powerpc/book3s64: Fix link stack flush on context switch

In commit ee13cb249fab ("powerpc/64s: Add support for software count
cache flush"), I added support for software to flush the count
cache (indirect branch cache) on context switch if firmware told us
that was the required mitigation for Spectre v2.

As part of that code we also added a software flush of the link
stack (return address stack), which protects against Spectre-RSB
between user processes.

That is all correct for CPUs that activate that mitigation, which is
currently Power9 Nimbus DD2.3.

What I got wrong is that on older CPUs, where firmware has disabled
the count cache, we also need to flush the link stack on context
switch.

To fix it we create a new feature bit which is not set by firmware,
which tells us we need to flush the link stack. We set that when
firmware tells us that either of the existing Spectre v2 mitigations
are enabled.

Then we adjust the patching code so that if we see that feature bit we
enable the link stack flush. If we're also told to flush the count
cache in software then we fall through and do that also.

On the older CPUs we don't need to do do the software count cache
flush, firmware has disabled it, so in that case we patch in an early
return after the link stack flush.

The naming of some of the functions is awkward after this patch,
because they're called "count cache" but they also do link stack. But
we'll fix that up in a later commit to ease backporting.

This is the fix for CVE-2019-18660.

Reported-by: Anthony Steinhauser <asteinhauser@google.com>
Fixes: ee13cb249fab ("powerpc/64s: Add support for software count cache flush")
Cc: stable@vger.kernel.org # v4.4+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

4 files changed, 54 insertions, 4 deletions

diff --git a/arch/powerpc/include/asm/asm-prototypes.h b/arch/powerpc/include/asm/asm-prototypes.h
index 8561498e653c..3ee92f692e64 100644
--- a/arch/powerpc/include/asm/asm-prototypes.h
+++ b/arch/powerpc/include/asm/asm-prototypes.h
@@ -152,6 +152,7 @@ void _kvmppc_save_tm_pr(struct kvm_vcpu *vcpu, u64 guest_msr);
 /* Patch sites */
 extern s32 patch__call_flush_count_cache;
 extern s32 patch__flush_count_cache_return;
+extern s32 patch__flush_link_stack_return;
 extern s32 patch__memset_nocache, patch__memcpy_nocache;
 
 extern long flush_count_cache;
diff --git a/arch/powerpc/include/asm/security_features.h b/arch/powerpc/include/asm/security_features.h
index 759597bf0fd8..ccf44c135389 100644
--- a/arch/powerpc/include/asm/security_features.h
+++ b/arch/powerpc/include/asm/security_features.h
@@ -81,6 +81,9 @@ static inline bool security_ftr_enabled(unsigned long feature)
 // Software required to flush count cache on context switch
 #define SEC_FTR_FLUSH_COUNT_CACHE	0x0000000000000400ull
 
+// Software required to flush link stack on context switch
+#define SEC_FTR_FLUSH_LINK_STACK	0x0000000000001000ull
+
 
 // Features enabled by default
 #define SEC_FTR_DEFAULT \
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S
index 6467bdab8d40..3fd3ef352e3f 100644
--- a/arch/powerpc/kernel/entry_64.S
+++ b/arch/powerpc/kernel/entry_64.S
@@ -537,6 +537,7 @@ flush_count_cache:
 	/* Save LR into r9 */
 	mflr	r9
 
+	// Flush the link stack
 	.rept 64
 	bl	.+4
 	.endr
@@ -546,6 +547,11 @@ flush_count_cache:
 	.balign 32
 	/* Restore LR */
 1:	mtlr	r9
+
+	// If we're just flushing the link stack, return here
+3:	nop
+	patch_site 3b patch__flush_link_stack_return
+
 	li	r9,0x7fff
 	mtctr	r9
 
diff --git a/arch/powerpc/kernel/security.c b/arch/powerpc/kernel/security.c
index 7cfcb294b11c..a3138e7d71bb 100644
--- a/arch/powerpc/kernel/security.c
+++ b/arch/powerpc/kernel/security.c
@@ -24,6 +24,7 @@ enum count_cache_flush_type {
 	COUNT_CACHE_FLUSH_HW	= 0x4,
 };
 static enum count_cache_flush_type count_cache_flush_type = COUNT_CACHE_FLUSH_NONE;
+static bool link_stack_flush_enabled;
 
 bool barrier_nospec_enabled;
 static bool no_nospec;
@@ -212,11 +213,19 @@ ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, c
 
 		if (ccd)
 			seq_buf_printf(&s, "Indirect branch cache disabled");
+
+		if (link_stack_flush_enabled)
+			seq_buf_printf(&s, ", Software link stack flush");
+
 	} else if (count_cache_flush_type != COUNT_CACHE_FLUSH_NONE) {
 		seq_buf_printf(&s, "Mitigation: Software count cache flush");
 
 		if (count_cache_flush_type == COUNT_CACHE_FLUSH_HW)
 			seq_buf_printf(&s, " (hardware accelerated)");
+
+		if (link_stack_flush_enabled)
+			seq_buf_printf(&s, ", Software link stack flush");
+
 	} else if (btb_flush_enabled) {
 		seq_buf_printf(&s, "Mitigation: Branch predictor state flush");
 	} else {
@@ -377,18 +386,40 @@ static __init int stf_barrier_debugfs_init(void)
 device_initcall(stf_barrier_debugfs_init);
 #endif /* CONFIG_DEBUG_FS */
 
+static void no_count_cache_flush(void)
+{
+	count_cache_flush_type = COUNT_CACHE_FLUSH_NONE;
+	pr_info("count-cache-flush: software flush disabled.\n");
+}
+
 static void toggle_count_cache_flush(bool enable)
 {
-	if (!enable || !security_ftr_enabled(SEC_FTR_FLUSH_COUNT_CACHE)) {
+	if (!security_ftr_enabled(SEC_FTR_FLUSH_COUNT_CACHE) &&
+	    !security_ftr_enabled(SEC_FTR_FLUSH_LINK_STACK))
+		enable = false;
+
+	if (!enable) {
 		patch_instruction_site(&patch__call_flush_count_cache, PPC_INST_NOP);
-		count_cache_flush_type = COUNT_CACHE_FLUSH_NONE;
-		pr_info("count-cache-flush: software flush disabled.\n");
+		pr_info("link-stack-flush: software flush disabled.\n");
+		link_stack_flush_enabled = false;
+		no_count_cache_flush();
 		return;
 	}
 
+	// This enables the branch from _switch to flush_count_cache
 	patch_branch_site(&patch__call_flush_count_cache,
 			  (u64)&flush_count_cache, BRANCH_SET_LINK);
 
+	pr_info("link-stack-flush: software flush enabled.\n");
+	link_stack_flush_enabled = true;
+
+	// If we just need to flush the link stack, patch an early return
+	if (!security_ftr_enabled(SEC_FTR_FLUSH_COUNT_CACHE)) {
+		patch_instruction_site(&patch__flush_link_stack_return, PPC_INST_BLR);
+		no_count_cache_flush();
+		return;
+	}
+
 	if (!security_ftr_enabled(SEC_FTR_BCCTR_FLUSH_ASSIST)) {
 		count_cache_flush_type = COUNT_CACHE_FLUSH_SW;
 		pr_info("count-cache-flush: full software flush sequence enabled.\n");
@@ -407,11 +438,20 @@ void setup_count_cache_flush(void)
 	if (no_spectrev2 || cpu_mitigations_off()) {
 		if (security_ftr_enabled(SEC_FTR_BCCTRL_SERIALISED) ||
 		    security_ftr_enabled(SEC_FTR_COUNT_CACHE_DISABLED))
-			pr_warn("Spectre v2 mitigations not under software control, can't disable\n");
+			pr_warn("Spectre v2 mitigations not fully under software control, can't disable\n");
 
 		enable = false;
 	}
 
+	/*
+	 * There's no firmware feature flag/hypervisor bit to tell us we need to
+	 * flush the link stack on context switch. So we set it here if we see
+	 * either of the Spectre v2 mitigations that aim to protect userspace.
+	 */
+	if (security_ftr_enabled(SEC_FTR_COUNT_CACHE_DISABLED) ||
+	    security_ftr_enabled(SEC_FTR_FLUSH_COUNT_CACHE))
+		security_ftr_set(SEC_FTR_FLUSH_LINK_STACK);
+
 	toggle_count_cache_flush(enable);
 }