runtime: optimize permission changes with mprotect

On Linux, both mprotect() and mmap() acquire the mmap_lock (in writer mode),
posing scalability challenges.

The mmap_lock (formerly called mmap_sem) is a reader/writer lock that controls
access to a process's address space; before making changes there (mapping in a
new range, for example), the kernel must acquire that lock.

Page-fault handling must also acquire mmap_lock (in reader mode) to ensure that
the address space doesn't change in surprising ways while a fault is being resolved.

A process can have a large address space and many threads running (and incurring
page faults) concurrently, turning mmap_lock into a significant bottleneck.

While both mmap() and mprotect() are protected by the mmap_lock, the shorter
duration of mprotect system call, due to their simpler nature, results in a reduced
locking time for the mmap_lock.

src/internal/runtime/syscall/defs_linux_386.go

@@ -6,6 +6,7 @@ package syscall
 
 const (
 	SYS_FCNTL         = 55
+	SYS_MPROTECT      = 125
 	SYS_EPOLL_CTL     = 255
 	SYS_EPOLL_PWAIT   = 319
 	SYS_EPOLL_CREATE1 = 329

src/internal/runtime/syscall/defs_linux_amd64.go

@@ -5,6 +5,7 @@
 package syscall
 
 const (
+	SYS_MPROTECT      = 10
 	SYS_FCNTL         = 72
 	SYS_EPOLL_CTL     = 233
 	SYS_EPOLL_PWAIT   = 281

src/internal/runtime/syscall/defs_linux_arm.go

@@ -6,6 +6,7 @@ package syscall
 
 const (
 	SYS_FCNTL         = 55
+	SYS_MPROTECT      = 125
 	SYS_EPOLL_CTL     = 251
 	SYS_EPOLL_PWAIT   = 346
 	SYS_EPOLL_CREATE1 = 357

src/internal/runtime/syscall/defs_linux_arm64.go

@@ -9,6 +9,7 @@ const (
 	SYS_EPOLL_CTL     = 21
 	SYS_EPOLL_PWAIT   = 22
 	SYS_FCNTL         = 25
+	SYS_MPROTECT      = 226
 	SYS_EPOLL_PWAIT2  = 441
 	SYS_EVENTFD2      = 19
 

src/internal/runtime/syscall/defs_linux_loong64.go

@@ -9,6 +9,7 @@ const (
 	SYS_EPOLL_CTL     = 21
 	SYS_EPOLL_PWAIT   = 22
 	SYS_FCNTL         = 25
+	SYS_MPROTECT      = 226
 	SYS_EPOLL_PWAIT2  = 441
 	SYS_EVENTFD2      = 19
 

src/internal/runtime/syscall/defs_linux_mips64x.go

@@ -7,6 +7,7 @@
 package syscall
 
 const (
+	SYS_MPROTECT      = 5010
 	SYS_FCNTL         = 5070
 	SYS_EPOLL_CTL     = 5208
 	SYS_EPOLL_PWAIT   = 5272

src/internal/runtime/syscall/defs_linux_mipsx.go

@@ -8,6 +8,7 @@ package syscall
 
 const (
 	SYS_FCNTL         = 4055
+	SYS_MPROTECT      = 4125
 	SYS_EPOLL_CTL     = 4249
 	SYS_EPOLL_PWAIT   = 4313
 	SYS_EPOLL_CREATE1 = 4326

src/internal/runtime/syscall/defs_linux_ppc64x.go

@@ -8,6 +8,7 @@ package syscall
 
 const (
 	SYS_FCNTL         = 55
+	SYS_MPROTECT      = 125
 	SYS_EPOLL_CTL     = 237
 	SYS_EPOLL_PWAIT   = 303
 	SYS_EPOLL_CREATE1 = 315

src/internal/runtime/syscall/defs_linux_riscv64.go

@@ -9,6 +9,7 @@ const (
 	SYS_EPOLL_CTL     = 21
 	SYS_EPOLL_PWAIT   = 22
 	SYS_FCNTL         = 25
+	SYS_MPROTECT      = 226
 	SYS_EPOLL_PWAIT2  = 441
 	SYS_EVENTFD2      = 19
 

src/internal/runtime/syscall/defs_linux_s390x.go

@@ -6,6 +6,7 @@ package syscall
 
 const (
 	SYS_FCNTL         = 55
+	SYS_MPROTECT      = 125
 	SYS_EPOLL_CTL     = 250
 	SYS_EPOLL_PWAIT   = 312
 	SYS_EPOLL_CREATE1 = 327

src/runtime/mem_linux.go

@@ -150,7 +150,8 @@ func sysFreeOS(v unsafe.Pointer, n uintptr) {
 }
 
 func sysFaultOS(v unsafe.Pointer, n uintptr) {
-	mmap(v, n, _PROT_NONE, _MAP_ANON|_MAP_PRIVATE|_MAP_FIXED, -1, 0)
+	mprotect(v, n, _PROT_NONE)
+	madvise(v, n, _MADV_DONTNEED)
 }
 
 func sysReserveOS(v unsafe.Pointer, n uintptr) unsafe.Pointer {

src/runtime/os_linux.go

@@ -891,3 +891,9 @@ func (c *sigctxt) sigFromUser() bool {
 	code := int32(c.sigcode())
 	return code == _SI_USER || code == _SI_TKILL
 }
+
+//go:nosplit
+func mprotect(addr unsafe.Pointer, n uintptr, prot int32) (ret int32, errno int32) {
+	r, _, err := syscall.Syscall6(syscall.SYS_MPROTECT, uintptr(addr), n, uintptr(prot), 0, 0, 0)
+	return int32(r), int32(err)
+}