Some code paths can lead to atomic accesses racing with memset()
on cpu->tb_jmp_cache, which can result in torn reads/writes
and is undefined behaviour in C11.
These torn accesses are unlikely to show up as bugs, but from code
inspection they seem possible. For example, tb_phys_invalidate does:
/* remove the TB from the hash list */
h = tb_jmp_cache_hash_func(tb->pc);
CPU_FOREACH(cpu) {
if (atomic_read(&cpu->tb_jmp_cache[h]) == tb) {
atomic_set(&cpu->tb_jmp_cache[h], NULL);
}
}
Here atomic_set might race with a concurrent memset (such as the
ones scheduled via "unsafe" async work, e.g. tlb_flush_page) and
therefore we might end up with a torn pointer (or who knows what,
because we are under undefined behaviour).
This patch converts parallel accesses to cpu->tb_jmp_cache to use
atomic primitives, thereby bringing these accesses back to defined
behaviour. The price to pay is to potentially execute more instructions
when clearing cpu->tb_jmp_cache, but given how infrequently they happen
and the small size of the cache, the performance impact I have measured
is within noise range when booting debian-arm.
Note that under "safe async" work (e.g. do_tb_flush) we could use memset
because no other vcpus are running. However I'm keeping these accesses
atomic as well to keep things simple and to avoid confusing analysis
tools such as ThreadSanitizer.
Backports commit f3ced3c59287dabc253f83f0c70aa4934470c15e from qemu
We are relying on cpu_env being defined as a global, yet most
targets (i.e. all but arm/a64) have it defined as a local variable.
Luckily all of them use the same "cpu_env" name, but really
compilation shouldn't break if the name of that local variable
changed.
Fix it by using tcg_ctx.tcg_env, which all targets set in their
translate_init function. This change also helps paving the way
for the upcoming "translation loop common to all targets" work.
Backports commit 53f6672bcf57d82b794a2cc3a3469be7d35c8653 from qemu
Add fsabs, fdabs, fsneg, fdneg, fsmove and fdmove.
The value is converted using the new floatx80_round() function.
Backports commit 77bdb2292492fafc4bc0fbb4d8c44fdd0ef1fa8e from qemu
Add a function to round a floatx80 to the defined precision
(floatx80_rounding_precision)
Backports commit 0f72129281765ed64d26353284059f2bdcde7a23 from qemu
fmovecr moves a floating point constant from the
FPU ROM to a floating point register.
Backports commit 9d403660d91229922c2786e81c23cc9dd8e644f1 from qemu
This may be used for deprecated object properties that are kept for
backwards compatibility.
Backports commit a733371214b68881d84725a3c71f60e2faf3b8e2 from qemu
This replaces env1 and page_index variables by env and index
so we can use VICTIM_TLB_HIT macro later.
Backports commit 3416343255cbe01fbe12e5e36cd4bb5042425b27 from qemu
Coldfire uses float64, but 680x0 use floatx80.
This patch introduces the use of floatx80 internally
and enables 680x0 80bits FPU.
Backports commit f83311e4764f1f25a8abdec2b32c64483be1759b from qemu
Switch to use QNum/uint where appropriate to remove i64 limitation.
The input visitor will cast i64 input to u64 for compatibility
reasons (existing json QMP client already use negative i64 for large
u64, and expect an implicit cast in qemu).
Note: before the patch, uint64_t values above INT64_MAX are sent over
json QMP as negative values, e.g. UINT64_MAX is sent as -1. After the
patch, they are sent unmodified. Clearly a bug fix, but we have to
consider compatibility issues anyway. libvirt should cope fine,
because its parsing of unsigned integers accepts negative values
modulo 2^64. There's hope that other clients will, too.
Backports commit 5923f85fb82df7c8c60a89458a5ae856045e5ab1 from qemu
In order to store integer values between INT64_MAX and UINT64_MAX, add
a uint64_t internal representation.
Backports commit 61a8f418b26a2d974e38e4ae55020aca8d402d88 from qemu
Before the previous commit, parameter promote_int = true made
visit_start_alternate() with an input visitor avoid QTYPE_QINT
variants and create QTYPE_QFLOAT variants instead. This was used
where QTYPE_QINT variants were invalid.
The previous commit fused QTYPE_QINT with QTYPE_QFLOAT, rendering
promote_int useless and unused.
Backports commit 60390d2dc85ffade8981ca41e02335cb07353a6d from qemu
We would like to use a same QObject type to represent numbers, whether
they are int, uint, or floats. Getters will allow some compatibility
between the various types if the number fits other representations.
Add a few more tests while at it.
Backports commit 01b2ffcedd94ad7b42bc870e4c6936c87ad03429 from qemu
QAPI_CLONE() returns a newly allocated QAPI object. Inconvenient when
we want to clone into an existing object. QAPI_CLONE_MEMBERS() does
exactly that.
Backports commit 4626a19c86c30d96cedbac2bd44ef8103303cb37 from qemu
Rather than making lots of callers wrap a scalar in a QInt, QString,
or QBool, provide helper macros that do the wrapping automatically.
Update the Coccinelle script to make mass conversions easy, although
the conversion itself will be done as a separate patches to ease
review and backport efforts.
Backports commit a92c21591b5bb9543996538f14854ca6b528318b from qemu
Visiting a list when input is the empty string should result in an
empty list, not an error. Noticed when commit 3d089ce belatedly added
tests, but simply accepted as weird then. It's actually a regression:
broken in commit 74f24cb, v2.7.0. Fix it, and throw in another test
case for empty string.
Backports commit d2788227c6185c72d88ef3127e9fed41686f8e39 from qemu
We can call tb_htable_lookup even when the tb_jmp_cache is completely
empty. Therefore, un-nest most of the code dependent on tb != NULL
from the read from the cache.
This improves the hit rate of lookup_tb_ptr; for instance, when booting
and immediately shutting down debian-arm, the hit rate improves from
93.2% to 99.4%.
Backports commit b97a879de980e99452063851597edb98e7e8039c from qemu
The new placement of the TB means that we can use one insn
to load the goto_tb destination directly from the TB.
Backports commit 308714e6bc945389c64faf1b9213e2c0d3f03391 from qemu
Since we're no longer using a direct branch, we have no
limit on the branch distance.
Backports commit acb0b292b6d0f49972dc98f742e79ed53973e438 from qemu
The new placement of the TB means that we can use one insn
to load the return value for exit_tb returning the TB pointer.
Backports commit cc74d332ff9a78684374847375ef63fc4bd10436 from qemu
We are partially initializing tb in tb_alloc. Instead, fully
initialize it in tb_gen_code, which is tb_alloc's only caller.
This saves an unnecessary write to tb->cflags.
Backports commit 2b48e10f888059a98043b4816769fa2a326a1d2c from qemu
Allocating an arbitrarily-sized array of tbs results in either
(a) a lot of memory wasted or (b) unnecessary flushes of the code
cache when we run out of TB structs in the array.
An obvious solution would be to just malloc a TB struct when needed,
and keep the TB array as an array of pointers (recall that tb_find_pc()
needs the TB array to run in O(log n)).
Perhaps a better solution, which is implemented in this patch, is to
allocate TB's right before the translated code they describe. This
results in some memory waste due to padding to have code and TBs in
separate cache lines--for instance, I measured 4.7% of padding in the
used portion of code_gen_buffer when booting aarch64 Linux on a
host with 64-byte cache lines. However, it can allow for optimizations
in some host architectures, since TCG backends could safely assume that
the TB and the corresponding translated code are very close to each
other in memory. See this message by rth for a detailed explanation:
https://lists.gnu.org/archive/html/qemu-devel/2017-03/msg05172.html
Subject: Re: GSoC 2017 Proposal: TCG performance enhancements
Backports commit 6e3b2bfd6af488a896f7936e99ef160f8f37e6f2 from qemu
Add helpers to gather cache info from the host at init-time.
For now, only export the host's I/D cache line sizes, which we
will use to improve cache locality to avoid false sharing.
Backports commit b255b2c8a5484742606e8760870ba3e14d0c9605 from qemu
V flag for subtraction is:
v = (res ^ src1) & (src1 ^ src2)
(see COMPUTE_CCR() in target/m68k/helper.c)
But gen_flush_flags() uses:
v = (res ^ src2) & (src1 ^ src2)
The problem has been found with the following program:
.global _start
_start:
move.l #-2147483648,%d0
subq.l #1,%d0
jvc 1f
move.l #1,%d1
move.l #1,%d0
trap #0
1:
move.l #0,%d1
move.l #1,%d0
trap #0
It works fine (exit(1)) on real hardware, and with "-singlestep".
"-singlestep" uses gen_helper_flush_flags(), whereas
without "-singlestep", V flag is computed directly in
gen_flush_flags().
This patch updates gen_flush_flags() to have the same result
as with gen_helper_flush_flags().
Backports commit 043b936ef6fe53396b3c6b8f5562ea3e238a071d from qemu
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.
Backports commit 5b003a40bb1ab14d0398e91f03393d3c6b9577cd from qemu
This speeds up SMM switches. Later on it may remove the need to take
the BQL, and it may also allow to reuse code between TCG and KVM.
Backports commit f8c45c6550b9ff1e1f0b92709ff3213a79870879 from qemu
It really only plays with the dispatchers, so the parameter list does
not need that complexity. This helps for readability at least.
Backports commit 003a0cf2cd1828a1141a874428571267b117f765 from qemu
