Trying to make GODEBUG=gcdead=1 work with liveness
and in particular ambiguously live variables.
1. In the liveness computation, mark all ambiguously live
variables as live for the entire function, except the entry.
They are zeroed directly after entry, and we need them not
to be poisoned thereafter.
2. In the liveness computation, compute liveness (and deadness)
for all parameters, not just pointer-containing parameters.
Otherwise gcdead poisons untracked scalar parameters and results.
3. Fix liveness debugging print for -live=2 to use correct bitmaps.
(Was not updated for compaction during compaction CL.)
4. Correct varkill during map literal initialization.
Was killing the map itself instead of the inserted value temp.
5. Disable aggressive varkill cleanup for call arguments if
the call appears in a defer or go statement.
6. In the garbage collector, avoid bug scanning empty
strings. An empty string is two zeros. The multiword
code only looked at the first zero and then interpreted
the next two bits in the bitmap as an ordinary word bitmap.
For a string the bits are 11 00, so if a live string was zero
length with a 0 base pointer, the poisoning code treated
the length as an ordinary word with code 00, meaning it
needed poisoning, turning the string into a poison-length
string with base pointer 0. By the same logic I believe that
a live nil slice (bits 11 01 00) will have its cap poisoned.
Always scan full multiword struct.
7. In the runtime, treat both poison words (PoisonGC and
PoisonStack) as invalid pointers that warrant crashes.
Manual testing as follows:
- Create a script called gcdead on your PATH containing:
#!/bin/bash
GODEBUG=gcdead=1 GOGC=10 GOTRACEBACK=2 exec "$@"
- Now you can build a test and then run 'gcdead ./foo.test'.
- More importantly, you can run 'go test -short -exec gcdead std'
to run all the tests.
Fixes#7676.
While here, enable the precise scanning of slices, since that was
disabled due to bugs like these. That now works, both with and
without gcdead.
Fixes#7549.
LGTM=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/83410044
The garbage collector poison pointers
(0x6969696969696969 and 0x6868686868686868)
are malformed addresses on amd64.
That is, they are not 48-bit addresses sign extended
to 64 bits. This causes a different kind of hardware fault
than the usual 'unmapped page' when accessing such
an address, and OS X 10.9.2 sends the resulting SIGSEGV
incorrectly, making it look like it was user-generated
rather than kernel-generated and does not include the
faulting address. This means that in GODEBUG=gcdead=1
mode, if there is a bug and something tries to dereference
a poisoned pointer, the runtime delivers the SIGSEGV to
os/signal and returns to the faulting code, which faults
again, causing the process to hang instead of crashing.
Fix by rewriting "user-generated" SIGSEGV on OS X to
look like a kernel-generated SIGSEGV with fault address
0xb01dfacedebac1e.
I chose that address because (1) when printed in hex
during a crash, it is obviously spelling out English text,
(2) there are no current Google hits for that pointer,
which will make its origin easy to find once this CL
is indexed, and (3) it is not an altogether inaccurate
description of the situation.
Add a test. Maybe other systems will break too.
LGTM=khr
R=golang-codereviews, khr
CC=golang-codereviews, iant, ken
https://golang.org/cl/83270049
Delaying the runtime.throw until here will print more information.
In particular it will print the signal and code values, which means
it will show the fault address.
The canpanic checks were added recently, in CL 75320043.
They were just not added in exactly the right place.
LGTM=iant
R=dvyukov, iant
CC=golang-codereviews
https://golang.org/cl/83980043
Brad has been asking for this for a while.
I have resisted because I wanted to find a more general way to
do this, one that would keep the performance of code introducing
variables the same as the performance of code that did not.
(See golang.org/issue/3512#c20).
I have not found the more general way, and recent changes to
remove ambiguously live temporaries have blown away the
property I was trying to preserve, so that's no longer a reason
not to make the change.
Fixes#3512.
LGTM=iant
R=iant
CC=bradfitz, golang-codereviews, khr, r
https://golang.org/cl/83740044
The software floating point runs with m->locks++
to avoid being preempted; recognize this case in panic
and undo it so that m->locks is maintained correctly
when panicking.
Fixes#7553.
LGTM=dvyukov
R=golang-codereviews, dvyukov
CC=golang-codereviews
https://golang.org/cl/84030043
The old limit of 5 was chosen because we didn't actually know how
many bytes of arguments there were; 5 was a halfway point between
printing some useful information and looking ridiculous.
Now we know how many bytes of arguments there are, and we stop
the printing when we reach that point, so the "looking ridiculous" case
doesn't happen anymore: we only print actual argument words.
The cutoff now serves only to truncate very long (but real) argument lists.
In multiple debugging sessions recently (completely unrelated bugs)
I have been frustrated by not seeing more of the long argument lists:
5 words is only 2.5 interface values or strings, and not even 2 slices.
Double the max amount we'll show.
LGTM=bradfitz
R=golang-codereviews, bradfitz
CC=golang-codereviews, iant, r
https://golang.org/cl/83850043
Reduce footprint of liveness bitmaps by about 5x.
1. Mark all liveness bitmap symbols as 4-byte aligned
(they were aligned to a larger size by default).
2. The bitmap data is a bitmap count n followed by n bitmaps.
Each bitmap begins with its own count m giving the number
of bits. All the m's are the same for the n bitmaps.
Emit this bitmap length once instead of n times.
3. Many bitmaps within a function have the same bit values,
but each call site was given a distinct bitmap. Merge duplicate
bitmaps so that no bitmap is written more than once.
4. Many functions end up with the same aggregate bitmap data.
We used to name the bitmap data funcname.gcargs and funcname.gclocals.
Instead, name it gclocals.<md5 of data> and mark it dupok so
that the linker coalesces duplicate sets. This cut the bitmap
data remaining after step 3 by 40%; I was not expecting it to
be quite so dramatic.
Applied to "go build -ldflags -w code.google.com/p/go.tools/cmd/godoc":
bitmaps pclntab binary on disk
before this CL 1326600 1985854 12738268
4-byte align 1154288 (0.87x) 1985854 (1.00x) 12566236 (0.99x)
one bitmap len 782528 (0.54x) 1985854 (1.00x) 12193500 (0.96x)
dedup bitmap 414748 (0.31x) 1948478 (0.98x) 11787996 (0.93x)
dedup bitmap set 245580 (0.19x) 1948478 (0.98x) 11620060 (0.91x)
While here, remove various dead blocks of code from plive.c.
Fixes#6929.
Fixes#7568.
LGTM=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/83630044
REP MOVSQ and REP STOSQ have a really high startup overhead.
Use a Duff's device to do the repetition instead.
benchmark old ns/op new ns/op delta
BenchmarkClearFat32 7.20 1.60 -77.78%
BenchmarkCopyFat32 6.88 2.38 -65.41%
BenchmarkClearFat64 7.15 3.20 -55.24%
BenchmarkCopyFat64 6.88 3.44 -50.00%
BenchmarkClearFat128 9.53 5.34 -43.97%
BenchmarkCopyFat128 9.27 5.56 -40.02%
BenchmarkClearFat256 13.8 9.53 -30.94%
BenchmarkCopyFat256 13.5 10.3 -23.70%
BenchmarkClearFat512 22.3 18.0 -19.28%
BenchmarkCopyFat512 22.0 19.7 -10.45%
BenchmarkCopyFat1024 36.5 38.4 +5.21%
BenchmarkClearFat1024 35.1 35.0 -0.28%
TODO: use for stack frame zeroing
TODO: REP prefixes are still used for "reverse" copying when src/dst
regions overlap. Might be worth fixing.
LGTM=rsc
R=golang-codereviews, rsc
CC=golang-codereviews, r
https://golang.org/cl/81370046
The old code was using the PC of the instruction after the CALL.
Variables live during the call but not live when it returns would
not be seen as live during the stack copy, which might lead to
corruption. The correct PC to use is the one just before the
return address. After this CL the lookup matches what mgc0.c does.
The only time this matters is if you have back to back CALL instructions:
CALL f1 // x live here
CALL f2 // x no longer live
If a stack copy occurs during the execution of f1, the old code will
use the liveness bitmap intended for the execution of f2 and will not
treat x as live.
The only way this situation can arise and cause a problem in a stack copy
is if x lives on the stack has had its address taken but the compiler knows
enough about the context to know that x is no longer needed once f1
returns. The compiler has never known that much, so using the f2 context
cannot currently cause incorrect execution. For the same reason, it is not
possible to write a test for this today.
CL 83090046 will make the compiler precise enough in some cases
that this distinction will start mattering. The existing stack growth tests
in package runtime will fail if that CL is submitted without this one.
While we're here, print the frame PC in debug mode and update the
bitmap interpretation strings.
LGTM=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/83250043
GODEBUG=allocfreetrace=1:
The allocfreetrace=1 mode prints a stack trace for each block
allocated and freed, and also a stack trace for each garbage collection.
It was implemented by reusing the heap profiling support: if allocfreetrace=1
then the heap profile was effectively running at 1 sample per 1 byte allocated
(always sample). The stack being shown at allocation was the stack gathered
for profiling, meaning it was derived only from the program counters and
did not include information about function arguments or frame pointers.
The stack being shown at free was the allocation stack, not the free stack.
If you are generating this log, you can find the allocation stack yourself, but
it can be useful to see exactly the sequence that led to freeing the block:
was it the garbage collector or an explicit free? Now that the garbage collector
runs on an m0 stack, the stack trace for the garbage collector was never interesting.
Fix all these problems:
1. Decouple allocfreetrace=1 from heap profiling.
2. Print the standard goroutine stack traces instead of a custom format.
3. Print the stack trace at time of allocation for an allocation,
and print the stack trace at time of free (not the allocation trace again)
for a free.
4. Print all goroutine stacks at garbage collection. Having all the stacks
means that you can see the exact point at which each goroutine was
preempted, which is often useful for identifying liveness-related errors.
GODEBUG=gcdead=1:
This mode overwrites dead pointers with a poison value.
Detect the poison value as an invalid pointer during collection,
the same way that small integers are invalid pointers.
LGTM=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/81670043
This is the same check we use during stack copying.
The check cannot be applied to C stack frames, even
though we do emit pointer bitmaps for the arguments,
because (1) the pointer bitmaps assume all arguments
are always live, not true of outputs during the prologue,
and (2) the pointer bitmaps encode interface values as
pointer pairs, not true of interfaces holding integers.
For the rest of the frames, however, we should hold ourselves
to the rule that a pointer marked live really is initialized.
The interface scanning already implicitly checks this
because it interprets the type word as a valid type pointer.
This may slow things down a little because of the extra loads.
Or it may speed things up because we don't bother enqueuing
nil pointers anymore. Enough of the rest of the system is slow
right now that we can't measure it meaningfully.
Enable for now, even if it is slow, to shake out bugs in the
liveness bitmaps, and then decide whether to turn it off
for the Go 1.3 release (issue 7650 reminds us to do this).
The new m->traceback field lets us force printing of fp=
values on all goroutine stack traces when we detect a
bad pointer. This makes it easier to understand exactly
where in the frame the bad pointer is, so that we can trace
it back to a specific variable and determine what is wrong.
Update #7650
LGTM=khr
R=khr
CC=golang-codereviews
https://golang.org/cl/80860044
The garbage collector will scan these pointers,
so make sure they are initialized.
LGTM=bradfitz, khr
R=khr, bradfitz
CC=golang-codereviews
https://golang.org/cl/80960047
m->moreargp/morebuf were not cleared in case of preemption and stack growing,
it can lead to persistent leaks of large memory blocks.
It seems to fix the sync.Pool finalizer failures. I've run the test 500'000 times
w/o a single failure; previously it would fail dozens of times.
Fixes#7633.
Fixes#7533.
LGTM=rsc
R=golang-codereviews
CC=golang-codereviews, khr, rsc
https://golang.org/cl/80480044
Update channel race annotations to support change in
cl/75130045: doc: allow buffered channel as semaphore without initialization
The new annotations are added only for channels with capacity 1.
Strictly saying it's possible to construct a counter-example that
will produce a false positive with capacity > 1. But it's hardly can
lead to false positives in real programs, at least I would like to see such programs first.
Any additional annotations also increase probability of false negatives,
so I would prefer to add them lazily.
LGTM=rsc
R=golang-codereviews
CC=golang-codereviews, iant, khr, rsc
https://golang.org/cl/76970043
Currently it's possible that bgsweep finishes before all spans
have been swept (we only know that sweeping of all spans has *started*).
In such case bgsweep may fail wake up runfinq goroutine when it needs to.
finq may still be nil at this point, but some finalizers may be queued later.
Make bgsweep to wait for sweeping to *complete*, then it can decide
whether it needs to wake up runfinq for sure.
Update #7533
LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/75960043
If we set obj, then it will be enqueued for marking at the end of the scanning loop.
This is not necessary, since we've already marked it.
This can wait for 1.4 if you wish.
LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/80030043
Change two-bit stack map entries to encode:
0 = dead
1 = scalar
2 = pointer
3 = multiword
If multiword, the two-bit entry for the following word encodes:
0 = string
1 = slice
2 = iface
3 = eface
That way, during stack scanning we can check if a string
is zero length or a slice has zero capacity. We can avoid
following the contained pointer in those cases. It is safe
to do so because it can never be dereferenced, and it is
desirable to do so because it may cause false retention
of the following block in memory.
Slice feature turned off until issue 7564 is fixed.
Update #7549
LGTM=rsc
R=golang-codereviews, bradfitz, rsc
CC=golang-codereviews
https://golang.org/cl/76380043
The existing code did not have a clear notion of whether
memory has been actually reserved. It checked based on
whether in 32-bit mode or 64-bit mode and (on GNU/Linux) the
requested address, but it confused the requested address and
the returned address.
LGTM=rsc
R=rsc, dvyukov
CC=golang-codereviews, michael.hudson
https://golang.org/cl/79610043
The nproc and ndone fields are uint32. This makes the type
consistent.
LGTM=minux.ma
R=golang-codereviews, minux.ma
CC=golang-codereviews
https://golang.org/cl/79340044
Structured Exception Handling (SEH) was the first way to handle
exceptions (memory faults, divides by zero) on Windows.
The S might as well stand for "stack-based": the implementation
interprets stack addresses in a few different ways, and it gets
subtly confused by Go's management of stacks. It's also something
that requires active maintenance during cgo switches, and we've
had bugs in that maintenance in the past.
We have recently come to believe that SEH cannot work with
Go's stack usage. See http://golang.org/issue/7325 for details.
Vectored Exception Handling (VEH) is more like a Unix signal
handler: you set it once for the whole process and forget about it.
This CL drops all the SEH code and replaces it with VEH code.
Many special cases and 7 #ifdefs disappear.
VEH was introduced in Windows XP, so Go on windows/386 will
now require Windows XP or later. The previous requirement was
Windows 2000 or later. Windows 2000 immediately preceded
Windows XP, so Windows 2000 is the only affected version.
Microsoft stopped supporting Windows 2000 in 2010.
See http://golang.org/s/win2000-golang-nuts for details.
Fixes#7325.
LGTM=alex.brainman, r
R=golang-codereviews, alex.brainman, stephen.gutekanst, dave
CC=golang-codereviews, iant, r
https://golang.org/cl/74790043
Also move generated code into a separate file,
because it's difficult to work with the file otherwise.
LGTM=khr
R=golang-codereviews, khr
CC=golang-codereviews
https://golang.org/cl/76080044
It was using the wrong offset and returned random values
making "runoutput" compiler tests crash.
LGTM=iant
R=golang-codereviews, iant
CC=golang-codereviews
https://golang.org/cl/76250043
They were rejected by NaCl due to AES instructions and
accesses to %gs:0x8, caused by wrong tlsoffset value.
LGTM=iant
R=rsc, dave, iant
CC=golang-codereviews
https://golang.org/cl/76050044
It's possible that bgsweep constantly does not catch up for some reason,
in this case runfinq was not woken at all.
R=rsc
CC=golang-codereviews
https://golang.org/cl/75940043
The problem was that spans end up in wrong lists after split
(e.g. in h->busy instead of h->central->empty).
Also the span can be non-swept before split,
I don't know what it can cause, but it's safer to operate on swept spans.
Fixes#7544.
R=golang-codereviews, rsc
CC=golang-codereviews, khr
https://golang.org/cl/76160043
Currently processes crash with obscure message.
Say that it's "out of memory".
LGTM=rsc
R=golang-codereviews
CC=golang-codereviews, khr, rsc
https://golang.org/cl/75820045
The Solaris network poller uses event ports, which are
level-triggered. As such, it has to re-arm itself after each
wakeup. The arming mechanism (which runs in its own thread) raced
with the closing of a file descriptor happening in a different
thread. When a network file descriptor is about to be closed,
the network poller is awaken to give it a chance to remove its
association with the file descriptor. Because the poller always
re-armed itself, it raced with code that closed the descriptor.
This change makes the network poller check before re-arming if
the file descriptor is about to be closed, in which case it will
ignore the re-arming request. It uses the per-PollDesc lock in
order to serialize access to the PollDesc.
This change also adds extensive documentation describing the
Solaris implementation of the network poller.
Fixes#7410.
LGTM=dvyukov, iant
R=golang-codereviews, bradfitz, iant, dvyukov, aram.h, gobot
CC=golang-codereviews
https://golang.org/cl/69190044
Mark free memory blocks as unused.
On amd64 it allows the process to eat all 128 GB of heap
without killing the machine.
LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/74070043
This is especially important for SetPanicOnCrash,
but also useful for e.g. nil deref in mallocgc.
Panics on such crashes can't lead to anything useful,
only to deadlocks, hangs and obscure crashes.
This is a copy of broken but already LGTMed
https://golang.org/cl/68540043/
TBR=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/75320043
When we copy stack, we check only new size of the top segment.
This is incorrect, because we can have other segments below it.
LGTM=khr
R=golang-codereviews, khr
CC=golang-codereviews, rsc
https://golang.org/cl/73980045
Calling runtime·cgocall could trigger a GC in the child while
gclock was held by the parent.
Fixes#7511
LGTM=bradfitz, dvyukov, dave
R=golang-codereviews, bradfitz, dvyukov, dave
CC=golang-codereviews, rsc
https://golang.org/cl/75210044
On Plan 9, the kernel disallows the use of floating point
instructions while handling a note. Previously, we worked
around this by using a simple loop in place of memmove.
When I added that work-around, I verified that all paths
from the note handler didn't end up calling memmove. Now
that memclr is using SSE instructions, the same process
will have to be done again.
Instead of doing that, however, this CL just punts and
uses unoptimized functions everywhere on Plan 9.
LGTM=rsc
R=rsc, 0intro
CC=golang-codereviews
https://golang.org/cl/73830044