Now it's failing on Windows:
panic: httptest: failed to listen on a port: listen tcp 127.0.0.1:0:
listen: An operation on a socket could not be performed because the
system lacked sufficient buffer space or because a queue was full.
Since we can't seem to understand what the test is trying to test,
and because it is causing problems on multiple systems,
delete it.
Fixes#7264.
TBR=bradfitz
CC=brainman, golang-codereviews
https://golang.org/cl/141210043
I am seeing deadlocks waiting on <-inHandler.
It seems to me that there is no guarantee that the
handler actually runs, if the client does
write header
close connection
fast enough. The server might see the EOF on the
connection before it manages to invoke the handler.
This change fixes the deadlock, but it may make
the test not actually test anything. Not sure.
LGTM=bradfitz
R=bradfitz, dvyukov
CC=golang-codereviews
https://golang.org/cl/140970043
This is one of those "how did this ever work?" bugs.
The current build failures are happening because
a fault comes up while executing on m->curg on a
system-created thread using an m obtained from needm,
but TLS is set to m->g0, not m->curg. On fault,
sigtramp starts executing, assumes r10 (g) might be
incorrect, reloads it from TLS, and gets m->g0, not
m->curg. Then sighandler dutifully pushes a call to
sigpanic onto the stack and returns to it.
We're now executing on the m->curg stack but with
g=m->g0. Sigpanic does a stack split check, sees that
the SP is not in range (50% chance depending on relative
ordering of m->g0's and m->curg's stacks), and then
calls morestack. Morestack sees that g=m->g0 and
crashes the program.
The fix is to replace every change of g in asm_arm.s
with a call to a function that both updates g and
saves the updated g to TLS.
Why did it start happening? That's unclear.
Unfortunately there were other bugs in the initial
checkin that mask exactly which of a sequence of
CLs started the behavior where sigpanic would end
up tripping the stack split.
Fixes arm build.
Fixes#8675.
LGTM=iant
R=golang-codereviews, iant
CC=dave, golang-codereviews, khr, minux, r
https://golang.org/cl/135570043
Syscall and everything it calls must be nosplit:
we cannot split a stack once Syscall has been invoked,
because we don't know which of its arguments are
pointers.
LGTM=khr, r, alex.brainman
R=dvyukov, iant, khr, r, bradfitz, alex.brainman
CC=golang-codereviews
https://golang.org/cl/133670043
Increase NOSPLIT reservation from 192 to 384 bytes.
The problem is that the non-Unix systems (Solaris and Windows)
just can't make system calls in a small amount of space,
and then worse they do things that are complex enough
to warrant calling runtime.throw on failure.
We don't have time to rewrite the code to use less stack.
I'm not happy about this, but it's still a small amount.
The good news is that we're doing this to get to only
using copying stacks for stack growth. Once that is true,
we can drop the default stack size from 8k to 4k, which
should more than make up for the bytes we're losing here.
LGTM=r
R=iant, r, bradfitz, aram.h
CC=golang-codereviews
https://golang.org/cl/140350043
Now that the calling conventions are the same,
there's no danger to using plain C for these.
LGTM=bradfitz
R=golang-codereviews, bradfitz
CC=dvyukov, golang-codereviews, iant, khr, r
https://golang.org/cl/134580043
The gp->panicwrap adjustment is just fatally flawed.
Now that there is a Panic.argp field, update that instead.
That can be done on entry only, so that unwinding doesn't
need to worry about undoing anything. The wrappers
emit a few more instructions in the prologue but everything
else in the system gets much simpler.
It also fixes (without trying) a broken test I never checked in.
Fixes#7491.
LGTM=khr
R=khr
CC=dvyukov, golang-codereviews, iant, r
https://golang.org/cl/135490044
testSchedLocal* tests need to malloc now because their
stack frames are too big to fit on the G0 stack.
LGTM=iant
R=golang-codereviews, iant, khr
CC=golang-codereviews
https://golang.org/cl/133660043
newstackcall creates a new stack segment, and we want to
be able to throw away all that code.
LGTM=khr
R=khr, iant
CC=dvyukov, golang-codereviews, r
https://golang.org/cl/139270043
We cannot let a real panic start there, because there is C code
on the stack, and worse, there is an assembly frame with a
saved copy of the registers and we have no idea which ones
are pointers.
Instead, detect the nil ptr load/store and return out of the C
and assembly into a stub that will start the call to sigpanic.
Fixes GOARM=5 build.
LGTM=iant
R=golang-codereviews, iant
CC=dave, golang-codereviews, minux, r
https://golang.org/cl/138130043
Minor changes to make logic clearer.
Observed while working on the conversion.
LGTM=iant, dvyukov
R=dvyukov, iant
CC=golang-codereviews
https://golang.org/cl/140250043
created panic1.go just so diffs were available.
After this CL is in, I'd like to move panic.go -> defer.go
and panic1.go -> panic.go.
LGTM=rsc
R=rsc, khr
CC=golang-codereviews
https://golang.org/cl/133530045
sigprof and setcpuprofilerate coordinate the enabling/disabling
of the handler using a Mutex. This has always been a bit dodgy:
setcpuprofilerate must be careful to turn off signals before acquiring
the lock to avoid a deadlock.
Now the lock implementations use onM, and onM isn't okay on the
signal stack. We know how to make it okay, but it's more work than
is probably worth doing.
Since this is super-dodgy anyway, replace the lock with a simple
cas loop. It is only contended if setcpuprofilerate is being called,
and that doesn't happen frequently enough to care about the
raw speed or about using futexes/semaphores.
TBR to fix freebsd/amd64 and dragonfly/amd64 builds.
Happy to make changes in a follow-up CL.
TBR=dvyukov
CC=golang-codereviews
https://golang.org/cl/141080044
The general kernel system call interface
takes 6 arguments: R0, R1, R2, R3, R4, R5.
Syscall is for calls that only need 3.
The amd64 and 386 versions zero the extra arg registers,
but the arm version does not.
func utimensat calls Syscall with 3 arguments.
The kernel expects a 4th argument.
That turns out to be whatever is in R3 at the time of the call.
CL 137160043 changed various pieces of code and apparently
changed the value left in R3 at the time of utimensat's Syscall.
This causes the kernel to return EINVAL.
Change linux/arm Syscall to zero R3, R4, R5, so that calls will
behave deterministically, even if they pass too few arguments.
Arguably, utimensat could be fixed too, but the predictable
zeroing is certainly worth doing, and once done utimensat's
use of Syscall is fine.
Fixes arm build.
TBR=bradfitz
CC=golang-codereviews
https://golang.org/cl/141080043
I did this just to clean things up, but it will be important
when we drop the pkg directory later.
LGTM=bradfitz
R=r, bradfitz
CC=golang-codereviews
https://golang.org/cl/132600043
Behavior before this CL:
1. If onM is called on a g0 stack, it just calls the given function.
2. If onM is called on a gsignal stack, it calls badonm.
3. If onM is called on a curg stack, it switches to the g0 stack
and then calls the function.
In cases 1 and 2, if the program then crashes (and badonm always does),
we want to see what called onM, but the traceback stops at onM.
In case 3, the traceback must stop at onM, because the g0
stack we are renting really does stop at onM.
The current code stops the traceback at onM to handle 3,
at the cost of making 1 and 2 crash with incomplete traces.
Change traceback to scan past onM but in case 3 make it look
like on the rented g0 stack, onM was called from mstart.
The traceback already knows that mstart is a top-of-stack function.
Alternate fix at CL 132610043 but I think this one is cleaner.
This CL makes 3 the exception, while that CL makes 1 and 2 the exception.
Submitting TBR to try to get better stack traces out of the
freebsd/amd64 builder, but happy to make changes in a
followup CL.
TBR=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/133620043
sysAlloc is the only mem function called from Go.
LGTM=iant, khr
R=golang-codereviews, khr, 0intro, iant
CC=dvyukov, golang-codereviews, r
https://golang.org/cl/139210043
Mostly NOSPLIT additions.
Had to rewrite atomic_arm.c in Go because it calls lock,
and lock is too complex.
With this CL, I find no Go -> C calls that can split the stack
on any system except Solaris and Windows.
Solaris and Windows need more work and will be done separately.
LGTM=iant, dave
R=golang-codereviews, bradfitz, iant, dave
CC=dvyukov, golang-codereviews, khr, r
https://golang.org/cl/137160043
Convert no-op race functions.
Everything else is tiny and gets NOSPLITs.
After this, all that is left on darwin is sysAlloc, panic, and gothrow (all pending).
There may be system-specific calls in other builds.
LGTM=iant
R=golang-codereviews, iant
CC=dvyukov, golang-codereviews, khr, r
https://golang.org/cl/140240044
When this code was written, there was no way for Go to
reuse the C function and enum values. Now there is.
LGTM=bradfitz
R=rlh, bradfitz
CC=dvyukov, golang-codereviews, iant, khr, r
https://golang.org/cl/139150045
The original conversion in CL 132090043 cut up
the function in an attempt to avoid converting most
of the code to Go. This contorts the control flow.
While debugging the onM signal stack bug,
I reconverted sigqueue.goc in its entirety.
This restores the original control flow, which is
much easier to understand.
The current conversion is correct, it's just complex
and will be hard to maintain. The new one is as
readable as the original code.
I uploaded sigqueue.goc as the initial copy of
sigqueue.go in the CL, so if you view the diffs
of sigqueue.go comparing against patch set 2 [sic]
it will show the actual starting point.
For example:
https://golang.org/cl/136160043/diff2/20001:60001/src/pkg/runtime/sigqueue.go
LGTM=dvyukov, iant
R=golang-codereviews, dvyukov, iant
CC=golang-codereviews, khr, r
https://golang.org/cl/136160043
The common code is converted, epoll and kqueue are converted.
Windows and solaris are still C.
LGTM=rsc
R=golang-codereviews, rsc, dave
CC=golang-codereviews, iant, khr, rsc
https://golang.org/cl/132910043
I had this right in one of my clients, but apparently not the one I submitted from.
Fixes 386 builds.
TBR=dfc
CC=golang-codereviews
https://golang.org/cl/138000045
The arm5 build breakage at CL 139110043 was caused by
calling funcPC on a lessstack defined as a struct{}.
That symbol ended up with a non-4-aligned address,
which caused the memory fault that broke the builders.
The definition of lessstack was fixed in CL 140880043.
Tracking that down suggested that it would be worth
looking for the same bug elsewhere in the directory.
This is the only one I found.
LGTM=bradfitz
R=golang-codereviews, dave, bradfitz
CC=dvyukov, golang-codereviews, iant, khr, r
https://golang.org/cl/134410043
Some things get converted.
Other things (too complex or too many C deps) get onM calls.
Other things (too simple) get #pragma textflag NOSPLIT.
After this CL, the offending function list is basically:
- panic.c
- netpoll.goc
- mem*.c
- race stuff
- readgstatus
- entersyscall/exitsyscall
LGTM=r, iant
R=golang-codereviews, r, iant
CC=dvyukov, golang-codereviews, khr
https://golang.org/cl/140930043
The implementation and use patterns of onM assume
that they run on either the m->curg or m->g0 stack.
Calling onM from m->gsignal has two problems:
(1) When not on g0, onM switches to g0 and then "back" to curg.
If we didn't start at curg, bad things happen.
(2) The use of scalararg/ptrarg to pass C arguments and results
assumes that there is only one onM call at a time.
If a gsignal starts running, it may have interrupted the
setup/teardown of the args for an onM on the curg or g0 stack.
Using scalararg/ptrarg itself would smash those.
We can fix (1) by remembering what g was running before the switch.
We can fix (2) by requiring that uses of onM that might happen
on a signal handling stack must save the old scalararg/ptrarg
and restore them after the call, instead of zeroing them.
The only sane way to do this is to introduce a separate
onM_signalsafe that omits the signal check, and then if you
see a call to onM_signalsafe you know the surrounding code
must preserve the old scalararg/ptrarg values.
(The implementation would be that onM_signalsafe just calls
fn if on the signal stack or else jumps to onM. It's not necessary
to have two whole copies of the function.)
(2) is not a problem if the caller and callee are both Go and
a closure is used instead of the scalararg/ptrarg slots.
For now, I think we can avoid calling onM from code executing
on gsignal stacks, so just reject it.
In the long term, (2) goes away (as do the scalararg/ptrarg slots)
once everything is in Go, and at that point fixing (1) would be
trivial and maybe worth doing just for regularity.
LGTM=iant
R=golang-codereviews, iant
CC=dvyukov, golang-codereviews, khr, r
https://golang.org/cl/135400043
Instead of making asmcgocall call asmcgocall_errno,
make both load args into registers and call a shared
assembly function.
On amd64, this costs 1 word in the asmcgocall_errno path
but saves 3 words in the asmcgocall path, and the latter
is what happens on critical nosplit paths on Windows.
On arm, this fixes build failures: asmcgocall was writing
the arguments for asmcgocall_errno into the wrong
place on the stack. Passing them in registers avoids the
decision entirely.
On 386, this isn't really needed, since the nosplit paths
have twice as many words to work with, but do it for consistency.
Update #8635
Fixes arm build (except GOARM=5).
TBR=iant
CC=golang-codereviews
https://golang.org/cl/134390043
I really hoped we could avoid this nonsense, but it appears not.
Should fix windows/amd64 build breakage.
TBR=iant
CC=golang-codereviews
https://golang.org/cl/137120043
