Lock contention in the hot path of moving between existing patched
TranslationBlocks is the main drag in multithreaded performance. This
patch pushes the tb_lock() usage down to the two places that really need
it:
- code generation (tb_gen_code)
- jump patching (tb_add_jump)
The rest of the code doesn't really need to hold a lock as it is either
using per-CPU structures, atomically updated or designed to be used in
concurrent read situations (qht_lookup).
To keep things simple I removed the #ifdef CONFIG_USER_ONLY stuff as the
locks become NOPs anyway until the MTTCG work is completed.
Backports commit 518615c6503ad78d3bb67ddf1cd848c4a41de02e from qemu
This ensures that if we find the TB on the slow path that tb->page_addr
is correctly set before being tested.
Backports commit 2e1ae44a4f4a6149fbb9dc812243522f07284700 from qemu
When invalidating a translation block, set an invalid flag into the
TranslationBlock structure first. It is also necessary to check whether
the target TB is still valid after acquiring 'tb_lock' but before calling
tb_add_jump() since TB lookup is to be performed out of 'tb_lock' in
future. Note that we don't have to check 'last_tb'; an already invalidated
TB will not be executed anyway and it is thus safe to patch it.
Backports commit 6d21e4208f382dd8ca1f7995a6dd9ea7ca281163 from qemu
Ensure atomicity and ordering of CPU's 'tb_flushed' access for future
translation block lookup out of 'tb_lock'.
This field can only be touched from another thread by tb_flush() in user
mode emulation. So the only access to be sequential atomic is:
* a single write in tb_flush();
* reads/writes out of 'tb_lock'.
In future, before enabling MTTCG in system mode, tb_flush() must be safe
and this field becomes unnecessary.
Backports commit 118b07308a8cedc16ef63d7ab243a95f1701db40 from qemu
Ensure atomicity of CPU's 'tb_jmp_cache' access for future translation
block lookup out of 'tb_lock'.
Note that this patch does *not* make CPU's TLB invalidation safe if it
is done from some other thread while the CPU is in its execution loop.
Backports commit 89a16b1e4294e3664667a151c2f70c84dfac6fd9 from qemu
This is a small clean up. tb_find_fast() is a final consumer of this
variable so no need to pass it by reference. 'last_tb' is always updated
by subsequent cpu_loop_exec_tb() in cpu_exec().
This change also simplifies calling cpu_exec_nocache() in
cpu_handle_exception().
Backports commit 4b7e69509df2fcbfdab8c62c294dbfcfdab8a6e1 from qemu
val is assigned twice; the second one should be combined with "|".
Reported by Coverity.
Backports commit 5ce747cfac697f61668ab4fa4a71c1dba15cc272 from qemu
There is no need to make sure that the memory is zeroed after the
allocation if we also immediatly fill the whole buffer afterwards
with memcpy(). Thus g_new0 should be g_new instead. But since we
are also doing a memcpy() here, we can also simply replace both
with g_memdup() instead.
Backports commit a337f295defad7eb977da4d6317cf70f7f2fa4b4 from qemu
QEMU's code relies on left shifts of signed integers always
being defined behaviour with the obvious 2s-complement
semantics. The only way to tell the compiler (and any
associated undefined-behaviour sanitizer) that we require a
C dialect with these semantics is to use the -fwrapv option.
This is a bit of a heavy hammer for the job as it also gives
us guaranteed semantics on integer arithmetic overflow which
in theory we don't require.
In an ideal world this would allow us to drop the warning
flag -Wno-shift-negative-value, but we must retain this to
avoid spurious warnings on clang versions predating the
fix to https://llvm.org/bugs/show_bug.cgi?id=25552.
Backports commit 2d31515bc0880a1cea86ce638d2a109f4f4e6f7d from qemu
Some software algorithms are based on the hardware's cache info, for example,
for x86 linux kernel, when cpu1 want to wakeup a task on cpu2, cpu1 will trigger
a resched IPI and told cpu2 to do the wakeup if they don't share low level
cache. Oppositely, cpu1 will access cpu2's runqueue directly if they share llc.
The relevant linux-kernel code as bellow:
static void ttwu_queue(struct task_struct *p, int cpu)
{
struct rq *rq = cpu_rq(cpu);
......
if (... && !cpus_share_cache(smp_processor_id(), cpu)) {
......
ttwu_queue_remote(p, cpu); /* will trigger RES IPI */
return;
}
......
ttwu_do_activate(rq, p, 0); /* access target's rq directly */
......
}
In real hardware, the cpus on the same socket share L3 cache, so one won't
trigger a resched IPIs when wakeup a task on others. But QEMU doesn't present a
virtual L3 cache info for VM, then the linux guest will trigger lots of RES IPIs
under some workloads even if the virtual cpus belongs to the same virtual socket.
For KVM, there will be lots of vmexit due to guest send IPIs.
The workload is a SAP HANA's testsuite, we run it one round(about 40 minuates)
and observe the (Suse11sp3)Guest's amounts of RES IPIs which triggering during
the period:
No-L3 With-L3(applied this patch)
cpu0: 363890 44582
cpu1: 373405 43109
cpu2: 340783 43797
cpu3: 333854 43409
cpu4: 327170 40038
cpu5: 325491 39922
cpu6: 319129 42391
cpu7: 306480 41035
cpu8: 161139 32188
cpu9: 164649 31024
cpu10: 149823 30398
cpu11: 149823 32455
cpu12: 164830 35143
cpu13: 172269 35805
cpu14: 179979 33898
cpu15: 194505 32754
avg: 268963.6 40129.8
The VM's topology is "1*socket 8*cores 2*threads".
After present virtual L3 cache info for VM, the amounts of RES IPIs in guest
reduce 85%.
For KVM, vcpus send IPIs will cause vmexit which is expensive, so it can cause
severe performance degradation. We had tested the overall system performance if
vcpus actually run on sparate physical socket. With L3 cache, the performance
improves 7.2%~33.1%(avg:15.7%).
Backports commit 14c985cffa6cb177fc01a163d8bcf227c104718c from qemu
If an alignment fault occurred and target EL is using AArch32,
then DFSR/IFSR bit LPAE[9] must be set correctly.
Backports commit e0fe723c24562c8f909bb40f131bfdbe75650677 from qemu
With a vfio assigned device we lay down a base MemoryRegion registered
as an IO region, giving us read & write accessors. If the region
supports mmap, we lay down a higher priority sub-region MemoryRegion
on top of the base layer initialized as a RAM device pointer to the
mmap. Finally, if we have any quirks for the device (ie. address
ranges that need additional virtualization support), we put another IO
sub-region on top of the mmap MemoryRegion. When this is flattened,
we now potentially have sub-page mmap MemoryRegions exposed which
cannot be directly mapped through KVM.
This is as expected, but a subtle detail of this is that we end up
with two different access mechanisms through QEMU. If we disable the
mmap MemoryRegion, we make use of the IO MemoryRegion and service
accesses using pread and pwrite to the vfio device file descriptor.
If the mmap MemoryRegion is enabled and results in one of these
sub-page gaps, QEMU handles the access as RAM, using memcpy to the
mmap. Using either pread/pwrite or the mmap directly should be
correct, but using memcpy causes us problems. I expect that not only
does memcpy not necessarily honor the original width and alignment in
performing a copy, but it potentially also uses processor instructions
not intended for MMIO spaces. It turns out that this has been a
problem for Realtek NIC assignment, which has such a quirk that
creates a sub-page mmap MemoryRegion access.
To resolve this, we disable memory_access_is_direct() for ram_device
regions since QEMU assumes that it can use memcpy for those regions.
Instead we access through MemoryRegionOps, which replaces the memcpy
with simple de-references of standard sizes to the host memory.
With this patch we attempt to provide unrestricted access to the RAM
device, allowing byte through qword access as well as unaligned
access. The assumption here is that accesses initiated by the VM are
driven by a device specific driver, which knows the device
capabilities. If unaligned accesses are not supported by the device,
we don't want them to work in a VM by performing multiple aligned
accesses to compose the unaligned access. A down-side of this
philosophy is that the xp command from the monitor attempts to use
the largest available access weidth, unaware of the underlying
device. Using memcpy had this same restriction, but at least now an
operator can dump individual registers, even if blocks of device
memory may result in access widths beyond the capabilities of a
given device (RTL NICs only support up to dword).
Backports commit 1b16ded6a512809f99c133a97f19026fe612b2de from qemu
Setting skip_dump on a MemoryRegion allows us to modify one specific
code path, but the restriction we're trying to address encompasses
more than that. If we have a RAM MemoryRegion backed by a physical
device, it not only restricts our ability to dump that region, but
also affects how we should manipulate it. Here we recognize that
MemoryRegions do not change to sometimes allow dumps and other times
not, so we replace setting the skip_dump flag with a new initializer
so that we know exactly the type of region to which we're applying
this behavior.
Backports commit ca83f87a66d19fdaabf23d4f5ebb49396fe232c1 from qemu
Rather than rely on recursion during the middle of register allocation,
lower indirect registers to loads and stores off the indirect base into
plain temps.
For an x86_64 host, with sufficient registers, this results in identical
code, modulo the actual register assignments.
For an i686 host, with insufficient registers, this means that temps can
be (temporarily) spilled to the stack in order to satisfy an allocation.
This as opposed to the possibility of not being able to spill, to allocate
a register for the indirect base, in order to perform a spill.
Backports commit 5a18407f55ade924aa6397c9a043a9ffd59645fe from qemu
We only need two bits per temporary. Fold the two bytes into one,
and reduce the memory and cachelines required during compilation.
Backports commit c70fbf0a9938baf3b4f843355a77c17a7e945b98 from qemu
Reduce the size of other bitfields to make room.
This reduces the cache footprint of compilation.
Backports commit bee158cb4dde35c41632a3a129c869f14a32f8f0 from qemu
Instead of using -1 as end of chain, use 0, and link through the 0
entry as a fully circular double-linked list.
Backports commit dcb8e75870e2de199db853697f8839cb603beefe from qemu
This reduces both memory usage and per-insn cacheline usage
during code generation.
Backports commit a1b3c48d2b23d6eaeb4529d3e1183d2648731bf8 from qemu
Make it obvious which macros are safe in which situations.
Useful since QEMU_ALIGN_UP and ROUND_UP both purport to do
the same thing, but differ on whether the alignment must be
a power of 2.
While implementing TLB invalidation feature we forgot to modify
part of code responsible for updating EntryHi during TLB exception.
Consequently EntryHi.EHINV is unexpectedly cleared on the exception.
Backports commit 701074a6fc7470d0ed54e4a4bcd4d491ad8da22e from qemu
If device doesn't have parent assined before its realize
is called, device_set_realized() will implicitly set parent
to '/machine/unattached'.
However device_set_realized() may fail after that point at
several other points leaving not realized object dangling
in '/machine/unattached' and as result caller of
obj = object_new()
obj->ref == 1
object_property_set_bool(obj,..., true, "realized",...)
obj->ref == 2
if (fail)
object_unref(obj);
obj->ref == 1
will get object leak instead of expected object destruction.
Fix it by making device_set_realized() to cleanup after itself
in case of failure.
Backports commit 69382d8b3e8600b349c191394d761dcb480502cf from qemu
object_property_add_child() silently fails with error that it can't
create duplicate propery 'apic' as we already have 'apic' property
registered for 'apic' feature. As result generic device_realize puts
apic into unattached container.
As it's programming error, abort if name collision happens in future
and fix property name for apic_state to 'lapic', this way apic is
a child of cpu instance.
Backports commit 6816b1b3811e839540df22855d975b6d76ae438b from qemu
These are both stored in CPUID[EAX=7,EBX=0].ECX. KVM is going to
be able to emulate both (albeit with a performance loss in the case
of RDPID, which therefore will be in KVM_GET_EMULATED_CPUID rather
than KVM_GET_SUPPORTED_CPUID).
It's also possible to implement both in TCG, but this is for 2.8.
Backports commit c2f193b538032accb9db504998bf2ea7c0ef65af from qemu
These properties will be used by as address where to plug
CPU with help -device/device_add commands.
Backports commit d89c2b8b98e097b9cad5104b0f178bde1cfa011b from qemu
Custom apic-id setter/getter doesn't do any property specific
checks anymore, so clean it up and use more compact static
property DEFINE_PROP_UINT32 instead.
Backports commit 2da00e3176abac34ca7a6aab1f5bbb94a0d03fc5 from qemu
Machine code knows about all possible APIC IDs so use that
instead of hack which does O(n^2) complexity duplicate
checks, interating over global CPUs list.
As result duplicate check is done only once with O(log n) complexity.
Backports commit 4ec60c76d5ab513e375f17b043d2b9cb849adf6c from qemu
Add the host-phys-bits boolean property, if true, take phys-bits
from the hosts physical bits value, overriding either the default
or the user specified value.
We can also use the value we read from the host to check the users
explicitly set value and warn them if it doesn't match.
Note:
a) We only read the hosts value in KVM mode (because on non-x86
we get an abort if we try)
b) We don't warn about trying to use host-phys-bits in TCG mode,
we just fall back to the TCG default. This allows the machine
type to set the host-phys-bits flag if it wants and then to
work in both TCG and KVM.
Backports commit 11f6fee576680a2d482123535da920f8ceb33eb5 from qemu
It's reverse of apicid_from_topo_ids() and will be used in follow up
patches to fill in data structures for query-hotpluggable-cpus and
for user friendly error reporting.
Backports commit 9f3aab58539b4cc716e42e772be8116dc2e7d159 from qemu
Redo 9886e834 (target-i386: Require APIC ID to be explicitly set before
CPU realize) in another way that doesn't use int64_t to detect
if apic-id property has been set.
Use the fact that 0xFFFFFFFF is the broadcast
value that a CPU can't have and set default
uint32_t apic_id to it instead of using int64_t.
Later uint32_t apic_id will be used to drop custom
property setter/getter in favor of static property.
Backports commit d9c84f196970f78d4b55ab87e03cbcad7c65f86f from qemu
Fill the bits between 51..number-of-physical-address-bits in the
MTRR_PHYSMASKn variable range mtrr masks so that they're consistent
in the migration stream irrespective of the physical address space
of the source VM in a migration.
Backports commit fcc35e7ccaed771790940524f3b0eef7aebfc9b1 from qemu
Currently QEMU sets the x86 number of physical address bits to the
magic number 40. This is only correct on some small AMD systems;
Intel systems tend to have 36, 39, 46 bits, and large AMD systems
tend to have 48.
Having the value different from your actual hardware is detectable
by the guest and in principal can cause problems;
The current limit of 40 stops TB VMs being created by those lucky
enough to have that much.
This patch lets you set the physical bits by a cpu property but
defaults to the same 40bits which matches TCGs setup.
I've removed the ancient warning about the 42 bit limit in exec.c;
I can't find that limit in there and no one else seems to know where
it is.
We use a magic value of 0 as the property default so that we can
later distinguish between the default and a user set value.
Backports commit af45907a132857cfd47acc998bf5f7c26cd13071 from qemu
Turn on the ability to pass command and event arguments in
a single boxed parameter, which must name a non-empty type
(although the type can be a struct with all optional members).
For structs, it makes it possible to pass a single qapi type
instead of a breakout of all struct members (useful if the
arguments are already in a struct or if the number of members
is large); for other complex types, it is now possible to use
a union or alternate as the data for a command or event.
The empty type may be technically feasible if needed down the
road, but it's easier to forbid it now and relax things to allow
it later, than it is to allow it now and have to special case
how the generated 'q_empty' type is handled (see commit 7ce106a9
for reasons why nothing is generated for the empty type). An
alternate type is never considered empty, but now that a boxed
type can be either an object or an alternate, we have to provide
a trivial QAPISchemaAlternateType.is_empty(). The new call to
arg_type.is_empty() during QAPISchemaCommand.check() requires
that we first check the type in question; but there is no chance
of introducing a cycle since objects do not refer back to commands.
We still have a split in syntax checking between ad-hoc parsing
up front (merely validates that 'boxed' has a sane value) and
during .check() methods (if 'boxed' is set, then 'data' must name
a non-empty user-defined type).
Generated code is unchanged, as long as no client uses the
new feature.
Backports commit c818408e449ea55371253bd4def1c1dc87b7bb03 from qemu
The next patch will add support for passing a qapi union type
as the 'data' of a command. But to do that, the user function
for implementing the command, as called by the generated
marshal command, must take the corresponding C struct as a
single boxed pointer, rather than a breakdown into one
parameter per member. Even without a union, being able to use
a C struct rather than a list of parameters can make it much
easier to handle coding with QAPI.
This patch adds the internal plumbing of a 'boxed' flag
associated with each command and event. In several cases,
this means adding indentation, with one new dead branch and
the remaining branch being the original code more deeply
nested; this was done so that the new implementation in the
next patch is easier to review without also being mixed with
indentation changes.
For this patch, no behavior or generated output changes, other
than the testsuite outputting the value of the new flag
(always False for now).
Backports commit 48825ca419fd9c8140d4fecb24e982d68ebca74f from qemu
Commit 7ce106a9 documented why we don't generated a visit_type_FOO()
for implicit types; and therefore events with an anonymous type for
'data' have to open-code a visit. Note that the open-coded visit in
qapi-event.c is slightly different from what is done in
qapi-visit.c for normal types, in part because we don't have to
check for *obj being NULL or free things on error. But where the
type is not implicit, it is nicer to reuse the normal visit instead
of open-coding a duplicate.
At the moment, the only event with a non-implicit 'data' is in the
testsuite, where test-qapi-event.c changes as follows:
|@@ -155,6 +155,7 @@ void qapi_event_send___org_qemu_x_event(
| __org_qemu_x_Struct param = {
| __org_qemu_x_member1, (char *)__org_qemu_x_member2, has_q_wchar_t, q_wchar_t
| };
|+ __org_qemu_x_Struct *arg = ¶m;
|
| emit = qmp_event_get_func_emit();
| if (!emit) {
|@@ -164,16 +165,7 @@ void qapi_event_send___org_qemu_x_event(
| qmp = qmp_event_build_dict("__ORG.QEMU_X-EVENT");
|
| v = qmp_output_visitor_new(&obj);
|-
|- visit_start_struct(v, "__ORG.QEMU_X-EVENT", NULL, 0, &err);
|- if (err) {
|- goto out;
|- }
|- visit_type___org_qemu_x_Struct_members(v, ¶m, &err);
|- if (!err) {
|- if (!err) {
|- visit_check_struct(v, &err);
|- }
|- visit_end_struct(v, NULL);
|+ visit_type___org_qemu_x_Struct(v, "__ORG.QEMU_X-EVENT", &arg, &err);
| if (err) {
| goto out;
| }
Backports commit 4d0b268fdb17a1fed10fe980e77fd388e5427bfd from qemu
Ever since commit 12f254f removed the last parameterization
of gen_err_check(), it no longer makes sense to hide the three
lines of generated C code behind a macro call. Just inline it
into the remaining users.
No change to generated code.
Backports commit fa274ed6fb788866ed3a2cfd54a2ddf78f04f2c0 from qemu
In the near future, we want to lift our artificial restriction of
no variants at the top level of an event, at which point the
currently open-coded check for empty members will become
insufficient. Factor it out into a new helper method is_empty()
now, and future-proof it by checking variants, too, along with an
assert that it is not used prior to the completion of .check().
Update places that were checking for (non-)empty .members to use
the new helper.
All of the current callers assert that there are no variants (either
directly, or by qapi.py asserting that base types have no variants),
so this is not a semantic change.
No change to generated code.
Backports commit b6167706829c6e0d3572daa2b6769594ced276f7 from qemu
