Cherry Zhang [Wed, 25 Mar 2020 21:10:16 +0000 (17:10 -0400)]
[dev.link] cmd/link: define _etext, etc. in the linker on Solaris
On Solaris, in the runtime it defines the external name of
runtime.etext as _etext (runtime/os3_solaris.go:13). In CL 224939
we changed to put external names in the ELF symbol table more
consistently. In this case it will contain _etext but not
runtime.etext.
To be conservative, this CL defines both runtime.etext and _text
in the linker.
Than McIntosh [Tue, 24 Mar 2020 13:05:19 +0000 (09:05 -0400)]
[dev.link] cmd/link: convert addexport to loader interfaces
Convert the linker's "addexport" function to use loader interfaces
for symbol manipulation instead of *sym.Symbol.
At the moment "addexport" happens after loadlibfull (there are other
chunks of functionality in the way that haven't been converted), so
this implementation contains temporary shim code to copy back the
contents of updated loader.Sym's into the corresponding sym.Symbol.
Than McIntosh [Fri, 20 Mar 2020 12:31:38 +0000 (08:31 -0400)]
[dev.link] cmd/link: move setupdynexp before loadlibfull
Rewrite setupdynexp to work with loader.Sym, and move the call to it
before the body of loadlibfull. After loadlibfull is complete,
construct the old *sym.Symbol version of dynexp, since not all all
clients that access this list are converted to the loader APIs.
Change-Id: I347d24958e2f3e2332fbe33f2eb6ec25cc126bdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/224382 Reviewed-by: Cherry Zhang <cherryyz@google.com> Reviewed-by: Jeremy Faller <jeremy@golang.org>
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Than McIntosh [Thu, 19 Mar 2020 15:04:37 +0000 (11:04 -0400)]
[dev.link] cmd/link: minor tweaks to PropagateLoaderChangesToSymbols
Update PropagateLoaderChangesToSymbols so that it no longer requires
a sym.Symbols pointer. The intent is to generalize it a little to
allow it to be used in more than just linker Dwarf generation.
Than McIntosh [Thu, 19 Mar 2020 12:17:54 +0000 (08:17 -0400)]
[dev.link] cmd/link: relocating Errorf() to ErrorReporter
Add an Errorf method to ErrorReporter. The hope is that we can
consolidate error handling/reporting in this helper, and eventually
do away with Link.Errorf and the global Errorf function (which
can be removed once we've eliminated enough uses of *sym.Symbol).
Change-Id: Ie1147020b8409b9c57acfd71c942b287b214afca
Reviewed-on: https://go-review.googlesource.com/c/go/+/224380 Reviewed-by: Jeremy Faller <jeremy@golang.org>
Cherry Zhang [Sat, 21 Mar 2020 20:06:57 +0000 (16:06 -0400)]
[dev.link] cmd/link: use Extname in doxcoff
On AIX, when external linking, we need to change the function
names to start with a dot and make function descriptors with the
names without the dot. Currently this is done through symbol
renaming, which is not friendly for switching to the loader.
In this CL we use symbol's external name for this. This allows us
to get rid of symbol renaming.
Cherry Zhang [Sat, 21 Mar 2020 17:50:46 +0000 (13:50 -0400)]
[dev.link] cmd/link: convert mangleTypeSym to new style
Use symbol's Extname, instead of symbol renaming, for the mangled
names.
The old symbol Rename has an interesting logic of "merging"
symbols, when a symbol is renamed to the name of an existing
symbol. It turns out that this is needed for linking against
shared libraries, where the Go object has a reference to a symbol
with the original name, and the shared libary provides a symbol
under the mangled name. Implement this logic with the loader.
Jeremy Faller [Mon, 9 Mar 2020 15:08:04 +0000 (11:08 -0400)]
[dev.link] cmd/link: remove global datap, moved to *Link
This change moves datap from global space into the link context. Rather
than having it exist in context, we could have it returned from dodata,
and pass it as a parameter, but it is used in awkward places in the
Arch functions. Easiest for now is just keeping it in the context, until
we more formally move it to slices of loader.Sym.
Than McIntosh [Thu, 19 Mar 2020 16:55:43 +0000 (12:55 -0400)]
[dev.link] cmd/link: set "local" attribute on symbols during preload
Set the "local" attribute on object file loader symbols during preload,
as opposed to the current strategy of only propagating the local attr
when we convert to sym.Symbols.
Change-Id: I18965e2fdaa9a0b9315d95466d17d9f41aa40a45
Reviewed-on: https://go-review.googlesource.com/c/go/+/224379
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Cherry Zhang [Fri, 20 Mar 2020 21:02:47 +0000 (17:02 -0400)]
[dev.link] cmd/link: invoke oldlink if old object format is chosen
Now we can choose the old object file format by setting
-gcflags=all=-go115newobj=false -asmflags=all=-go115newobj=false -ldflags=all=-go115newobj=false
Tested that setting all three to default false and it still works.
Change-Id: I9514b62a676916cc383b8afa389489fe7b8fa2bf
Reviewed-on: https://go-review.googlesource.com/c/go/+/224625 Reviewed-by: Than McIntosh <thanm@google.com>
It is not used for now. A flag will be introduced in followup
CLs, which will allow to switch between the old and new linker.
Some adjustments to make it still build:
- Adjust accessor of NoSplit attribute to match the change in
goobj2 package (cmd/oldlink/internal/loader/loader.go:1206).
- Change sym.Symbol's Len method to Length to match the new
interface in cmd/internal/dwarf
(cmd/oldlink/internal/sym/symbol.go:102).
- Add a TODO for deletion (cmd/oldlink/main.go:5).
Austin Clements [Fri, 20 Mar 2020 20:54:10 +0000 (16:54 -0400)]
[dev.link] cmd/link: finish phase CPU profile before running GC
Currently, phase profiling runs GC at the end of a phase before
stopping the CPU profile. Rearrange things so we stop the CPU profile
right when we collect the end time-stamp and before dealing with GCs
and heap profiles.
Austin Clements [Fri, 20 Mar 2020 19:52:43 +0000 (15:52 -0400)]
[dev.link] cmd/link: record per-phase memory profile
We already have an option to record per-phase CPU profiles. If we're
in "mem" benchmark mode, then it also makes sense to collect a heap
profile of the live heap at the end of a phase. This CL adds that
profile and changes the extensions of the profiles to "cpuprof" and
"memprof" to make the distinction clear.
Than McIntosh [Fri, 20 Mar 2020 14:34:58 +0000 (10:34 -0400)]
[dev.link] cmd/link: minor perf tweak for PropagateLoaderChangesToSymbols
When fixing up relocations in PropagateLoaderChangesToSymbols, don't
reallocate the target sym.Symbol relocation slice if it already has
the desired size (this gets rid of some unneeded allocations).
Change-Id: I05287772c18cab861c2df805fa9497103fb00dcc
Reviewed-on: https://go-review.googlesource.com/c/go/+/224420 Reviewed-by: Jeremy Faller <jeremy@golang.org>
Than McIntosh [Thu, 12 Mar 2020 18:10:09 +0000 (14:10 -0400)]
[dev.link] cmd/compile, cmd/link: move away from DWARF file renumbering
This patch moves the compiler and linker away from the current scheme
used to generate file references in DWARF subprogram dies.
Up until now the scheme has been to have the compiler emit a special
relocation on a DIE file reference that points to the file symbol in
question. The linker then reads this relocation and updates the addend
to the index of the appropriate file in the line table of the
compilation unit of the DIE (the linker emits the comp unit file
table, so it knows at that point what number use). The drawback of
this scheme is that it requires a lot of relocation processing.
With this patch, we switch to having the compiler emit the file index
directly, and then have the linker use the compiler-generated file
table to emit the line table file section (no renumbering, no
relocations, etc).
Than McIntosh [Mon, 16 Mar 2020 12:13:16 +0000 (08:13 -0400)]
[dev.link] cmd/compile, cmd/link: move DWARF info sym to anonymous aux data
Switch the primary subprogram die DWARF symbol emitted by the compiler
from named+dupOK to anonymous aux. This should help performance wise
by not having to add these symbols to the linker's symbol name lookup
tables.
Than McIntosh [Mon, 16 Mar 2020 19:24:48 +0000 (15:24 -0400)]
[dev.link] cmd/compile: refactor aux handling in newobj sym traversal
Generalize symbol traversal code for aux symbols to allow for client
control over whether the walk incldues symbols referenced by
relocations on visited aux syms. This is not needed just yet but will
be required in order to support anonymous aux syms that have
relocations.
Than McIntosh [Mon, 16 Mar 2020 17:56:42 +0000 (13:56 -0400)]
[dev.link] cmd/link: cleanup of attribute handling code
Minor cleanup of symbol attributes. Specifically:
- if a symbol originally begins life as an object file symbol,
then is converted to external in cloneToExternal, use the
previously recorded object file index for the sym to figure
out if it has read-only data (in the case that there is no
entry for it in the map in question).
- remove SetAttrShared; only the loader should be populating this
attribute at symbol creation (it never gets updated later)
Cherry Zhang [Fri, 13 Mar 2020 20:56:00 +0000 (16:56 -0400)]
[dev.link] cmd/link: don't set unreachable string variables in addstrdata
If the variable is not reachable, don't bother setting it.
The old behavior was to set it but not mark it reachable, nor the
string data it points to. This was changed in CL 219226, which
changed the variable and the string data to always reachable.
Typically it shouldn't matter much besides some waste of binary
size. But it does matter on AIX (crash in make.bash). I haven't
looked into why.
Fix AIX build.
Change-Id: I546a0c94ad77b10485ceb66e1288a408e2a2a3e2
Reviewed-on: https://go-review.googlesource.com/c/go/+/223380
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org> Reviewed-by: Than McIntosh <thanm@google.com>
Cherry Zhang [Fri, 13 Mar 2020 15:43:35 +0000 (11:43 -0400)]
[dev.link] cmd/link: don't truncate external relocation type
In CL 222244, we store external relocations in the same format as
Go object files. In Go object file format, the relocation type is
a uint8. However, for external relocations the type may not
always fit in a uint8. Truncating it will result in a bad
relocation. Fix this by storing the external reloc type on the
side. (An alternative is to extend the Go object file format to
use a uint16, but it is not necessary for Go relocations and
will waste some binary size.)
zikaeroh [Thu, 12 Mar 2020 00:02:50 +0000 (17:02 -0700)]
cmd/cover: skip function declarations with blank names
Function declarations with blank ("_") names do not introduce a binding,
and therefore cannot be referenced or executed (in fact, they do not
make it into the final compiled binary at all). As such, counters
defined while annotating their bodies will always be zero.
These types of functions are commonly used to create compile-time
checks (e.g., stringer) which are not expected to be executed.
Skip over these functions when annotating a file, preventing the unused
counters from being generated and appearing as uncovered lines in
coverage reports.
Fixes #36264
Change-Id: I6b516cf43c430a6248d68d5f483a3902253fbdab
Reviewed-on: https://go-review.googlesource.com/c/go/+/223117 Reviewed-by: Bryan C. Mills <bcmills@google.com>
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Lynn Boger [Tue, 10 Mar 2020 14:08:02 +0000 (10:08 -0400)]
cmd/internal/obj/ppc64: clean up some opcodes
This does some clean up of the ppc64 opcodes to remove names
from the opcode list that don't actually assemble. At one time
names were added to this list to represent opcode "classes" to
organize other opcodes that have the same set of operand
combinations. Since this is not documented, it is confusing as
to which opcodes can be used in an asm file and which can't, and
which opcodes should be supported in the disassembler. It is
clearer for the user if the list of Go opcodes are all opcodes
that can be assembled with names that match the ppc64 opcode
where possible.
I found this when trying to use Go opcode XXLAND in an asm file
which seems like it should map to ppc64 xxland but when used it
gets this error:
go tool asm test_xxland.s
asm: bad r/r, r/r/r or r/r/r/r opcode XXLAND
asm: assembly failed
This change removes the opcodes that are only used for opcode
"classes" and fixes the case statement where they are referenced.
This also fixes XXLAND and XXPERM which are opcodes that should
assemble to their corresponding ppc64 opcode but do not.
Ian Lance Taylor [Thu, 12 Mar 2020 05:03:50 +0000 (22:03 -0700)]
runtime: leave cleantimers early if G is being preempted
The cleantimers can run for a while in some unlikely cases.
If the GC is trying to preempt the G, it is forced to wait as the
G is holding timersLock. To avoid introducing a GC delay,
return from cleantimers if the G has a preemption request.
Fixes #37779
Change-Id: Id9a567f991e26668e2292eefc39e2edc56efa4e0
Reviewed-on: https://go-review.googlesource.com/c/go/+/223122
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Michael Knyszek <mknyszek@google.com> Reviewed-by: Cherry Zhang <cherryyz@google.com>
David Chase [Wed, 11 Mar 2020 17:36:42 +0000 (13:36 -0400)]
cmd/objdump: guard against out-of-range lines from directives.
//line bogo.go:9999999 will cause 'go tool objdump' to crash
unless bogo.go has that many lines. Guard the array index
and return innocuous values (nil, nil) from the file cache.
Walt Della [Sun, 2 Feb 2020 05:58:37 +0000 (21:58 -0800)]
cmd/go: improve pseudo-version timestamp error
The previous "invalid pseudo-version: does not match version-control
timestamp" error message used a different timestamp format than the
format used in go.mod and go.sum. For cut-and-paste-ability this patch
makes the two consistent.
Fixes #36974
Change-Id: I21f344ab9898cc584c0bcf4a75d74275a703c650
Reviewed-on: https://go-review.googlesource.com/c/go/+/217437
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan C. Mills <bcmills@google.com> Reviewed-by: Michael Matloob <matloob@golang.org> Reviewed-by: Jay Conrod <jayconrod@google.com>
Than McIntosh [Wed, 11 Mar 2020 15:16:19 +0000 (11:16 -0400)]
[dev.link] cmd/link: demote DWARF line symbols to anonymous aux
Convert DWARF .debug_line symbols to anonymous aux syms, so as
to save space in object files and reduce the number of symbols
that have to be added to the linker's lookup tables.
Than McIntosh [Wed, 11 Mar 2020 15:18:00 +0000 (11:18 -0400)]
[dev.link] cmd/internal/obj: add dump of aux symbols for -S=2
For compiler developers interested in seeing DWARF generation details,
this patch provides symbol "debug asm" dumps for DWARF aux symbols
when -S=2 is in effect.
Than McIntosh [Mon, 9 Mar 2020 15:42:03 +0000 (11:42 -0400)]
[dev.link] cmd/link: demote dwarf {range,loc} sub-symbols to aux
When the compiler emits DWARF for a function F, in addition to the
text symbol for F, it emits a set of sibling or child symbols that
carry the various DWARF bits for F (for example, go.info.F,
go.ranges.F, go.loc.F, and so on).
Prior to the linker modernization work, name lookup was the way you
made your way from a function symbol to one of its child DWARF
symbols. We now have a new mechanism (aux symbols), so there is really
no need for the DWARF sub-symbols to be named or to be dupok.
This patch converts DWARF "range" and "loc" sub-symbols to be pure aux
syms: unnamed, and connected to their parent text symbol only via aux
data. This should presumably have performance benefits in that we add
fewer symbols to the linker lookup tables.
Other related DWARF sub-symbols (ex: go.line.*) will be handled in a
subsequent patch.
Than McIntosh [Thu, 12 Mar 2020 15:04:36 +0000 (11:04 -0400)]
[dev.link] cmd/link: remove unnecessary file processing from writelines
The linker DWARF-gen's line table writing routine contains a loop that
walks all abstract function DIEs looking for files that aren't
referenced in concrete function DIEs. Turns out this loop is no longer
necessary, most likely because the compiler emits an explicit DWARF
file table into the object file.
This patch removes the offending loop. This is a prelude to some
additional work that will hopefully get rid of file renumbering in
writelines altogether (still WIP).
Change-Id: I3b3a9acce1bae7dda878ab6de2d3436de302712e
Reviewed-on: https://go-review.googlesource.com/c/go/+/223145
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Cherry Zhang [Thu, 12 Mar 2020 02:24:01 +0000 (22:24 -0400)]
[dev.link] cmd/link: refer to .got instead of .got.plt on S390X
S390X uses .got instead of .got.plt. It is changed accidentally
in CL 222977. This CL fixes it.
Also, on S390X, we need to set the relocation "variant" of
R_PCREL relocation. In the old code AddPCRelPlus has the magic.
Here we use the equivalent R_PCRELDBL, as the loader doesn't
have variant.
Rob Pike [Thu, 12 Mar 2020 00:25:39 +0000 (11:25 +1100)]
io: add a comment about how to turn a Reader into ByteReader
Offered as an alternative to CL 221380, which was more
tutorial than necessary.
Update #37344
Change-Id: Ide673b0b97983c2c2319a9311dc3d0a10567e6c4
Reviewed-on: https://go-review.googlesource.com/c/go/+/223097 Reviewed-by: Ian Lance Taylor <iant@golang.org>
Robert Griesemer [Thu, 5 Mar 2020 22:14:54 +0000 (14:14 -0800)]
cmd/compile/internal/syntax: various cleanups following CL 221603
1) Introduced setLit method to uniformly set the scanner state for
literals instead of directly manipulating the scanner fields.
2) Use a local variable 'ok' to track validity of literals instead
of relying on the side-effect of error reporters setting s.bad.
More code but clearer because it is local and explicit.
3) s/litname/baseName/ and use this function uniformly, also for
escapes. Consequently we now report always "hexadecimal" and
not "hex" (in the case of invalid escapes).
4) Added TestDirectives verifying that we get the correct directive
string (even if that string contains '%').
Verified that lines/s parsing performance is unchanged by comparing
go test -run StdLib -fast -skip "syntax/(scanner|scanner_test)\.go"
before and after (no relevant difference).
Change-Id: I143e4648fdaa31d1c365fb794a1cae4bc1c3f5ba
Reviewed-on: https://go-review.googlesource.com/c/go/+/222258
Run-TryBot: Robert Griesemer <gri@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Jay Conrod [Tue, 25 Feb 2020 16:00:08 +0000 (11:00 -0500)]
cmd/go: extract module zip files in place
Previously, we extracted module zip files to temporary directories
with random names, then renamed them to their final locations. This
failed with ERROR_ACCESS_DENIED on Windows if any file in the
temporary was open. Antivirus programs did this occasionally. Retrying
the rename did not work (CL 220978).
With this change, we extract module zip files in place. We create a
.partial file alongside the .lock file to indicate a directory is not
fully populated, and we delete this at the end of the process.
Updates #36568
Change-Id: I75c09df879a602841f3459322c021896292b2fdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/221157
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Bryan C. Mills <bcmills@google.com> Reviewed-by: Michael Matloob <matloob@golang.org>
Jay Conrod [Fri, 28 Feb 2020 21:31:19 +0000 (16:31 -0500)]
cmd/go: make module zip extraction more robust
Currently, we extract module zip files to temporary directories, then
atomically rename them into place. On Windows, this can fail with
ERROR_ACCESS_DENIED if another process (antivirus) has files open
before the rename. In CL 220978, we repeated the rename operation in a
loop over 500 ms, but this didn't solve the problem for everyone.
A better solution will extract module zip files to their permanent
locations in the cache and will keep a ".partial" marker file,
indicating when a module hasn't been fully extracted (CL 221157).
This approach is not safe if current versions of Go access the module
cache concurrently, since the module directory is detected with a
single os.Stat.
In the interim, this CL makes two changes:
1. Flaky file system operations are repeated over 2000 ms to reduce
the chance of this error occurring.
2. cmd/go will now check for .partial files created by future
versions. If a .partial file is found, it will lock the lock file,
then remove the .partial file and directory if needed.
After some time has passed and Go versions lacking this CL are no
longer supported, we can start extracting module zip files in place.
Updates #36568
Change-Id: I467ee11aa59a90b63cf0e3e761c4fec89d57d3b6
Reviewed-on: https://go-review.googlesource.com/c/go/+/221820
Run-TryBot: Jay Conrod <jayconrod@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Michael Matloob <matloob@golang.org> Reviewed-by: Bryan C. Mills <bcmills@google.com>
Russ Cox [Wed, 11 Mar 2020 04:02:04 +0000 (00:02 -0400)]
cmd/compile: fix buggy AMD64 rewrite from CL 213058
CL 213058's "bonus optimization I noticed while working on this"
turns out to be buggy. It would be correct for CMP, but not TEST.
Fix it to use TEST semantics instead.
This was breaking compilation with the upcoming Spectre mode.
Change-Id: If2d4c3798ed182f35f0244febe74e68c61e4c61b
Reviewed-on: https://go-review.googlesource.com/c/go/+/222853 Reviewed-by: Keith Randall <khr@golang.org>
Bryan C. Mills [Wed, 11 Mar 2020 13:39:54 +0000 (09:39 -0400)]
runtime: skip TestSignalIgnoreSIGTRAP on known-flaky OpenBSD builders
This test is flaky, and the cause is suspected to be an OpenBSD kernel bug.
Since there is no obvious workaround on the Go side, skip the test on
builders whose versions are known to be affected.
Fixes #17496
Change-Id: Ifa70061eb429e1d949f0fa8a9e25d177afc5c488
Reviewed-on: https://go-review.googlesource.com/c/go/+/222856
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Alexander Rakoczy <alex@golang.org>
This change forces CFGs to take the full width of their column
and allows them to be as tall as necessary.
In my (recent) experience, this makes them far less likely to
be cropped, which makes them much more useful.
On rare occasions, this can lead to gigantic CFGs,
but if you've bothered to explicitly request a CFG,
this is still better than an irrevocably truncated CFG.
erifan01 [Sun, 23 Dec 2018 02:15:44 +0000 (02:15 +0000)]
strings, bytes: improve IndexAny and LastIndexAny performance
For the case of a pattern containing multi-byte rune, the time complexity of the
previous algorithm is O(nm), and if both input arguments are long, the search
performance will be poor. This CL improves the searching performance for these
cases by using IndexRune, which is mainly implemented with IndexByte and Index.
As IndexByte and Index are specially optimized with some powerful instructions
for short patterns (an UTF8 rune is 1 to 4 bytes), so they can help to reduce the
runtime complexity of IndexAny and LastIndexAny.
Another optimization method is using hash table, however, the actual test results
show that using indexrune is better, and the space complexity is lower.
There are two fast paths in IndexAny and LastIndexAny for cases where the length
of the input arguements are 1, and their locations are not exactly the same, which
is determined based on the actual test results.
Change-Id: Ie05e306f8b184b989701868cb161ce8b3f18203b
Reviewed-on: https://go-review.googlesource.com/c/go/+/156998
Run-TryBot: eric fang <eric.fang@arm.com>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Dan Scales [Thu, 5 Mar 2020 20:46:04 +0000 (12:46 -0800)]
runtime: fix problem with repeated panic/recover/re-panics and open-coded defers
In the open-code defer implementation, we add defer struct entries to the defer
chain on-the-fly at panic time to represent stack frames that contain open-coded
defers. This allows us to process non-open-coded and open-coded defers in the
correct order. Also, we need somewhere to be able to store the 'started' state of
open-coded defers. However, if a recover succeeds, defers will now be processed
inline again (unless another panic happens). Any defer entry representing a frame
with open-coded defers will become stale once we run the corresponding defers
inline and exit the associated stack frame. So, we need to remove all entries for
open-coded defers at recover time.
The current code was only removing the top-most open-coded defer from the defer
chain during recovery. However, with recursive functions that do repeated
panic-recover-repanic, multiple stale entries can accumulate on the chain. So, we
just adjust the loop to process the entire chain. Since this is at panic/recover
case, it is fine to scan through the entire chain (which should usually have few
elements in it, since most defers are open-coded).
The added test fails with a SEGV without the fix, because it tries to run a stale
open-code defer entry (and the stack has changed).
Fixes #37664.
Change-Id: I8e3da5d610b5e607411451b66881dea887f7484d
Reviewed-on: https://go-review.googlesource.com/c/go/+/222420
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Keith Randall <khr@golang.org>
Than McIntosh [Fri, 7 Feb 2020 19:00:26 +0000 (14:00 -0500)]
cmd/link: fix for package name attr testpoint in dwarf_test.go
Tighten up a testpoint that looks for the compile unit
DW_AT_go_package_name attribute. The linker code that injects this
attribute was accidentally broken on the dev.link branch, but in a way
that wasn't detected by the test (attr was generated, but always with
an empty string). The new test will fail if the attr is an empty
string, or if we can't find the attribute for the runtime package.
Change-Id: I8b065e7eb3486646364d0eaf48a73db6acffbd18
Reviewed-on: https://go-review.googlesource.com/c/go/+/218483
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Jeremy Faller <jeremy@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com>
Than McIntosh [Fri, 6 Mar 2020 17:43:59 +0000 (12:43 -0500)]
[dev.link] cmd/asm: new -p option, changes to DWARF generation
Adds a new "-p" option to the assembler, for specifying the import
path of the package being compiled. DWARF generation is now conditional
on having a valid package path -- if we don't know the package path,
then don't emit DWARF.
This is intended to lay the groundwork for removing the various
"patchDWARFname" hacks in the linker.
Change-Id: I5f8315c0881791eb8fe1f2ba32f5bb0ae76f6b98
Reviewed-on: https://go-review.googlesource.com/c/go/+/222718
Run-TryBot: Than McIntosh <thanm@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Cherry Zhang <cherryyz@google.com> Reviewed-by: Jeremy Faller <jeremy@golang.org>
Bryan C. Mills [Thu, 5 Mar 2020 21:17:45 +0000 (16:17 -0500)]
cmd/go/internal/lockedfile: use a retry loop to suppress EDEADLK on AIX and Solaris
AIX, Solaris, and Illumos all appear to implement fcntl deadlock
detection at the granularity of processes. However, we are acquiring
and releasing file locks on individual goroutines running
concurrently: our locking occurs at a much finer granularity. As a
result, these platforms occasionally fail with EDEADLK errors, when
they detect locks that would be _misordered_ in a single-threaded
program but are safely _unordered_ in a multi-threaded context.
To work around the spurious errors, we treat EDEADLK as always
spurious, and retry the failing system call with a bounded exponential
backoff. This approach may introduce substantial latency since we no
longer benefit from kernel-scheduled wakeups in case of collisions,
but high-latency operations seem better than spurious failures.
Updates #33974
Updates #35618
Fixes #32817
Change-Id: I58b2c6a0f143bce55d6460fd4ddc3db83577ada7
Reviewed-on: https://go-review.googlesource.com/c/go/+/222277 Reviewed-by: Jay Conrod <jayconrod@google.com>
Than McIntosh [Tue, 10 Mar 2020 14:27:13 +0000 (10:27 -0400)]
[dev.link] cmd/link: fix buglet in compilationUnitByStartPC
The methods of compilationUnitByStartPC (used in DWARF generation)
were looking at comp unit sym.Symbols instead of loader.Sym's, which
will not be viable once the wavefront reaches DWARF gen phase two.
Rewrite the methods to use only loader.Sym.
Than McIntosh [Tue, 10 Mar 2020 14:08:50 +0000 (10:08 -0400)]
[dev.link] cmd/link: use shared reloc slice in writelines method
Move to a shared/reused slice of loader.Reloc's in a couple of places
in the linker's DWARF writelines method, as opposed to allocating a
new slice each time. Small performance improvement.
Than McIntosh [Tue, 10 Mar 2020 14:01:11 +0000 (10:01 -0400)]
[dev.link] cmd/link: fix up stray references to legacy "newdie" function
In a couple of places in the DWARF type generation code there were
calls to the older sym.Symbol "newdie" funtion as opposed to the
loader.Sym based method. This patch converts these to method calls.
Keith Randall [Fri, 6 Mar 2020 22:01:26 +0000 (14:01 -0800)]
runtime: make typehash match compiler generated hashes exactly
If typehash (used by reflect) does not match the built-in map's hash,
then problems occur. If a map is built using reflect, and then
assigned to a variable of map type, the hash function can change. That
causes very bad things.
This issue is rare. MapOf consults a cache of all types that occur in
the binary before making a new one. To make a true new map type (with
a hash function derived from typehash) that map type must not occur in
the binary anywhere. But to cause the bug, we need a variable of that
type in order to assign to it. The only way to make that work is to
use a named map type for the variable, so it is distinct from the
unnamed version that MapOf looks for.