runtime: use vDSO for getrandom() on linux

Linux 6.11 supports calling getrandom() from the vDSO. It operates on a
thread-local opaque state allocated with mmap using flags specified by
the vDSO.

Opaque states are allocated in chunks, ideally ncpu at a time as a hint,
rounding up to as many fit in a complete page. On first use, a state is
assigned to an m, which owns that state, until the m exits, at which
point it is given back to the pool.

Performance appears to be quite good:

           │    sec/op    │   sec/op       vs base                 │
Read/4-16    222.45n ± 3%   27.13n   ± 6%  -87.80% (p=0.000 n=10)
           │     B/s      │      B/s       vs base                 │
Read/4-16    17.15Mi ± 3%   140.61Mi ± 6%  +719.82% (p=0.000 n=10)

Fixes #69577.

Change-Id: Ib6f44e8f2f3940c94d970eaada0eb566ec297dc7
Reviewed-on: https://go-review.googlesource.com/c/go/+/614835
Reviewed-by: Filippo Valsorda <filippo@golang.org>
Reviewed-by: Cuong Manh Le <cuong.manhle.vn@gmail.com>
Auto-Submit: Jason Donenfeld <Jason@zx2c4.com>
Reviewed-by: Paul Murphy <murp@ibm.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Michael Pratt <mpratt@google.com>

13 files changed, 364 insertions, 8 deletions

diff --git a/src/runtime/os_linux.go b/src/runtime/os_linux.go
index e18ef8e776..979761cc6a 100644
--- a/src/runtime/os_linux.go
+++ b/src/runtime/os_linux.go
@@ -31,6 +31,10 @@ type mOS struct {
 	// needPerThreadSyscall indicates that a per-thread syscall is required
 	// for doAllThreadsSyscall.
 	needPerThreadSyscall atomic.Uint8
+
+	// This is a pointer to a chunk of memory allocated with a special
+	// mmap invocation in vgetrandomGetState().
+	vgetrandomState uintptr
 }
 
 //go:noescape
@@ -344,6 +348,7 @@ func osinit() {
 	ncpu = getproccount()
 	physHugePageSize = getHugePageSize()
 	osArchInit()
+	vgetrandomInit()
 }
 
 var urandom_dev = []byte("/dev/urandom\x00")
@@ -400,6 +405,10 @@ func unminit() {
 // Called from exitm, but not from drop, to undo the effect of thread-owned
 // resources in minit, semacreate, or elsewhere. Do not take locks after calling this.
 func mdestroy(mp *m) {
+	if mp.vgetrandomState != 0 {
+		vgetrandomPutState(mp.vgetrandomState)
+		mp.vgetrandomState = 0
+	}
 }
 
 // #ifdef GOARCH_386
diff --git a/src/runtime/sys_linux_amd64.s b/src/runtime/sys_linux_amd64.s
index b6c64dc095..941f70b0e8 100644
--- a/src/runtime/sys_linux_amd64.s
+++ b/src/runtime/sys_linux_amd64.s
@@ -704,3 +704,36 @@ TEXT runtime·sbrk0(SB),NOSPLIT,$0-8
 	SYSCALL
 	MOVQ	AX, ret+0(FP)
 	RET
+
+// func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+TEXT runtime·vgetrandom1<ABIInternal>(SB),NOSPLIT,$16-48
+	MOVQ	SI, R8 // stateSize
+	MOVL	CX, DX // flags
+	MOVQ	DI, CX // state
+	MOVQ	BX, SI // length
+	MOVQ	AX, DI // buf
+
+	MOVQ	SP, R12
+
+	MOVQ	runtime·vdsoGetrandomSym(SB), AX
+	MOVQ	g_m(R14), BX
+
+	MOVQ	m_vdsoPC(BX), R9
+	MOVQ	R9, 0(SP)
+	MOVQ	m_vdsoSP(BX), R9
+	MOVQ	R9, 8(SP)
+	LEAQ	buf+0(FP), R9
+	MOVQ	R9, m_vdsoSP(BX)
+	MOVQ	-8(R9), R9
+	MOVQ	R9, m_vdsoPC(BX)
+
+	ANDQ	$~15, SP
+
+	CALL	AX
+
+	MOVQ	R12, SP
+	MOVQ	8(SP), R9
+	MOVQ	R9, m_vdsoSP(BX)
+	MOVQ	0(SP), R9
+	MOVQ	R9, m_vdsoPC(BX)
+	RET
diff --git a/src/runtime/sys_linux_arm64.s b/src/runtime/sys_linux_arm64.s
index 51c87bea05..7a81d5479e 100644
--- a/src/runtime/sys_linux_arm64.s
+++ b/src/runtime/sys_linux_arm64.s
@@ -785,3 +785,48 @@ TEXT runtime·sbrk0(SB),NOSPLIT,$0-8
 	SVC
 	MOVD	R0, ret+0(FP)
 	RET
+
+// func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+TEXT runtime·vgetrandom1<ABIInternal>(SB),NOSPLIT,$16-48
+	MOVD	RSP, R20
+
+	MOVD	runtime·vdsoGetrandomSym(SB), R8
+	MOVD	g_m(g), R21
+
+	MOVD	m_vdsoPC(R21), R9
+	MOVD	R9, 8(RSP)
+	MOVD	m_vdsoSP(R21), R9
+	MOVD	R9, 16(RSP)
+	MOVD	LR, m_vdsoPC(R21)
+	MOVD	$buf-8(FP), R9
+	MOVD	R9, m_vdsoSP(R21)
+
+	MOVD	RSP, R9
+	BIC	$15, R9
+	MOVD	R9, RSP
+
+	MOVBU	runtime·iscgo(SB), R9
+	CBNZ	R9, nosaveg
+	MOVD	m_gsignal(R21), R9
+	CBZ	R9, nosaveg
+	CMP	g, R9
+	BEQ	nosaveg
+	MOVD	(g_stack+stack_lo)(R9), R22
+	MOVD	g, (R22)
+
+	BL	(R8)
+
+	MOVD	ZR, (R22)
+	B	restore
+
+nosaveg:
+	BL	(R8)
+
+restore:
+	MOVD	R20, RSP
+	MOVD	16(RSP), R1
+	MOVD	R1, m_vdsoSP(R21)
+	MOVD	8(RSP), R1
+	MOVD	R1, m_vdsoPC(R21)
+	NOP	R0 // Satisfy go vet, since the return value comes from the vDSO function.
+	RET
diff --git a/src/runtime/sys_linux_loong64.s b/src/runtime/sys_linux_loong64.s
index eba8e1f24c..0a25d56047 100644
--- a/src/runtime/sys_linux_loong64.s
+++ b/src/runtime/sys_linux_loong64.s
@@ -657,3 +657,46 @@ TEXT runtime·socket(SB),$0-20
 	MOVV	R0, 2(R0) // unimplemented, only needed for android; declared in stubs_linux.go
 	MOVW	R0, ret+16(FP) // for vet
 	RET
+
+// func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+TEXT runtime·vgetrandom1<ABIInternal>(SB),NOSPLIT,$16-48
+	MOVV	R3, R23
+
+	MOVV	runtime·vdsoGetrandomSym(SB), R12
+
+	MOVV	g_m(g), R24
+
+	MOVV	m_vdsoPC(R24), R13
+	MOVV	R13, 8(R3)
+	MOVV	m_vdsoSP(R24), R13
+	MOVV	R13, 16(R3)
+	MOVV	R1, m_vdsoPC(R24)
+	MOVV    $buf-8(FP), R13
+	MOVV	R13, m_vdsoSP(R24)
+
+	AND	$~15, R3
+
+	MOVBU	runtime·iscgo(SB), R13
+	BNE	R13, nosaveg
+	MOVV	m_gsignal(R24), R13
+	BEQ	R13, nosaveg
+	BEQ	g, R13, nosaveg
+	MOVV	(g_stack+stack_lo)(R13), R25
+	MOVV	g, (R25)
+
+	JAL	(R12)
+
+	MOVV	R0, (R25)
+	JMP	restore
+
+nosaveg:
+	JAL	(R12)
+
+restore:
+	MOVV	R23, R3
+	MOVV	16(R3), R25
+	MOVV	R25, m_vdsoSP(R24)
+	MOVV	8(R3), R25
+	MOVV	R25, m_vdsoPC(R24)
+	NOP	R4 // Satisfy go vet, since the return value comes from the vDSO function.
+	RET
diff --git a/src/runtime/sys_linux_ppc64x.s b/src/runtime/sys_linux_ppc64x.s
index ba4988b723..8735b932ed 100644
--- a/src/runtime/sys_linux_ppc64x.s
+++ b/src/runtime/sys_linux_ppc64x.s
@@ -757,3 +757,53 @@ TEXT runtime·socket(SB),$0-20
 	MOVD	R0, 0(R0) // unimplemented, only needed for android; declared in stubs_linux.go
 	MOVW	R0, ret+16(FP) // for vet
 	RET
+
+// func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+TEXT runtime·vgetrandom1<ABIInternal>(SB),NOSPLIT,$16-48
+	MOVD	R1, R15
+
+	MOVD	runtime·vdsoGetrandomSym(SB), R12
+	MOVD	R12, CTR
+	MOVD	g_m(g), R21
+
+	MOVD	m_vdsoPC(R21), R22
+	MOVD	R22, 32(R1)
+	MOVD	m_vdsoSP(R21), R22
+	MOVD	R22, 40(R1)
+	MOVD	LR, m_vdsoPC(R21)
+	MOVD	$buf-FIXED_FRAME(FP), R22
+	MOVD	R22, m_vdsoSP(R21)
+
+	RLDICR  $0, R1, $59, R1
+
+	MOVBZ	runtime·iscgo(SB), R22
+	CMP	R22, $0
+	BNE	nosaveg
+	MOVD	m_gsignal(R21), R22
+	CMP	R22, $0
+	BEQ	nosaveg
+	CMP	R22, g
+	BEQ	nosaveg
+	MOVD	(g_stack+stack_lo)(R22), R22
+	MOVD	g, (R22)
+
+	BL	(CTR)
+
+	MOVD	$0, (R22)
+	JMP	restore
+
+nosaveg:
+	BL	(CTR)
+
+restore:
+	MOVD	$0, R0
+	MOVD	R15, R1
+	MOVD	40(R1), R22
+	MOVD	R22, m_vdsoSP(R21)
+	MOVD	32(R1), R22
+	MOVD	R22, m_vdsoPC(R21)
+
+	BVC	out
+	NEG	R3, R3
+out:
+	RET
diff --git a/src/runtime/sys_linux_s390x.s b/src/runtime/sys_linux_s390x.s
index adf5612c3c..7da4a52729 100644
--- a/src/runtime/sys_linux_s390x.s
+++ b/src/runtime/sys_linux_s390x.s
@@ -604,3 +604,53 @@ TEXT runtime·socket(SB),$0-20
 	MOVD	$0, 2(R0) // unimplemented, only needed for android; declared in stubs_linux.go
 	MOVW	R0, ret+16(FP)
 	RET
+
+// func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+TEXT runtime·vgetrandom1(SB),NOSPLIT,$16-48
+	MOVD	buf+0(FP), R2
+	MOVD	length+8(FP), R3
+	MOVW	flags+16(FP), R4
+	MOVD	state+24(FP), R5
+	MOVD	stateSize+32(FP), R6
+
+	MOVD	R15, R7
+
+	MOVD	runtime·vdsoGetrandomSym(SB), R1
+	MOVD	g_m(g), R9
+
+	MOVD	m_vdsoPC(R9), R12
+	MOVD	R12, 8(R15)
+	MOVD	m_vdsoSP(R9), R12
+	MOVD	R12, 16(R15)
+	MOVD	R14, m_vdsoPC(R9)
+	MOVD	$buf+0(FP), R12
+	MOVD	R12, m_vdsoSP(R9)
+
+	SUB	$160, R15
+	MOVD	$~7, R12
+	AND	R12, R15
+
+	MOVB	runtime·iscgo(SB), R12
+	CMPBNE	R12, $0, nosaveg
+	MOVD	m_gsignal(R9), R12
+	CMPBEQ	R12, $0, nosaveg
+	CMPBEQ	g, R12, nosaveg
+	MOVD	(g_stack+stack_lo)(R12), R12
+	MOVD	g, (R12)
+
+	BL	R1
+
+	MOVD	$0, (R12)
+	JMP	restore
+
+nosaveg:
+	BL	R1
+
+restore:
+	MOVD	R7, R15
+	MOVD	16(R15), R12
+	MOVD	R12, m_vdsoSP(R9)
+	MOVD	8(R15), R12
+	MOVD	R12, m_vdsoPC(R9)
+	MOVD	R2, ret+40(FP)
+	RET
diff --git a/src/runtime/vdso_linux_amd64.go b/src/runtime/vdso_linux_amd64.go
index 9c56409137..8a89771ca4 100644
--- a/src/runtime/vdso_linux_amd64.go
+++ b/src/runtime/vdso_linux_amd64.go
@@ -17,11 +17,13 @@ var vdsoLinuxVersion = vdsoVersionKey{"LINUX_2.6", 0x3ae75f6}
 var vdsoSymbolKeys = []vdsoSymbolKey{
 	{"__vdso_gettimeofday", 0x315ca59, 0xb01bca00, &vdsoGettimeofdaySym},
 	{"__vdso_clock_gettime", 0xd35ec75, 0x6e43a318, &vdsoClockgettimeSym},
+	{"__vdso_getrandom", 0x25425d, 0x84a559bf, &vdsoGetrandomSym},
 }
 
 var (
 	vdsoGettimeofdaySym uintptr
 	vdsoClockgettimeSym uintptr
+	vdsoGetrandomSym    uintptr
 )
 
 // vdsoGettimeofdaySym is accessed from the syscall package.
diff --git a/src/runtime/vdso_linux_arm64.go b/src/runtime/vdso_linux_arm64.go
index f5959525af..21f875d0df 100644
--- a/src/runtime/vdso_linux_arm64.go
+++ b/src/runtime/vdso_linux_arm64.go
@@ -15,7 +15,10 @@ var vdsoLinuxVersion = vdsoVersionKey{"LINUX_2.6.39", 0x75fcb89}
 
 var vdsoSymbolKeys = []vdsoSymbolKey{
 	{"__kernel_clock_gettime", 0xb0cd725, 0xdfa941fd, &vdsoClockgettimeSym},
+	{"__kernel_getrandom", 0x9800c0d, 0x540d4e24, &vdsoGetrandomSym},
 }
 
-// initialize to fall back to syscall
-var vdsoClockgettimeSym uintptr = 0
+var (
+	vdsoClockgettimeSym uintptr
+	vdsoGetrandomSym    uintptr
+)
diff --git a/src/runtime/vdso_linux_loong64.go b/src/runtime/vdso_linux_loong64.go
index e00ef952b3..37ee02dcb0 100644
--- a/src/runtime/vdso_linux_loong64.go
+++ b/src/runtime/vdso_linux_loong64.go
@@ -19,9 +19,10 @@ var vdsoLinuxVersion = vdsoVersionKey{"LINUX_5.10", 0xae78f70}
 
 var vdsoSymbolKeys = []vdsoSymbolKey{
 	{"__vdso_clock_gettime", 0xd35ec75, 0x6e43a318, &vdsoClockgettimeSym},
+	{"__vdso_getrandom", 0x25425d, 0x84a559bf, &vdsoGetrandomSym},
 }
 
-// initialize to fall back to syscall
 var (
-	vdsoClockgettimeSym uintptr = 0
+	vdsoClockgettimeSym uintptr
+	vdsoGetrandomSym    uintptr
 )
diff --git a/src/runtime/vdso_linux_ppc64x.go b/src/runtime/vdso_linux_ppc64x.go
index 09c8d9d20e..939da7b54c 100644
--- a/src/runtime/vdso_linux_ppc64x.go
+++ b/src/runtime/vdso_linux_ppc64x.go
@@ -16,9 +16,10 @@ var vdsoLinuxVersion = vdsoVersionKey{"LINUX_2.6.15", 0x75fcba5}
 
 var vdsoSymbolKeys = []vdsoSymbolKey{
 	{"__kernel_clock_gettime", 0xb0cd725, 0xdfa941fd, &vdsoClockgettimeSym},
+	{"__kernel_getrandom", 0x9800c0d, 0x540d4e24, &vdsoGetrandomSym},
 }
 
-// initialize with vsyscall fallbacks
 var (
-	vdsoClockgettimeSym uintptr = 0
+	vdsoClockgettimeSym uintptr
+	vdsoGetrandomSym    uintptr
 )
diff --git a/src/runtime/vdso_linux_s390x.go b/src/runtime/vdso_linux_s390x.go
index 970ecd3ce6..113152ff0a 100644
--- a/src/runtime/vdso_linux_s390x.go
+++ b/src/runtime/vdso_linux_s390x.go
@@ -16,9 +16,10 @@ var vdsoLinuxVersion = vdsoVersionKey{"LINUX_2.6.29", 0x75fcbb9}
 
 var vdsoSymbolKeys = []vdsoSymbolKey{
 	{"__kernel_clock_gettime", 0xb0cd725, 0xdfa941fd, &vdsoClockgettimeSym},
+	{"__kernel_getrandom", 0x9800c0d, 0x540d4e24, &vdsoGetrandomSym},
 }
 
-// initialize with vsyscall fallbacks
 var (
-	vdsoClockgettimeSym uintptr = 0
+	vdsoClockgettimeSym uintptr
+	vdsoGetrandomSym    uintptr
 )
diff --git a/src/runtime/vgetrandom_linux.go b/src/runtime/vgetrandom_linux.go
new file mode 100644
index 0000000000..1e8c8ceaf0
--- /dev/null
+++ b/src/runtime/vgetrandom_linux.go
@@ -0,0 +1,100 @@
+// Copyright 2024 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build linux && (amd64 || arm64 || arm64be || ppc64 || ppc64le || loong64 || s390x)
+
+package runtime
+
+import "unsafe"
+
+func vgetrandom1(buf *byte, length uintptr, flags uint32, state uintptr, stateSize uintptr) int
+
+var vgetrandomAlloc struct {
+	states     []uintptr
+	statesLock mutex
+	stateSize  uintptr
+	mmapProt   int32
+	mmapFlags  int32
+}
+
+func vgetrandomInit() {
+	if vdsoGetrandomSym == 0 {
+		return
+	}
+
+	var params struct {
+		SizeOfOpaqueState uint32
+		MmapProt          uint32
+		MmapFlags         uint32
+		reserved          [13]uint32
+	}
+	if vgetrandom1(nil, 0, 0, uintptr(unsafe.Pointer(&params)), ^uintptr(0)) != 0 {
+		return
+	}
+	vgetrandomAlloc.stateSize = uintptr(params.SizeOfOpaqueState)
+	vgetrandomAlloc.mmapProt = int32(params.MmapProt)
+	vgetrandomAlloc.mmapFlags = int32(params.MmapFlags)
+
+	lockInit(&vgetrandomAlloc.statesLock, lockRankLeafRank)
+}
+
+func vgetrandomGetState() uintptr {
+	lock(&vgetrandomAlloc.statesLock)
+	if len(vgetrandomAlloc.states) == 0 {
+		num := uintptr(ncpu) // Just a reasonable size hint to start.
+		allocSize := (num*vgetrandomAlloc.stateSize + physPageSize - 1) &^ (physPageSize - 1)
+		num = (physPageSize / vgetrandomAlloc.stateSize) * (allocSize / physPageSize)
+		p, err := mmap(nil, allocSize, vgetrandomAlloc.mmapProt, vgetrandomAlloc.mmapFlags, -1, 0)
+		if err != 0 {
+			unlock(&vgetrandomAlloc.statesLock)
+			return 0
+		}
+		newBlock := uintptr(p)
+		if vgetrandomAlloc.states == nil {
+			vgetrandomAlloc.states = make([]uintptr, 0, num)
+		}
+		for i := uintptr(0); i < num; i++ {
+			if (newBlock&(physPageSize-1))+vgetrandomAlloc.stateSize > physPageSize {
+				newBlock = (newBlock + physPageSize - 1) &^ (physPageSize - 1)
+			}
+			vgetrandomAlloc.states = append(vgetrandomAlloc.states, newBlock)
+			newBlock += vgetrandomAlloc.stateSize
+		}
+	}
+	state := vgetrandomAlloc.states[len(vgetrandomAlloc.states)-1]
+	vgetrandomAlloc.states = vgetrandomAlloc.states[:len(vgetrandomAlloc.states)-1]
+	unlock(&vgetrandomAlloc.statesLock)
+	return state
+}
+
+func vgetrandomPutState(state uintptr) {
+	lock(&vgetrandomAlloc.statesLock)
+	vgetrandomAlloc.states = append(vgetrandomAlloc.states, state)
+	unlock(&vgetrandomAlloc.statesLock)
+}
+
+// This is exported for use in internal/syscall/unix as well as x/sys/unix.
+//
+//go:linkname vgetrandom
+func vgetrandom(p []byte, flags uint32) (ret int, supported bool) {
+	if vgetrandomAlloc.stateSize == 0 {
+		return -1, false
+	}
+
+	mp := acquirem()
+	if mp.vgetrandomState == 0 {
+		state := vgetrandomGetState()
+		if state == 0 {
+			releasem(mp)
+			return -1, false
+		}
+		mp.vgetrandomState = state
+	}
+
+	ret = vgetrandom1(unsafe.SliceData(p), uintptr(len(p)), flags, mp.vgetrandomState, vgetrandomAlloc.stateSize)
+	supported = true
+
+	releasem(mp)
+	return
+}
diff --git a/src/runtime/vgetrandom_unsupported.go b/src/runtime/vgetrandom_unsupported.go
new file mode 100644
index 0000000000..070392cfaa
--- /dev/null
+++ b/src/runtime/vgetrandom_unsupported.go
@@ -0,0 +1,18 @@
+// Copyright 2024 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !(linux && (amd64 || arm64 || arm64be || ppc64 || ppc64le || loong64 || s390x))
+
+package runtime
+
+import _ "unsafe"
+
+//go:linkname vgetrandom
+func vgetrandom(p []byte, flags uint32) (ret int, supported bool) {
+	return -1, false
+}
+
+func vgetrandomPutState(state uintptr) {}
+
+func vgetrandomInit() {}