runtime: update and restore g0 stack bounds at cgocallback

Currently, at a cgo callback where there is already a Go frame on
the stack (i.e. C->Go->C->Go), we require that at the inner Go
callback the SP is within the g0's stack bounds set by a previous
callback. This is to prevent that the C code switches stack while
having a Go frame on the stack, which we don't really support. But
this could also happen when we cannot get accurate stack bounds,
e.g. when pthread_getattr_np is not available. Since the stack
bounds are just estimates based on the current SP, if there are
multiple C->Go callbacks with various stack depth, it is possible
that the SP of a later callback falls out of a previous call's
estimate. This leads to runtime throw in a seemingly reasonable
program.

This CL changes it to save the old g0 stack bounds at cgocallback,
update the bounds, and restore the old bounds at return. So each
callback will get its own stack bounds based on the current SP,
and when it returns, the outer callback has the its old stack
bounds restored.

Also, at a cgo callback when there is no Go frame on the stack,
we currently always get new stack bounds. We do this because if
we can only get estimated bounds based on the SP, and the stack
depth varies a lot between two C->Go calls, the previous
estimates may be off and we fall out or nearly fall out of the
previous bounds. But this causes a performance problem: the
pthread API to get accurate stack bounds (pthread_getattr_np) is
very slow when called on the main thread. Getting the stack bounds
every time significantly slows down repeated C->Go calls on the
main thread.

This CL fixes it by "caching" the stack bounds if they are
accurate. I.e. at the second time Go calls into C, if the previous
stack bounds are accurate, and the current SP is in bounds, we can
be sure it is the same stack and we don't need to update the bounds.
This avoids the repeated calls to pthread_getattr_np. If we cannot
get the accurate bounds, we continue to update the stack bounds
based on the SP, and that operation is very cheap.

On a Linux/AMD64 machine with glibc:

name                     old time/op  new time/op  delta
CgoCallbackMainThread-8  96.4µs ± 3%   0.1µs ± 2%  -99.92%  (p=0.000 n=10+9)

Fixes #68285.
Fixes #68587.

Change-Id: I3422badd5ad8ff63e1a733152d05fb7a44d5d435
Reviewed-on: https://go-review.googlesource.com/c/go/+/600296
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Michael Pratt <mpratt@google.com>

1 files changed, 42 insertions, 41 deletions

diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index 34aefd4c47..5e75e154e6 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -530,47 +530,48 @@ type m struct {
 	_       uint32 // align next field to 8 bytes
 
 	// Fields not known to debuggers.
-	procid        uint64            // for debuggers, but offset not hard-coded
-	gsignal       *g                // signal-handling g
-	goSigStack    gsignalStack      // Go-allocated signal handling stack
-	sigmask       sigset            // storage for saved signal mask
-	tls           [tlsSlots]uintptr // thread-local storage (for x86 extern register)
-	mstartfn      func()
-	curg          *g       // current running goroutine
-	caughtsig     guintptr // goroutine running during fatal signal
-	p             puintptr // attached p for executing go code (nil if not executing go code)
-	nextp         puintptr
-	oldp          puintptr // the p that was attached before executing a syscall
-	id            int64
-	mallocing     int32
-	throwing      throwType
-	preemptoff    string // if != "", keep curg running on this m
-	locks         int32
-	dying         int32
-	profilehz     int32
-	spinning      bool // m is out of work and is actively looking for work
-	blocked       bool // m is blocked on a note
-	newSigstack   bool // minit on C thread called sigaltstack
-	printlock     int8
-	incgo         bool          // m is executing a cgo call
-	isextra       bool          // m is an extra m
-	isExtraInC    bool          // m is an extra m that is not executing Go code
-	isExtraInSig  bool          // m is an extra m in a signal handler
-	freeWait      atomic.Uint32 // Whether it is safe to free g0 and delete m (one of freeMRef, freeMStack, freeMWait)
-	needextram    bool
-	traceback     uint8
-	ncgocall      uint64        // number of cgo calls in total
-	ncgo          int32         // number of cgo calls currently in progress
-	cgoCallersUse atomic.Uint32 // if non-zero, cgoCallers in use temporarily
-	cgoCallers    *cgoCallers   // cgo traceback if crashing in cgo call
-	park          note
-	alllink       *m // on allm
-	schedlink     muintptr
-	lockedg       guintptr
-	createstack   [32]uintptr // stack that created this thread, it's used for StackRecord.Stack0, so it must align with it.
-	lockedExt     uint32      // tracking for external LockOSThread
-	lockedInt     uint32      // tracking for internal lockOSThread
-	nextwaitm     muintptr    // next m waiting for lock
+	procid          uint64            // for debuggers, but offset not hard-coded
+	gsignal         *g                // signal-handling g
+	goSigStack      gsignalStack      // Go-allocated signal handling stack
+	sigmask         sigset            // storage for saved signal mask
+	tls             [tlsSlots]uintptr // thread-local storage (for x86 extern register)
+	mstartfn        func()
+	curg            *g       // current running goroutine
+	caughtsig       guintptr // goroutine running during fatal signal
+	p               puintptr // attached p for executing go code (nil if not executing go code)
+	nextp           puintptr
+	oldp            puintptr // the p that was attached before executing a syscall
+	id              int64
+	mallocing       int32
+	throwing        throwType
+	preemptoff      string // if != "", keep curg running on this m
+	locks           int32
+	dying           int32
+	profilehz       int32
+	spinning        bool // m is out of work and is actively looking for work
+	blocked         bool // m is blocked on a note
+	newSigstack     bool // minit on C thread called sigaltstack
+	printlock       int8
+	incgo           bool          // m is executing a cgo call
+	isextra         bool          // m is an extra m
+	isExtraInC      bool          // m is an extra m that is not executing Go code
+	isExtraInSig    bool          // m is an extra m in a signal handler
+	freeWait        atomic.Uint32 // Whether it is safe to free g0 and delete m (one of freeMRef, freeMStack, freeMWait)
+	needextram      bool
+	g0StackAccurate bool // whether the g0 stack has accurate bounds
+	traceback       uint8
+	ncgocall        uint64        // number of cgo calls in total
+	ncgo            int32         // number of cgo calls currently in progress
+	cgoCallersUse   atomic.Uint32 // if non-zero, cgoCallers in use temporarily
+	cgoCallers      *cgoCallers   // cgo traceback if crashing in cgo call
+	park            note
+	alllink         *m // on allm
+	schedlink       muintptr
+	lockedg         guintptr
+	createstack     [32]uintptr // stack that created this thread, it's used for StackRecord.Stack0, so it must align with it.
+	lockedExt       uint32      // tracking for external LockOSThread
+	lockedInt       uint32      // tracking for internal lockOSThread
+	nextwaitm       muintptr    // next m waiting for lock
 
 	mLockProfile mLockProfile // fields relating to runtime.lock contention
 	profStack    []uintptr    // used for memory/block/mutex stack traces