Skip to contentSkip to conversation

x86/speculation: Clarify Spectre-v2 mitigation when STIBP/IBPB features are unsupported

RFC PATCH In review Vitaly Kuznetsov 3 participants 5 messages Jan 21, 2020

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 x86/speculation: Clarify Spectre-v2 mitigation when STIBP/IBPB features are unsupported

When STIBP/IBPB features are not supported (no microcode update, AWS/Azure/... instances deliberately hiding SPEC_CTRL for performance reasons,...) /sys/devices/system/cpu/vulnerabilities/spectre_v2 looks like

Mitigation: Full generic retpoline, STIBP: disabled, RSB filling

and this looks imperfect. In particular, STIBP is 'disabled' and 'IBPB' is not mentioned while both features are just not supported. Also, for STIBP the 'disabled' state (SPECTRE_V2_USER_NONE) can represent both the absence of hardware support and deliberate user's choice (spectre_v2_user=off)

Make the following adjustments:
- Output 'unsupported' for both STIBP/IBPB when there's no support in hardware.
- Output 'unneeded' for STIBP when SMT is disabled/missing (and this switch_to_cond_stibp is off).

RFC. Some tools out there may be looking at this information so by changing the output we're breaking them. Also, it may make sense to separate kernel and userspace protections and switch to something like

Mitigation: Kernel: Full generic retpoline, RSB filling; Userspace:
Vulnerable

for the above mentioned case.

---
 Documentation/admin-guide/hw-vuln/spectre.rst | 3 +++
 arch/x86/kernel/cpu/bugs.c                    | 9 +++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)
Documentation/admin-guide/hw-vuln/spectre.rst +3 -0
index e05e581af5cf..2b8a42d0c57b 100644
@@ -385,6 +385,7 @@ The possible values in this file are:
     an x86 only feature. For more details see below.
 
   ===================   ========================================================
+  'IBPB: unsupported'   IBPB is not supported by hardware
   'IBPB: disabled'      IBPB unused
   'IBPB: always-on'     Use IBPB on all tasks
   'IBPB: conditional'   Use IBPB on SECCOMP or indirect branch restricted tasks
@@ -396,6 +397,8 @@ The possible values in this file are:
     only feature. For more details see below.
 
   ====================  ========================================================
+  'STIBP: unsupported'  STIBP is not supported by hardware
+  'STIBP: unneeded'     STIBP is not needed because SMT is disabled
   'STIBP: disabled'     STIBP unused
   'STIBP: forced'       Use STIBP on all tasks
   'STIBP: conditional'  Use STIBP on SECCOMP or indirect branch restricted tasks
arch/x86/kernel/cpu/bugs.c +7 -2
index 8bf64899f56a..d72a36fe042b 100644
@@ -1496,7 +1496,10 @@ static char *stibp_state(void)
 
 	switch (spectre_v2_user) {
 	case SPECTRE_V2_USER_NONE:
-		return ", STIBP: disabled";
+		if (boot_cpu_has(X86_FEATURE_STIBP))
+			return ", STIBP: disabled";
+		else
+			return ", STIBP: unsupported";
 	case SPECTRE_V2_USER_STRICT:
 		return ", STIBP: forced";
 	case SPECTRE_V2_USER_STRICT_PREFERRED:
@@ -1505,6 +1508,8 @@ static char *stibp_state(void)
 	case SPECTRE_V2_USER_SECCOMP:
 		if (static_key_enabled(&switch_to_cond_stibp))
 			return ", STIBP: conditional";
+		else
+			return ", STIBP: unneeded";
 	}
 	return "";
 }
@@ -1518,7 +1523,7 @@ static char *ibpb_state(void)
 			return ", IBPB: conditional";
 		return ", IBPB: disabled";
 	}
-	return "";
+	return ", IBPB: unsupported";
 }
 
 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
Signature · 1 line
2.24.1
maintainer
#1 lore raw
Quoted 25 lines · Vitaly Kuznetsov

When STIBP/IBPB features are not supported (no microcode update, AWS/Azure/... instances deliberately hiding SPEC_CTRL for performance reasons,...) /sys/devices/system/cpu/vulnerabilities/spectre_v2 looks like

Mitigation: Full generic retpoline, STIBP: disabled, RSB filling

and this looks imperfect. In particular, STIBP is 'disabled' and 'IBPB' is not mentioned while both features are just not supported. Also, for STIBP the 'disabled' state (SPECTRE_V2_USER_NONE) can represent both the absence of hardware support and deliberate user's choice (spectre_v2_user=off)

Make the following adjustments:
- Output 'unsupported' for both STIBP/IBPB when there's no support in hardware.
- Output 'unneeded' for STIBP when SMT is disabled/missing (and this switch_to_cond_stibp is off).

RFC. Some tools out there may be looking at this information so by changing the output we're breaking them. Also, it may make sense to separate kernel and userspace protections and switch to something like

Mitigation: Kernel: Full generic retpoline, RSB filling; Userspace:
Vulnerable

Quoted 7 lines

for the above mentioned case.

---
 Documentation/admin-guide/hw-vuln/spectre.rst | 3 +++
 arch/x86/kernel/cpu/bugs.c                    | 9 +++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)

There's another attempt to fix similar aspects of this whole deal going on ATM:

https://lkml.kernel.org/r/[email protected]

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#2 lore raw
Quoted 34 lines · Borislav Petkov
Quoted 25 lines · Vitaly Kuznetsov

When STIBP/IBPB features are not supported (no microcode update, AWS/Azure/... instances deliberately hiding SPEC_CTRL for performance reasons,...) /sys/devices/system/cpu/vulnerabilities/spectre_v2 looks like

Mitigation: Full generic retpoline, STIBP: disabled, RSB filling

and this looks imperfect. In particular, STIBP is 'disabled' and 'IBPB' is not mentioned while both features are just not supported. Also, for STIBP the 'disabled' state (SPECTRE_V2_USER_NONE) can represent both the absence of hardware support and deliberate user's choice (spectre_v2_user=off)

Make the following adjustments:
- Output 'unsupported' for both STIBP/IBPB when there's no support in hardware.
- Output 'unneeded' for STIBP when SMT is disabled/missing (and this switch_to_cond_stibp is off).

RFC. Some tools out there may be looking at this information so by changing the output we're breaking them. Also, it may make sense to separate kernel and userspace protections and switch to something like

Mitigation: Kernel: Full generic retpoline, RSB filling; Userspace:
Vulnerable

Quoted 7 lines

for the above mentioned case.

---
 Documentation/admin-guide/hw-vuln/spectre.rst | 3 +++
 arch/x86/kernel/cpu/bugs.c                    | 9 +++++++--
 2 files changed, 10 insertions(+), 2 deletions(-)
Quoted 4 lines

There's another attempt to fix similar aspects of this whole deal going on ATM:

https://lkml.kernel.org/r/[email protected]

Missed that, thanks! (Cc: Anthony)

This patch seem to address my STIBP: disabled/unsupported concern but not 'unneeded'. And not IBPB.

Signature · 1 line
Vitaly
#3 lore raw
In reply to Vitaly Kuznetsov
Quoted 12 lines · Vitaly Kuznetsov

When STIBP/IBPB features are not supported (no microcode update, AWS/Azure/... instances deliberately hiding SPEC_CTRL for performance reasons,...) /sys/devices/system/cpu/vulnerabilities/spectre_v2 looks like

Mitigation: Full generic retpoline, STIBP: disabled, RSB filling

and this looks imperfect. In particular, STIBP is 'disabled' and 'IBPB' is not mentioned while both features are just not supported. Also, for STIBP the 'disabled' state (SPECTRE_V2_USER_NONE) can represent both the absence of hardware support and deliberate user's choice (spectre_v2_user=off)

Quoted 5 lines

Make the following adjustments:
- Output 'unsupported' for both STIBP/IBPB when there's no support in hardware.
- Output 'unneeded' for STIBP when SMT is disabled/missing (and this switch_to_cond_stibp is off).

I support outputting "unsupported" when the microcode doesn't support it. However, I am not sure if "unneeded" is really necessary or not. STIBP is not needed when SMT is disabled or when Enhanced IBRS is available and used. Your patch handles the first case, but not the second. I think it may be easier to just leave it out in case it is not needed.

Cheers,
Longman

#4 lore raw
Quoted 19 lines · Waiman Long
Quoted 12 lines · Vitaly Kuznetsov

When STIBP/IBPB features are not supported (no microcode update, AWS/Azure/... instances deliberately hiding SPEC_CTRL for performance reasons,...) /sys/devices/system/cpu/vulnerabilities/spectre_v2 looks like

Mitigation: Full generic retpoline, STIBP: disabled, RSB filling

and this looks imperfect. In particular, STIBP is 'disabled' and 'IBPB' is not mentioned while both features are just not supported. Also, for STIBP the 'disabled' state (SPECTRE_V2_USER_NONE) can represent both the absence of hardware support and deliberate user's choice (spectre_v2_user=off)

Quoted 5 lines

Make the following adjustments:
- Output 'unsupported' for both STIBP/IBPB when there's no support in hardware.
- Output 'unneeded' for STIBP when SMT is disabled/missing (and this switch_to_cond_stibp is off).

Quoted 6 lines

I support outputting "unsupported" when the microcode doesn't support it. However, I am not sure if "unneeded" is really necessary or not. STIBP is not needed when SMT is disabled or when Enhanced IBRS is available and used. Your patch handles the first case, but not the second. I think it may be easier to just leave it out in case it is not needed.

Makes sense. Or, alternatively, we can output 'unneeded' in both cases to make things explicit and to distinguish it from the current state of IBPB where missing means 'unsupported by hardware'.

Signature · 1 line
Vitaly