Merge tag 'x86_fsgsbase_for_v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fsgsbase updates from Borislav Petkov:
 "Misc minor cleanups and corrections to the fsgsbase code and
  respective selftests"

* tag 'x86_fsgsbase_for_v5.10' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  selftests/x86/fsgsbase: Test PTRACE_PEEKUSER for GSBASE with invalid LDT GS
  selftests/x86/fsgsbase: Reap a forgotten child
  x86/fsgsbase: Replace static_cpu_has() with boot_cpu_has()
  x86/entry/64: Correct the comment over SAVE_AND_SET_GSBASE

arch/x86/entry/entry_64.S

@@ -842,8 +842,9 @@ SYM_CODE_START_LOCAL(paranoid_entry)
 	 * retrieve and set the current CPUs kernel GSBASE. The stored value
 	 * has to be restored in paranoid_exit unconditionally.
 	 *
-	 * The MSR write ensures that no subsequent load is based on a
+	 * The unconditional write to GS base below ensures that no subsequent
-	 * mispredicted GSBASE. No extra FENCE required.
+	 * loads based on a mispredicted GS base can happen, therefore no LFENCE
+	 * is needed here.
 	 */
 	SAVE_AND_SET_GSBASE scratch_reg=%rax save_reg=%rbx
 	ret

arch/x86/include/asm/fsgsbase.h

@@ -57,7 +57,7 @@ static inline unsigned long x86_fsbase_read_cpu(void)
 {
 	unsigned long fsbase;
 
-	if (static_cpu_has(X86_FEATURE_FSGSBASE))
+	if (boot_cpu_has(X86_FEATURE_FSGSBASE))
 		fsbase = rdfsbase();
 	else
 		rdmsrl(MSR_FS_BASE, fsbase);
@@ -67,7 +67,7 @@ static inline unsigned long x86_fsbase_read_cpu(void)
 
 static inline void x86_fsbase_write_cpu(unsigned long fsbase)
 {
-	if (static_cpu_has(X86_FEATURE_FSGSBASE))
+	if (boot_cpu_has(X86_FEATURE_FSGSBASE))
 		wrfsbase(fsbase);
 	else
 		wrmsrl(MSR_FS_BASE, fsbase);

arch/x86/kernel/process_64.c

@@ -407,7 +407,7 @@ unsigned long x86_gsbase_read_cpu_inactive(void)
 {
 	unsigned long gsbase;
 
-	if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
+	if (boot_cpu_has(X86_FEATURE_FSGSBASE)) {
 		unsigned long flags;
 
 		local_irq_save(flags);
@@ -422,7 +422,7 @@ unsigned long x86_gsbase_read_cpu_inactive(void)
 
 void x86_gsbase_write_cpu_inactive(unsigned long gsbase)
 {
-	if (static_cpu_has(X86_FEATURE_FSGSBASE)) {
+	if (boot_cpu_has(X86_FEATURE_FSGSBASE)) {
 		unsigned long flags;
 
 		local_irq_save(flags);
@@ -439,7 +439,7 @@ unsigned long x86_fsbase_read_task(struct task_struct *task)
 
 	if (task == current)
 		fsbase = x86_fsbase_read_cpu();
-	else if (static_cpu_has(X86_FEATURE_FSGSBASE) ||
+	else if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
 		 (task->thread.fsindex == 0))
 		fsbase = task->thread.fsbase;
 	else
@@ -454,7 +454,7 @@ unsigned long x86_gsbase_read_task(struct task_struct *task)
 
 	if (task == current)
 		gsbase = x86_gsbase_read_cpu_inactive();
-	else if (static_cpu_has(X86_FEATURE_FSGSBASE) ||
+	else if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
 		 (task->thread.gsindex == 0))
 		gsbase = task->thread.gsbase;
 	else

tools/testing/selftests/x86/fsgsbase.c

@@ -443,6 +443,68 @@ static void test_unexpected_base(void)
 
 #define USER_REGS_OFFSET(r) offsetof(struct user_regs_struct, r)
 
+static void test_ptrace_write_gs_read_base(void)
+{
+	int status;
+	pid_t child = fork();
+
+	if (child < 0)
+		err(1, "fork");
+
+	if (child == 0) {
+		printf("[RUN]\tPTRACE_POKE GS, read GSBASE back\n");
+
+		printf("[RUN]\tARCH_SET_GS to 1\n");
+		if (syscall(SYS_arch_prctl, ARCH_SET_GS, 1) != 0)
+			err(1, "ARCH_SET_GS");
+
+		if (ptrace(PTRACE_TRACEME, 0, NULL, NULL) != 0)
+			err(1, "PTRACE_TRACEME");
+
+		raise(SIGTRAP);
+		_exit(0);
+	}
+
+	wait(&status);
+
+	if (WSTOPSIG(status) == SIGTRAP) {
+		unsigned long base;
+		unsigned long gs_offset = USER_REGS_OFFSET(gs);
+		unsigned long base_offset = USER_REGS_OFFSET(gs_base);
+
+		/* Read the initial base.  It should be 1. */
+		base = ptrace(PTRACE_PEEKUSER, child, base_offset, NULL);
+		if (base == 1) {
+			printf("[OK]\tGSBASE started at 1\n");
+		} else {
+			nerrs++;
+			printf("[FAIL]\tGSBASE started at 0x%lx\n", base);
+		}
+
+		printf("[RUN]\tSet GS = 0x7, read GSBASE\n");
+
+		/* Poke an LDT selector into GS. */
+		if (ptrace(PTRACE_POKEUSER, child, gs_offset, 0x7) != 0)
+			err(1, "PTRACE_POKEUSER");
+
+		/* And read the base. */
+		base = ptrace(PTRACE_PEEKUSER, child, base_offset, NULL);
+
+		if (base == 0 || base == 1) {
+			printf("[OK]\tGSBASE reads as 0x%lx with invalid GS\n", base);
+		} else {
+			nerrs++;
+			printf("[FAIL]\tGSBASE=0x%lx (should be 0 or 1)\n", base);
+		}
+	}
+
+	ptrace(PTRACE_CONT, child, NULL, NULL);
+
+	wait(&status);
+	if (!WIFEXITED(status))
+		printf("[WARN]\tChild didn't exit cleanly.\n");
+}
+
 static void test_ptrace_write_gsbase(void)
 {
 	int status;
@@ -517,6 +579,9 @@ static void test_ptrace_write_gsbase(void)
 
 END:
 	ptrace(PTRACE_CONT, child, NULL, NULL);
+	wait(&status);
+	if (!WIFEXITED(status))
+		printf("[WARN]\tChild didn't exit cleanly.\n");
 }
 
 int main()
@@ -526,6 +591,9 @@ int main()
 	shared_scratch = mmap(NULL, 4096, PROT_READ | PROT_WRITE,
 			      MAP_ANONYMOUS | MAP_SHARED, -1, 0);
 
+	/* Do these tests before we have an LDT. */
+	test_ptrace_write_gs_read_base();
+
 	/* Probe FSGSBASE */
 	sethandler(SIGILL, sigill, 0);
 	if (sigsetjmp(jmpbuf, 1) == 0) {