| Age | Commit message (Collapse) | Author |
|---|
|
When allocating arrays, scan size should be the position of the last pointer in the last object.
Small array allocations containing only pointers (for example, the backing store for a []*int)
have a fast path in the allocator for unrolling their bits. However, this fast path doesn't correctly
calculate the scan size for the object, and the result is that only the first pointer is actually counted.
This might lead the GC to think it has less work to do than it actually does.
Fixes #77573
Change-Id: I4b9db6db069e3fb64e4fcbbbc89b68585a71d483
Reviewed-on: https://go-review.googlesource.com/c/go/+/744840
Reviewed-by: Mark Freeman <markfreeman@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Wayne Zuo <wdvxdr1123@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
Proposal #74609
Change-Id: I97a754b128aac1bc5b7b9ab607fcd5bb390058c8
GitHub-Last-Rev: 60f2a192badf415112246de8bc6c0084085314f6
GitHub-Pull-Request: golang/go#74622
Reviewed-on: https://go-review.googlesource.com/c/go/+/688335
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: t hepudds <thepudds1460@gmail.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
|
|
This is another case very similar to CL 684015 and #74375.
In spans with type headers, mallocgc always writes to the page before
returning the allocated memory. This initial write is done by
runtime.heapSetTypeSmallHeader. Prior to the write, the compiler inserts
a nil check, implemented as a dummy instruction reading from memory.
On a freshly mapped page, this read triggers a page fault, mapping the
zero page read-only. Immediately afterwards, the write triggers another
page fault, copying to a writeable page and performing a TLB flush.
This problem is exacerbated as the process scales up. At GOMAXPROCS=6,
the tile38 sweet benchmark spends around 0.1% of cycles directly
handling these page faults. On the same machine at GOMAXPROCS=192, it
spends about 2.7% of cycles directly handling these page faults.
Replacing the read with an explicit nil check reduces the direct cost of
these page faults down to around 0.1% at GOMAXPROCS=192. There are
additional positive side-effects due to reduced contention, so the
overall time spent in page faults drops from around 12.8% to 6.8%. Most
of the remaining time in page faults is spent on automatic NUMA page
migration (completely unrelated to this issue).
Impact on the tile38 benchmark results:
│ baseline │ cl704755 │
│ sec/op │ sec/op vs base │
Tile38QueryLoad-192 1.638m ± 3% 1.494m ± 5% -8.79% (p=0.002 n=6)
│ baseline │ cl704755 │
│ average-RSS-bytes │ average-RSS-bytes vs base │
Tile38QueryLoad-192 5.384Gi ± 3% 5.399Gi ± 3% ~ (p=0.818 n=6)
│ baseline │ cl704755 │
│ peak-RSS-bytes │ peak-RSS-bytes vs base │
Tile38QueryLoad-192 5.818Gi ± 1% 5.864Gi ± 2% ~ (p=0.394 n=6)
│ baseline │ cl704755 │
│ peak-VM-bytes │ peak-VM-bytes vs base │
Tile38QueryLoad-192 7.121Gi ± 1% 7.180Gi ± 2% ~ (p=0.818 n=6)
│ baseline │ cl704755 │
│ p50-latency-sec │ p50-latency-sec vs base │
Tile38QueryLoad-192 343.2µ ± 1% 313.2µ ± 3% -8.73% (p=0.002 n=6)
│ baseline │ cl704755 │
│ p90-latency-sec │ p90-latency-sec vs base │
Tile38QueryLoad-192 1.662m ± 2% 1.603m ± 5% ~ (p=0.093 n=6)
│ baseline │ cl704755 │
│ p99-latency-sec │ p99-latency-sec vs base │
Tile38QueryLoad-192 41.56m ± 8% 35.26m ± 18% -15.17% (p=0.026 n=6)
│ baseline │ cl704755 │
│ ops/s │ ops/s vs base │
Tile38QueryLoad-192 87.89k ± 3% 96.36k ± 4% +9.64% (p=0.002 n=6)
Updates #74375.
Change-Id: I6a6a636c1a16261b6d5076f2e1b08524a6544d33
Reviewed-on: https://go-review.googlesource.com/c/go/+/704755
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Michael Pratt <mpratt@google.com>
|
|
This info makes more sense in the flags instead of as a high
bit of the kind. This makes kind access simpler because we now
don't need to mask anything.
Cleaned up most direct field accesses to use methods instead.
(reflect making new types is the only remaining direct accessor.)
IfaceIndir -> !IsDirectIface everywhere.
gocore has been updated to handle the new location. So has delve.
TODO: any other tools need updating?
Change-Id: I123f97a4d4bdd0bff1641ee7e276d1cc0bd7e8eb
Reviewed-on: https://go-review.googlesource.com/c/go/+/681936
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: David Chase <drchase@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
This is long overdue.
Change-Id: I891b114cb581e82b903c20d1c455bbbdad548fe8
Reviewed-on: https://go-review.googlesource.com/c/go/+/690535
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
|
|
CL 614257 refactored mallocgc but lost an optimization: if a span for a
large object is already backed by memory fresh from the OS (and thus
zeroed), we don't need to zero it. CL 614257 unconditionally zeroed
spans for large objects that contain pointers.
This change restores the optimization from before CL 614257, which seems
to matter in some real-world programs.
While we're here, let's also fix a hole with the garbage collector being
able to observe uninitialized memory of the large object is observed
by the conservative scanner before being published. The gory details are
in a comment in heapSetTypeLarge. In short, this change makes
span.largeType an atomic variable, such that the GC can only observe
initialized memory if span.largeType != nil.
Fixes #72991.
Change-Id: I2048aeb220ab363d252ffda7d980b8788e9674dc
Reviewed-on: https://go-review.googlesource.com/c/go/+/659956
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Felix Geisendörfer <felix.geisendoerfer@datadoghq.com>
|
|
Currently, it's possible for asynchronous preemption to observe a
partially initialized object. The sequence of events goes like this:
- The GC is in the mark phase.
- Thread T1 is allocating object O1.
- Thread T1 zeroes the allocation, runs the publication barrier, and
updates freeIndexForScan. It has not yet updated the mark bit on O1.
- Thread T2 is conservatively scanning some stack frame.
That stack frame has a dead pointer with the same address as O1.
- T2 picks up the pointer, checks isFree (which checks
freeIndexForScan without an import barrier), and sees that O1 is
allocated. It marks and queues O1.
- T2 then goes to scan O1, and observes uninitialized memory.
Although a publication barrier was executed, T2 did not have an import
barrier. T2 may thus observe T1's writes to zero the object out-of-order
with the write to freeIndexForScan.
Normally this would be impossible if T2 got a pointer to O1 from
somewhere written by T1. The publication barrier guarantees that if the
read side is data-dependent on the write side then we'd necessarily
observe all writes to O1 before T1 published it. However, T2 got the
pointer 'out of thin air' by scanning a stack frame with a dead pointer
on it.
One fix to this problem would be to add the import barrier in the
conservative scanner. We would then also need to put freeIndexForScan
behind the publication barrier, or make the write to freeIndexForScan
exactly that barrier.
However, there's a simpler way. We don't actually care if conservative
scanning observes a stale freeIndexForScan during the mark phase.
Newly-allocated memory is always marked at the point of allocation (the
allocate-black policy part of the GC's design). So it doesn't actually
matter that if the garbage collector scans that memory or not.
This change modifies the allocator to only update freeIndexForScan
outside the mark phase. This means freeIndexForScan is essentially
a snapshot of freeindex at the point the mark phase started. Because
there's no more race between conservative scanning and newly-allocated
objects, the complicated scenario above is no longer a possibility.
One thing we do have to be careful of is other callers of isFree.
Previously freeIndexForScan would always track freeindex, now it no
longer does. This change thus introduces isFreeOrNewlyAllocated which is
used by the conservative scanner, and uses freeIndexForScan. Meanwhile
isFree goes back to using freeindex like it used to. This change also
documents the requirement on isFree that the caller must have obtained
the pointer not 'out of thin air' but after the object was published.
isFree is not currently used anywhere particularly sensitive (heap dump
and checkmark mode, where the world is stopped in both cases) so using
freeindex is both conceptually simple and also safe.
Change-Id: If66b8c536b775971203fb4358c17d711c2944723
Reviewed-on: https://go-review.googlesource.com/c/go/+/672340
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
On every arch except amd64, it is faster to do x&(x-1) than x^(1<<n).
Most archs need 3 instructions for the latter: MOV $1, R; SLL n, R;
ANDN R, x. Maybe 4 if there's no ANDN.
Most archs need only 2 instructions to do x&(x-1). It takes 3 on
x86/amd64 because NEG only works in place.
Only amd64 can do x^(1<<n) in a single instruction.
(We could on 386 also, but that's currently not implemented.)
Change-Id: I3b74b7a466ab972b20a25dbb21b572baf95c3467
Reviewed-on: https://go-review.googlesource.com/c/go/+/672956
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
Our current parallel mark algorithm suffers from frequent stalls on
memory since its access pattern is essentially random. Small objects
are the worst offenders, since each one forces pulling in at least one
full cache line to access even when the amount to be scanned is far
smaller than that. Each object also requires an independent access to
per-object metadata.
The purpose of this change is to improve garbage collector performance
by scanning small objects in batches to obtain better cache locality
than our current approach. The core idea behind this change is to defer
marking and scanning small objects, and then scan them in batches
localized to a span.
This change adds scanned bits to each small object (<=512 bytes) span in
addition to mark bits. The scanned bits indicate that the object has
been scanned. (One way to think of them is "grey" bits and "black" bits
in the tri-color mark-sweep abstraction.) Each of these spans is always
8 KiB and if they contain pointers, the pointer/scalar data is already
packed together at the end of the span, allowing us to further optimize
the mark algorithm for this specific case.
When the GC encounters a pointer, it first checks if it points into a
small object span. If so, it is first marked in the mark bits, and then
the object is queued on a work-stealing P-local queue. This object
represents the whole span, and we ensure that a span can only appear at
most once in any queue by maintaining an atomic ownership bit for each
span. Later, when the pointer is dequeued, we scan every object with a
set mark that doesn't have a corresponding scanned bit. If it turns out
that was the only object in the mark bits since the last time we scanned
the span, we scan just that object directly, essentially falling back to
the existing algorithm. noscan objects have no scan work, so they are
never queued.
Each span's mark and scanned bits are co-located together at the end of
the span. Since the span is always 8 KiB in size, it can be found with
simple pointer arithmetic. Next to the marks and scans we also store the
size class, eliminating the need to access the span's mspan altogether.
The work-stealing P-local queue is a new source of GC work. If this
queue gets full, half of it is dumped to a global linked list of spans
to scan. The regular scan queues are always prioritized over this queue
to allow time for darts to accumulate. Stealing work from other Ps is a
last resort.
This change also adds a new debug mode under GODEBUG=gctrace=2 that
dumps whole-span scanning statistics by size class on every GC cycle.
A future extension to this CL is to use SIMD-accelerated scanning
kernels for scanning spans with high mark bit density.
For #19112. (Deadlock averted in GOEXPERIMENT.)
For #73581.
Change-Id: I4bbb4e36f376950a53e61aaaae157ce842c341bc
Reviewed-on: https://go-review.googlesource.com/c/go/+/658036
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
These constants are needed by some future generator programs.
Change-Id: I5dccd009cbb3b2f321523bc0d8eaeb4c82e5df81
Reviewed-on: https://go-review.googlesource.com/c/go/+/655276
Reviewed-by: Cherry Mui <cherryyz@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
Instead, have the runtime build the gc bitmaps on demand
at runtime.
Change-Id: If7a245bc62e4bce3ce80972410b0ed307d921abe
Reviewed-on: https://go-review.googlesource.com/c/go/+/616255
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@google.com>
|
|
This change finally fully fixes mallocgc for asan after the recent
refactoring. Here is everything that changed:
Fix the accounting for the alloc header; large objects don't have them.
Mask out extra bits set from unrolling the bitmap for slice backing
stores in writeHeapBitsSmall. The redzone in asan mode makes it so that
dataSize is no longer an exact multiple of typ.Size_ in this case (a
new assumption I have recently discovered) but we didn't mask out any
extra bits, so we'd accidentally set bits in other allocations. Oops.
Move the initHeapBits optimization for the 8-byte scan sizeclass on
64-bit platforms up to mallocgc, out from writeHeapBitsSmall. So, this
actually caused a problem with asan when the optimization first landed,
but we missed it. The issue was then masked once we started passing the
redzone down into writeHeapBitsSmall, since the optimization would no
longer erroneously fire on asan. What happened was that dataSize would
be 8 (because that was the user-provided alloc size) so we'd skip
writing heap bits, but it would turn out the redzone bumped the size
class, so we'd actually *have* to write the heap bits for that size
class. This is not really a problem now *but* it caused problems for me
when debugging, since I would try to remove the red zone from dataSize
and this would trigger this bug again. Ultimately, this whole situation
is confusing because the check in writeHeapBitsSmall is *not* the same
as the check in initHeapBits. By moving this check up to mallocgc, we
can make the checks align better by matching on the sizeclass, so this
should be less error-prone in the future.
Change-Id: I1e9819223be23f722f3bf21e63e812f5fb557194
Reviewed-on: https://go-review.googlesource.com/c/go/+/622041
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
Last CL we separated mallocgc into several specialized paths. Let's
split up heapSetType too. This will make the specialized heapSetType
functions inlineable and cut out some branches as well as a function
call.
Microbenchmark results at this point in the stack:
│ before.out │ after-5.out │
│ sec/op │ sec/op vs base │
Malloc8-4 13.52n ± 3% 12.15n ± 2% -10.13% (p=0.002 n=6)
Malloc16-4 21.49n ± 2% 18.32n ± 4% -14.75% (p=0.002 n=6)
MallocTypeInfo8-4 27.12n ± 1% 18.64n ± 2% -31.30% (p=0.002 n=6)
MallocTypeInfo16-4 28.71n ± 3% 21.63n ± 5% -24.65% (p=0.002 n=6)
geomean 21.81n 17.31n -20.64%
Change-Id: I5de9ac5089b9eb49bf563af2a74e6dc564420e05
Reviewed-on: https://go-review.googlesource.com/c/go/+/614795
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
|
|
This change brings back a minor optimization lost in the Go 1.22 cycle
wherein the 8-byte pointer-ful span class spans would have the pointer
bitmap written ahead of time in bulk, because there's only one possible
pattern.
│ before │ after │
│ sec/op │ sec/op vs base │
MallocTypeInfo8-4 25.13n ± 1% 23.59n ± 2% -6.15% (p=0.002 n=6)
Change-Id: I135b84bb1d5b7e678b841b56430930bc73c0a038
Reviewed-on: https://go-review.googlesource.com/c/go/+/614256
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
For whatever reason, span.heapBits is kind of slow. It accounts for
about a quarter of the cost of writeHeapBitsSmall, which is absurd. We
get a nice speed improvement for small allocations by eliminating this
call.
│ before │ after │
│ sec/op │ sec/op vs base │
MallocTypeInfo16-4 29.47n ± 1% 27.02n ± 1% -8.31% (p=0.002 n=6)
Change-Id: I6270e26902e5a9254cf1503fac81c3c799c59d6a
Reviewed-on: https://go-review.googlesource.com/c/go/+/614255
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
|
|
Cleanup and friction reduction
For #65355.
Change-Id: Ia14c9dc584a529a35b97801dd3e95b9acc99a511
Reviewed-on: https://go-review.googlesource.com/c/go/+/600436
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@golang.org>
|
|
For #67401.
Change-Id: Icc10ede72547d8020c0ba45e89d954822a4b2455
Reviewed-on: https://go-review.googlesource.com/c/go/+/587218
Auto-Submit: Russ Cox <rsc@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
This change makes it possible for the runtime to preempt the zeroing of
large objects that contain pointers. It turns out this is fairly
straightforward with allocation headers, since we can just temporarily
tell the GC that there's nothing to scan for a large object with a
single pointer write (as opposed to trying to zero a whole bunch of
bits, as we would've had to do once upon a time).
Fixes #31222.
Change-Id: I10d0dcfa3938c383282a3eb485a6f00070d07bd2
Reviewed-on: https://go-review.googlesource.com/c/go/+/577495
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
|
|
This change removes the allocheaders, deleting all the old code and
merging mbitmap_allocheaders.go back into mbitmap.go.
This change also deletes the SetType benchmarks which were already
broken in the new GOEXPERIMENT (it's harder to set up than before). We
weren't really watching these benchmarks at all, and they don't provide
additional test coverage.
Change-Id: I135497201c3259087c5cd3722ed3fbe24791d25d
Reviewed-on: https://go-review.googlesource.com/c/go/+/567200
Reviewed-by: Keith Randall <khr@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
|
|
For #59670
Change-Id: Id66e102f13e529dd041b68ce869026a56f0a1b9b
GitHub-Last-Rev: 43aa9376f72bc02a9d86518cdc99494a6b2f8573
GitHub-Pull-Request: golang/go#65564
Reviewed-on: https://go-review.googlesource.com/c/go/+/562298
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Auto-Submit: Austin Clements <austin@google.com>
|
|
For #65355
Change-Id: I65dd090fb99de9b231af2112c5ccb0eb635db2be
Reviewed-on: https://go-review.googlesource.com/c/go/+/560155
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Ibrahim Bazoka <ibrahimbazoka729@gmail.com>
Auto-Submit: Emmanuel Odeke <emmanuel@orijtech.com>
|
|
fix typo in `countAlloc` doc
Change-Id: I9f0752412b7a7dfae4915870edeab4ac52e38b2d
GitHub-Last-Rev: 6080d3c03ba6cacb1874af9724cfeb7cae27b78f
GitHub-Pull-Request: golang/go#64357
Reviewed-on: https://go-review.googlesource.com/c/go/+/544755
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Hiro Hamada <laciferin@gmail.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
This change adds the allocation headers GOEXPERIMENT which is a no-op.
It forks two runtime files temporarily to make the GOEXPERIMENT easier
to maintain. The forked files are mbitmap.go and msize.go.
Change-Id: I60202c00e614e4517de7dd000029cf80dd0121ef
Reviewed-on: https://go-review.googlesource.com/c/go/+/537980
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Keith Randall <khr@golang.org>
|
|
The writeBarrier "needed" struct member has the exact same
value as "enabled", and used interchangeably.
I'm not sure if we plan to make a distinction between the
two at some point, but today they are effectively the same,
so dedup it and keep only "enabled".
Change-Id: I65e596f174e1e820dc471a45ff70c0ef4efbc386
GitHub-Last-Rev: f8c805a91606d42c8d5b178ddd7d0bec7aaf9f55
GitHub-Pull-Request: golang/go#63814
Reviewed-on: https://go-review.googlesource.com/c/go/+/538495
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Mauri de Souza Meneguzzo <mauri870@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
|
This will make the upcoming GOEXPERIMENT easier to implement, since this
function relies on a lot of heap bitmap internals.
Change-Id: I2ab76e928e7bfd383dcdb5bfe72c9b23c2002a4e
Reviewed-on: https://go-review.googlesource.com/c/go/+/537979
Reviewed-by: Cherry Mui <cherryyz@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
We're going to want to fork this data in the near future for a
GOEXPERIMENT, so break it out now.
Change-Id: Ia7ded850bb693c443fe439c6b7279dcac638512c
Reviewed-on: https://go-review.googlesource.com/c/go/+/537978
Reviewed-by: Keith Randall <khr@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Auto-Submit: Michael Knyszek <mknyszek@google.com>
|
|
mspan.freeindex and nelems can fit into uint16 for all possible
values. Use uint16 instead of uintptr.
Change-Id: Ifce20751e81d5022be1f6b5cbb5fbe4fd1728b1b
Reviewed-on: https://go-review.googlesource.com/c/go/+/451359
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
|
|
Now that pcvalue keeps its cache on the M, we can drop all of the
stack-allocated pcvalueCaches and stop carefully passing them around
between lots of operations. This significantly simplifies a fair
amount of code and makes several structures smaller.
This series of changes has no statistically significant effect on any
runtime Stack benchmarks.
I also experimented with making the cache larger, now that the impact
is limited to the M struct, but wasn't able to measure any
improvements.
This is a re-roll of CL 515277
Change-Id: Ia27529302f81c1c92fb9c3a7474739eca80bfca1
Reviewed-on: https://go-review.googlesource.com/c/go/+/520064
Auto-Submit: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
|
|
This reverts CL 515277
Change-Id: Ie10378eed4993cb69f4a9b43a38af32b9d743155
Reviewed-on: https://go-review.googlesource.com/c/go/+/516855
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
|
|
Now that pcvalue keeps its cache on the M, we can drop all of the
stack-allocated pcvalueCaches and stop carefully passing them around
between lots of operations. This significantly simplifies a fair
amount of code and makes several structures smaller.
This series of changes has no statistically significant effect on any
runtime Stack benchmarks.
I also experimented with making the cache larger, now that the impact
is limited to the M struct, but wasn't able to measure any
improvements.
Change-Id: I4719ebf347c7150a05e887e75a238e23647c20cd
Reviewed-on: https://go-review.googlesource.com/c/go/+/515277
TryBot-Result: Gopher Robot <gobot@golang.org>
Auto-Submit: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
|
|
Some C APIs require the use or structures that contain pointers to
buffers (iovec, io_uring, ...). The pointer passing rules would
require that these buffers are allocated in C memory and to process
this data with Go libraries it would need to be copied.
In order to provide a zero-copy way to use these C APIs, this CL
implements a Pinner API that allows to pin Go objects, which
guarantees that the garbage collector does not move these objects
while pinned. This allows to relax the pointer passing rules so that
pinned pointers can be stored in C allocated memory or can be
contained in Go memory that is passed to C functions.
The Pin() method accepts pointers to objects of any type and
unsafe.Pointer. Slices and arrays can be pinned by calling Pin()
with the pointer to the first element. Pinning of maps is not
supported.
If the GC collects unreachable Pinner holding pinned objects it
panics. If Pin() is called with the other non-pointer types it
panics as well.
Performance considerations: This change has no impact on execution
time on existing code, because checks are only done in code paths,
that would panic otherwise. The memory footprint on existing code is
one pointer per memory span.
Fixes: #46787
Signed-off-by: Sven Anderson <sven@anderson.de>
Change-Id: I110031fe789b92277ae45a9455624687bd1c54f2
Reviewed-on: https://go-review.googlesource.com/c/go/+/367296
Auto-Submit: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
|
|
Change-Id: I1f031f0f83a94bebe41d3978a91a903dc5bcda66
Reviewed-on: https://go-review.googlesource.com/c/go/+/489276
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
|
Change-Id: I1c478b704d84811caa209006c657dda82d9c4cf9
Reviewed-on: https://go-review.googlesource.com/c/go/+/488435
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
|
|
This touches a lot of files, which is bad, but it is also good,
since there's N copies of this information commoned into 1.
The new files in internal/abi are copied from the end of the stack;
ultimately this will all end up being used.
Change-Id: Ia252c0055aaa72ca569411ef9f9e96e3d610889e
Reviewed-on: https://go-review.googlesource.com/c/go/+/462995
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Run-TryBot: David Chase <drchase@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
|
|
This is a really nice simplification for all of these call sites.
It also achieves a nice performance improvement for stack copying:
goos: linux
goarch: amd64
pkg: runtime
cpu: Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz
│ before │ after │
│ sec/op │ sec/op vs base │
StackCopyPtr-48 89.25m ± 1% 79.78m ± 1% -10.62% (p=0.000 n=20)
StackCopy-48 83.48m ± 2% 71.88m ± 1% -13.90% (p=0.000 n=20)
StackCopyNoCache-48 2.504m ± 2% 2.195m ± 1% -12.32% (p=0.000 n=20)
StackCopyWithStkobj-48 21.66m ± 1% 21.02m ± 2% -2.95% (p=0.000 n=20)
geomean 25.21m 22.68m -10.04%
Updates #54466.
Change-Id: I31715b7b6efd65726940041d3052bb1c0a1186f3
Reviewed-on: https://go-review.googlesource.com/c/go/+/468297
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>
|
|
Future CLs will remove the invariant that pointers are always put in
the write barrier in pairs.
The behavior of the assembly code changes a bit, where instead of writing
the pointers unconditionally and then checking for overflow, check for
overflow first and then write the pointers.
Also changed the write barrier flush function to not take the src/dst
as arguments.
Change-Id: I2ef708038367b7b82ea67cbaf505a1d5904c775c
Reviewed-on: https://go-review.googlesource.com/c/go/+/447779
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Bypass: Keith Randall <khr@golang.org>
|
|
Move this knob from a binary-startup thing to a build-time thing.
This will enable followon optmizations to the write barrier.
Change-Id: Ic3323348621c76a7dc390c09ff55016b19c43018
Reviewed-on: https://go-review.googlesource.com/c/go/+/447778
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
|
|
When the GC is scanning some memory (possibly conservatively),
finding a pointer, while concurrently another goroutine is
allocating an object at the same address as the found pointer, the
GC may see the pointer before the object and/or the heap bits are
initialized. This may cause the GC to see bad pointers and
possibly crash.
To prevent this, we make it that the scanner can only see the
object as allocated after the object and the heap bits are
initialized. Currently the allocator uses freeindex to find the
next available slot, and that code is coupled with updating the
free index to a new slot past it. The scanner also uses the
freeindex to determine if an object is allocated. This is somewhat
racy. This CL makes the scanner use a different field, which is
only updated after the object initialization (and a memory
barrier).
Fixes #54596.
Change-Id: I2a57a226369926e7192c253dd0d21d3faf22297c
Reviewed-on: https://go-review.googlesource.com/c/go/+/449017
Reviewed-by: Austin Clements <austin@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
|
This reverts commit bed2b7cf41471e1521af5a83ae28bd643eb3e038.
Reason for revert: I clicked submit by accident on the wrong CL.
Change-Id: Iddf128cb62f289d472510eb30466e515068271b2
Reviewed-on: https://go-review.googlesource.com/c/go/+/449501
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
|
|
When the GC is scanning some memory (possibly conservatively),
finding a pointer, while concurrently another goroutine is
allocating an object at the same address as the found pointer, the
GC may see the pointer before the object and/or the heap bits are
initialized. This may cause the GC to see bad pointers and
possibly crash.
To prevent this, we make it that the scanner can only see the
object as allocated after the object and the heap bits are
initialized. As the scanner uses the freeindex to determine if an
object is allocated, we delay the increment of freeindex after the
initialization.
As currently in some code path finding the next free index and
updating the free index to a new slot past it is coupled, this
needs a small refactoring. In the new code mspan.nextFreeIndex
return the next free index but not update it (although allocCache
is updated). mallocgc will update it at a later time.
Fixes #54596.
Change-Id: I6dd5ccf743f2d2c46a1ed67c6a8237fe09a71260
Reviewed-on: https://go-review.googlesource.com/c/go/+/427619
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
|
|
Change-Id: I4be0b1e612dcc21ca6bb7d4395f1c0aa52480759
GitHub-Last-Rev: 032480c4c9ddb2bedea26b01bb80b8a079bfdcf3
GitHub-Pull-Request: golang/go#55993
Reviewed-on: https://go-review.googlesource.com/c/go/+/437518
Reviewed-by: hopehook <hopehook@golangcn.org>
Reviewed-by: Keith Randall <khr@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: hopehook <hopehook@golangcn.org>
|
|
Replace all uses of Ctz64/32/8 with TrailingZeros64/32/8, because they
are the same and maybe duplicated. Also renamed CtzXX functions in 386
assembly code.
Change-Id: I19290204858083750f4be589bb0923393950ae6d
Reviewed-on: https://go-review.googlesource.com/c/go/+/438935
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Auto-Submit: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
|
|
This change adds an API to the runtime for arenas. A later CL can
potentially export it as an experimental API, but for now, just the
runtime implementation will suffice.
The purpose of arenas is to improve efficiency, primarily by allowing
for an application to manually free memory, thereby delaying garbage
collection. It comes with other potential performance benefits, such as
better locality, a better allocation strategy, and better handling of
interior pointers by the GC.
This implementation is based on one by danscales@google.com with a few
significant differences:
* The implementation lives entirely in the runtime (all layers).
* Arena chunks are the minimum of 8 MiB or the heap arena size. This
choice is made because in practice 64 MiB appears to be way too large
of an area for most real-world use-cases.
* Arena chunks are not unmapped, instead they're placed on an evacuation
list and when there are no pointers left pointing into them, they're
allowed to be reused.
* Reusing partially-used arena chunks no longer tries to find one used
by the same P first; it just takes the first one available.
* In order to ensure worst-case fragmentation is never worse than 25%,
only types and slice backing stores whose sizes are 1/4th the size of
a chunk or less may be used. Previously larger sizes, up to the size
of the chunk, were allowed.
* ASAN, MSAN, and the race detector are fully supported.
* Sets arena chunks to fault that were deferred at the end of mark
termination (a non-public patch once did this; I don't see a reason
not to continue that).
For #51317.
Change-Id: I83b1693a17302554cb36b6daa4e9249a81b1644f
Reviewed-on: https://go-review.googlesource.com/c/go/+/423359
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Michael Knyszek <mknyszek@google.com>
|
|
This places getStackMap alongside argBytes and argMapInternal as
another method of stkframe.
For #54466, albeit rather indirectly.
Change-Id: I411dda3605dd7f996983706afcbefddf29a68a85
Reviewed-on: https://go-review.googlesource.com/c/go/+/424515
Reviewed-by: Michael Pratt <mpratt@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Austin Clements <austin@google.com>
Auto-Submit: Austin Clements <austin@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
|
Some code paths in the runtime (cgo, heapdump) request heap bits
without first checking that the span is !noscan. Instead of trying
to find and work around all those cases, just set the pointer bits
of noscan spans correctly. It's somewhat safer than ensuring we
caught all the possible cases.
Fixes #54557
Fixes #54558
Change-Id: Ibd476e6cdea77c962e4d15aad26f29df66fd94e8
Reviewed-on: https://go-review.googlesource.com/c/go/+/425194
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
|
|
Updates #46731
Change-Id: Ic2208c8bb639aa1e390be0d62e2bd799ecf20654
Reviewed-on: https://go-review.googlesource.com/c/go/+/421878
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
|
|
Change-Id: I1226ff66fd0c64984939793eb8ef96c08d030fa1
Reviewed-on: https://go-review.googlesource.com/c/go/+/424399
Reviewed-by: Robert Griesemer <gri@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>
Auto-Submit: Michael Pratt <mpratt@google.com>
Run-TryBot: hopehook <hopehook@qq.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
|
[This is a retry of CL 407036 + its revert CL 422394. The only
content change is the 1-line change in cmd/internal/obj/objfile.go.]
Read the bitmaps one uintptr at a time instead of one byte at a time.
Performance so far:
Allocation heavy, no retention: ~30% faster in heapBitsSetType
Scan heavy, ~no allocation: ~even in scanobject
Change-Id: I04d899e1dbd23e989e9f552cdc1880318779c14c
Reviewed-on: https://go-review.googlesource.com/c/go/+/422635
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
|
|
[this is a retry of CL 407035 + its revert CL 422395. The content is unchanged]
Use just 1 bit per word to record the ptr/nonptr bitmap.
Use word-sized operations to manipulate the bitmap, so we can operate
on up to 64 ptr/nonptr bits at a time.
Use a separate bitmap, one bit per word of the ptr/nonptr bitmap,
to encode a no-more-pointers signal. Since we can check 64 ptr/nonptr
bits at once, knowing the exact last pointer location is not necessary.
As a followon CL, we should make the gcdata bitmap an array of
uintptr instead of an array of byte, so we can load 64 bits of it at once.
Similarly for the processing of gc programs.
Change-Id: Ica5eb622f5b87e647be64f471d67b02732ef8be6
Reviewed-on: https://go-review.googlesource.com/c/go/+/422634
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
|
|
This reverts commit b589208c8cc6e08239868f47e12c1449cd797bac.
Reason for revert: Bug somewhere in this code, causing wasm and maybe linux/386 to fail.
Change-Id: I5e1e501d839584e0219271bb937e94348f83c11f
Reviewed-on: https://go-review.googlesource.com/c/go/+/422395
Reviewed-by: Than McIntosh <thanm@google.com>
Run-TryBot: Keith Randall <khr@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
|
