The use of &[1]uintptr{fn} was causing sigfwd to allocate memory, even
though it is being compiled for the runtime package. That is a bad
idea for this function, which is invoked by a signal handler. Rewrite
it to use only constructs that do not allocate memory when compiled
for the runtime package.
The test for this is misc/cgo/testcarchive in the main repo, which we
don't yet test.
Reviewed-on: https://go-review.googlesource.com/37454
From-SVN: r245777
The quoting was causing us to never add the system-specific signals.
The test for this is misc/cgo/testcarchive in the master repo, which
we don't yet run for gccgo.
Reviewed-on: https://go-review.googlesource.com/37453
From-SVN: r245776
Along with the update this fixes a problem that was always present but
only showed up with the new reflect test. When a program used a
**unsafe.Pointer and stored the value in an interface type, the
generated type descriptor pointed to the GC data for *unsafe.Pointer.
It did that by name, but we were not generating a variable with the
right name.
Reviewed-on: https://go-review.googlesource.com/37144
From-SVN: r245535
When I updated to Go 1.8, the initsig function moved from
signal1_unix.go to signal_unix.go, and lost a gccgo-specific change in
the move. Add it back.
This would have been caught by the misc/cgo/testcarchive tests in the
gc repository, but we don't run those in the gccgo repository. We
should fix that, somehow.
Reviewed-on: https://go-review.googlesource.com/35839
From-SVN: r244947
I typoed the argument passed to getcontext in getTraceback, and the
error was hidden by ucontext_arg. This would have been caught by some
of the runtime package tests, but we don't run most of them because
they rely on `go build`, and the go tool is not available while
running the libgo testsuite. We should fix the libgo testsuite,
somehow, so that they run.
Reviewed-on: https://go-review.googlesource.com/35837
From-SVN: r244946
Fix a bug in the generation of the hash value in reflect.FuncOf.
The merge script missed a bunch of testdata files over the years.
Copy them over.
Reviewed-on: https://go-review.googlesource.com/35570
From-SVN: r244835
PR go/79037
compiler, runtime: align gc data for m68k
The current GC requires that the gc data be aligned to at least a 4
byte boundary, because it uses the lower two bits of the address for
flags (see LOOP and PRECISE in runtime/mgc0.c). As the gc data is
stored as a [...]uintptr, that is normally always true. However, on
m68k, that only guarantees 2 byte alignment. Fix it by forcing the
alignment.
The parfor code used by the current GC requires that the parfor data
be aligned to at least an 8 byte boundary. The code in parfor.c
verifies this. This is normally true, as the data uses uint64_t
values, but, again, this must be enforced explicitly on m68k.
Fixes GCC PR 79037.
Reviewed-on: https://go-review.googlesource.com/35478
From-SVN: r244824
Some fixes that permit misc/cgo/test in the master gc repository to
pass using the current gccgo.
Install testing/internal/testdeps.gox; it is needed by `go test`.
Export runtime.lockedOSThread to enable calling via go:linkname; it is
used by misc/cgo/test.
Loop on EAGAIN when creating a new thread; this is what the gc code
does, and misc/cgo/test tests that it works.
Reviewed-on: https://go-review.googlesource.com/35479
From-SVN: r244733
PR go/79146
math/big: fix build on s390x
Don't build arith_decl_s390x.go for gccgo; it is only for assembly
code that has not yet been ported to gccgo.
For GCC PR 79146.
From-SVN: r244731
PR go/79146
crypto/elliptic: explicitly ignore p256_s390x.go
The file only works if used in conjunction with assembly code not
(yet) written for gccgo.
Fixes GCC PR 79146.
Reviewed-on: https://go-review.googlesource.com/35477
From-SVN: r244679
Since the name siginfo winds up in runtime.inc, this avoids a name
collision on systems that define "struct siginfo" in the system header
files.
Reviewed-on: https://go-review.googlesource.com/35239
From-SVN: r244484
Compiler changes:
* Change map assignment to use mapassign and assign value directly.
* Change string iteration to use decoderune, faster for ASCII strings.
* Change makeslice to take int, and use makeslice64 for larger values.
* Add new noverflow field to hmap struct used for maps.
Unresolved problems, to be fixed later:
* Commented out test in go/types/sizes_test.go that doesn't compile.
* Commented out reflect.TestStructOf test for padding after zero-sized field.
Reviewed-on: https://go-review.googlesource.com/35231
gotools/:
Updates for Go 1.8rc1.
* Makefile.am (go_cmd_go_files): Add bug.go.
(s-zdefaultcc): Write defaultPkgConfig.
* Makefile.in: Rebuild.
From-SVN: r244456
I looked at a diff of proc.go between Go 1.7 and gccgo, and copied
over all the easy stuff.
Reviewed-on: https://go-review.googlesource.com/35090
From-SVN: r244291
Drop the size arguments for the hash/equal functions stored in type
descriptors. Types know what size they are. To make this work,
generate hash/equal functions for types that can use an identity
comparison but are not a standard size and alignment.
Drop the multiplications by 33 in the generated hash code and the
reflect package hash code. They are not necessary since we started
passing a seed value around, as the seed includes the hash of the
earlier values.
Copy the algorithms for standard types from the Go 1.7 runtime,
replacing the C functions.
Reviewed-on: https://go-review.googlesource.com/34983
From-SVN: r244256
This started by moving procresize from C to Go so that we can pass the
right type to the memory allocator when allocating a p, which forced
the gomaxprocs variable to move from C to Go, and everything else
followed from that.
Reviewed-on: https://go-review.googlesource.com/34916
From-SVN: r244236
PR go/78978
libgo: build with -Wa,-nH if possible on Solaris
By default the Solaris assembler records the required hardware
capability in the object file. This means that the AES hashing code
breaks on systems that do not support AES, even though the code uses a
runtime check to only actually invoke the AES instructions on systems
that support it. An earlier fix for the problem only fixed the shared
library, not the static libgo.a. Fix the problem for real by using an
assembler option to not record the hardware capability.
For GCC PR 78978.
Patch by Rainer Orth.
Reviewed-on: https://go-review.googlesource.com/34910
From-SVN: r244165
PR go/78789
runtime: don't build aeshash.c if the assembler doesn't support it
This is for CentOS 5, whose assembler does not know the aesinc
instruction.
Fixes GCC PR 78789.
Patch by Uros Bizjak.
Reviewed-on: https://go-review.googlesource.com/34796
From-SVN: r244031
Remove support for _cgo_allocate. It was removed from the gc
toolchain in Go 1.5, so it is unlikely that anybody is trying to use it.
Reviewed-on: https://go-review.googlesource.com/34557
From-SVN: r243805
Don't put m[0-4] in runtime.inc.
Pass -mclear-hwcap to the linker if supported.
From Rainer Orth.
Reviewed-on: https://go-review.googlesource.com/34331
From-SVN: r243619
Rewrite the AES hashing code from gc assembler to C code using
intrinsics. The resulting code generates the same hash code for the
same input as the gc code--that doesn't matter as such, but testing it
ensures that the C code does something useful.
Also change mips64pe32le to mips64p32le in configure script--noticed
during CL review.
Reviewed-on: https://go-review.googlesource.com/34022
From-SVN: r243445
The gc library allocates a _panic struct on the stack. This does not
work for gccgo, because when a deferred function recovers the panic we
unwind the stack up to that point so that returning from the function
will work correctly.
Allocating on the stack fine if the panic is not recovered, and it
works fine if the panic is recovered by a function that
returns. However, it fails if the panic is recovered by a function
that itself panics, and if that second panic is then recovered by a
function higher up on the stack. When we unwind the stack to that
second panic, the g will wind up pointing at a panic farther down on
the stack. Even then everything will often work fine, except when the
deferred function catching the second panic makes a bunch of calls
that use stack space before returning. In that case the code can
overwrite the panic struct, which will then cause disaster when we
remove the struct from the linked list, as the link field will be
garbage. This case is rare enough that all the x86 tests were passing,
but there was a failure on ppc64le.
Before https://golang.org/cl/33414 we allocated the panic struct on
the heap, so go back to doing that again.
Fixesgolang/go#18228.
Reviewed-on: https://go-review.googlesource.com/34027
From-SVN: r243444
The library initialization code in go-libmain.c sets the C variable
runtime_isarchive but failed to set the Go variable runtime.isarchive.
We don't currently have a way to let C code access an unexported Go
variable, but fortunately the only time the Go function initsig is
called with an argument of true is exactly where we want to set
isarchive. So let initsig do it.
Reviewed-on: https://go-review.googlesource.com/33753
From-SVN: r243094
Since gccgo can trace back through C code as easily as Go code, we
should print C functions in the traceback.
This worked before https://golang.org/cl/31230 for a dumb reason. The
default value for runtime.traceback_cache was, and is, 2 << 2, meaning
to print all functions. The old C code for runtime_parsedebugvars
would return immediately and do nothing if the environment variable
GODEBUG was not set (if GODEBUG was set it would later call
setTraceback. The new Go code for runtime.parsedebugvars does not
return immediately if GODEBUG is not set, and always calls
setTraceback. Either way, if GOTRACEBACK is not set, setTraceback
would set traceback_cache to 1 << 2, meaning to only print non-runtime
functions and having the effect of not printing plain C functions.
Keep the current handling of GODEBUG/GOTRACEBACK, which matches the gc
library, but add an extra check to print C functions by default.
Reviewed-on: https://go-review.googlesource.com/33717
From-SVN: r243083
With -buildmode=c-archive, initsig is called before the memory
allocator has been initialized. The code was doing a memory
allocation because of the call to funcPC(sigtramp). When escape
analysis is fully implemented, that call should not allocate. For
now, finesse the issue by calling a C function to get the C function
pointer value of sigtramp.
When returning from a call from C to a Go function, a deferred
function is run to go back to syscall mode. When the call occurs on a
non-Go thread, that call sets g to nil, making it impossible to add
the _defer struct back to the pool. Just drop it and let the garbage
collector clean it up.
Reviewed-on: https://go-review.googlesource.com/33675
From-SVN: r242992
The actual stack unwind code is still in C, but the rest of the code,
notably all the memory allocation, is now in Go. The names are changed
to the names used in the Go 1.7 runtime, but the code is necessarily
somewhat different.
The __go_makefunc_can_recover function is dropped, as the uses of it
were removed in https://golang.org/cl/198770044.
Reviewed-on: https://go-review.googlesource.com/33414
From-SVN: r242715
This doesn't change any actual code, it just starts using the Go
definition of the schedt type and the sched variable rather than the C
definitions.
The schedt type is tweaked slightly for gccgo. We aren't going to
release goroutine stacks, so we don't need separate gfreeStack and
gfreeNostack lists. We only have one size of defer function, so we
don't need a list of 5 different pools.
Reviewed-on: https://go-review.googlesource.com/33364
From-SVN: r242600
Correct gcc/go/gofrontend/lex.cc and libgo/aclocal.m4 to the versions
in the gofrontend repo, which is supposed to be the master copy.
Remove a few files in libgo that somehow were not deleted in the past.
From-SVN: r242583
Update a few binary files that were changed in the master gc repo,
copied into the gofrontend repo, but not correctly copied into the GCC
repo. The changes are all minor and do not affect any actual tests.
Two instances of "http" changed to "https", and two timestamps were
zeroed out.
From-SVN: r242582
As we move toward the Go 1.7 garbage collector, it's essential that all
allocation of values that can contain Go pointers be done using the
correct type descriptor. That is simplest if we do all such allocation
in Go code. This rewrites the code that converts from a Go type to a
libffi CIF into Go.
Reviewed-on: https://go-review.googlesource.com/33353
From-SVN: r242578
A step toward eliminating goc2c.
Drop the exported parfor code; it was needed for tests in the past, but
no longer is. The Go 1.7 runtime no longer uses parfor.
Reviewed-on: https://go-review.googlesource.com/33324
From-SVN: r242509
Apparently on Solaris 10 a SA_SIGINFO signal handler can be invoked with
a nil info argument. I would not have believed it but I've now seen it
happen, and the sigaction man page actually says "If the second argument
is not equal to NULL, it points to a siginfo_t structure...." So, if
that happens, don't crash.
Also fix another case where we want to make sure that &T{} does not
allocate.
Reviewed-on: https://go-review.googlesource.com/33150
From-SVN: r242403
Add a little shell script to auto-generate runtime.sigtable from the
known signal names.
Force the main package to always import the runtime package. Otherwise
some runtime package global variables may never be initialized.
Set the syscallsp and syscallpc fields of g when entering a syscall, so
that the runtime package knows when a g is executing a syscall.
Fix runtime.funcPC to avoid dead store elimination of the interface
value when the function is inlined.
Reviewed-on: https://go-review.googlesource.com/33025
From-SVN: r242060
PR go/78172.
libgo: avoid confusion in upcase_fields in mksysinfo.sh
The mksysinfo.sh script could get confused when there were multiple
types starting with the same name. I believe this is the underlying
cause of GCC PR 78172.
Also redirect a grep to /dev/null to avoid extraneous messages during
the build.
Reviewed-on: https://go-review.googlesource.com/32821
From-SVN: r241868
