A half-shuffle operation takes a word with zeros in the high half:
0000 0000 0000 0000 ABCD EFGH IJKL MNOP
and spreads the bits out so they are in every other bit of the word:
0A0B 0C0D 0E0F 0G0H 0I0J 0K0L 0M0N 0O0P
A half-unshuffle performs the reverse operation.
Provide functions in bitops.h which implement these operations
for 32-bit and 64-bit inputs, and add tests for them.
Backports commit b355438de52d0782983bf4bdc47936189a0c988b from qemu
Add an API object_type_get_size(const char *typename) that returns the
instance_size of the give typename.
Backports commit 3f97b53a682d2595747c926c00d78b9d406f1be0 from qemu
For some workloads such as arm bootup, tb_phys_hash is performance-critical.
The is due to the high frequency of accesses to the hash table, originated
by (frequent) TLB flushes that wipe out the cpu-private tb_jmp_cache's.
More info:
https://lists.nongnu.org/archive/html/qemu-devel/2016-03/msg05098.html
To dig further into this I modified an arm image booting debian jessie to
immediately shut down after boot. Analysis revealed that quite a bit of time
is unnecessarily spent in tb_phys_hash: the cause is poor hashing that
results in very uneven loading of chains in the hash table's buckets;
the longest observed chain had ~550 elements.
The appended addresses this with two changes:
1) Use xxhash as the hash table's hash function. xxhash is a fast,
high-quality hashing function.
2) Feed the hashing function with not just tb_phys, but also pc and flags.
This improves performance over using just tb_phys for hashing, since that
resulted in some hash buckets having many TB's, while others getting very few;
with these changes, the longest observed chain on a single hash bucket is
brought down from ~550 to ~40.
Tests show that the other element checked for in tb_find_physical,
cs_base, is always a match when tb_phys+pc+flags are a match,
so hashing cs_base is wasteful. It could be that this is an ARM-only
thing, though. UPDATE:
On Tue, Apr 05, 2016 at 08:41:43 -0700, Richard Henderson wrote:
> The cs_base field is only used by i386 (in 16-bit modes), and sparc (for a TB
> consisting of only a delay slot).
> It may well still turn out to be reasonable to ignore cs_base for hashing.
BTW, after this change the hash table should not be called "tb_hash_phys"
anymore; this is addressed later in this series.
This change gives consistent bootup time improvements. I tested two
host machines:
- Intel Xeon E5-2690: 11.6% less time
- Intel i7-4790K: 19.2% less time
Increasing the number of hash buckets yields further improvements. However,
using a larger, fixed number of buckets can degrade performance for other
workloads that do not translate as many blocks (600K+ for debian-jessie arm
bootup). This is dealt with later in this series.
Backports commit 42bd32287f3a18d823f2258b813824a39ed7c6d9 from qemu
This will be used by upcoming changes for hashing the tb hash.
Add this into a separate file to include the copyright notice from
xxhash.
Backports commit dc8b295d05ec35a8c032f9abca421772347ba5d4 from qemu
The function cpu_resume_from_signal() is now always called with a
NULL puc argument, and is rather misnamed since it is never called
from a signal handler. It is essentially forcing an exit to the
top level cpu loop but without raising any exception, so rename
it to cpu_loop_exit_noexc() and drop the useless unused argument.
Backports commit 6886b98036a8f8f5bce8b10756ce080084cef11b from qemu
The WORDS_ALIGNED #define is not used anywhere, and hasn't been since
2013 when commit 612d590ebc6cef rewrote the various ld<type>_<endian>_p
functions to not use it. Remove the #define and the comment describing it.
Also remove the line in the comment about TARGET_WORDS_ALIGNED, since
it has never actually existed.
Backports commit 0d5c21f2b3bf1e0b562a2c74e353d2e03f2f50ef from qemu
Let users of qemu_get_ram_ptr and qemu_ram_ptr_length pass in an
address that is relative to the MemoryRegion. This basically means
what address_space_translate returns.
Because the semantics of the second parameter change, rename the
function to qemu_map_ram_ptr.
Backports commit 0878d0e11ba8013dd759c6921cbf05ba6a41bd71 from qemu
Move the old qemu_ram_addr_from_host to memory_region_from_host and
make it return an offset within the region. For qemu_ram_addr_from_host
return the ram_addr_t directly, similar to what it was before
commit 1b5ec23 ("memory: return MemoryRegion from qemu_ram_addr_from_host",
2013-07-04).
Backports commit 07bdaa4196b51bc7ffa7c3f74e9e4a9dc8a7966a from qemu
Of the two callers, one does not use it, and the other can compute
it itself based on the other output argument (offset) and the RAMBlock.
Backports commit f615f39616c4fd1a3a3b078af8d75bb4be6390de from qemu
Remove direct uses of ram_addr_t and optimize memory_region_{get,set}_fd
now that a MemoryRegion knows its RAMBlock directly.
Backports commit 4ff87573df3606856a92c14eef3393a63d736d11 from qemu
Currently we emit a consume-load in atomic_rcu_read. Because of
limitations in current compilers, this is overkill for non-Alpha hosts
and it is only useful to make Thread Sanitizer work.
This patch leaves the consume-load in atomic_rcu_read when
compiling with Thread Sanitizer enabled, and resorts to a
relaxed load + smp_read_barrier_depends otherwise.
On an RMO host architecture, such as aarch64, the performance
improvement of this change is easily measurable. For instance,
qht-bench performs an atomic_rcu_read on every lookup. Performance
before and after applying this patch:
$ tests/qht-bench -d 5 -n 1
Before: 9.78 MT/s
After: 10.96 MT/s
Backports commit 15487aa132109891482f79d78a30d6cfd465a391 from qemu
For correctness, smp_read_barrier_depends() is only required to
emit a barrier on Alpha hosts. However, we are currently emitting
a consume fence unconditionally, and most compilers currently treat
consume and acquire fences as equivalent.
Fix it by keeping the consume fence if we're compiling with Thread
Sanitizer, since this might help prevent false warnings. Otherwise,
only emit the barrier for Alpha hosts. Note that we still guarantee
that smp_read_barrier_depends() is a compiler barrier.
Backports commit c983895258a771f8a5e4a53950bfb7fd2216651c from qemu
The macro will be used by code that will stop calling
qemu_hw_version() at runtime and just need a constant value.
Backports commit d494352c2f7818aeba184a8ef757569083740bb2 from qemu
The collision check does nothing and hasn't been used. Remove the
variable together with related code.
Backports commit b61359781958759317ee6fd1a45b59be0b7dbbe1 from qemu
On the one hand, we have already qemu_get_ram_block() whose function
is similar. On the other hand, we can directly use mr->ram_block but
searching RAMblock by ram_addr which is a kind of waste.
Backports commit fa53a0e53efdc7002497ea4a76aacf6cceb170ef from qemu
Include qom/object.h and exec/memory.h instead of exec/ioport.h;
exec/ioport.h was almost everywhere required only for those two
includes, not for the content of the header itself.
Remove block/aio.h, everybody is already including it through
another path.
With this change, include/hw/hw.h is freed from qemu-common.h.
Backports commit df43d49cb8708b9c88a20afe0d1a3089b550a5b8 from qemu
pio_addr_t is almost unused, because these days I/O ports are simply
accessed through the address space. cpu_{in,out}[bwl] themselves are
almost unused; monitor.c and xen-hvm.c could use address_space_read/write
directly, since they have an integer size at hand. This leaves qtest as
the only user of those functions.
On the other hand even portio_* functions use this type; the only
interesting use of pio_addr_t thus is include/hw/sysbus.h. I guess I
could move it there, but I don't see much benefit in that either. Using
uint32_t is enough and avoids the need to include ioport.h everywhere.
Backports commit 89a80e7400f7225d9401b35ef32454b4ab29dc67 from qemu
exec-all.h contains TCG-specific definitions. It is not needed outside
TCG-specific files such as translate.c, exec.c or *helper.c.
One generic function had snuck into include/exec/exec-all.h; move it to
include/qom/cpu.h.
Backports commit 63c915526d6a54a95919ebece83fa9ca631b2508 from qemu
TCG backends do not need most of exec-all.h; extract what they actually
need to a separate file or move it directly to tcg.h. The next patch
will stop including exec-all.h from everywhere.
Backports commit 00f6da6a1a5d1ce085334eccbb50ec899ceed513 from qemu
Move it to the actual users. There are still a few includes of
qemu/bswap.h in headers; removing them is left for future work.
Backports commit 58369e22cf971448411bfbc8c894b2addebe2111 from qemu
Disentangle cpu-common.h and memory.h from NEED_CPU_H. Prototypes are
not defined for !NEED_CPU_H, so remove them from poison.h too. Only
macros need poisoning.
Backports commit a7d6039cb35592683ecc56d2b37817da2d2f8b00 from qemu
Return the negated value of accel_initialised is meaningless,
and the caller vl doesn't check it.
Backports commit bdc3f61dec2f9c227235bb5f677a0272e1184c82 from qemu
This field was used for telling cpu_interrupt() to unlink a chain of TBs
being executed when it worked that way. Now, cpu_interrupt() don't do
this anymore. So we don't need this field anymore.
Backports commit 3213525f8ab48742db09dab18cb9ae6f36a6c921 from qemu
'tb_invalidated_flag' was meant to catch two events:
* some TB has been invalidated by tb_phys_invalidate();
* the whole translation buffer has been flushed by tb_flush().
Then it was checked:
* in cpu_exec() to ensure that the last executed TB can be safely
linked to directly call the next one;
* in cpu_exec_nocache() to decide if the original TB should be provided
for further possible invalidation along with the temporarily
generated TB.
It is always safe to patch an invalidated TB since it is not going to be
used anyway. It is also safe to call tb_phys_invalidate() for an already
invalidated TB. Thus, setting this flag in tb_phys_invalidate() is
simply unnecessary. Moreover, it can prevent from pretty proper linking
of TBs, if any arbitrary TB has been invalidated. So just don't touch it
in tb_phys_invalidate().
If this flag is only used to catch whether tb_flush() has been called
then rename it to 'tb_flushed'. Declare it as 'bool' and stick to using
only 'true' and 'false' to set its value. Also, instead of setting it in
tb_gen_code(), just after tb_flush() has been called, do it right inside
of tb_flush().
In cpu_exec(), this flag is used to track if tb_flush() has been called
and have made 'next_tb' (a reference to the last executed TB) invalid
for linking it to directly call the next TB. tb_flush() can be called
during the CPU execution loop from tb_gen_code(), during TB execution or
by another thread while 'tb_lock' is released. Catch for translation
buffer flush reliably by resetting this flag once before first TB lookup
and each time we find it set before trying to add a direct jump. Don't
touch in in tb_find_physical().
Each vCPU has its own execution loop in multithreaded mode and thus
should have its own copy of the flag to be able to reset it with its own
'next_tb' and don't affect any other vCPU execution thread. So make this
flag per-vCPU and move it to CPUState.
In cpu_exec_nocache(), we only need to check if tb_flush() has been
called from tb_gen_code() called by cpu_exec_nocache() itself. To do
this reliably, preserve the old value of the flag, reset it before
calling tb_gen_code(), check afterwards, and combine the saved value
back to the flag.
This patch is based on the patch "tcg: move tb_invalidated_flag to
CPUState" from Paolo Bonzini <pbonzini@redhat.com>.
Backports commit 6f789be56d3f38e9214dafcfab3bf9be7191f370 from qemu
The check is to make sure that another thread hasn't already done the
same while we were outside of tb_lock. Mention this in a comment.
Backports commit 9962c478b153a18fe88a6509fe58cd178aff8abc from qemu
These fields do not contain pure pointers to a TranslationBlock
structure. So uintptr_t is the most appropriate type for them.
Also put some asserts to assure that the two least significant bits of
the pointer are always zero before assigning it to jmp_list_first.
Backports commit c37e6d7e3589ecb96914faa21025ad7ba6654aea from qemu
Briefly describe in a comment how direct block chaining is done. It
should help in understanding of the following data fields.
Rename some fields in TranslationBlock and TCGContext structures to
better reflect their purpose (dropping excessive 'tb_' prefix in
TranslationBlock but keeping it in TCGContext):
tb_next_offset => jmp_reset_offset
tb_jmp_offset => jmp_insn_offset
tb_next => jmp_target_addr
jmp_next => jmp_list_next
jmp_first => jmp_list_first
Avoid using a magic constant as an invalid offset which is used to
indicate that there's no n-th jump generated.
Backports commit f309101c26b59641fc1aa8fb2a98a5441cdaea03 from qemu
Ensure direct jump patching in ARM is atomic by using
atomic_read()/atomic_set() for code patching.
Backports commit 7d14e0e2d661479985197203589c38840e1066df from qemu
Ensure direct jump patching in s390 is atomic by:
* naturally aligning a location of direct jump address;
* using atomic_read()/atomic_set() for code patching.
Backports commit ed3d51ecd7fe248d3959e469d53890ac9ffe0cd2 from qemu
Ensure direct jump patching in i386 is atomic by:
* naturally aligning a location of direct jump address;
* using atomic_read()/atomic_set() for code patching.
Backports commit 0d07abf05e98903c7faf204a9a90f7d45b7554dc from qemu
These macros provide a convenient way to n-byte align pointers up and
down and check if a pointer is n-byte aligned.
Backports commit 6b587d3cda48e7ba26de8d30bf0d8a7063970715 from qemu
We are inconsistent with the type of tb->flags: usage varies loosely
between int and uint64_t. Settle to uint32_t everywhere, which is
superior to both: at least one target (aarch64) uses the most significant
bit in the u32, and uint64_t is wasteful.
Compile-tested for all targets.
Backports commit 89fee74a0f066dfd73830a7b5fa137e87888c870 from qemu
Use tcg_set_insn_param() instead of directly accessing internal
tcg data structures to update an insn param.
Backports commit 25caa94c4a26daaab1e65c6d887e2972aeb5749e from qemu
Returning a partial object on error is an invitation for a careless
caller to leak memory. We already fixed things in an earlier
patch to guarantee NULL if visit_start fails ("qapi: Guarantee
NULL obj on input visitor callback error"), but that does not
help the case where visit_start succeeds but some other failure
happens before visit_end, such that we leak a partially constructed
object outside visit_type_FOO(). As no one outside the testsuite
was actually relying on these semantics, it is cleaner to just
document and guarantee that ALL pointer-based visit_type_FOO()
functions always leave a safe value in *obj during an input visitor
(either the new object on success, or NULL if an error is
encountered), so callers can now unconditionally use
qapi_free_FOO() to clean up regardless of whether an error occurred.
The decision is done by adding visit_is_input(), then updating the
generated code to check if additional cleanup is needed based on
the type of visitor in use.
Note that we still leave *obj unchanged after a scalar-based
visit_type_FOO(); I did not feel like auditing all uses of
visit_type_Enum() to see if the callers would tolerate a specific
sentinel value (not to mention having to decide whether it would
be better to use 0 or ENUM__MAX as that sentinel).
Backports commit 68ab47e4b4ecc1c4649362b8cc1e49794d1a6537 from qemu
The semantics of the list visit are somewhat baroque, with the
following pseudocode when FooList is used:
start()
for (prev = head; cur = next(prev); prev = &cur) {
visit(&cur->value)
}
Note that these semantics (advance before visit) requires that
the first call to next() return the list head, while all other
calls return the next element of the list; that is, every visitor
implementation is required to track extra state to decide whether
to return the input as-is, or to advance. It also requires an
argument of 'GenericList **' to next(), solely because the first
iteration might need to modify the caller's GenericList head, so
that all other calls have to do a layer of dereferencing.
Thankfully, we only have two uses of list visits in the entire
code base: one in spapr_drc (which completely avoids
visit_next_list(), feeding in integers from a different source
than uint8List), and one in qapi-visit.py. That is, all other
list visitors are generated in qapi-visit.c, and share the same
paradigm based on a qapi FooList type, so we can refactor how
lists are laid out with minimal churn among clients.
We can greatly simplify things by hoisting the special case
into the start() routine, and flipping the order in the loop
to visit before advance:
start(head)
for (tail = *head; tail; tail = next(tail)) {
visit(&tail->value)
}
With the simpler semantics, visitors have less state to track,
the argument to next() is reduced to 'GenericList *', and it
also becomes obvious whether an input visitor is allocating a
FooList during visit_start_list() (rather than the old way of
not knowing if an allocation happened until the first
visit_next_list()). As a minor drawback, we now allocate in
two functions instead of one, and have to pass the size to
both functions (unless we were to tweak the input visitors to
cache the size to start_list for reuse during next_list, but
that defeats the goal of less visitor state).
The signature of visit_start_list() is chosen to match
visit_start_struct(), with the new parameters after 'name'.
The spapr_drc case is a virtual visit, done by passing NULL for
list, similarly to how NULL is passed to visit_start_struct()
when a qapi type is not used in those visits. It was easy to
provide these semantics for qmp-output and dealloc visitors,
and a bit harder for qmp-input (several prerequisite patches
refactored things to make this patch straightforward). But it
turned out that the string and opts visitors munge enough other
state during visit_next_list() to make it easier to just
document and require a GenericList visit for now; an assertion
will remind us to adjust things if we need the semantics in the
future.
Several pre-requisite cleanup patches made the reshuffling of
the various visitors easier; particularly the qmp input visitor.
Backports commit d9f62dde1303286b24ac8ce88be27e2b9b9c5f46 from qemu
As mentioned in previous patches, we want to call visit_end_struct()
functions unconditionally, so that visitors can release resources
tied up since the matching visit_start_struct() without also having
to worry about error priority if more than one error occurs.
Even though error_propagate() can be safely used to ignore a second
error during cleanup caused by a first error, it is simpler if the
cleanup cannot set an error. So, split out the error checking
portion (basically, input visitors checking for unvisited keys) into
a new function visit_check_struct(), which can be safely skipped if
any earlier errors are encountered, and leave the cleanup portion
(which never fails, but must be called unconditionally if
visit_start_struct() succeeded) in visit_end_struct().
Generated code in qapi-visit.c has diffs resembling:
|@@ -59,10 +59,12 @@ void visit_type_ACPIOSTInfo(Visitor *v,
| goto out_obj;
| }
| visit_type_ACPIOSTInfo_members(v, obj, &err);
|- error_propagate(errp, err);
|- err = NULL;
|+ if (err) {
|+ goto out_obj;
|+ }
|+ visit_check_struct(v, &err);
| out_obj:
|- visit_end_struct(v, &err);
|+ visit_end_struct(v);
| out:
and in qapi-event.c:
@@ -47,7 +47,10 @@ void qapi_event_send_acpi_device_ost(ACP
| goto out;
| }
| visit_type_q_obj_ACPI_DEVICE_OST_arg_members(v, ¶m, &err);
|- visit_end_struct(v, err ? NULL : &err);
|+ if (!err) {
|+ visit_check_struct(v, &err);
|+ }
|+ visit_end_struct(v);
| if (err) {
| goto out;
Backports commit 15c2f669e3fb2bc97f7b42d1871f595c0ac24af8 from qemu
Right now, qmp-output-visitor happens to produce a QNull result
if nothing is actually visited between the creation of the visitor
and the request for the resulting QObject. A stronger protocol
would require that a QMP output visit MUST visit something. But
to still be able to produce a JSON 'null' output, we need a new
visitor function that states our intentions. Yes, we could say
that such a visit must go through visit_type_any(), but that
feels clunky.
So this patch introduces the new visit_type_null() interface and
its no-op interface in the dealloc visitor, and stubs in the
qmp visitors (the next patch will finish the implementation).
For the visitors that will not implement the callback, document
the situation. The code in qapi-visit-core unconditionally
dereferences the callback pointer, so that a segfault will inform
a developer if they need to implement the callback for their
choice of visitor.
Note that JSON has a primitive null type, with the single value
null; likewise with the QNull type for QObject; but for QAPI,
we just have the 'null' value without a null type. We may
eventually want to add more support in QAPI for null (most likely,
we'd use it via an alternate type that permits 'null' or an
object); but we'll create that usage when we need it.
Backports commit 3bc97fd5924561d92f32758c67eaffd2e4e25038 from qemu
The visitor interface for mapping between QObject/QemuOpts/string
and QAPI is scandalously under-documented, making changes to visitor
core, individual visitors, and users of visitors difficult to
coordinate. Among other questions: when is it safe to pass NULL,
vs. when a string must be provided; which visitors implement which
callbacks; the difference between concrete and virtual visits.
Correct this by retrofitting proper contracts, and document where some
of the interface warts remain (for example, we may want to modify
visit_end_* to require the same 'obj' as the visit_start counterpart,
so the dealloc visitor can be simplified). Later patches in this
series will tackle some, but not all, of these warts.
Add assertions to (partially) enforce the contract. Some of these
were only made possible by recent cleanup commits.
Backports commit adfb264c9ed04bfc694921b72173be8e29e90024 from qemu
After recent changes, the only remaining use of
visit_start_implicit_struct() is for allocating the space needed
when visiting an alternate. Since the term 'implicit struct' is
hard to explain, rename the function to its current usage. While
at it, we can merge the functionality of visit_get_next_type()
into the same function, making it more like visit_start_struct().
Generated code is now slightly smaller:
| {
| Error *err = NULL;
|
|- visit_start_implicit_struct(v, (void**) obj, sizeof(BlockdevRef), &err);
|+ visit_start_alternate(v, name, (GenericAlternate **)obj, sizeof(**obj),
|+ true, &err);
| if (err) {
| goto out;
| }
|- visit_get_next_type(v, name, &(*obj)->type, true, &err);
|- if (err) {
|- goto out_obj;
|- }
| switch ((*obj)->type) {
| case QTYPE_QDICT:
| visit_start_struct(v, name, NULL, 0, &err);
...
| }
|-out_obj:
|- visit_end_implicit_struct(v);
|+ visit_end_alternate(v);
| out:
| error_propagate(errp, err);
| }
Backports commit dbf11922622685934bfb41e7cf2be9bd4a0405c0 from qemu
Rather than having two separate ways to create a QMP input
visitor, where the safer approach has the more verbose name,
it is better to consolidate things into a single function
where the caller must explicitly choose whether to be strict
or to ignore excess input. This patch is the strictly
mechanical conversion; the next patch will then audit which
uses can be made stricter.
Backports commit fc471c18d5d2ec713d5a019f9530398675494bc8 from qemu
By sticking the next pointer first, we don't need a union with
64-bit padding for smaller types. On 32-bit platforms, this
can reduce the size of uint8List from 16 bytes (or 12, depending
on whether 64-bit ints can tolerate 4-byte alignment) down to 8.
It has no effect on 64-bit platforms (where alignment still
dictates a 16-byte struct); but fewer anonymous unions is still
a win in my book.
It requires visit_next_list() to gain a size parameter, to know
what size element to allocate; comparable to the size parameter
of visit_start_struct().
I debated about going one step further, to allow for fewer casts,
by doing:
typedef GenericList GenericList;
struct GenericList {
GenericList *next;
};
struct FooList {
GenericList base;
Foo *value;
};
so that you convert to 'GenericList *' by '&foolist->base', and
back by 'container_of(generic, GenericList, base)' (as opposed to
the existing '(GenericList *)foolist' and '(FooList *)generic').
But doing that would require hoisting the declaration of
GenericList prior to inclusion of qapi-types.h, rather than its
current spot in visitor.h; it also makes iteration a bit more
verbose through 'foolist->base.next' instead of 'foolist->next'.
Note that for lists of objects, the 'value' payload is still
hidden behind a boxed pointer. Someday, it would be nice to do:
struct FooList {
FooList *next;
Foo value;
};
for one less level of malloc for each list element. This patch
is a step in that direction (now that 'next' is no longer at a
fixed non-zero offset within the struct, we can store more than
just a pointer's-worth of data as the value payload), but the
actual conversion would be a task for another series, as it will
touch a lot of code.
Backports commit e65d89bf1a4484e0db0f3dc820a8b209f2fb1e8b from qemu
We have three classes of QAPI visitors: input, output, and dealloc.
Currently, all implementations of these visitors have one thing in
common based on their visitor type: the implementation used for the
visit_type_enum() callback. But since we plan to add more such
common behavior, in relation to documenting and further refining
the semantics, it makes more sense to have the visitor
implementations advertise which class they belong to, so the common
qapi-visit-core code can use that information in multiple places.
A later patch will better document the types of visitors directly
in visitor.h.
For this patch, knowing the class of a visitor implementation lets
us make input_type_enum() and output_type_enum() become static
functions, by replacing the callback function Visitor.type_enum()
with the simpler enum member Visitor.type. Share a common
assertion in qapi-visit-core as part of the refactoring.
Move comments in opts-visitor.c to match the refactored layout.
Backports commit 983f52d4b3f86fb9dc9f8b142132feb5a8723016 from qemu
