[release-branch.go1.15] runtime: fix crash during VDSO calls on PowerPC

This patch reinstates a fix for PowerPC with regard to making VDSO calls
while receiving a signal, and subsequently crashing. The crash happens
because certain VDSO calls can modify the r30 register, which is where g
is stored. This change was reverted for PowerPC because r30 is supposed
to be a non-volatile register. This is true, but that only makes a
guarantee across function calls, but not "within" a function call. This
patch was seemingly fine before because the Linux kernel still had hand
rolled assembly VDSO function calls, however with a recent change to C
function calls it seems the compiler used can generate instructions
which temporarily clobber r30. This means that when we receive a signal
during one of these calls the value of r30 will not be the g as the
runtime expects, causing a segfault.

You can see from this assembly dump how the register is clobbered during
the call:

(the following is from a 5.13rc2 kernel)

```
Dump of assembler code for function __cvdso_clock_gettime_data:
   0x00007ffff7ff0700 <+0>:     cmplwi  r4,15
   0x00007ffff7ff0704 <+4>:     bgt     0x7ffff7ff07f0 <__cvdso_clock_gettime_data+240>
   0x00007ffff7ff0708 <+8>:     li      r9,1
   0x00007ffff7ff070c <+12>:    slw     r9,r9,r4
   0x00007ffff7ff0710 <+16>:    andi.   r10,r9,2179
   0x00007ffff7ff0714 <+20>:    beq     0x7ffff7ff0810 <__cvdso_clock_gettime_data+272>
   0x00007ffff7ff0718 <+24>:    rldicr  r10,r4,4,59
   0x00007ffff7ff071c <+28>:    lis     r9,32767
   0x00007ffff7ff0720 <+32>:    std     r30,-16(r1)
   0x00007ffff7ff0724 <+36>:    std     r31,-8(r1)
   0x00007ffff7ff0728 <+40>:    add     r6,r3,r10
   0x00007ffff7ff072c <+44>:    ori     r4,r9,65535
   0x00007ffff7ff0730 <+48>:    lwz     r8,0(r3)
   0x00007ffff7ff0734 <+52>:    andi.   r9,r8,1
   0x00007ffff7ff0738 <+56>:    bne     0x7ffff7ff07d0 <__cvdso_clock_gettime_data+208>
   0x00007ffff7ff073c <+60>:    lwsync
   0x00007ffff7ff0740 <+64>:    mftb    r30      <---- RIGHT HERE
=> 0x00007ffff7ff0744 <+68>:    ld      r12,40(r6)
```

What I believe is happening is that the kernel changed the PowerPC VDSO
calls to use standard C calls instead of using hand rolled assembly. The
hand rolled assembly calls never touched r30, so this change was safe to
roll back. That does not seem to be the case anymore as on the 5.13rc2
kernel the compiler *is* generating assembly which modifies r30, making
this change again unsafe and causing a crash when the program receives a
signal during these calls (which will happen often due to async
preempt). This change happened here:
https://lwn.net/ml/linux-kernel/235e5571959cfa89ced081d7e838ed5ff38447d2.1601365870.git.christophe.leroy@csgroup.eu/.

I realize this was reverted due to unexplained hangs in PowerPC
builders, but I think we should reinstate this change and investigate
those issues separately:
https://github.com/golang/go/commit/f4ca3c1e0a2066ca4f7bd6203866d282ed34acf2

Fixes #46857

Change-Id: Ib18d7bbfc80a1a9cb558f0098878d41081324b52
GitHub-Last-Rev: c3002bcfca3ef58b27485e31328e6297b7a9dfe7
GitHub-Pull-Request: golang/go#46767
Reviewed-on: https://go-review.googlesource.com/c/go/+/328110
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Trust: Lynn Boger <laboger@linux.vnet.ibm.com>
(cherry picked from commit 16e82be454cbf41299e6a055d54d489ca4612ee0)
Reviewed-on: https://go-review.googlesource.com/c/go/+/334411
Run-TryBot: Carlos Amedee <carlos@golang.org>

2 files changed, 74 insertions, 14 deletions

diff --git a/src/runtime/signal_unix.go b/src/runtime/signal_unix.go
index 003c7b0bc8..8e1eb9e186 100644
--- a/src/runtime/signal_unix.go
+++ b/src/runtime/signal_unix.go
@@ -393,7 +393,7 @@ func preemptM(mp *m) {
 //go:nosplit
 func sigFetchG(c *sigctxt) *g {
 	switch GOARCH {
-	case "arm", "arm64":
+	case "arm", "arm64", "ppc64", "ppc64le":
 		if !iscgo && inVDSOPage(c.sigpc()) {
 			// When using cgo, we save the g on TLS and load it from there
 			// in sigtramp. Just use that.
diff --git a/src/runtime/sys_linux_ppc64x.s b/src/runtime/sys_linux_ppc64x.s
index fd69ee70a5..7be8c4c724 100644
--- a/src/runtime/sys_linux_ppc64x.s
+++ b/src/runtime/sys_linux_ppc64x.s
@@ -215,15 +215,45 @@ TEXT runtime·walltime1(SB),NOSPLIT,$16-12
 	MOVD	(g_sched+gobuf_sp)(R7), R1	// Set SP to g0 stack
 
 noswitch:
-	SUB	$16, R1			// Space for results
-	RLDICR	$0, R1, $59, R1		// Align for C code
+	SUB	$16, R1                 // Space for results
+	RLDICR	$0, R1, $59, R1         // Align for C code
 	MOVD	R12, CTR
 	MOVD	R1, R4
-	BL	(CTR)			// Call from VDSO
-	MOVD	$0, R0			// Restore R0
-	MOVD	0(R1), R3		// sec
-	MOVD	8(R1), R5		// nsec
-	MOVD	R15, R1			// Restore SP
+
+	// Store g on gsignal's stack, so if we receive a signal
+	// during VDSO code we can find the g.
+	// If we don't have a signal stack, we won't receive signal,
+	// so don't bother saving g.
+	// When using cgo, we already saved g on TLS, also don't save
+	// g here.
+	// Also don't save g if we are already on the signal stack.
+	// We won't get a nested signal.
+	MOVBZ	runtime·iscgo(SB), R22
+	CMP	R22, $0
+	BNE	nosaveg
+	MOVD	m_gsignal(R21), R22	// g.m.gsignal
+	CMP	R22, $0
+	BEQ	nosaveg
+
+	CMP	g, R22
+	BEQ	nosaveg
+	MOVD	(g_stack+stack_lo)(R22), R22 // g.m.gsignal.stack.lo
+	MOVD	g, (R22)
+
+	BL	(CTR)	// Call from VDSO
+
+	MOVD	$0, (R22)	// clear g slot, R22 is unchanged by C code
+
+	JMP	finish
+
+nosaveg:
+	BL	(CTR)	// Call from VDSO
+
+finish:
+	MOVD	$0, R0		// Restore R0
+	MOVD	0(R1), R3	// sec
+	MOVD	8(R1), R5	// nsec
+	MOVD	R15, R1		// Restore SP
 
 	// Restore vdsoPC, vdsoSP
 	// We don't worry about being signaled between the two stores.
@@ -235,7 +265,7 @@ noswitch:
 	MOVD	32(R1), R6
 	MOVD	R6, m_vdsoPC(R21)
 
-finish:
+return:
 	MOVD	R3, sec+0(FP)
 	MOVW	R5, nsec+8(FP)
 	RET
@@ -246,7 +276,7 @@ fallback:
 	SYSCALL $SYS_clock_gettime
 	MOVD	32(R1), R3
 	MOVD	40(R1), R5
-	JMP	finish
+	JMP	return
 
 TEXT runtime·nanotime1(SB),NOSPLIT,$16-8
 	MOVD	$1, R3		// CLOCK_MONOTONIC
@@ -282,7 +312,37 @@ noswitch:
 	RLDICR	$0, R1, $59, R1		// Align for C code
 	MOVD	R12, CTR
 	MOVD	R1, R4
-	BL	(CTR)			// Call from VDSO
+
+	// Store g on gsignal's stack, so if we receive a signal
+	// during VDSO code we can find the g.
+	// If we don't have a signal stack, we won't receive signal,
+	// so don't bother saving g.
+	// When using cgo, we already saved g on TLS, also don't save
+	// g here.
+	// Also don't save g if we are already on the signal stack.
+	// We won't get a nested signal.
+	MOVBZ	runtime·iscgo(SB), R22
+	CMP	R22, $0
+	BNE	nosaveg
+	MOVD	m_gsignal(R21), R22	// g.m.gsignal
+	CMP	R22, $0
+	BEQ	nosaveg
+
+	CMP	g, R22
+	BEQ	nosaveg
+	MOVD	(g_stack+stack_lo)(R22), R22 // g.m.gsignal.stack.lo
+	MOVD	g, (R22)
+
+	BL	(CTR)	// Call from VDSO
+
+	MOVD	$0, (R22)	// clear g slot, R22 is unchanged by C code
+
+	JMP	finish
+
+nosaveg:
+	BL	(CTR)	// Call from VDSO
+
+finish:
 	MOVD	$0, R0			// Restore R0
 	MOVD	0(R1), R3		// sec
 	MOVD	8(R1), R5		// nsec
@@ -298,7 +358,7 @@ noswitch:
 	MOVD	32(R1), R6
 	MOVD	R6, m_vdsoPC(R21)
 
-finish:
+return:
 	// sec is in R3, nsec in R5
 	// return nsec in R3
 	MOVD	$1000000000, R4
@@ -313,7 +373,7 @@ fallback:
 	SYSCALL $SYS_clock_gettime
 	MOVD	32(R1), R3
 	MOVD	40(R1), R5
-	JMP	finish
+	JMP	return
 
 TEXT runtime·rtsigprocmask(SB),NOSPLIT|NOFRAME,$0-28
 	MOVW	how+0(FP), R3
@@ -366,7 +426,7 @@ TEXT sigtramp<>(SB),NOSPLIT,$64
 	// this might be called in external code context,
 	// where g is not set.
 	MOVBZ	runtime·iscgo(SB), R6
-	CMP 	R6, $0
+	CMP	R6, $0
 	BEQ	2(PC)
 	BL	runtime·load_g(SB)