x86/speculation: Fill RSB on vmexit for IBRS

Prevent RSB underflow/poisoning attacks with RSB.  While at it, add a
bunch of comments to attempt to document the current state of tribal
knowledge about RSB attacks and what exactly is being mitigated.

Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Borislav Petkov <bp@suse.de>

arch/x86/include/asm/cpufeatures.h

@@ -204,7 +204,7 @@
 #define X86_FEATURE_XCOMPACTED		( 7*32+10) /* "" Use compacted XSTATE (XSAVES or XSAVEC) */
 #define X86_FEATURE_PTI			( 7*32+11) /* Kernel Page Table Isolation enabled */
 #define X86_FEATURE_KERNEL_IBRS		( 7*32+12) /* "" Set/clear IBRS on kernel entry/exit */
-/* FREE!				( 7*32+13) */
+#define X86_FEATURE_RSB_VMEXIT		( 7*32+13) /* "" Fill RSB on VM-Exit */
 #define X86_FEATURE_INTEL_PPIN		( 7*32+14) /* Intel Processor Inventory Number */
 #define X86_FEATURE_CDP_L2		( 7*32+15) /* Code and Data Prioritization L2 */
 #define X86_FEATURE_MSR_SPEC_CTRL	( 7*32+16) /* "" MSR SPEC_CTRL is implemented */

arch/x86/kernel/cpu/bugs.c

@@ -1401,16 +1401,69 @@ static void __init spectre_v2_select_mitigation(void)
 	pr_info("%s\n", spectre_v2_strings[mode]);
 
 	/*
-	 * If spectre v2 protection has been enabled, unconditionally fill
-	 * RSB during a context switch; this protects against two independent
-	 * issues:
+	 * If Spectre v2 protection has been enabled, fill the RSB during a
+	 * context switch.  In general there are two types of RSB attacks
+	 * across context switches, for which the CALLs/RETs may be unbalanced.
 	 *
-	 *	- RSB underflow (and switch to BTB) on Skylake+
-	 *	- SpectreRSB variant of spectre v2 on X86_BUG_SPECTRE_V2 CPUs
+	 * 1) RSB underflow
+	 *
+	 *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
+	 *    speculated return targets may come from the branch predictor,
+	 *    which could have a user-poisoned BTB or BHB entry.
+	 *
+	 *    AMD has it even worse: *all* returns are speculated from the BTB,
+	 *    regardless of the state of the RSB.
+	 *
+	 *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
+	 *    scenario is mitigated by the IBRS branch prediction isolation
+	 *    properties, so the RSB buffer filling wouldn't be necessary to
+	 *    protect against this type of attack.
+	 *
+	 *    The "user -> user" attack scenario is mitigated by RSB filling.
+	 *
+	 * 2) Poisoned RSB entry
+	 *
+	 *    If the 'next' in-kernel return stack is shorter than 'prev',
+	 *    'next' could be tricked into speculating with a user-poisoned RSB
+	 *    entry.
+	 *
+	 *    The "user -> kernel" attack scenario is mitigated by SMEP and
+	 *    eIBRS.
+	 *
+	 *    The "user -> user" scenario, also known as SpectreBHB, requires
+	 *    RSB clearing.
+	 *
+	 * So to mitigate all cases, unconditionally fill RSB on context
+	 * switches.
+	 *
+	 * FIXME: Is this pointless for retbleed-affected AMD?
 	 */
 	setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
 	pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
 
+	/*
+	 * Similar to context switches, there are two types of RSB attacks
+	 * after vmexit:
+	 *
+	 * 1) RSB underflow
+	 *
+	 * 2) Poisoned RSB entry
+	 *
+	 * When retpoline is enabled, both are mitigated by filling/clearing
+	 * the RSB.
+	 *
+	 * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
+	 * prediction isolation protections, RSB still needs to be cleared
+	 * because of #2.  Note that SMEP provides no protection here, unlike
+	 * user-space-poisoned RSB entries.
+	 *
+	 * eIBRS, on the other hand, has RSB-poisoning protections, so it
+	 * doesn't need RSB clearing after vmexit.
+	 */
+	if (boot_cpu_has(X86_FEATURE_RETPOLINE) ||
+	    boot_cpu_has(X86_FEATURE_KERNEL_IBRS))
+		setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
+
 	/*
 	 * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
 	 * and Enhanced IBRS protect firmware too, so enable IBRS around

arch/x86/kvm/vmx/vmenter.S

@@ -194,15 +194,15 @@ SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL)
 	 * IMPORTANT: RSB filling and SPEC_CTRL handling must be done before
 	 * the first unbalanced RET after vmexit!
 	 *
-	 * For retpoline, RSB filling is needed to prevent poisoned RSB entries
-	 * and (in some cases) RSB underflow.
+	 * For retpoline or IBRS, RSB filling is needed to prevent poisoned RSB
+	 * entries and (in some cases) RSB underflow.
 	 *
 	 * eIBRS has its own protection against poisoned RSB, so it doesn't
 	 * need the RSB filling sequence.  But it does need to be enabled
 	 * before the first unbalanced RET.
          */
 
-	FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RETPOLINE
+	FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT
 
 	pop %_ASM_ARG2	/* @flags */
 	pop %_ASM_ARG1	/* @vmx */