2020-10-30 06:30:23

by Anand K. Mistry

Subject: [PATCH] proc: Provide details on indirect branch speculation

Similar to speculation store bypass, show information about the indirect
branch speculation mode of a task in /proc/$pid/status.

Signed-off-by: Anand K Mistry <[email protected]>
---

Documentation/filesystems/proc.rst | 2 ++
fs/proc/array.c | 28 ++++++++++++++++++++++++++++
2 files changed, 30 insertions(+)

diff --git a/Documentation/filesystems/proc.rst b/Documentation/filesystems/proc.rst
index 533c79e8d2cd..710dd69614b9 100644
--- a/Documentation/filesystems/proc.rst
+++ b/Documentation/filesystems/proc.rst
@@ -210,6 +210,7 @@ read the file /proc/PID/status::
NoNewPrivs: 0
Seccomp: 0
Speculation_Store_Bypass: thread vulnerable
+ Speculation_Indirect_Branch: conditional enabled
voluntary_ctxt_switches: 0
nonvoluntary_ctxt_switches: 1

@@ -292,6 +293,7 @@ It's slow but very precise.
NoNewPrivs no_new_privs, like prctl(PR_GET_NO_NEW_PRIV, ...)
Seccomp seccomp mode, like prctl(PR_GET_SECCOMP, ...)
Speculation_Store_Bypass speculative store bypass mitigation status
+ Speculation_Indirect_Branch indirect branch speculation mode
Cpus_allowed mask of CPUs on which this process may run
Cpus_allowed_list Same as previous, but in "list format"
Mems_allowed mask of memory nodes allowed to this process
diff --git a/fs/proc/array.c b/fs/proc/array.c
index 65ec2029fa80..ce4fa948c9dd 100644
--- a/fs/proc/array.c
+++ b/fs/proc/array.c
@@ -368,6 +368,34 @@ static inline void task_seccomp(struct seq_file *m, struct task_struct *p)
seq_puts(m, "vulnerable");
break;
}
+
+ seq_puts(m, "\nSpeculation_Indirect_Branch:\t");
+ switch (arch_prctl_spec_ctrl_get(p, PR_SPEC_INDIRECT_BRANCH)) {
+ case -EINVAL:
+ seq_puts(m, "unsupported");
+ break;
+ case PR_SPEC_NOT_AFFECTED:
+ seq_puts(m, "not affected");
+ break;
+ case PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE:
+ seq_puts(m, "conditional force disabled");
+ break;
+ case PR_SPEC_PRCTL | PR_SPEC_DISABLE:
+ seq_puts(m, "conditional disabled");
+ break;
+ case PR_SPEC_PRCTL | PR_SPEC_ENABLE:
+ seq_puts(m, "conditional enabled");
+ break;
+ case PR_SPEC_ENABLE:
+ seq_puts(m, "always enabled");
+ break;
+ case PR_SPEC_DISABLE:
+ seq_puts(m, "always disabled");
+ break;
+ default:
+ seq_puts(m, "unknown");
+ break;
+ }
seq_putc(m, '\n');
}

--
2.29.1.341.ge80a0c044ae-goog

2020-10-31 20:08:39

by Andrew Morton

Subject: Re: [PATCH] proc: Provide details on indirect branch speculation

On Fri, 30 Oct 2020 17:27:54 +1100 Anand K Mistry <[email protected]> wrote:

> Similar to speculation store bypass, show information about the indirect
> branch speculation mode of a task in /proc/$pid/status.

Why is this considered useful?

2020-11-05 06:40:40

by Anand K. Mistry

Subject: Re: [PATCH] proc: Provide details on indirect branch speculation

On Sun, 1 Nov 2020 at 07:05, Andrew Morton <[email protected]> wrote:
>
> On Fri, 30 Oct 2020 17:27:54 +1100 Anand K Mistry <[email protected]> wrote:
>
> > Similar to speculation store bypass, show information about the indirect
> > branch speculation mode of a task in /proc/$pid/status.
>
> Why is this considered useful?

For testing/benchmarking, I needed to see whether IB (Indirect Branch)
speculation (see Spectre-v2) is enabled on a task, to see whether an
IBPB instruction should be executed on an address space switch.
Unfortunately, this information isn't available anywhere else and
currently the only way to get it is to hack the kernel to expose it
(like this change). It also helped expose a bug with conditional IB
speculation on certain CPUs.

Another place this could be useful is to audit the system when using
sanboxing. With this change, I can confirm that seccomp-enabled
process have IB speculation force disabled as expected when the kernel
command line parameter `spectre_v2_user=seccomp`.

Since there's already a 'Speculation_Store_Bypass' field, I used that
as precedence for adding this one.

--
Anand K. Mistry
Software Engineer
Google Australia