diff --git a/src/runtime/proc.go b/src/runtime/proc.go
index aa0a1fa2be..7d2ff2748b 100644
--- a/src/runtime/proc.go
+++ b/src/runtime/proc.go
@@ -4166,6 +4166,21 @@ func (pp *p) destroy() {
 	gfpurge(pp)
 	traceProcFree(pp)
 	if raceenabled {
+		if pp.timerRaceCtx != 0 {
+			// The race detector code uses a callback to fetch
+			// the proc context, so arrange for that callback
+			// to see the right thing.
+			// This hack only works because we are the only
+			// thread running.
+			mp := getg().m
+			phold := mp.p.ptr()
+			mp.p.set(pp)
+
+			racectxend(pp.timerRaceCtx)
+			pp.timerRaceCtx = 0
+
+			mp.p.set(phold)
+		}
 		raceprocdestroy(pp.raceprocctx)
 		pp.raceprocctx = 0
 	}
diff --git a/src/runtime/race.go b/src/runtime/race.go
index d2fc6a3c47..d11dc9b5bf 100644
--- a/src/runtime/race.go
+++ b/src/runtime/race.go
@@ -459,6 +459,11 @@ func racegoend() {
 	racecall(&__tsan_go_end, getg().racectx, 0, 0, 0)
 }
 
+//go:nosplit
+func racectxend(racectx uintptr) {
+	racecall(&__tsan_go_end, racectx, 0, 0, 0)
+}
+
 //go:nosplit
 func racewriterangepc(addr unsafe.Pointer, sz, callpc, pc uintptr) {
 	_g_ := getg()
@@ -506,6 +511,14 @@ func raceacquireg(gp *g, addr unsafe.Pointer) {
 	racecall(&__tsan_acquire, gp.racectx, uintptr(addr), 0, 0)
 }
 
+//go:nosplit
+func raceacquirectx(racectx uintptr, addr unsafe.Pointer) {
+	if !isvalidaddr(addr) {
+		return
+	}
+	racecall(&__tsan_acquire, racectx, uintptr(addr), 0, 0)
+}
+
 //go:nosplit
 func racerelease(addr unsafe.Pointer) {
 	racereleaseg(getg(), addr)
diff --git a/src/runtime/race0.go b/src/runtime/race0.go
index f1d3706231..6f26afa854 100644
--- a/src/runtime/race0.go
+++ b/src/runtime/race0.go
@@ -29,6 +29,7 @@ func racereadrangepc(addr unsafe.Pointer, sz, callerpc, pc uintptr)         { th
 func racewriterangepc(addr unsafe.Pointer, sz, callerpc, pc uintptr)        { throw("race") }
 func raceacquire(addr unsafe.Pointer)                                       { throw("race") }
 func raceacquireg(gp *g, addr unsafe.Pointer)                               { throw("race") }
+func raceacquirectx(racectx uintptr, addr unsafe.Pointer)                   { throw("race") }
 func racerelease(addr unsafe.Pointer)                                       { throw("race") }
 func racereleaseg(gp *g, addr unsafe.Pointer)                               { throw("race") }
 func racereleasemerge(addr unsafe.Pointer)                                  { throw("race") }
@@ -38,3 +39,4 @@ func racemalloc(p unsafe.Pointer, sz uintptr)                               { th
 func racefree(p unsafe.Pointer, sz uintptr)                                 { throw("race") }
 func racegostart(pc uintptr) uintptr                                        { throw("race"); return 0 }
 func racegoend()                                                            { throw("race") }
+func racectxend(racectx uintptr)                                            { throw("race") }
diff --git a/src/runtime/race_amd64.s b/src/runtime/race_amd64.s
index 4ed9533bfb..758d543203 100644
--- a/src/runtime/race_amd64.s
+++ b/src/runtime/race_amd64.s
@@ -416,9 +416,11 @@ rest:
 	// Set g = g0.
 	get_tls(R12)
 	MOVQ	g(R12), R13
-	MOVQ	g_m(R13), R13
-	MOVQ	m_g0(R13), R14
-	MOVQ	R14, g(R12)	// g = m->g0
+	MOVQ	g_m(R13), R14
+	MOVQ	m_g0(R14), R15
+	CMPQ	R13, R15
+	JEQ	noswitch	// branch if already on g0
+	MOVQ	R15, g(R12)	// g = m->g0
 	PUSHQ	RARG1	// func arg
 	PUSHQ	RARG0	// func arg
 	CALL	runtime·racecallback(SB)
@@ -430,6 +432,7 @@ rest:
 	MOVQ	g_m(R13), R13
 	MOVQ	m_curg(R13), R14
 	MOVQ	R14, g(R12)	// g = m->curg
+ret:
 	// Restore callee-saved registers.
 	POPQ	R15
 	POPQ	R14
@@ -440,3 +443,12 @@ rest:
 	POPQ	BP
 	POPQ	BX
 	RET
+
+noswitch:
+	// already on g0
+	PUSHQ	RARG1	// func arg
+	PUSHQ	RARG0	// func arg
+	CALL	runtime·racecallback(SB)
+	POPQ	R12
+	POPQ	R12
+	JMP	ret
diff --git a/src/runtime/race_arm64.s b/src/runtime/race_arm64.s
index 00a67e8602..46224f8d73 100644
--- a/src/runtime/race_arm64.s
+++ b/src/runtime/race_arm64.s
@@ -448,7 +448,10 @@ rest:
 	// restore R0
 	MOVD	R13, R0
 	MOVD	g_m(g), R13
-	MOVD	m_g0(R13), g
+	MOVD	m_g0(R13), R14
+	CMP	R14, g
+	BEQ	noswitch	// branch if already on g0
+	MOVD	R14, g
 
 	MOVD	R0, 8(RSP)	// func arg
 	MOVD	R1, 16(RSP)	// func arg
@@ -457,6 +460,7 @@ rest:
 	// All registers are smashed after Go code, reload.
 	MOVD	g_m(g), R13
 	MOVD	m_curg(R13), g	// g = m->curg
+ret:
 	// Restore callee-saved registers.
 	MOVD	0(RSP), LR
 	LDP	24(RSP), (R19, R20)
@@ -467,5 +471,12 @@ rest:
 	ADD	$112, RSP
 	JMP	(LR)
 
+noswitch:
+	// already on g0
+	MOVD	R0, 8(RSP)	// func arg
+	MOVD	R1, 16(RSP)	// func arg
+	BL	runtime·racecallback(SB)
+	JMP	ret
+
 // tls_g, g value for each thread in TLS
 GLOBL runtime·tls_g+0(SB), TLSBSS+DUPOK, $8
diff --git a/src/runtime/race_ppc64le.s b/src/runtime/race_ppc64le.s
index 0486bb338b..7421d539ca 100644
--- a/src/runtime/race_ppc64le.s
+++ b/src/runtime/race_ppc64le.s
@@ -507,20 +507,29 @@ rest:
 	FMOVD   F30, 312(R1)
 	FMOVD   F31, 320(R1)
 
+	MOVD	R3, FIXED_FRAME+0(R1)
+	MOVD	R4, FIXED_FRAME+8(R1)
+
 	MOVD    runtime·tls_g(SB), R10
 	MOVD    0(R13)(R10*1), g
 
 	MOVD	g_m(g), R7
-	MOVD	m_g0(R7), g // set g = m-> g0
-	MOVD	R3, FIXED_FRAME+0(R1)
-	MOVD	R4, FIXED_FRAME+8(R1)
+	MOVD	m_g0(R7), R8
+	CMP	g, R8
+	BEQ	noswitch
+
+	MOVD	R8, g // set g = m-> g0
+
 	BL	runtime·racecallback(SB)
+
 	// All registers are clobbered after Go code, reload.
 	MOVD    runtime·tls_g(SB), R10
 	MOVD    0(R13)(R10*1), g
 
 	MOVD	g_m(g), R7
 	MOVD	m_curg(R7), g // restore g = m->curg
+
+ret:
 	MOVD    328(R1), R14
 	MOVD    48(R1), R15
 	MOVD    56(R1), R16
@@ -565,5 +574,9 @@ rest:
 	MOVD    R10, LR
 	RET
 
+noswitch:
+	BL      runtime·racecallback(SB)
+	JMP     ret
+
 // tls_g, g value for each thread in TLS
 GLOBL runtime·tls_g+0(SB), TLSBSS+DUPOK, $8
diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index b57ae75baf..2e6c3d9d79 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -613,6 +613,9 @@ type p struct {
 	// such as timerModifying.
 	adjustTimers uint32
 
+	// Race context used while executing timer functions.
+	timerRaceCtx uintptr
+
 	pad cpu.CacheLinePad
 }
 
diff --git a/src/runtime/time.go b/src/runtime/time.go
index 4bc819f023..fea5d6871c 100644
--- a/src/runtime/time.go
+++ b/src/runtime/time.go
@@ -9,6 +9,7 @@ package runtime
 import (
 	"internal/cpu"
 	"runtime/internal/atomic"
+	"runtime/internal/sys"
 	"unsafe"
 )
 
@@ -1095,6 +1096,13 @@ func runtimer(pp *p, now int64) int64 {
 // runOneTimer runs a single timer.
 // The caller must have locked the timers for pp.
 func runOneTimer(pp *p, t *timer, now int64) {
+	if raceenabled {
+		if pp.timerRaceCtx == 0 {
+			pp.timerRaceCtx = racegostart(funcPC(runtimer) + sys.PCQuantum)
+		}
+		raceacquirectx(pp.timerRaceCtx, unsafe.Pointer(t))
+	}
+
 	f := t.f
 	arg := t.arg
 	seq := t.seq
@@ -1119,10 +1127,24 @@ func runOneTimer(pp *p, t *timer, now int64) {
 		}
 	}
 
+	if raceenabled {
+		// Temporarily use the P's racectx for g0.
+		gp := getg()
+		if gp.racectx != 0 {
+			throw("runOneTimer: unexpected racectx")
+		}
+		gp.racectx = pp.timerRaceCtx
+	}
+
 	// Note that since timers are locked here, f may not call
 	// addtimer or resettimer.
 
 	f(arg, seq)
+
+	if raceenabled {
+		gp := getg()
+		gp.racectx = 0
+	}
 }
 
 func timejump() *p {