The rdpmc instruction allows reading performance counters directly
from usersapce. Prior to Linux 4.0 any process could use this
instruction when a perf event was running, even if the process itself
did not have any open. The following changesets changed the default
behavior so that only processes with active events can use rdpmc.
Note this change broke the ABI. Previously:
/sys/bus/event_source/devices/cpu/rdpmc
Set to "1" meant allow across whole system.
After the change "2" means the whole system, and "1" means per-process.
Probably a better change would have been to add "2" to mean per-process
and make that the default setting. Probably too late to fix that now.
commit a66734297f78707ce39d756b656bfae861d53f62
Author: Andy Lutomirski <[email protected]>
perf/x86: Add /sys/devices/cpu/rdpmc=2 to allow rdpmc for all tasks
commit 7911d3f7af14a614617e38245fedf98a724e46a9
Author: Andy Lutomirski <[email protected]>
perf/x86: Only allow rdpmc if a perf_event is mapped
Signed-off-by: Andy Lutomirski <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Andrea Arcangeli <[email protected]>
Cc: Vince Weaver <[email protected]>
Cc: "hillf.zj" <[email protected]>
Cc: Valdis Kletnieks <[email protected]>
Cc: Linus Torvalds <[email protected]>
Link: http://lkml.kernel.org/r/caac3c1c707dcca48ecbc35f4def21495856f479.1414190806.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <[email protected]>
Signed-off-by: Vince Weaver <[email protected]>
diff --git a/man2/perf_event_open.2 b/man2/perf_event_open.2
index 01ee579..c854d21 100644
--- a/man2/perf_event_open.2
+++ b/man2/perf_event_open.2
@@ -2377,6 +2377,16 @@ Support for this can be detected with the
.I cap_usr_rdpmc
field in the mmap page; documentation on how
to calculate event values can be found in that section.
+
+Originally when rdpmc support was enabled, any process (not just ones
+with an active perf event) could use the rdpmc instruction to access
+the counters.
+Starting with Linux 4.0
+.\" 7911d3f7af14a614617e38245fedf98a724e46a9
+rdpmc support is only enabled if an event is currently enabled
+in a process' context.
+To restore the old behavior, write the value 2 to
+.IR /sys/devices/cpu/rdpmc .
.SS perf_event ioctl calls
.PP
Various ioctls act on
@@ -2552,11 +2562,18 @@ field of
.I perf_event_attr
to indicate that you wish to use this PMU.
.TP
-.IR /sys/bus/event_source/devices/*/rdpmc " (since Linux 3.4)"
+.IR /sys/bus/event_source/devices/cpu/rdpmc " (since Linux 3.4)"
.\" commit 0c9d42ed4cee2aa1dfc3a260b741baae8615744f
If this file is 1, then direct user-space access to the
performance counter registers is allowed via the rdpmc instruction.
This can be disabled by echoing 0 to the file.
+
+As of Linux 4.0
+.\" a66734297f78707ce39d756b656bfae861d53f62
+.\" 7911d3f7af14a614617e38245fedf98a724e46a9
+the behavior has changed, so that 1 now means only allow access
+to processes with active perf events, with 2 indicating the old
+allow-anyone-access behavior.
.TP
.IR /sys/bus/event_source/devices/*/format/ " (since Linux 3.4)"
.\" commit 641cc938815dfd09f8fa1ec72deb814f0938ac33
On 04/16/2015 11:20 AM, Vince Weaver wrote:
>
> The rdpmc instruction allows reading performance counters directly
> from usersapce. Prior to Linux 4.0 any process could use this
> instruction when a perf event was running, even if the process itself
> did not have any open. The following changesets changed the default
> behavior so that only processes with active events can use rdpmc.
>
> Note this change broke the ABI. Previously:
> /sys/bus/event_source/devices/cpu/rdpmc
> Set to "1" meant allow across whole system.
>
> After the change "2" means the whole system, and "1" means per-process.
>
> Probably a better change would have been to add "2" to mean per-process
> and make that the default setting. Probably too late to fix that now.
Good point. I wish you'd thought of that sooner :(
--Andy
>
> commit a66734297f78707ce39d756b656bfae861d53f62
> Author: Andy Lutomirski <[email protected]>
>
> perf/x86: Add /sys/devices/cpu/rdpmc=2 to allow rdpmc for all tasks
>
> commit 7911d3f7af14a614617e38245fedf98a724e46a9
> Author: Andy Lutomirski <[email protected]>
>
> perf/x86: Only allow rdpmc if a perf_event is mapped
>
> Signed-off-by: Andy Lutomirski <[email protected]>
> Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> Cc: Paul Mackerras <[email protected]>
> Cc: Arnaldo Carvalho de Melo <[email protected]>
> Cc: Kees Cook <[email protected]>
> Cc: Andrea Arcangeli <[email protected]>
> Cc: Vince Weaver <[email protected]>
> Cc: "hillf.zj" <[email protected]>
> Cc: Valdis Kletnieks <[email protected]>
> Cc: Linus Torvalds <torvalds-de/[email protected]>
> Link: http://lkml.kernel.org/r/caac3c1c707dcca48ecbc35f4def21495856f479.1414190806.git.luto-kltTT9wpgjJwATOyAt5JVQ@public.gmane.org
> Signed-off-by: Ingo Molnar <[email protected]>
>
> Signed-off-by: Vince Weaver <[email protected]>
>
> diff --git a/man2/perf_event_open.2 b/man2/perf_event_open.2
> index 01ee579..c854d21 100644
> --- a/man2/perf_event_open.2
> +++ b/man2/perf_event_open.2
> @@ -2377,6 +2377,16 @@ Support for this can be detected with the
> .I cap_usr_rdpmc
> field in the mmap page; documentation on how
> to calculate event values can be found in that section.
> +
> +Originally when rdpmc support was enabled, any process (not just ones
> +with an active perf event) could use the rdpmc instruction to access
> +the counters.
> +Starting with Linux 4.0
> +.\" 7911d3f7af14a614617e38245fedf98a724e46a9
> +rdpmc support is only enabled if an event is currently enabled
> +in a process' context.
> +To restore the old behavior, write the value 2 to
> +.IR /sys/devices/cpu/rdpmc .
> .SS perf_event ioctl calls
> .PP
> Various ioctls act on
> @@ -2552,11 +2562,18 @@ field of
> .I perf_event_attr
> to indicate that you wish to use this PMU.
> .TP
> -.IR /sys/bus/event_source/devices/*/rdpmc " (since Linux 3.4)"
> +.IR /sys/bus/event_source/devices/cpu/rdpmc " (since Linux 3.4)"
> .\" commit 0c9d42ed4cee2aa1dfc3a260b741baae8615744f
> If this file is 1, then direct user-space access to the
> performance counter registers is allowed via the rdpmc instruction.
> This can be disabled by echoing 0 to the file.
> +
> +As of Linux 4.0
> +.\" a66734297f78707ce39d756b656bfae861d53f62
> +.\" 7911d3f7af14a614617e38245fedf98a724e46a9
> +the behavior has changed, so that 1 now means only allow access
> +to processes with active perf events, with 2 indicating the old
> +allow-anyone-access behavior.
> .TP
> .IR /sys/bus/event_source/devices/*/format/ " (since Linux 3.4)"
> .\" commit 641cc938815dfd09f8fa1ec72deb814f0938ac33
> --
> To unsubscribe from this list: send the line "unsubscribe linux-man" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
On Fri, 17 Apr 2015, Andy Lutomirski wrote:
> On 04/16/2015 11:20 AM, Vince Weaver wrote:
> > Probably a better change would have been to add "2" to mean per-process
> > and make that the default setting. Probably too late to fix that now.
>
> Good point. I wish you'd thought of that sooner :(
Well it's been a busy semester and so I've had less time than normal to
play perf_event ABI police.
I also was surprised this particular patch managed to get into 4.0 so
quickly; I thought it was going to be a 4.1 change when I added it to my
"changesets to look at more closely" list.
In any case most people won't notice this change. It will be a minor
inconvenience for HPC people doing complicated homebrew self monitoring
(especially if RHEL backports the patch and then suddenly things break and
a simple "well are you running 4.0 or newer" is meaningless). Using
the value in /sys/devices/cpu/rdpmc to help debug the problem wouldn't
have worked anyway because it looks like non-root can't do that.
Vince