Peter Maydell [Thu, 27 Oct 2016 11:45:45 +0000 (12:45 +0100)]
Merge remote-tracking branch 'remotes/jasowang/tags/net-pull-request' into staging
# gpg: Signature made Wed 26 Oct 2016 03:19:06 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:
colo-proxy: fix memory leak
net: rtl8139: limit processing of ring descriptors
net: vmxnet: initialise local tx descriptor
e1000e: Don't zero out buffer address in rx descriptor
net: rocker: set limit to DMA buffer size
net: eepro100: fix memory leak in device uninit
tap-bsd: OpenBSD uses tap(4) now
net: pcnet: fix source formatting and indentation
net: pcnet: check rx/tx descriptor ring length
* remotes/vivier/tags/m68k-part1-pull-request: (23 commits)
target-m68k: Optimize gen_flush_flags
target-m68k: Optimize some comparisons
target-m68k: Use setcond for scc
target-m68k: Introduce DisasCompare
target-m68k: Reorg flags handling
target-m68k: Remove incorrect clearing of cc_x
target-m68k: Some fixes to SR and flags management
target-m68k: Print flags properly
target-m68k: update CPU flags management
target-m68k: don't update cc_dest in helpers
target-m68k: update move to/from ccr/sr
target-m68k: remove m68k_cpu_exec_enter() and m68k_cpu_exec_exit()
target-m68k: Replace helper_xflag_lt with setcond
target-m68k: allow to update flags with operation on words and bytes
target-m68k: REG() macro cleanup
target-m68k: set PAGE_BITS to 12 for m68k
target-m68k: define operand sizes
target-m68k: set disassembler mode to 680x0 or coldfire
target-m68k: introduce read_imXX() functions
target-m68k: manage scaled index
...
Emulating LL/SC with cmpxchg is not correct, since it can
suffer from the ABA problem. However, portable parallel
code is written assuming only cmpxchg which means that in
practice this is a viable alternative.
Emilio G. Cota [Mon, 27 Jun 2016 19:02:13 +0000 (15:02 -0400)]
target-arm: emulate aarch64's LL/SC using cmpxchg helpers
Emulating LL/SC with cmpxchg is not correct, since it can
suffer from the ABA problem. Portable parallel code, however,
is written assuming only cmpxchg--and not LL/SC--is available.
This means that in practice emulating LL/SC with cmpxchg is
a viable alternative.
The appended emulates LL/SC pairs in aarch64 with cmpxchg helpers.
This works in both user and system mode. In usermode, it avoids
pausing all other CPUs to perform the LL/SC pair. The subsequent
performance and scalability improvement is significant, as the
plots below show. They plot the throughput of atomic_add-bench
compiled for ARM and executed on a 64-core x86 machine.
Emilio G. Cota [Mon, 27 Jun 2016 19:02:08 +0000 (15:02 -0400)]
target-arm: emulate LL/SC using cmpxchg helpers
Emulating LL/SC with cmpxchg is not correct, since it can
suffer from the ABA problem. Portable parallel code, however,
is written assuming only cmpxchg--and not LL/SC--is available.
This means that in practice emulating LL/SC with cmpxchg is
a viable alternative.
The appended emulates LL/SC pairs in ARM with cmpxchg helpers.
This works in both user and system mode. In usermode, it avoids
pausing all other CPUs to perform the LL/SC pair. The subsequent
performance and scalability improvement is significant, as the
plots below show. They plot the throughput of atomic_add-bench
compiled for ARM and executed on a 64-core x86 machine.
target-arm: Rearrange aa32 load and store functions
Stop specializing on TARGET_LONG_BITS == 32; unconditionally allocate
a temp and expand with tcg_gen_extu_i32_tl. Split out gen_aa32_addr,
gen_aa32_frob64, gen_aa32_ld_i32 and gen_aa32_st_i32 as separate interfaces.
Emilio G. Cota [Mon, 27 Jun 2016 19:02:05 +0000 (15:02 -0400)]
tests: add atomic_add-bench
With this microbenchmark we can measure the overhead of emulating atomic
instructions with a configurable degree of contention.
The benchmark spawns $n threads, each performing $o atomic ops (additions)
in a loop. Each atomic operation is performed on a different cache line
(assuming lines are 64b long) that is randomly selected from a range [0, $r).
Emilio G. Cota [Mon, 27 Jun 2016 19:02:06 +0000 (15:02 -0400)]
target-i386: remove helper_lock()
It's been superseded by the atomic helpers.
The use of the atomic helpers provides a significant performance and scalability
improvement. Below is the result of running the atomic_add-test microbenchmark with:
$ x86_64-linux-user/qemu-x86_64 tests/atomic_add-bench -o 5000000 -r $r -n $n
, where $n is the number of threads and $r is the allowed range for the additions.
The scenarios measured are:
- atomic: implements x86' ADDL with the atomic_add helper (i.e. this patchset)
- cmpxchg: implement x86' ADDL with a TCG loop using the cmpxchg helper
- master: before this patchset
Results sorted in ascending range, i.e. descending degree of contention.
Y axis is Throughput in Mops/s. Tests are run on an AMD machine with 64
Opteron 6376 cores.
For master I stopped measuring master after 8 threads, because there is little
point in measuring the well-known performance collapse of a contended lock.
Allow qemu to build on 32-bit hosts without 64-bit atomic ops.
Even if we only allow 32-bit hosts to multi-thread emulate 32-bit
guests, we still need some way to handle the 32-bit guest using a
64-bit atomic operation. Do so by dropping back to single-step.
Wikipedia suggests that only very old AMD64 (circa 2004) did not have
this instruction. Further, it's required by Windows 8 so no new cpus
will ever omit it.
If we truely care about these, then we could check this at startup time
and then avoid executing paths that use it.
Add all of cmpxchg, op_fetch, fetch_op, and xchg.
Handle both endian-ness, and sizes up to 8.
Handle expanding non-atomically, when emulating in serial.
Alex Bennée [Wed, 5 Oct 2016 18:13:04 +0000 (11:13 -0700)]
linux-user: enable parallel code generation on clone
The variable parallel_cpus controls the generation of thread aware
atomic code. We only need to set it once we clone our first thread.
At this point any existing translations need to be thrown away.
While the check against sizeof(void *) is appropriate for
normal usage within qemu, there are places in which we want
wider operaions and have checked for their existance.
net: rtl8139: limit processing of ring descriptors
RTL8139 ethernet controller in C+ mode supports multiple
descriptor rings, each with maximum of 64 descriptors. While
processing transmit descriptor ring in 'rtl8139_cplus_transmit',
it does not limit the descriptor count and runs forever. Add
check to avoid it.
Li Qiang [Wed, 10 Aug 2016 19:12:20 +0000 (00:42 +0530)]
net: vmxnet: initialise local tx descriptor
In Vmxnet3 device emulator while processing transmit(tx) queue,
when it reaches end of packet, it calls vmxnet3_complete_packet.
In that local 'txcq_descr' object is not initialised, which could
leak host memory bytes a guest.
Kevin Wolf [Sun, 16 Oct 2016 22:35:07 +0000 (00:35 +0200)]
e1000e: Don't zero out buffer address in rx descriptor
The e1000e emulation zeroes out any used rx descriptor and then writes a
completely newly constructed value there. By doing this, it doesn't only
update the write-back area of the descriptors (as it's supposed to do),
but it also clears the buffer address, which real hardware doesn't do.
The spec explicitly mentions in chapter 7.1.8 that it is valid for a
driver to reuse a descriptor and only update the status field while
doing so, i.e. reusing the old buffer address:
If software statically allocates buffers, and uses memory read to
check for completed descriptors, it simply has to zero the status
byte in the descriptor to make it ready for reuse by hardware.
This patch fixes the behaviour to leave the buffer address in
descriptors unchanged even after the descriptor has been used.
Rocker network switch emulator has test registers to help debug
DMA operations. While testing host DMA access, a buffer address
is written to register 'TEST_DMA_ADDR' and its size is written to
register 'TEST_DMA_SIZE'. When performing TEST_DMA_CTRL_INVERT
test, if DMA buffer size was greater than 'INT_MAX', it leads to
an invalid buffer access. Limit the DMA buffer size to avoid it.
Li Qiang [Sat, 8 Oct 2016 12:07:25 +0000 (05:07 -0700)]
net: eepro100: fix memory leak in device uninit
The exit dispatch of eepro100 network card device doesn't free
the 's->vmstate' field which was allocated in device realize thus
leading a host memory leak. This patch avoid this.
The AMD PC-Net II emulator has set of control and status(CSR)
registers. Of these, CSR76 and CSR78 hold receive and transmit
descriptor ring length respectively. This ring length could range
from 1 to 65535. Setting ring length to zero leads to an infinite
loop in pcnet_rdra_addr() or pcnet_transmit(). Add check to avoid it.
Separate all ccr bits. Continue to batch updates via cc_op.
Signed-off-by: Richard Henderson <[email protected]>
Fix gen_logic_cc() to really extend the size of the result.
Fix gen_get_ccr(): update cc_op as it is used by the helper.
Factorize flags computing and src/ccr cleanup
Laurent Vivier [Wed, 24 Jun 2015 00:35:06 +0000 (02:35 +0200)]
target-m68k: manage scaled index
Scaled index is not supported by 68000, 68008, and 68010.
EA = (bd + PC) + Xn.SIZE*SCALE + od
Ignore it:
M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL
2.4 BRIEF EXTENSION WORD FORMAT COMPATIBILITY
"If the MC68000 were to execute an instruction that
encoded a scaling factor, the scaling factor would be
ignored and would not access the desired memory address.
The earlier microprocessors do not recognize the brief
extension word formats implemented by newer processors.
Although they can detect illegal instructions, they do not
decode invalid encodings of the brief extension word formats
as exceptions."
Peter Maydell [Tue, 25 Oct 2016 16:03:11 +0000 (17:03 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-qapi-2016-10-25' into staging
QAPI patches for 2016-10-25
# gpg: Signature made Tue 25 Oct 2016 16:56:27 BST
# gpg: using RSA key 0x3870B400EB918653
# gpg: Good signature from "Markus Armbruster <[email protected]>"
# gpg: aka "Markus Armbruster <[email protected]>"
# Primary key fingerprint: 354B C8B3 D7EB 2A6B 6867 4E5F 3870 B400 EB91 8653
* remotes/armbru/tags/pull-qapi-2016-10-25:
qdict: implement a qdict_crumple method for un-flattening a dict
qapi: don't pass two copies of TestInputVisitorData to tests
qapi: rename QmpOutputVisitor to QObjectOutputVisitor
qapi: rename QmpInputVisitor to QObjectInputVisitor
qapi: rename *qmp-*-visitor* to *qobject-*-visitor*
qapi: add trace events for visitor
trivial: Restore blank line in qapi-schema
qdict: implement a qdict_crumple method for un-flattening a dict
The qdict_flatten() method will take a dict whose elements are
further nested dicts/lists and flatten them by concatenating
keys.
The qdict_crumple() method aims to do the reverse, taking a flat
qdict, and turning it into a set of nested dicts/lists. It will
apply nesting based on the key name, with a '.' indicating a
new level in the hierarchy. If the keys in the nested structure
are all numeric, it will create a list, otherwise it will create
a dict.
If the keys are a mixture of numeric and non-numeric, or the
numeric keys are not in strictly ascending order, an error will
be reported.
The intent of this function is that it allows a set of QemuOpts
to be turned into a nested data structure that mirrors the nesting
used when the same object is defined over QMP.
qapi: don't pass two copies of TestInputVisitorData to tests
The input_visitor_test_add() method was accepting an instance
of 'TestInputVisitorData' and passing it as the 'user_data'
parameter to test functions. The main 'TestInputVisitorData'
instance that was actually used, was meanwhile being allocated
automatically by the test framework fixture setup.
The 'user_data' parameter is going to be needed for tests
added in later patches, so getting rid of the current mistaken
usage now allows this.
qapi: rename QmpOutputVisitor to QObjectOutputVisitor
The QmpOutputVisitor has no direct dependency on QMP. It is
valid to use it anywhere that one wants a QObject. Rename it
to better reflect its functionality as a generic QAPI
to QObject converter.
The commit before previous renamed the files, this one renames C
identifiers.
qapi: rename QmpInputVisitor to QObjectInputVisitor
The QmpInputVisitor has no direct dependency on QMP. It is
valid to use it anywhere that one has a QObject. Rename it
to better reflect its functionality as a generic QObject
to QAPI converter.
The previous commit renamed the files, this one renames C identifiers.
qapi: rename *qmp-*-visitor* to *qobject-*-visitor*
The QMP visitors have no direct dependency on QMP. It is
valid to use them anywhere that one has a QObject. Rename them
to better reflect their functionality as a generic QObject
to QAPI converter.
This is the first of three parts: rename the files. The next two
parts will rename C identifiers. The split is necessary to make git
rename detection work.
Eric Blake [Mon, 17 Oct 2016 21:29:54 +0000 (16:29 -0500)]
trivial: Restore blank line in qapi-schema
Commit de63ab6 accidentally undid part of commit a43edcf,
because the two patches were written in parallel, and the
blank line was not noticed as a casualty of merge conflicts.
Peter Maydell [Tue, 25 Oct 2016 09:25:27 +0000 (10:25 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/x86-pull-request' into staging
x86 and CPU queue, 2016-10-24
x2APIC support to APIC code, cpu_exec_init() refactor on all
architectures, and other x86 changes.
# gpg: Signature made Mon 24 Oct 2016 20:51:14 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:
exec: call cpu_exec_exit() from a CPU unrealize common function
exec: move cpu_exec_init() calls to realize functions
exec: split cpu_exec_init()
pc: q35: Bump max_cpus to 288
pc: Require IRQ remapping and EIM if there could be x2APIC CPUs
pc: Add 'etc/boot-cpus' fw_cfg file for machine with more than 255 CPUs
Increase MAX_CPUMASK_BITS from 255 to 288
pc: Clarify FW_CFG_MAX_CPUS usage comment
pc: kvm_apic: Pass APIC ID depending on xAPIC/x2APIC mode
pc: apic_common: Reset APIC ID to initial ID when switching into x2APIC mode
pc: apic_common: Restore APIC ID to initial ID on reset
pc: apic_common: Extend APIC ID property to 32bit
pc: Leave max apic_id_limit only in legacy cpu hotplug code
acpi: cphp: Force switch to modern cpu hotplug if APIC ID > 254
pc: acpi: x2APIC support for SRAT table
pc: acpi: x2APIC support for MADT table and _MAT method
Laurent Vivier [Thu, 20 Oct 2016 11:26:04 +0000 (13:26 +0200)]
exec: call cpu_exec_exit() from a CPU unrealize common function
As cpu_exec_exit() mirrors the cpu_exec_realizefn(),
rename it as cpu_exec_unrealizefn().
Create and register a cpu_common_unrealizefn() function for
the CPU device class and call cpu_exec_unrealizefn() from
this function.
Remove cpu_exec_exit() from cpu_common_finalize()
(which mirrors init, not realize), and as x86_cpu_unrealizefn()
and ppc_cpu_unrealizefn() overwrite the device class unrealize function,
add a call to a parent_unrealize pointer.
Laurent Vivier [Thu, 20 Oct 2016 11:26:03 +0000 (13:26 +0200)]
exec: move cpu_exec_init() calls to realize functions
Modify all CPUs to call it from XXX_cpu_realizefn() function.
Remove all the cannot_destroy_with_object_finalize_yet as
unsafe references have been moved to cpu_exec_realizefn().
(tested with QOM command provided by commit 4c315c27)
for arm:
Setting of cpu->mp_affinity is moved from arm_cpu_initfn()
to arm_cpu_realizefn() as setting of cpu_index is now done
in cpu_exec_realizefn(). To avoid to overwrite an user defined
value, we set it to an invalid value by default, and update
it in realize function only if the value is still invalid.
Laurent Vivier [Thu, 20 Oct 2016 11:26:02 +0000 (13:26 +0200)]
exec: split cpu_exec_init()
Put in cpu_exec_initfn() what initializes the CPU,
and leave in cpu_exec_init() what adds it to the environment.
As cpu_exec_initfn() is called by all XX_cpu_initfn(), call it
directly in cpu_common_initfn().
cpu_exec_init() is now a realize function, it will be renamed
to cpu_exec_realizefn() and moved to the XX_cpu_realizefn()
function in a following patch.
Igor Mammedov [Thu, 20 Oct 2016 14:58:42 +0000 (16:58 +0200)]
pc: Add 'etc/boot-cpus' fw_cfg file for machine with more than 255 CPUs
Currently firmware uses 1 byte at 0x5F offset in RTC CMOS
to get number of CPUs present at boot. However 1 byte is
not enough to handle more than 255 CPUs. So add a new
fw_cfg file that would allow QEMU to tell it.
For compat reasons add file only for machine types that
support more than 255 CPUs.
Igor Mammedov [Wed, 19 Oct 2016 12:05:37 +0000 (14:05 +0200)]
pc: apic_common: Reset APIC ID to initial ID when switching into x2APIC mode
SDM: x2APIC State Transitions:
State Changes From xAPIC Mode to x2APIC Mode
"
Any APIC ID value written to the memory-mapped
local APIC ID register is not preserved
"
Igor Mammedov [Wed, 19 Oct 2016 12:05:33 +0000 (14:05 +0200)]
acpi: cphp: Force switch to modern cpu hotplug if APIC ID > 254
Switch to modern cpu hotplug at machine startup time if
a cpu present at boot has apic-id in range unsupported
by legacy cpu hotplug interface (i.e. > 254), to avoid
killing QEMU from legacy cpu hotplug code with error:
"acpi: invalid cpu id: #apic-id#"
Peter Maydell [Mon, 24 Oct 2016 18:37:33 +0000 (19:37 +0100)]
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20161024' into staging
target-arm queue:
* support variable (runtime-determined) page sizes, for a
nearly-20% speedup of TCG for ARMv7 and v8 CPUs with 4K pages
* ptimer: add tests, support more flexible behaviour around
what happens on the "zero" tick, use ptimer for a9gtimer
* virt: ACPI: Add IORT Structure definition
* i2c: Fix SMBus read transactions to avoid double events
* timer: stm32f2xx_timer: add check for prescaler value
* QOMify musicpal, pxa2xx_gpio, strongarm, pl110
* target-arm: Implement new HLT trap for semihosting
* i2c: Add asserts for second smbus i2c_start_transfer()
Peter Maydell [Mon, 24 Oct 2016 17:26:59 +0000 (18:26 +0100)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block layer patches
# gpg: Signature made Mon 24 Oct 2016 17:02:47 BST
# gpg: using RSA key 0x7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74 56FE 7F09 B272 C88F 2FD6
* remotes/kevin/tags/for-upstream: (23 commits)
block/replication: Clarify 'top-id' parameter usage
block: More operations for meta dirty bitmap
tests: Add test code for hbitmap serialization
block: BdrvDirtyBitmap serialization interface
hbitmap: serialization
block: Assert that bdrv_release_dirty_bitmap succeeded
block: Add two dirty bitmap getters
block: Support meta dirty bitmap
tests: Add test code for meta bitmap
HBitmap: Introduce "meta" bitmap to track bit changes
block: Hide HBitmap in block dirty bitmap interface
quorum: do not allocate multiple iovecs for FIFO strategy
quorum: change child_iter to children_read
iotests: Do not rely on unavailable domains in 162
iotests: Remove raciness from 162
qemu-nbd: Add --fork option
qemu-iotests: Test I/O in a single drive from a throttling group
throttle: Correct access to wrong BlockBackendPublic structures
qapi: fix memory leak in bdrv_image_info_specific_dump
block: improve error handling in raw_open
...
projects / qemu.git / log