Eric Blake [Sat, 27 May 2017 03:04:21 +0000 (22:04 -0500)]
nbd: Fully initialize client in case of failed negotiation
If a non-NBD client connects to qemu-nbd, we would end up with
a SIGSEGV in nbd_client_put() because we were trying to
unregister the client's association to the export, even though
we skipped inserting the client into that list. Easy trigger
in two terminals:
nmap claims that it thinks it connected to a pago-services1
server (which probably means nmap could be updated to learn the
NBD protocol and give a more accurate diagnosis of the open
port - but that's not our problem), then terminates immediately,
so our call to nbd_negotiate() fails. The fix is to reorder
nbd_co_client_start() to ensure that all initialization occurs
before we ever try talking to a client in nbd_negotiate(), so
that the teardown sequence on negotiation failure doesn't fault
while dereferencing a half-initialized object.
While debugging this, I also noticed that nbd_update_server_watch()
called by nbd_client_closed() was still adding a channel to accept
the next client, even when the state was no longer RUNNING. That
is fixed by making nbd_can_accept() pay attention to the current
state.
sockets: improve error reporting if UNIX socket path is too long
The 'struct sockaddr_un' only allows 108 bytes for the socket
path.
If the user supplies a path, QEMU uses snprintf() to silently
truncate it when too long. This is undesirable because the user
will then be unable to connect to the path they asked for.
If the user doesn't supply a path, QEMU builds one based on
TMPDIR, but if that leads to an overlong path, it mistakenly
uses error_setg_errno() with a stale errno value, because
snprintf() does not set errno on truncation.
In solving this the code needed some refactoring to ensure we
don't pass 'un.sun_path' directly to any APIs which expect
NUL-terminated strings, because the path is not required to
be terminated.
Mihail Abakumov [Fri, 19 May 2017 09:36:15 +0000 (12:36 +0300)]
i386: fix read/write cr with icount option
Running Windows with icount causes a crash in instruction of write cr.
This patch fixes it.
Reading and writing cr cause an icount read because there are called
cpu_get_apic_tpr and cpu_set_apic_tpr functions. So, there is need
gen_io_start()/gen_io_end() calls.
Fam Zheng [Thu, 18 May 2017 10:28:08 +0000 (18:28 +0800)]
virtio-scsi: Unset hotplug handler when unrealize
This matches the qbus_set_hotplug_handler in realize, and it releases
the final reference to the embedded VirtIODevice so that it is
properly finalized.
A use-after-free is fixed with this patch, indirectly:
virtio_device_instance_finalize wasn't called at hot-unplug, and the
vdev->listener would be a dangling pointer in the global and the per
address space listener list. See also RHBZ 1449031.
nbd: add errp to read_sync, write_sync and drop_sync
There a lot of calls of these functions, which already have errp, which
they are filling themselves. On the other hand, nbd_wr_syncv has errp
parameter too, so it would be great to connect them.
functions read_sync, drop_sync, write_sync, and also
nbd_negotiate_write, nbd_negotiate_read, nbd_negotiate_drop_sync
returns number of processed bytes. But what this number can be,
except requested number of bytes?
Actually, underlying nbd_wr_syncv function returns a value >= 0 and
!= requested_bytes only on eof on read operation. So, firstly, it is
impossible on write (let's add an assert) and on read it actually
means, that communication is broken (except nbd_receive_reply, see
below).
Most of callers operate like this:
if (func(..., size) != size) {
/* error path */
}
, i.e.:
1. They are not interested in partial success
2. Extra duplications in code (especially bad are duplications of
magic numbers)
3. User doesn't see actual error message, as return code is lost.
(this patch doesn't fix this point, but it makes fixing easier)
Several callers handles ret >= 0 and != requested-size separately, by
just returning EINVAL in this case. This patch makes read_sync and
friends return EINVAL in this case, so final behavior is the same.
And only one caller - nbd_receive_reply() does something not so
obvious. It returns EINVAL for ret > 0 and != requested-size, like
previous group, but for ret == 0 it returns 0. The only caller of
nbd_receive_reply() - nbd_read_reply_entry() handles ret == 0 in the
same way as ret < 0, so for now it doesn't matter. However, in
following commits error path handling will be improved and we'll need
to distinguish success from fail in this case too. So, this patch adds
separate helper for this case - read_sync_eof.
nbd_wr_syncv is called either from coroutine or from client negotiation
code, when socket is in blocking mode. So, -EAGAIN is impossible.
Furthermore, EAGAIN is confusing, as, what to read/write again? With
EAGAIN as a return code we don't know how much data is already
read or written by the function, so in case of EAGAIN the whole
communication is broken.
Peter Xu [Tue, 9 May 2017 06:00:44 +0000 (14:00 +0800)]
kvm: irqchip: skip update msi when disabled
It's possible that one device kept its irqfd/virq there even when
MSI/MSIX was disabled globally for that device. One example is
virtio-net-pci (see commit f1d0f15a6 and virtio_pci_vq_vector_mask()).
It is used as a fast path to avoid allocate/release irqfd/virq
frequently when guest enables/disables MSIX.
However, this fast path brought a problem to msi_route_list, that the
device MSIRouteEntry is still dangling there even if MSIX disabled -
then we cannot know which message to fetch, even if we can, the messages
are meaningless. In this case, we can just simply ignore this entry.
It's safe, since when MSIX is enabled again, we'll rebuild them no
matter what.
Xiao Guangrong [Sat, 27 May 2017 02:53:01 +0000 (10:53 +0800)]
qtest: add rtc periodic timer test
It tests the accuracy of rtc periodic timer which is recently
improved & fixed by commit 7ffcb539a3 ("mc146818rtc: precisely count
the clock for periodic timer", 2017-05-19).
Xiao Guangrong [Wed, 10 May 2017 08:32:59 +0000 (16:32 +0800)]
mc146818rtc: embrace all x86 specific code
Introduce a function, rtc_policy_slew_deliver_irq(), which delivers
irq if LOST_TICK_POLICY_SLEW is used, as which is only supported on
x86, other platforms call it will trigger a assert
After that, we can move the x86 specific code to the common place
Xiao Guangrong [Wed, 10 May 2017 08:32:57 +0000 (16:32 +0800)]
mc146818rtc: ensure LOST_TICK_POLICY_SLEW is only enabled on TARGET_I386
Any tick policy specified on other platforms rather on TARGET_I386
will fall back to LOST_TICK_POLICY_DISCARD silently, this patch makes
sure only TARGET_I386 can enable LOST_TICK_POLICY_SLEW
After that, we can enable LOST_TICK_POLICY_SLEW in the common code
which need not use '#ifdef TARGET_I386' to make these code be x86
specific anymore
Tai Yunfang [Wed, 10 May 2017 08:32:56 +0000 (16:32 +0800)]
mc146818rtc: precisely count the clock for periodic timer
There are two issues in current code:
1) If the period is changed by re-configuring RegA, the coalesced
irq will be scaled to reflect the new period, however, it
calculates the new interrupt number like this:
s->irq_coalesced = (s->irq_coalesced * s->period) / period;
There are some clocks will be lost if they are not enough to
be squeezed to a single new period that will cause the VM clock
slower
In order to fix the issue, we calculate the interrupt window
based on the precise clock rather than period, then the clocks
lost during period is scaled can be compensated properly
2) If periodic_timer_update() is called due to RegA reconfiguration,
i.e, the period is updated, current time is not the start point
for the next periodic timer, instead, which should start from the
last interrupt, otherwise, the clock in VM will become slow
This patch takes the clocks from last interrupt to current clock
into account and compensates the clocks for the next interrupt,
especially if a complete interrupt was lost in this window, the
time can be caught up by LOST_TICK_POLICY_SLEW
Xiao Guangrong [Wed, 10 May 2017 08:32:55 +0000 (16:32 +0800)]
mc146818rtc: update periodic timer only if it is needed
Currently, the timer is updated whenever RegA or RegB is written
even if the periodic timer related configuration is not changed
This patch optimizes it slightly to make the update happen only
if its period or enable-status is changed, also later patches are
depend on this optimization
Peter Maydell [Tue, 6 Jun 2017 16:00:12 +0000 (17:00 +0100)]
Merge remote-tracking branch 'remotes/xtensa/tags/20170606-xtensa' into staging
target/xtensa fixes:
- fix read/write simcall mapping flags and return value;
- use -serial option to direct console output of sim machine to QEMU chardev;
- fix handling of unknown registers in the gdbstub.
* remotes/xtensa/tags/20170606-xtensa:
target/xtensa: handle unknown registers in gdbstub
target/xtensa: support output to chardev console
target/xtensa: fix return value of read/write simcalls
target/xtensa: fix mapping direction in read/write simcalls
Peter Maydell [Tue, 6 Jun 2017 14:37:53 +0000 (15:37 +0100)]
Merge remote-tracking branch 'remotes/armbru/tags/pull-misc-2017-06-06' into staging
Miscellaneous patches for 2017-06-06
# gpg: Signature made Tue 06 Jun 2017 08:30:43 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-misc-2017-06-06:
monitor: fix object_del for command-line-created objects
tests: check-qom-proplist: add checks for cmdline-created objects
virtio-scsi-test: Use scsi-hd instead of legacy scsi-disk
block: Clarify documentation of BlockInfo member io-status
Peter Maydell [Tue, 6 Jun 2017 13:30:06 +0000 (14:30 +0100)]
Merge remote-tracking branch 'remotes/dgibson/tags/ppc-for-2.10-20170606' into staging
ppc patch queue 2017-06-06
Accumulated patches for ppc targets and the pseries machine type.
The big thing in this batch is a start on a substantial cleanup of the
pseries hotplug mechanisms, which were pretty confusing. For now
these shouldn't cause substantial behavioural changes, but I am hoping
these lead to clearer code and eventually to fixes for the bugs we
have in hotplug handling, particularly when hotplug and migration are
combined.
* remotes/dgibson/tags/ppc-for-2.10-20170606:
spapr: Remove some non-useful properties on DRC objects
spapr: Eliminate spapr_drc_get_type_str()
spapr: Move configure-connector state into DRC
spapr: Clean up spapr_dr_connector_by_*()
spapr: Introduce DRC subclasses
spapr/drc: don't migrate DRC of cold-plugged CPUs and LMBs
spapr: Allow boot from vhost-*-scsi backends
ppc/pnv: check the return value of fdt_setprop()
spapr_nvram: Check return value from blk_getlength()
target/ppc: Fixup set_spr error in h_register_process_table
target-ppc: Fix openpic timer read register offset
spapr: Make DRC get_index and get_type methods into plain functions
spapr: Abolish DRC set_configured method
spapr: Abolish DRC get_fdt method
spapr: Move DRC RTAS calls into spapr_drc.c
migration: Mark CPU states dirty before incoming migration/loadvm
migration: remove register_savevm()
Max Filippov [Sat, 3 Jun 2017 09:11:13 +0000 (02:11 -0700)]
target/xtensa: handle unknown registers in gdbstub
Xtensa cores may have registers of types/sizes not supported by the
gdbstub accessors. Ignore writes to such registers and return zero on
read, but always return correct register size, so that gdb on the other
side is able to access all registers in the packet holding unsupported
registers in the middle. This fixes gdb interaction with cores that have
vector/custom TIE registers.
Max Filippov [Fri, 12 May 2017 18:09:14 +0000 (11:09 -0700)]
target/xtensa: support output to chardev console
In semihosting mode QEMU allows guest to read and write host file
descriptors directly, including descriptors 0..2, a.k.a. stdin, stdout
and stderr. Sometimes it's desirable to have semihosting console
controlled by -serial option, e.g. to connect it to network.
Add semihosting console to xtensa-semi.c, open it in the 'sim' machine
in the presence of -serial option and direct stdout and stderr to it
when it's present.
Max Filippov [Fri, 12 May 2017 19:05:23 +0000 (12:05 -0700)]
target/xtensa: fix return value of read/write simcalls
Return value of read/write simcalls is not calculated correctly in case
of operations crossing page boundary and in case of short reads/writes.
Read and write simcalls should return the size of data actually
read/written or -1 in case of error.
Max Filippov [Fri, 12 May 2017 18:17:01 +0000 (11:17 -0700)]
target/xtensa: fix mapping direction in read/write simcalls
Read and write simcalls map physical memory to access I/O buffers, but
'read' simcall need to map it for writing and 'write' simcall need to
map it for reading, i.e. the opposite of what they do now. Fix that.
Peter Maydell [Tue, 6 Jun 2017 09:00:34 +0000 (10:00 +0100)]
Merge remote-tracking branch 'remotes/ehabkost/tags/x86-and-machine-pull-request' into staging
x86 and machine queue, 2017-06-05
# gpg: Signature made Mon 05 Jun 2017 19:58:01 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-and-machine-pull-request:
scripts: Test script to look for -device crashes
qemu.py: Add QEMUMachine.exitcode() method
qemu.py: Don't set _popen=None on error/shutdown
spapr: cleanup spapr_fixup_cpu_numa_dt() usage
numa: move numa_node from CPUState into target specific classes
numa: make hmp 'info numa' fetch numa nodes from qmp_query_cpus() result
numa: make sure that all cpus have has_node_id set if numa is enabled
numa: move default mapping init to machine
numa: consolidate cpu_preplug fixups/checks for pc/arm/spapr
pc: Use "min-[x]level" on compat_props
Michael Roth [Sat, 3 Jun 2017 23:13:32 +0000 (18:13 -0500)]
monitor: fix object_del for command-line-created objects
Currently objects specified on the command-line are only partially
cleaned up when 'object_del' is issued in either HMP or QMP: the
object itself is fully finalized, but the QemuOpts are not removed.
This results in the following behavior:
QEMU 2.7.91 monitor - type 'help' for more information
(qemu) object_del ram1
(qemu) object_del ram1
object 'ram1' not found
(qemu) object_add memory-backend-ram,id=ram1,size=256M
Duplicate ID 'ram1' for object
Try "help object_add" for more information
which can be an issue for use-cases like memory hotplug.
This happens on the HMP side because hmp_object_add() attempts to
create a temporary QemuOpts entry with ID 'ram1', which ends up
conflicting with the command-line-created entry, since it was never
cleaned up during the previous hmp_object_del() call.
We address this by adding a check in user_creatable_del(), which
is called by both qmp_object_del() and hmp_object_del() to handle
the actual object cleanup, to determine whether an option group entry
matching the object's ID is present and removing it if it is.
Note that qmp_object_add() never attempts to create a temporary
QemuOpts entry, so it does not encounter the duplicate ID error,
which is why this isn't generally visible in libvirt.
Michael Roth [Sat, 3 Jun 2017 23:13:31 +0000 (18:13 -0500)]
tests: check-qom-proplist: add checks for cmdline-created objects
check-qom-proplist originally added tests for verifying that
object-creation helpers object_new_with_{props,propv} behaved in
similar fashion to the "traditional" method involving setting each
individual property separately after object creation rather than
via a single call.
Another similar "helper" for creating Objects exists in the form of
objects specified via -object command-line parameters. By that
rationale, we extend check-qom-proplist to include similar checks
for command-line-created objects by employing the same
qemu_opts_parse()-based parsing the vl.c employs.
This parser has a side-effect of parsing the object's options into
a QemuOpt structure and registering this in the global QemuOptsList
using the Object's ID. This can conflict with future Object instances
that attempt to use the same ID if we don't ensure this is cleaned
up as part of Object finalization, so we include a FIXME stub to test
for this case, which will then be resolved in a subsequent patch.
block: Clarify documentation of BlockInfo member io-status
Say "SCSI except scsi-generic" instead of "scsi-disk", because
scsi-disk could mean either scsi-disk.c (which is correct) or device
model scsi-disk (which would be incorrect).
David Gibson [Sun, 4 Jun 2017 10:28:04 +0000 (20:28 +1000)]
spapr: Remove some non-useful properties on DRC objects
* 'connector_type' is easily derived from the 'index' property, so there's
no point to it (it's also implicit in the QOM type of the DRC)
* 'isolation-state', 'indicator-state' and 'allocation-state' are
part of the transaction between qemu and guest during PAPR hotplug
operations, and outside tools really have no business looking at it
(especially not changing, and these were RW properties)
* 'entity-sense' is basically just a weird PAPR encoding of whether there
is a device connected to this DRC
Strictly speaking removing these properties is breaking the qemu interface.
However, I'm pretty sure no management tools have ever used these. For
debugging there are better alternatives. Therefore, I think removing these
broken interfaces is the better option.
David Gibson [Sun, 4 Jun 2017 10:26:54 +0000 (20:26 +1000)]
spapr: Eliminate spapr_drc_get_type_str()
This function was used in generating the device tree. However, now that
we have different QOM types for different DRC types we can easily store
the information we need in the class structure and avoid this specialized
lookup function.
David Gibson [Sun, 4 Jun 2017 10:26:25 +0000 (20:26 +1000)]
spapr: Move configure-connector state into DRC
Currently the sPAPRMachineState contains a list of sPAPRConfigureConnector
structures which store intermediate state for the ibm,configure-connector
RTAS call.
This was an attempt to separate this state from the core of the DRC state.
However the configure connector process is intimately tied to the DRC
model, so there's really no point trying to have two levels of interface
here.
Moving the configure-connector state into its corresponding DRC allows
removal of a number of helpers for maintaining the anciliary list.
David Gibson [Sun, 4 Jun 2017 10:26:03 +0000 (20:26 +1000)]
spapr: Clean up spapr_dr_connector_by_*()
* Change names to something less ludicrously verbose
* Now that we have QOM subclasses for the different DRC types, use a QOM
typename instead of a PAPR type value parameter
The latter allows removal of the get_type_shift() helper.
David Gibson [Sun, 4 Jun 2017 10:25:17 +0000 (20:25 +1000)]
spapr: Introduce DRC subclasses
Currently we only have a single QOM type for all DRCs, but lots of
places where we switch behaviour based on the DRC's PAPR defined type.
This is a poor use of our existing type system.
So, instead create QOM subclasses for each PAPR defined DRC type. We
also introduce intermediate subclasses for physical and logical DRCs,
a division which will be useful later on.
Instead of being stored in the DRC object itself, the PAPR type is now
stored in the class structure. There are still many places where we
switch directly on the PAPR type value, but this at least provides the
basis to start to remove those.
Greg Kurz [Fri, 2 Jun 2017 10:09:35 +0000 (12:09 +0200)]
spapr/drc: don't migrate DRC of cold-plugged CPUs and LMBs
As explained in commit 5c0139a8c2f0 ("spapr: fix default DRC state for
coldplugged LMBs"), guests expect cold-plugged LMBs to be pre-allocated
and unisolated. The same goes for cold-plugged CPUs.
While here, let's convert g_assert(false) to the better self documenting
g_assert_not_reached().
The current implementation of spapr_get_fw_dev_path() doesn't take into
consideration vhost-*-scsi devices. This makes said devices unbootable
on PPC as SLOF is unable to work out the path to scan boot disks.
This makes VMs bootable on spapr when using vhost-*-scsi by implementing
a disk path for VHostSCSICommon (which currently includes both
vhost-user-scsi and vhost-scsi).
Peter Maydell [Mon, 5 Jun 2017 15:14:17 +0000 (16:14 +0100)]
spapr_nvram: Check return value from blk_getlength()
The blk_getlength() function can return an error value if the
image size cannot be determined. Check for this rather than
ploughing on and trying to g_malloc0() a negative number.
(Spotted by Coverity, CID 1288484.)
target/ppc: Fixup set_spr error in h_register_process_table
set_spr is used in the function h_register_process_table() to update the
LPCR_GTSE and LPCR_UPRT values based on the flags passed by the guest.
The set_spr function takes the last two arguments mask and value used to
mask and set the value of the spr respectively.
The current call site passes these arguments in the wrong order and thus
bot GTSE and UPRT will be set irrespective, which is obviously
incorrect.
Rearrange the function call so that these arguments are passed in the
correct order and the correct behaviour is exhibited.
It is worth noting that this wasn't detected earlier since these were
always both set in all cases where this H_CALL was made.
Fixes: 6de833070ca2 ("target/ppc: Set UPRT and GTSE on all cpus in H_REGISTER_PROCESS_TABLE") Signed-off-by: Suraj Jitindar Singh <[email protected]> Signed-off-by: David Gibson <[email protected]>
openpic_tmr_read() is incorrectly computing register offset of the
TCCR, TBCR, TVPR, and TDR registers when accessing the open pic timer
registers. Specifically the offset of timer registers for
openpic_tmr_read() is not accounting for the timer frequency reporting
register (TFFR) which is the first register in the "tmr" memory
region.
openpic_tmr_write() *is* correctly computing the offset by adding
0x10f0 to the address prior to computing the register index. This
patch instead subtracts 0x10 in both the read and write routines and
eliminates some other gratuitous differences between the functions.
David Gibson [Fri, 2 Jun 2017 03:36:10 +0000 (13:36 +1000)]
spapr: Abolish DRC set_configured method
DRConnectorClass has a set_configured method, however:
* There is only one implementation, and only ever likely to be one
* There's exactly one caller, and that's (now) local
* The implementation is very straightforward
So abolish the method entirely, and just open-code what we need.
David Gibson [Thu, 1 Jun 2017 00:36:24 +0000 (10:36 +1000)]
spapr: Abolish DRC get_fdt method
The DRConnectorClass includes a get_fdt method. However
* There's only one implementation, and there's only likely to ever be one
* Both callers are local to spapr_drc
* Each caller only uses one half of the actual implementation
So abolish get_fdt() entirely, and just open-code what we need.
David Gibson [Thu, 1 Jun 2017 00:30:00 +0000 (10:30 +1000)]
spapr: Move DRC RTAS calls into spapr_drc.c
Currently implementations of the RTAS calls related to DRCs are in
spapr_rtas.c. They belong better in spapr_drc.c - that way they're closer
to related code, and we'll be able to make some more things local.
spapr_rtas.c was intended to contain the RTAS infrastructure and core calls
that don't belong anywhere else, not every RTAS implementation.
David Gibson [Fri, 26 May 2017 04:46:28 +0000 (14:46 +1000)]
migration: Mark CPU states dirty before incoming migration/loadvm
As a rule, CPU internal state should never be updated when
!cpu->kvm_vcpu_dirty (or the HAX equivalent). If that is done, then
subsequent calls to cpu_synchronize_state() - usually safe and idempotent -
will clobber state.
However, we routinely do this during a loadvm or incoming migration.
Usually this is called shortly after a reset, which will clear all the cpu
dirty flags with cpu_synchronize_all_post_reset(). Nothing is expected
to set the dirty flags again before the cpu state is loaded from the
incoming stream.
This means that it isn't safe to call cpu_synchronize_state() from a
post_load handler, which is non-obvious and potentially inconvenient.
We could cpu_synchronize_all_state() before the loadvm, but that would be
overkill since a) we expect the state to already be synchronized from the
reset and b) we expect to completely rewrite the state with a call to
cpu_synchronize_all_post_init() at the end of qemu_loadvm_state().
To clear this up, this patch introduces cpu_synchronize_pre_loadvm() and
associated helpers, which simply marks the cpu state as dirty without
actually changing anything. i.e. it says we want to discard any existing
KVM (or HAX) state and replace it with what we're going to load.
Laurent Vivier [Wed, 24 May 2017 12:10:48 +0000 (14:10 +0200)]
migration: remove register_savevm()
We can replace the four remaining calls of register_savevm() by
calls to register_savevm_live(). So we can remove the function and
as we don't allocate anymore the ops pointer with g_new0()
we don't have to free it then.
Eduardo Habkost [Fri, 26 May 2017 18:12:00 +0000 (15:12 -0300)]
scripts: Test script to look for -device crashes
Test code to check if we can crash QEMU using -device. It will
test all accel/machine/device combinations by default, which may
take a few hours (it's more than 90k test cases). There's a "-r"
option that makes it test a random sample of combinations.
The scripts contains a whitelist for: 1) known error messages
that make QEMU exit cleanly; 2) known QEMU crashes.
This is the behavior when the script finds a failure:
* Known clean (exitcode=1) errors generate DEBUG messages
(hidden by default)
* Unknown clean (exitcode=1) errors will generate INFO messages
(visible by default)
* Known crashes generate error messages, but are not fatal
(unless --strict mode is used)
* Unknown crashes generate fatal error messages
Having an updated whitelist of known clean errors is useful to make the
script less verbose and run faster when in --quick mode, but the
whitelist doesn't need to be always up to date.
Igor Mammedov [Tue, 30 May 2017 16:24:01 +0000 (18:24 +0200)]
spapr: cleanup spapr_fixup_cpu_numa_dt() usage
even though spapr_fixup_cpu_numa_dt() has no effect on FDT
if numa is disabled, don't call it uselessly. It makes it
obvious at call sites that function is needed only when numa
is enabled.
Igor Mammedov [Tue, 30 May 2017 16:24:00 +0000 (18:24 +0200)]
numa: move numa_node from CPUState into target specific classes
Move vcpu's associated numa_node field out of generic CPUState
into inherited classes that actually care about cpu<->numa mapping,
i.e: ARMCPU, PowerPCCPU, X86CPU.
Igor Mammedov [Tue, 30 May 2017 16:23:59 +0000 (18:23 +0200)]
numa: make hmp 'info numa' fetch numa nodes from qmp_query_cpus() result
HMP command 'info numa' is the last external user that access
CPUState::numa_node field directly. In order to move it to CPU
classes that actually use it, eliminate direct access and use
an alternative approach by using result of qmp_query_cpus(),
which provides topology properties CPU threads are associated
with (including node-id).
Igor Mammedov [Tue, 30 May 2017 16:23:58 +0000 (18:23 +0200)]
numa: make sure that all cpus have has_node_id set if numa is enabled
It fixes/add missing _PXM object for non mapped CPU (x86)
and missing fdt node (virt-arm).
It ensures that possible_cpus contains complete mapping if
numa is enabled by the time machine_init() is executed.
As result non completely mapped CPUs:
1) appear in ACPI/fdt blobs
2) QMP query-hotpluggable-cpus command shows bound nodes for such CPUs
3) allows to drop checks for has_node_id in numa only code,
reducing number of invariants incomplete mapping could produce
4) moves fixup/implicit node init from runtime numa_cpu_pre_plug()
(when CPU object is created) to machine_numa_finish_init() which
helps to fix [1, 2] and make possible_cpus complete source
of numa mapping available even before CPUs are created.
Igor Mammedov [Tue, 30 May 2017 16:23:57 +0000 (18:23 +0200)]
numa: move default mapping init to machine
there is no need use cpu_index_to_instance_props() for setting
default cpu -> node mapping. Generic machine code can do it
without cpu_index by just enabling already preset defaults
in possible_cpus.
PS:
as bonus it makes one less user of cpu_index_to_instance_props()
Eduardo Habkost [Mon, 5 Jun 2017 15:19:27 +0000 (12:19 -0300)]
pc: Use "min-[x]level" on compat_props
Since the automatic cpuid-level code was introduced in commit c39c0edf9bb3b968ba95484465a50c7b19f4aa3a ("target-i386: Automatically
set level/xlevel/xlevel2 when needed"), the CPU model tables just define
the default CPUID level code (set using "min-level"). Setting
"[x]level" forces CPUID level to a specific value and disable the
automatic-level logic.
But the PC compat code was not updated and the existing "[x]level"
compat properties broke compatibility for people using features that
triggered the auto-level code. To keep previous behavior, we should set
"min-[x]level" instead of "[x]level" on compat_props.
This was not a problem for most cases, because old machine-types don't
have full-cpuid-auto-level enabled. The only common use case it broke
was the CPUID[7] auto-level code, that was already enabled since the
first CPUID[7] feature was introduced (in QEMU 1.4.0).
This causes the regression reported at:
https://bugzilla.redhat.com/show_bug.cgi?id=1454641
Change the PC compat code to use "min-[x]level" instead of "[x]level" on
compat_props, and add new test cases to ensure we don't break this
again.
In theory this would re-enable usage of QEMU on an armv4 host.
Whether this is worthwhile is debatable -- we've been unconditionally
issuing the armv5t BX instruction in the prologue since 2011 without
complaint. Possibly we should simply require an armv6 host.
tb-hash: improve tb_jmp_cache hash function in user mode
Optimizations to cross-page chaining and indirect branches make
performance more sensitive to the hit rate of tb_jmp_cache.
The constraint of reserving some bits for the page number
lowers the achievable quality of the hashing function.
However, user-mode does not have this requirement. Thus,
with this change we use for user-mode a hashing function that
is both faster and of better quality than the previous one.
Measurements:
Note: baseline (i.e. speedup == 1x) is QEMU v2.9.0.
tcg: Introduce goto_ptr opcode and tcg_gen_lookup_and_goto_ptr
Instead of exporting goto_ptr directly to TCG frontends, export
tcg_gen_lookup_and_goto_ptr(), which calls goto_ptr with the pointer
returned by the lookup_tb_ptr() helper. This is the only use case
we have for goto_ptr and lookup_tb_ptr, so having this function is
very convenient. Furthermore, it trivially allows us to avoid calling
the lookup helper if goto_ptr is not implemented by the backend.
tcg/sparc: Use the proper compilation flags for 32-bit
We have required a v9 cpu since 9b9c37c36439ee0452632253dac7a31897f27f70.
However, the flags we were using did not reliably enable v8plus, which
meant that the compiler didn't know it could inline 64-bit atomics.
* remotes/mjt/tags/trivial-patches-fetch: (21 commits)
hw/core: nmi.c can be compiled as common-obj nowadays
dump: fix memory_mapping_filter leak
ide-test: check return of fwrite
help: Add newline to end of thread option help text
qemu-ga: remove useless allocation
scsi/lsi53c895a: Remove unused lsi_mem_*() return value
qapi: Fix some QMP documentation regressions
hw/mips: add missing include
register: display register prefix (name) since it is available
hw/sparc: use ARRAY_SIZE() macro
hw/xtensa: sim: use g_string/g_new
target/arm: add data cache invalidation cp15 instruction to cortex-r5
block: Correct documentation for BLOCK_WRITE_THRESHOLD
trivial: Remove unneeded ifndef in memory.h
altera_timer: fix incorrect memset
configure: Detect native NetBSD curses(3)
tests/libqtest: Print error instead of aborting when env variable is missing
docs/qdev-device-use.txt: update section Default Devices
docs qemu-doc: Avoid ide-drive, it's deprecated
qemu-doc: Add hyperlinks to further license information
...
Peter Maydell [Mon, 5 Jun 2017 09:09:14 +0000 (10:09 +0100)]
Merge remote-tracking branch 'remotes/elmarco/tags/chrfe-pull-request' into staging
# gpg: Signature made Fri 02 Jun 2017 20:12:48 BST
# gpg: using RSA key 0xDAE8E10975969CE5
# gpg: Good signature from "Marc-André Lureau <[email protected]>"
# gpg: aka "Marc-André Lureau <[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: 87A9 BD93 3F87 C606 D276 F62D DAE8 E109 7596 9CE5
* remotes/elmarco/tags/chrfe-pull-request:
char: move char devices to chardev/
char: make chr_fe_deinit() optionaly delete backend
char: rename functions that are not part of fe
char: move CharBackend handling in char-fe unit
char: generalize qemu_chr_write_all()
be-hci: use backend functions
chardev: serial & parallel declaration to own headers
chardev: move headers to include/chardev
Remove/replace sysemu/char.h inclusion
char-win: close file handle except with console
char-win: rename hcom->file
char-win: rename win_chr_init/poll win_chr_serial_init/poll
char-win: remove WinChardev.len
char-win: simplify win_chr_read()
char: cast ARRAY_SIZE() as signed to silent warning on empty array
help: Add newline to end of thread option help text
The help text for the thread sub option of the accel option is missing
a newline at the end. This is annoying as it makes it hard to see the
help text for the next option.
Add the new line so that the following option help text (-smp) is
displayed on a new line rather on the same line and directly after
the thread help.
Before patch:
-accel [accel=]accelerator[,thread=single|multi]
select accelerator (kvm, xen, hax or tcg; use 'help' for a list)
thread=single|multi (enable multi-threaded TCG)-smp [cpus=]n[,maxcpus=cpus][,cores=cores][,threads=threads][,sockets=sockets]
set the number of CPUs to 'n' [default=1]
maxcpus= maximum number of total cpus, including
offline CPUs for hotplug, etc
cores= number of CPU cores on one socket
threads= number of threads on one CPU core
sockets= number of discrete sockets in the system
After patch:
-accel [accel=]accelerator[,thread=single|multi]
select accelerator (kvm, xen, hax or tcg; use 'help' for a list)
thread=single|multi (enable multi-threaded TCG)
-smp [cpus=]n[,maxcpus=cpus][,cores=cores][,threads=threads][,sockets=sockets]
set the number of CPUs to 'n' [default=1]
maxcpus= maximum number of total cpus, including
offline CPUs for hotplug, etc
cores= number of CPU cores on one socket
threads= number of threads on one CPU core
sockets= number of discrete sockets in the system
Eric Blake [Tue, 23 May 2017 17:44:20 +0000 (12:44 -0500)]
qapi: Fix some QMP documentation regressions
In the process of getting rid of docs/qmp-commands.txt, we managed
to regress on some of the text that changed after the point where
the move was first branched and when the move actually occurred.
For example, commit 3282eca for blockdev-snapshot re-added the
extra "options" layer which had been cleaned up in commit 0153d2f.
This clears up all regressions identified over the range 02b351d..bd6092e:
https://lists.gnu.org/archive/html/qemu-devel/2017-05/msg05127.html
as well as a cleanup to x-blockdev-remove-medium to prefer
'id' over 'device' (matching the cleanup for 'eject').
projects / qemu.git / log