Skip to contentSkip to conversation

x86/speculation: Allow IBPB to be conditionally enabled on CPUs with always-on STIBP

PATCH In review Anand K Mistry 2 participants 3 messages Nov 12, 2020 – Nov 17, 2020

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

PATCH 1/1 x86/speculation: Allow IBPB to be conditionally enabled on CPUs with always-on STIBP

commit 1978b3a53a74e3230cd46932b149c6e62e832e9a upstream.

On AMD CPUs which have the feature X86_FEATURE_AMD_STIBP_ALWAYS_ON, STIBP is set to on and

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

At the same time, IBPB can be set to conditional.

However, this leads to the case where it's impossible to turn on IBPB for a process because in the PR_SPEC_DISABLE case in ib_prctl_set() the

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

condition leads to a return before the task flag is set. Similarly, ib_prctl_get() will return PR_SPEC_DISABLE even though IBPB is set to conditional.

More generally, the following cases are possible:

1. STIBP = conditional && IBPB = on for spectre_v2_user=seccomp,ibpb
2. STIBP = on && IBPB = conditional for AMD CPUs with X86_FEATURE_AMD_STIBP_ALWAYS_ON

The first case functions correctly today, but only because spectre_v2_user_ibpb isn't updated to reflect the IBPB mode.

At a high level, this change does one thing. If either STIBP or IBPB is set to conditional, allow the prctl to change the task flag. Also, reflect that capability when querying the state. This isn't perfect since it doesn't take into account if only STIBP or IBPB is unconditionally on. But it allows the conditional feature to work as expected, without affecting the unconditional one.

[ bp: Massage commit message and comment; space out statements for
better readability. ]

Conflicts:

	arch/x86/kernel/cpu/bugs.c

Superfluous newline which was removed in upstream commit a5ce9f2bb665

---
The one conflict is a newline in a comment which was removed in upstream commit a5ce9f2bb665, but was not merged into the stable trees.

This patch applies cleanly on the stable trees 5.4, 4.19, 4.14, 4.9, and 4.4, which are affected by this bug.

 arch/x86/kernel/cpu/bugs.c | 52 ++++++++++++++++++++++++--------------
 1 file changed, 33 insertions(+), 19 deletions(-)
arch/x86/kernel/cpu/bugs.c +33 -19
index acbf3dbb8bf2..bdc1ed7ff669 100644
@@ -1252,6 +1252,14 @@ static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
 	return 0;
 }
 
+static bool is_spec_ib_user_controlled(void)
+{
+	return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
+		spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
+		spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
+		spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
+}
+
 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
 {
 	switch (ctrl) {
@@ -1259,17 +1267,26 @@ static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
 			return 0;
-		/*
-		 * Indirect branch speculation is always disabled in strict
-		 * mode. It can neither be enabled if it was force-disabled
-		 * by a  previous prctl call.
 
+		/*
+		 * With strict mode for both IBPB and STIBP, the instruction
+		 * code paths avoid checking this task flag and instead,
+		 * unconditionally run the instruction. However, STIBP and IBPB
+		 * are independent and either can be set to conditionally
+		 * enabled regardless of the mode of the other.
+		 *
+		 * If either is set to conditional, allow the task flag to be
+		 * updated, unless it was force-disabled by a previous prctl
+		 * call. Currently, this is possible on an AMD CPU which has the
+		 * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
+		 * kernel is booted with 'spectre_v2_user=seccomp', then
+		 * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
+		 * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
 		 */
-		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
-		    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
-		    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ||
+		if (!is_spec_ib_user_controlled() ||
 		    task_spec_ib_force_disable(task))
 			return -EPERM;
+
 		task_clear_spec_ib_disable(task);
 		task_update_spec_tif(task);
 		break;
@@ -1282,10 +1299,10 @@ static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
 		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
 		    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
 			return -EPERM;
-		if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
-		    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
-		    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
+
+		if (!is_spec_ib_user_controlled())
 			return 0;
+
 		task_set_spec_ib_disable(task);
 		if (ctrl == PR_SPEC_FORCE_DISABLE)
 			task_set_spec_ib_force_disable(task);
@@ -1350,20 +1367,17 @@ static int ib_prctl_get(struct task_struct *task)
 	if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
 	    spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
 		return PR_SPEC_ENABLE;
-	else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
-	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
-	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
-		return PR_SPEC_DISABLE;
-	else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
-	    spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
-	    spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
-	    spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP) {
+	else if (is_spec_ib_user_controlled()) {
 		if (task_spec_ib_force_disable(task))
 			return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
 		if (task_spec_ib_disable(task))
 			return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
 		return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
-	} else
+	} else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
+	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
+	    spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
+		return PR_SPEC_DISABLE;
+	else
 		return PR_SPEC_NOT_AFFECTED;
 }
 
Signature · 1 line
2.29.2.222.g5d2a92d10f8-goog
#1 lore raw
Quoted 44 lines · Anand K Mistry

commit 1978b3a53a74e3230cd46932b149c6e62e832e9a upstream.

On AMD CPUs which have the feature X86_FEATURE_AMD_STIBP_ALWAYS_ON, STIBP is set to on and

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

At the same time, IBPB can be set to conditional.

However, this leads to the case where it's impossible to turn on IBPB for a process because in the PR_SPEC_DISABLE case in ib_prctl_set() the

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

condition leads to a return before the task flag is set. Similarly, ib_prctl_get() will return PR_SPEC_DISABLE even though IBPB is set to conditional.

More generally, the following cases are possible:

1. STIBP = conditional && IBPB = on for spectre_v2_user=seccomp,ibpb
2. STIBP = on && IBPB = conditional for AMD CPUs with X86_FEATURE_AMD_STIBP_ALWAYS_ON

The first case functions correctly today, but only because spectre_v2_user_ibpb isn't updated to reflect the IBPB mode.

At a high level, this change does one thing. If either STIBP or IBPB is set to conditional, allow the prctl to change the task flag. Also, reflect that capability when querying the state. This isn't perfect since it doesn't take into account if only STIBP or IBPB is unconditionally on. But it allows the conditional feature to work as expected, without affecting the unconditional one.

[ bp: Massage commit message and comment; space out statements for
better readability. ]

Quoted 4 lines

Conflicts:

	arch/x86/kernel/cpu/bugs.c

Superfluous newline which was removed in upstream commit a5ce9f2bb665

In the future, No need for the conflicts or this line in the changelog text, that just requires me to edit it out :(

thanks,

greg k-h

#2 lore raw
In reply to Anand K Mistry
Quoted 47 lines · Anand K Mistry

commit 1978b3a53a74e3230cd46932b149c6e62e832e9a upstream.

On AMD CPUs which have the feature X86_FEATURE_AMD_STIBP_ALWAYS_ON, STIBP is set to on and

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

At the same time, IBPB can be set to conditional.

However, this leads to the case where it's impossible to turn on IBPB for a process because in the PR_SPEC_DISABLE case in ib_prctl_set() the

spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED

condition leads to a return before the task flag is set. Similarly, ib_prctl_get() will return PR_SPEC_DISABLE even though IBPB is set to conditional.

More generally, the following cases are possible:

1. STIBP = conditional && IBPB = on for spectre_v2_user=seccomp,ibpb
2. STIBP = on && IBPB = conditional for AMD CPUs with X86_FEATURE_AMD_STIBP_ALWAYS_ON

The first case functions correctly today, but only because spectre_v2_user_ibpb isn't updated to reflect the IBPB mode.

At a high level, this change does one thing. If either STIBP or IBPB is set to conditional, allow the prctl to change the task flag. Also, reflect that capability when querying the state. This isn't perfect since it doesn't take into account if only STIBP or IBPB is unconditionally on. But it allows the conditional feature to work as expected, without affecting the unconditional one.

[ bp: Massage commit message and comment; space out statements for
better readability. ]

Conflicts:

	arch/x86/kernel/cpu/bugs.c
Quoted 8 lines

Superfluous newline which was removed in upstream commit a5ce9f2bb665

---
The one conflict is a newline in a comment which was removed in upstream commit a5ce9f2bb665, but was not merged into the stable trees.

This patch applies cleanly on the stable trees 5.4, 4.19, 4.14, 4.9, and 4.4, which are affected by this bug.

Now queued up, thanks.

greg k-h