Currently, Pool stores each per-P shard's overflow in a slice
protected by a Mutex. In order to store to the overflow or steal from
another shard, a P must lock that shard's Mutex. This allows for
simple synchronization between Put and Get, but has unfortunate
consequences for clearing pools.
Pools are cleared during STW sweep termination, and hence rely on
pinning a goroutine to its P to synchronize between Get/Put and
clearing. This makes the Get/Put fast path extremely fast because it
can rely on quiescence-style coordination, which doesn't even require
atomic writes, much less locking.
The catch is that a goroutine cannot acquire a Mutex while pinned to
its P (as this could deadlock). Hence, it must drop the pin on the
slow path. But this means the slow path is not synchronized with
clearing. As a result,
1) It's difficult to reason about races between clearing and the slow
path. Furthermore, this reasoning often depends on unspecified nuances
of where preemption points can occur.
2) Clearing must zero out the pointer to every object in every Pool to
prevent a concurrent slow path from causing all objects to be
retained. Since this happens during STW, this has an O(# objects in
Pools) effect on STW time.
3) We can't implement a victim cache without making clearing even
slower.
This CL solves these problems by replacing the locked overflow slice
with a lock-free structure. This allows Gets and Puts to be pinned the
whole time they're manipulating the shards slice (Pool.local), which
eliminates the races between Get/Put and clearing. This, in turn,
eliminates the need to zero all object pointers, reducing clearing to
O(# of Pools) during STW.
In addition to significantly reducing STW impact, this also happens to
speed up the Get/Put fast-path and the slow path. It somewhat
increases the cost of PoolExpensiveNew, but we'll fix that in the next
CL.
name old time/op new time/op delta
Pool-12 3.00ns ± 0% 2.21ns ±36% -26.32% (p=0.000 n=18+19)
PoolOverflow-12 600ns ± 1% 587ns ± 1% -2.21% (p=0.000 n=16+18)
PoolSTW-12 71.0µs ± 2% 5.6µs ± 3% -92.15% (p=0.000 n=20+20)
PoolExpensiveNew-12 3.14ms ± 5% 3.69ms ± 7% +17.67% (p=0.000 n=19+20)
name old p50-ns/STW new p50-ns/STW delta
PoolSTW-12 70.7k ± 1% 5.5k ± 2% -92.25% (p=0.000 n=20+20)
name old p95-ns/STW new p95-ns/STW delta
PoolSTW-12 73.1k ± 2% 6.7k ± 4% -90.86% (p=0.000 n=18+19)
name old GCs/op new GCs/op delta
PoolExpensiveNew-12 0.38 ± 1% 0.39 ± 1% +2.07% (p=0.000 n=20+18)
name old New/op new New/op delta
PoolExpensiveNew-12 33.9 ± 6% 40.0 ± 6% +17.97% (p=0.000 n=19+20)
(https://perf.golang.org/search?q=upload:20190311.1)
Fixes#22331.
For #22950.
Change-Id: Ic5cd826e25e218f3f8256dbc4d22835c1fecb391
Reviewed-on: https://go-review.googlesource.com/c/go/+/166960
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This adds two benchmarks that will highlight two problems in Pool that
we're about to address.
The first benchmark measures the impact of large Pools on GC STW time.
Currently, STW time is O(# of items in Pools), and this benchmark
demonstrates 70µs STW times.
The second benchmark measures the impact of fully clearing all Pools
on each GC. Typically this is a problem in heavily-loaded systems
because it causes a spike in allocation. This benchmark stresses this
by simulating an expensive "New" function, so the cost of creating new
objects is reflected in the ns/op of the benchmark.
For #22950, #22331.
Change-Id: I0c8853190d23144026fa11837b6bf42adc461722
Reviewed-on: https://go-review.googlesource.com/c/go/+/166959
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This adds a dynamically sized, lock-free, single-producer,
multi-consumer queue that will be used in the new Pool stealing
implementation. It's built on top of the fixed-size queue added in the
previous CL.
For #22950, #22331.
Change-Id: Ifc0ca3895bec7e7f9289ba9fb7dd0332bf96ba5a
Reviewed-on: https://go-review.googlesource.com/c/go/+/166958
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
This is the first step toward fixing multiple issues with sync.Pool.
This adds a fixed size, lock-free, single-producer, multi-consumer
queue that will be used in the new Pool stealing implementation.
For #22950, #22331.
Change-Id: I50e85e3cb83a2ee71f611ada88e7f55996504bb5
Reviewed-on: https://go-review.googlesource.com/c/go/+/166957
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
fdopendir takes ownership of its file descriptor argument.
Getdirentries shouldn't do that, so dup the file descriptor
before passing to fdopendir.
Fixes#31269
Change-Id: Ie36be8fd6c59eb339dcc9f40228d4191fc1e5850
Reviewed-on: https://go-review.googlesource.com/c/go/+/170698
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Upon unrecovered panic and program termination, process exit code is hard-coded
to 2, not set to the parameter passed to panic().
Change-Id: If64b75493227b4fd69c0bbb529f84e6df2d1b93f
Reviewed-on: https://go-review.googlesource.com/c/go/+/167709
Reviewed-by: Robert Griesemer <gri@golang.org>
load.PackageAndErrors now preloads data used to build load.Package
structures. Multiple packages may be preloaded in parallel, so this
parallelizes most of the package loading work.
The actual package construction and error-checking process is still
sequential, since this process needs to detect and report cycles.
Fixes#29758
Change-Id: Icf37e6669836ce8aad076e34fd895f97f4f3f9e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/161397
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
If x does not have an inverse modulo m, and a negative exponent is used,
return nil just like ModInverse does now.
Change-Id: I8fa72f7a851e8cf77c5fab529ede88408740626f
Reviewed-on: https://go-review.googlesource.com/c/go/+/170757
Run-TryBot: Filippo Valsorda <filippo@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
This adds an internal runtime debug log. It uses per-M time-stamped
ring buffers of binary log records. On panic, these buffers are
collected, interleaved, and printed.
The entry-point to the debug log is a new "dlog" function. dlog is
designed so it can be used even from very constrained corners of the
runtime such as signal handlers or inside the write barrier.
The facility is only enabled if the debuglog build tag is set.
Otherwise, it compiles away to a no-op implementation.
The debug log format is also designed so it would be reasonable to
decode from a core dump, though this hasn't been implemented.
Change-Id: I6e2737c286358e97a0d8091826498070b95b66a3
Reviewed-on: https://go-review.googlesource.com/c/go/+/157997
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
This change adds directives to link the res_search function in libSystem.
The corresponding Go function is then used in `lookup_darwin.go` for
resolution when cgo is disabled. This makes DNS resolution logic more
reliable as macOS has some unique quirks such as the `/etc/resolver/`
directory for specifying nameservers.
Fixes#12524
Change-Id: I367263c4951383965b3ef6491196152f78e614b1
GitHub-Last-Rev: 3c3ff6bfa7
GitHub-Pull-Request: golang/go#30686
Reviewed-on: https://go-review.googlesource.com/c/go/+/166297
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
When building as part of the bootstrap process, avoid
use of "go:linkname" applied to variables, since this
feature is ill-defined/unsupported for gccgo.
Updates #30771.
Change-Id: Id44d01b5c98d292702e5075674117518cb59e2d0
Reviewed-on: https://go-review.googlesource.com/c/go/+/170737
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This commit adds a KeepAlive field to ListenConfig and uses it
analogously to Dialer.KeepAlive to set TCP KeepAlives per default on
Accept()
Fixes#23378
Change-Id: I57eaf9508c979e7f0e2b8c5dd8e8901f6eb27fd6
GitHub-Last-Rev: e9e035d53e
GitHub-Pull-Request: golang/go#31242
Reviewed-on: https://go-review.googlesource.com/c/go/+/170678
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Before this fix, a mistaken := in a (const/type/var) declaration
ended that declaration with an error from which the parser didn't
recover well. This low-cost change will provide a better error
message and lets the parser recover perfectly.
Fixes#31092.
Change-Id: Ic4f94dc5e29dd00b7ef6d53a80dded638e3cea80
Reviewed-on: https://go-review.googlesource.com/c/go/+/169958
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Getdirentries is implemented with the __getdirentries64 function
in libSystem.dylib. That function works, but it's on Apple's
can't-be-used-in-an-app-store-application list.
Implement Getdirentries using the underlying fdopendir/readdir_r/closedir.
The simulation isn't faithful, and could be slow, but it should handle
common cases.
Don't use Getdirentries in the stdlib, use fdopendir/readdir_r/closedir
instead (via (*os.File).readdirnames).
Fixes#30933
Update #28984
RELNOTE=yes
Change-Id: Ia6b5d003e5bfe43ba54b1e1d9cfa792cc6511717
Reviewed-on: https://go-review.googlesource.com/c/go/+/168479
Reviewed-by: Emmanuel Odeke <emm.odeke@gmail.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
These new calls should not prevent NOSPLIT promotion, like the old ones.
These new calls should not prevent racefuncenter/exit removal.
(The latter was already true, as the new calls are not yet lowered
to StaticCalls at the point where racefuncenter/exit removal is done.)
Add tests to make sure we don't regress (again).
Fixes#31219
Change-Id: I3fb6b17cdd32c425829f1e2498defa813a5a9ace
Reviewed-on: https://go-review.googlesource.com/c/go/+/170639
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
Some var declarations return "extra expression" or "missing expression"
errors when they should return “assignment mismatch” instead. Change
the returned error messages to exhibit the desired behavior.
Fixes#30085.
Change-Id: I7189355fbb0f976d70100779db4f81a9ae64fb11
Reviewed-on: https://go-review.googlesource.com/c/go/+/161558
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Robert Griesemer <gri@golang.org>
We rely on SSL certificates to verify the identity of origin servers.
If an HTTPS server redirects through a plain-HTTP URL, that hop can be
compromised. We should allow it only if the user set the -insecure
flag explicitly.
Fixes#29591
Change-Id: I00639541cca2ca034c01c464385a43b3aa8ee84f
Reviewed-on: https://go-review.googlesource.com/c/go/+/156838
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
nanomsg.org currently performs an HTTPS-to-HTTP redirect, so this case
fails after the fix for #29591.
Updates #29591
Change-Id: I6306d378ef213e98c0271258bbc6669bb33f9021
Reviewed-on: https://go-review.googlesource.com/c/go/+/170637
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
mvs.BuildList and functions that invoke it directly (UpgradeAll) now
return an *mvs.BuildListError when there is an error retrieving the
requirements for a module. This new error prints the chain of
requirements from the main module to the module where the error
occurred.
These errors come up most commonly when a go.mod file has an
unexpected module path or can't be parsed for some other reason. It's
currently difficult to debug these errors because it's not clear where
the "bad" module is required from. Tools like "go list -m" and
"go mod why" don't work without the build graph.
Fixes#30661
Change-Id: I3c9d4683dcd9a5d7c259e5e4cc7e1ee209700b10
Reviewed-on: https://go-review.googlesource.com/c/go/+/166984
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
The compiler (and race detector) don't interpret locking a file as a
synchronization operation, so we add an explicit (and redundant)
sync.Mutex to make that property clear.
The additional synchronization makes it safe to parallelize the tests
in cmd/go/internal/modfetch/coderepo_test.go, which cuts the wall time
of that test by around 50%.
Updates #30550
Change-Id: Ief3479020ebf9e0fee524a4aae5568697727c683
Reviewed-on: https://go-review.googlesource.com/c/go/+/170597
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Reviewed-by: Jay Conrod <jayconrod@google.com>
As ld on AIX doesn't keep the same layout in .text section,
-Wl,-bnoobjreoder must be passed to gcc when building a C program with a
Go archive.
Change-Id: I89b584cce43ab5792f315192b073923c10d5690e
Reviewed-on: https://go-review.googlesource.com/c/go/+/170538
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Decoder.Decode and Unmarshal actually scan the input bytes twice - the
first time to check for syntax errors and the length of the value, and
the second to perform the decoding.
It's in the second scan that we actually tokenize the bytes. Since
syntax errors aren't a possibility, we can take shortcuts.
In particular, literals such as quoted strings are very common in JSON,
so we can avoid a lot of work by special casing them.
name old time/op new time/op delta
CodeDecoder-8 10.3ms ± 1% 9.1ms ± 0% -11.89% (p=0.002 n=6+6)
UnicodeDecoder-8 342ns ± 0% 283ns ± 0% -17.25% (p=0.000 n=6+5)
DecoderStream-8 239ns ± 0% 230ns ± 0% -3.90% (p=0.000 n=6+5)
CodeUnmarshal-8 11.0ms ± 0% 9.8ms ± 0% -11.45% (p=0.002 n=6+6)
CodeUnmarshalReuse-8 10.3ms ± 0% 9.0ms ± 0% -12.72% (p=0.004 n=5+6)
UnmarshalString-8 104ns ± 0% 92ns ± 0% -11.35% (p=0.002 n=6+6)
UnmarshalFloat64-8 93.2ns ± 0% 87.6ns ± 0% -6.01% (p=0.010 n=6+4)
UnmarshalInt64-8 74.5ns ± 0% 71.5ns ± 0% -3.91% (p=0.000 n=5+6)
name old speed new speed delta
CodeDecoder-8 189MB/s ± 1% 214MB/s ± 0% +13.50% (p=0.002 n=6+6)
UnicodeDecoder-8 40.9MB/s ± 0% 49.5MB/s ± 0% +20.96% (p=0.002 n=6+6)
CodeUnmarshal-8 176MB/s ± 0% 199MB/s ± 0% +12.93% (p=0.002 n=6+6)
Updates #28923.
Change-Id: I7a5e2aef51bd4ddf2004aad24210f6f50e01eaeb
Reviewed-on: https://go-review.googlesource.com/c/go/+/151042
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
(*mvsReqs).Required assumes that it is safe to mutate the slice
returned by (*mvsReqs).required. In most cases, that was true, but in
the case of -mod=vendor it resulted in unsynchronized (and
potentially interfering) writes to the global vendorList.
Fixes#30550
Change-Id: I99bcc2037e0182418b7dfda1002f8b540dbf3a1d
Reviewed-on: https://go-review.googlesource.com/c/go/+/170598
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
splittests already contains most of the tests that cover explode. Add
the missing ones and skip the append test for empty results which would
otherwise lead to an "index out of range" panic.
Change-Id: I2cb922282d2676be9ef85f186513075ae17c0243
Reviewed-on: https://go-review.googlesource.com/c/go/+/170126
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This commit changes sendmsg, recvmsg to use nsendmsg, nrecvmsg on AIX.
These syscalls support the new msghdr structure (with Control
and Controllen) which is needed for golang.org/x/net.
Also define SockaddrDataLink.
Change-Id: I233fbd24f9eb86648e0d4d50c2b56da3626292d0
Reviewed-on: https://go-review.googlesource.com/c/go/+/170537
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Tobias Klauser <tobias.klauser@gmail.com>
On darwin/amd64 the runtime method sigctxt.fixsigcode changes SIGSEGV
signals so that they are never marked SI_USER. CL 169120 changed the
signal handler to call fixsigcode even when the signal is delivered to
a non-Go thread. This breaks TestSignalForwardingExternal, so skip it.
Change-Id: I6740fb5a8f4f854ca69793537a983a696da3b495
Reviewed-on: https://go-review.googlesource.com/c/go/+/170446
Run-TryBot: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
The presence of the android_get_device_api_level symbol is used to
detect Android Q or later. Use the suggestion by Ryan Prichard and
look for it in libc.so and not in the entire program where someone
else might have defined it.
Manually tested on an Android Q amd64 emulator and arm64 Pixel.
Updates #29674
Change-Id: Iaef35d8f8910037b3690aa21f319e216a05a9a73
Reviewed-on: https://go-review.googlesource.com/c/go/+/170127
Run-TryBot: Elias Naur <mail@eliasnaur.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Package unsafe's safety rules require that pointers converted to
uintptr must be converted back to pointer-type before being stored
into memory. In particular, storing a pointer into a non-pointer-typed
expression does not guarantee the pointer stays valid, even if the
expression refers to a pointer-typed variable.
wasm's StorepNoWB implementation violates these rules by storing a
pointer through a uintptr-typed expression.
This happens to work today because esc.go is lenient in its
implementation of package unsafe's rules, but my escape analysis
rewrite follows them more rigorously, which causes val to be treated
as a non-leaking parameter.
This CL fixes the issue by using a *T-typed expression, where T is
marked //go:notinheap so that the compiler still omits the write
barrier as appropriate.
Updates #23109.
Change-Id: I49bc5474dbaa95729e5c93201493afe692591bc8
Reviewed-on: https://go-review.googlesource.com/c/go/+/170323
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The certificates argument to verifyServerCertificate must contain
at least one certificate. Simplify the intermediate certificate
handling code accordingly.
Change-Id: I8292cdfb51f418e011d6d97f47d10b4e631aa932
Reviewed-on: https://go-review.googlesource.com/c/go/+/169657
Reviewed-by: Filippo Valsorda <filippo@golang.org>
- Respect the NETRC environment variable if set.
- Ignore lines that contain macro definitions.
- Associate the 'machine' token with only the tokens that follow (not
precede) it.
Updates #29888
Updates #26232
Change-Id: I3128b7d6da2d6492df7c864e165eea1a27384f0f
Reviewed-on: https://go-review.googlesource.com/c/go/+/161698
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Jay Conrod <jayconrod@google.com>
"leaking closure reference" is redundant for similar reasons as "&x
escapes to heap" for OADDR nodes: the reference itself does not
allocate, and we already report when the referenced variable is moved
to heap.
"mark escaped content" is redundant with "leaking param content".
Updates #23109.
Change-Id: I1ab599cb1e8434f1918dd80596a70cba7dc8a0cf
Reviewed-on: https://go-review.googlesource.com/c/go/+/170321
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Add a new custom attribute to compile units containing the package name
of the package (i.e. the name after the 'package' keyword), so that
debuggers can know it when it's different from the last segment
of the package path.
Change-Id: Ieadaab6f47091aabf2f4dc42c8524452eaa6715b
Reviewed-on: https://go-review.googlesource.com/c/go/+/163677
Run-TryBot: Alessandro Arzilli <alessandro.arzilli@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
For most nodes (e.g., OPTRLIT, OMAKESLICE, OCONVIFACE), escape
analysis prints "escapes to heap" or "does not escape" to indicate
whether that node's allocation can be heap or stack allocated.
These messages are also emitted for OADDR, even though OADDR does not
actually allocate anything itself. Moreover, it's redundant because
escape analysis already prints "moved to heap" diagnostics when an
OADDR node like "&x" causes x to require heap allocation.
Because OADDR nodes don't allocate memory, my escape analysis rewrite
doesn't naturally emit the "escapes to heap" / "does not escape"
diagnostics for them. It's also non-trivial to replicate the exact
semantics esc.go uses for OADDR.
Since there are so many of these messages, I'm disabling them in this
CL by themselves. I modified esc.go to suppress the Warnl calls
without any other behavior changes, and then used a shell script to
automatically remove any ERROR messages mentioned by run.go in
"missing error" or "no match for" lines.
Fixes#16300.
Updates #23109.
Change-Id: I3993e2743c3ff83ccd0893f4e73b366ff8871a57
Reviewed-on: https://go-review.googlesource.com/c/go/+/170319
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: David Chase <drchase@google.com>
AVX-512 instructions that use RIP-relative addressing and require the
R bit of the EVEX prefix to be zero, i.e., instructions that use Z8-Z15 or
Z24-Z31, are incorrectly encoded by the assembler. The reason is that
the location of the offset at which the relative address is to be written
is incorrectly computed when the R bit is clear.
For example,
VMOVUPS bInitX<>+0(SB), Z0
encodes correctly to
62 f1 7c 48 10 05 66 e9 02 00
whereas
VMOVUPS bInitX<>+0(SB), Z8
encodes incorrectly to
62 71 7c 48 10 05 00 56 e9 02 00
Note the extra zero byte between the ModR/M byte (05) and the relative
address starting with 56. This error results in the first byte of the
following instruction being overwritten and typically, a program crash.
This commit fixes the issue in the same way that is fixed for VEX encoded
instructions, by simply not incrementing the offset for EVEX instructions.
Existing test code created for a similar VEX encoding issue (19518) has
been modified to also test for the issue addressed by this commit.
Fixes#31001
Change-Id: If84719ac22ebb5fb3c42ff96cd32b611ad497414
Reviewed-on: https://go-review.googlesource.com/c/go/+/168562
Run-TryBot: Ilya Tocar <ilya.tocar@intel.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ilya Tocar <ilya.tocar@intel.com>
