changes on translated set 2 (the most commonly used):
- change of PRINT make code from 0xe0 0x37 to 0xe0 0x2a 0xe0 0x37
- change of PRINT break code from 0xe0 0xb7 to 0xe0 0xb7 0xe0 0xaa
- change of PAUSE key: new make code = old make code + old break code, no more break code
s390x/css: {c,h,t,r,x}sch: require enable AND device number valid
According to the PoP, subchannels are only considered operational if
they are enabled _and_ the device number is valid. With the current
checks being enabled _or_ having a valid device number was
sufficient. This caused qemu to allow IO on subchannels that were not
enabled.
pc-bios/s390-ccw: enable subchannel for IPL I/O devices
IPL should cause the IPL I/O device to become enabled. So when handling
the IPL program, we should set the E (Enable) bit. However, virtio-ccw
does not know whether it's dealing with an IPL device or not. Since
trying to perform I/O on a disabled device doesn't make any sense,
let's just always enable it. At the same time we can remove the
SCSW_FCTL_START_FUNC flag as it is ignored for msch anyway and did
not enable the device as intended.
s390x/kvm: Fix potential deadlock in sigp handling
If two VCPUs exit at the same time and target each other
with a sigp, both could run into a deadlock as run_on_cpu
on CPU0 will free the BQL when starting the CPU1 target routine.
CPU1 will run its sigp initiater for CPU0 before handling
the run_on_cpu requests, thus resulting in a dead lock.
As all qemu SIGPs are slow path anway we can use a big sigp
lock and allow only one SIGP for the guest at a time. We will
return condition code 2 (BUSY) on contention to the guest.
Yi Min Zhao [Tue, 6 Sep 2016 06:00:44 +0000 (14:00 +0800)]
s390x/pci: assign msix io region for each pci device
For efficiency we now assign one msix io region for each pci device
and provide it with the pointer to the zPCI device as opaque
parameter. In addition, we remove msix address space and add msix io
region as a subregion to the root memory region of pci device.
Alex Bennée [Thu, 22 Sep 2016 10:13:08 +0000 (11:13 +0100)]
ui/vnc-enc-tight: remove switch and have single return
When enabling the sanitizer build it will complain about control
reaching a non-void function. Normally the compiler should detect that
there is only one possible exit given a static VNC_SERVER_FB_BYTES.
As we always expect a static VNC_SERVER_FB_BYTES I've added a compile
time assert and just called the sub-function directly.
This switches over spice (in opengl mode) to render DisplaySurface
updates into a opengl texture, using the helper functions in
ui/console-gl.c. With this patch applied spice (with gl=on) will
stop using qxl rendering ops, it will use dma-buf passing all the
time, i.e. for bios/bootloader (before virtio-gpu driver is loaded)
too.
This should improve performance even using spice (with gl=on) with
non-accelerated stdvga because we stop squeezing all display updates
through a unix/tcp socket and basically using a shared memory transport
instead.
Keep track of gl_block state (added in bba19b8 console: block rendering
until client is done) in QemuConsole and allow to query it. This way
we can avoid state inconsistencies in case different code paths make use
of this.
virtio-gpu does a set-scanout at each frame (it might be a driver
regression). qemu_console_resize() recreate a surface even if the size
didn't change, and this shows up in profiling reports because the
surface is cleared. With this patch, I get a +15-20% glmark2
improvement.
As epoll whether enabled or not is a global setting, we can just
check it only once rather than checking it with every node iteration.
Through this we can avoid a lot of checks when epoll is not enabled.
Rabin Vincent [Tue, 23 Aug 2016 14:50:18 +0000 (16:50 +0200)]
tests: cris: remove check_time1
This test, borrowed from the GDB simulator test suite, checks that every
syscall increments the time returned by gettimeofday() by exactly 1 ms.
This is not guaranteed or even desirable on QEMU so remove this test.
Rabin Vincent [Tue, 23 Aug 2016 14:44:15 +0000 (16:44 +0200)]
tests: cris: remove openpf4 test
This test, borrowed from the GDB simulator test suite, is meant to test
the GDB simulator's --sysroot feature and always fails in QEMU. Remove
it. openpf3 tests the same sequence of system calls (without assuming
the precence of --sysroot).
Rabin Vincent [Wed, 24 Aug 2016 07:08:21 +0000 (09:08 +0200)]
tests: cris: fix syscall inline asm
Add the appropriate register constraints for the inline asm for the
write and exit system calls. Without the correct constraints for the
write() function, correct failure messages are not printed succesfully
on newer version of GCC.
Rabin Vincent [Tue, 23 Aug 2016 14:34:59 +0000 (16:34 +0200)]
tests: cris: force inlining
The CRIS tests expect that functions marked inline are always inline.
With newer versions of GCC, building them results warnings like the
following and spurious failures when they are run.
In file included from tests/tcg/cris/check_moveq.c:5:0:
tests/tcg/cris/crisutils.h:66:20: warning: inlining failed in call to
'cris_tst_cc.constprop.0': call is unlikely and code size would grow [-Winline]
tests/tcg/cris/check_moveq.c:28:13: warning: called from here [-Winline]
Use the always_inline attribute when building them to fix this.
qdisk - hw/block/xen_disk: grant copy implementation
Copy data operated on during request from/to local buffers to/from
the grant references.
Before grant copy operation local buffers must be allocated what is
done by calling ioreq_init_copy_buffers. For the 'read' operation,
first, the qemu device invokes the read operation on local buffers
and on the completion grant copy is called and buffers are freed.
For the 'write' operation grant copy is performed before invoking
write by qemu device.
A new value 'feature_grant_copy' is added to recognize when the
grant copy operation is supported by a guest.
Peter Maydell [Tue, 27 Sep 2016 22:10:12 +0000 (23:10 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging
x86 and machine queue, 2016-09-27
# gpg: Signature made Tue 27 Sep 2016 21:10:06 BST
# gpg: using RSA key 0x2807936F984DC5A6
# gpg: Good signature from "Eduardo Habkost <[email protected]>"
# Primary key fingerprint: 5A32 2FD5 ABC4 D3DB ACCF D1AA 2807 936F 984D C5A6
* remotes/ehabkost/tags/x86-pull-request:
sysbus: Remove ignored return value of FindSysbusDeviceFunc
target-i386: Remove has_msr_* global vars for KVM features
target-i386: Clear KVM CPUID features if KVM is disabled
target-i386: Remove has_msr_hv_tsc global variable
target-i386: Remove has_msr_hv_apic global variable
target-i386: Remove has_msr_mtrr global variable
target-i386: Move xsave component mask to features array
target-i386: xsave: Calculate set of xsave components on realize
target-i386: xsave: Helper function to calculate xsave area size
target-i386: xsave: Simplify CPUID[0xD,0].{EAX,EDX} calculation
target-i386: xsave: Calculate enabled components only once
target-i386: Don't try to enable PT State xsave component
target-i386: Move feature name arrays inside FeatureWordInfo
linux-user: remove #define smp_{cores, threads}
target-i386: Enable CPUID[0x8000000A] if SVM is enabled
target-i386: Automatically set level/xlevel/xlevel2 when needed
tests: Test CPUID level handling for old machines
tests: Add test code for CPUID level/xlevel handling
target-i386: Add a marker to end of the region zeroed on reset
target-i386: Remove unused X86CPUDefinition::xlevel2 field
David Gibson [Wed, 21 Sep 2016 05:23:53 +0000 (15:23 +1000)]
sysbus: Remove ignored return value of FindSysbusDeviceFunc
Functions of type FindSysbusDeviceFunc currently return an integer.
However, this return value is always ignored by the caller in
find_sysbus_device().
This changes the function type to return void, to avoid confusion over
the function semantics.
target-i386: Remove has_msr_hv_tsc global variable
The global variable is not necessary because we can check
cpu->hyperv_time directly.
We just need to ensure cpu->hyperv_time will be cleared if the
feature is not really being exposed to the guest due to missing
KVM_CAP_HYPERV_TIME capability.
target-i386: xsave: Calculate set of xsave components on realize
Instead of doing complex calculations and calling
kvm_arch_get_supported_cpuid() inside cpu_x86_cpuid(), calculate
the set of required XSAVE components earlier, at realize time.
target-i386: xsave: Helper function to calculate xsave area size
Move the xsave area size calculation from cpu_x86_cpuid() inside
its own function. While doing it, change it to use the XSAVE area
struct sizes for the initial size, instead of the magic 0x240
number.
target-i386: xsave: Calculate enabled components only once
Instead of checking both env->features and ena_mask at two
different places in the CPUID code, initialize ena_mask based on
the features that are enabled for the CPU, and then clear
unsupported bits based on kvm_arch_get_supported_cpuid().
The results should be exactly the same, but it will make it
easier to move the mask calculation elsewhare, and reuse
x86_cpu_filter_features() for the kvm_arch_get_supported_cpuid()
check.
target-i386: Don't try to enable PT State xsave component
The code that calculates the set of supported XSAVE components on
CPUID looks at ext_save_areas to find out which components should
be enabled. However, if there are zeroed entries in the
ext_save_areas array, the
((env->features[esa->feature] & esa->bits) == esa->bits)
check will always succeed and QEMU will unconditionally try to
enable the component.
Luckily this never caused any problems because the only missing
entry in ext_save_areas is the PT State component (bit 8), and
KVM currently doesn't support it (so it was cleared on ena_mask).
But the code was still incorrect and would break if KVM starts
returning CPUID[EAX=0xD,ECX=0].EAX[bit 8] as supported on
GET_SUPPORTED_CPUID.
Fix the problem by changing the code to not enable a XSAVE
component if ExtSaveArea::bits is zero.
target-i386: Enable CPUID[0x8000000A] if SVM is enabled
SVM needs CPUID[0x8000000A] to be available. So if SVM is enabled
in a CPU model or explicitly in the command-line, adjust CPUID
xlevel to expose the CPUID[0x8000000A] leaf.
target-i386: Automatically set level/xlevel/xlevel2 when needed
Instead of requiring users and management software to be aware of
required CPUID level/xlevel/xlevel2 values for each feature,
automatically increase those values when features need them.
This was already done for CPUID[7].EBX, and is now made generic
for all CPUID feature flags. Unit test included, to make sure we
don't break ABI on older machine-types and don't mess with the
CPUID level values if they are explicitly set by the user.
We're going to change the way level/xlevel/xlevel2 are handled
when enabling features, but we need to keep the old behavior on
existing machine types. Add test cases for that.
target-i386: Add a marker to end of the region zeroed on reset
Instead of using cpuid_level, use an empty struct as a marker
(like we already did with {start,end}_init_save). This will avoid
accidentaly resetting the wrong fields if we change the field
ordering on CPUX86State.
target-i386: Remove unused X86CPUDefinition::xlevel2 field
No CPU model in builtin_x86_defs has xlevel2 set, so it is always
zero. Delete the field.
Note that this is not an user-visible change. It doesn't remove
the ability to set xlevel2 on the command-line, it just removes
an unused field in builtin_x86_defs.
Peter Maydell [Tue, 27 Sep 2016 15:23:08 +0000 (16:23 +0100)]
Merge remote-tracking branch 'remotes/jasowang/tags/net-pull-request' into staging
# gpg: Signature made Tue 27 Sep 2016 11:05:56 BST
# gpg: using RSA key 0xEF04965B398D6211
# gpg: Good signature from "Jason Wang (Jason Wang on RedHat) <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 215D 46F4 8246 689E C77F 3562 EF04 965B 398D 6211
* remotes/jasowang/tags/net-pull-request: (27 commits)
imx_fec: fix error in qemu_send_packet argument
mcf_fec: fix error in qemu_send_packet argument
net: mcf: limit buffer descriptor count
e1000e: Fix EIAC register implementation
e1000e: Fix spurious RX TCP ACK interrupts
e1000e: Fix OTHER interrupts processing for MSI-X
e1000e: Fix PBACLR implementation
e1000e: Fix CTRL_EXT.EIAME behavior
e1000e: Flush receive queues on link up
e1000e: Flush all receive queues on receive enable
net: limit allocation in nc_sendv_compat
tap: Allow specifying a bridge
e1000: fix buliding complaint
docs: Add documentation for COLO-proxy
MAINTAINERS: add maintainer for COLO-proxy
filter-rewriter: rewrite tcp packet to keep secondary connection
filter-rewriter: track connection and parse packet
filter-rewriter: introduce filter-rewriter initialization
colo-compare: add TCP, UDP, ICMP packet comparison
colo-compare: introduce packet comparison thread
...
Paolo Bonzini [Wed, 31 Aug 2016 19:33:58 +0000 (21:33 +0200)]
cpus-common: lock-free fast path for cpu_exec_start/end
Set cpu->running without taking the cpu_list lock, only requiring it if
there is a concurrent exclusive section. This requires adding a new
field to CPUState, which records whether a running CPU is being counted
in pending_cpus.
When an exclusive section is started concurrently with cpu_exec_start,
cpu_exec_start can use the new field to determine if it has to wait for
the end of the exclusive section. Likewise, cpu_exec_end can use it to
see if start_exclusive is waiting for that CPU.
This a separate patch for easier bisection of issues.
Sergey Fedorov [Tue, 2 Aug 2016 17:27:43 +0000 (18:27 +0100)]
tcg: Make tb_flush() thread safe
Use async_safe_run_on_cpu() to make tb_flush() thread safe. This is
possible now that code generation does not happen in the middle of
execution.
It can happen that multiple threads schedule a safe work to flush the
translation buffer. To keep statistics and debugging output sane, always
check if the translation buffer has already been flushed.
Paolo Bonzini [Fri, 2 Sep 2016 21:33:38 +0000 (23:33 +0200)]
cpus-common: simplify locking for start_exclusive/end_exclusive
It is not necessary to hold qemu_cpu_list_mutex throughout the
exclusive section, because no other exclusive section can run
while pending_cpus != 0.
exclusive_idle() is called in cpu_exec_start(), and that prevents
any CPUs created after start_exclusive() from entering cpu_exec()
during an exclusive section.
Paolo Bonzini [Fri, 2 Sep 2016 19:02:10 +0000 (21:02 +0200)]
cpus-common: remove redundant call to exclusive_idle()
No need to call exclusive_idle() from cpu_exec_end since it is done
immediately afterwards in cpu_exec_start. Any exclusive section could
run as soon as cpu_exec_end leaves, because cpu->running is false and the
mutex is not taken, so the call does not add any protection either.
Paolo Bonzini [Wed, 31 Aug 2016 16:03:39 +0000 (18:03 +0200)]
cpus-common: always defer async_run_on_cpu work items
async_run_on_cpu is only called from the I/O thread, not from CPU threads,
so it doesn't make any difference. It will make a difference however
for async_safe_run_on_cpu.
Paolo Bonzini [Wed, 31 Aug 2016 14:56:04 +0000 (16:56 +0200)]
cpus-common: move exclusive work infrastructure from linux-user
This will serve as the base for async_safe_run_on_cpu. Because
start_exclusive uses CPU_FOREACH, merge exclusive_lock with
qemu_cpu_list_lock: together with a call to exclusive_idle (via
cpu_exec_start/end) in cpu_list_add, this protects exclusive work
against concurrent CPU addition and removal.
Paolo Bonzini [Sun, 28 Aug 2016 01:45:14 +0000 (03:45 +0200)]
cpus-common: move CPU list management to common code
Add a mutex for the CPU list to system emulation, as it will be used to
manage safe work. Abstract manipulation of the CPU list in new functions
cpu_list_add and cpu_list_remove.
Sergey Fedorov [Tue, 2 Aug 2016 17:27:37 +0000 (18:27 +0100)]
linux-user: Use QemuMutex and QemuCond
Convert pthread_mutex_t and pthread_cond_t to QemuMutex and QemuCond.
This will allow to make some locks and conditional variables common
between user and system mode emulation.
Alex Bennée [Tue, 2 Aug 2016 17:27:33 +0000 (18:27 +0100)]
cpus: pass CPUState to run_on_cpu helpers
CPUState is a fairly common pointer to pass to these helpers. This means
if you need other arguments for the async_run_on_cpu case you end up
having to do a g_malloc to stuff additional data into the routine. For
the current users this isn't a massive deal but for MTTCG this gets
cumbersome when the only other parameter is often an address.
This adds the typedef run_on_cpu_func for helper functions which has an
explicit CPUState * passed as the first parameter. All the users of
run_on_cpu and async_run_on_cpu have had their helpers updated to use
CPUState where available.
projects / qemu.git / log