On 07.02.2019 2:58, Jonathan Corbet wrote: > On Fri, 1 Feb 2019 10:29:11 +0300 > Alexey Budankov <alexey.budan...@linux.intel.com> wrote: > >> Extend perf-security.rst file with perf_events/Perf resource control >> section describing RLIMIT_NOFILE and perf_event_mlock_kb settings for >> performance monitoring user processes. >> >> Signed-off-by: Alexey Budankov <alexey.budan...@linux.intel.com> > > Overall these patches seem reasonable, though I have some nits to pick. > I'm happy to apply them but wouldn't mind an ack from the perf camp. > > Alexey, could you wrap your paragraphs at 72-75 columns?
Sure, let's have it as the forth patch in the series in order not to mix the actual content with formatting. > >> --- >> Documentation/admin-guide/perf-security.rst | 36 +++++++++++++++++++++ >> 1 file changed, 36 insertions(+) >> >> diff --git a/Documentation/admin-guide/perf-security.rst >> b/Documentation/admin-guide/perf-security.rst >> index f73ebfe9bfe2..ff6832191577 100644 >> --- a/Documentation/admin-guide/perf-security.rst >> +++ b/Documentation/admin-guide/perf-security.rst >> @@ -84,6 +84,40 @@ governed by perf_event_paranoid [2]_ setting: >> locking limit is imposed but ignored for unprivileged processes with >> CAP_IPC_LOCK capability. >> >> +perf_events/Perf resource control >> +--------------------------------- >> + >> +perf_events system call API [2]_ allocates file descriptors for every >> configured > > *The* perf_events system call API Accepted. > >> +PMU event. Open file descriptors are a per-process accountable *resource* >> governed >> +by RLIMIT_NOFILE [11]_ limit (ulimit -n), which is usually derived from the >> login > > by *the* RLIMIT_NOFILE Accepted. > >> +shell process. When configuring Perf collection for a long list of events >> on a >> +large server system, this limit can be easily hit preventing required >> monitoring >> +configuration. RLIMIT_NOFILE limit can be increased on per-user basis >> modifying >> +content of limits.conf file [12]_ on some systems. Ordinary Perf sampling >> session > > of *the* limits.conf file > > Ordinarily, a Perf Accepted. > >> +(perf record) requires an amount of open perf_event file descriptors that >> is not >> +less than a number of monitored events multiplied by a number of monitored >> CPUs. >> + >> +An amount of memory available to user processes for capturing performance >> monitoring >> +data is governed by perf_event_mlock_kb [2]_ setting. This perf_event >> specific > > by *the* perf_event_mlock_kb Accepted. > >> +*resource* setting defines overall per-cpu limits of memory allowed for >> mapping > > Why the *emphasis* here? Avoided emphasis here and in the other places of this paragraphs. > >> +by the user processes to execute performance monitoring. The setting >> essentially >> +extends RLIMIT_MEMLOCK [11]_ limit but only for memory regions mapped >> specially > > extends *the* RMLIMIT_MEMLOCK limit *,* but only Accepted. > >> +for capturing monitored performance events and related data. >> + >> +For example, if a machine has eight cores and perf_event_mlock_kb limit is >> set >> +to 516 KiB then a user process is provided with 516 KiB * 8 = 4128 KiB of >> memory > > Kib, then Comma accepted, Kib = 1024 bits and this is not what is meant here - KiB=1024 Bytes. > >> +above RLIMIT_MEMLOCK limit (ulimit -l) for perf_event mmap buffers. In >> particular > > above *the* RLIMIT_MEMLOCK > > particular, Accepted. > >> +this means that if the user wants to start two or more performance >> monitoring > > that, if Accepted. > >> +processes, it is required to manually distribute available 4128 KiB between >> the > > s/it is/they are/ Not sure what you mean here. I want to say that the users is required to distribute memory among the processes using --mmap-pages. Replaced 'it' with 'the user'. > >> +monitoring processes, for example, using --mmap-pages Perf record mode >> option. > > using *the* --mmap-pages option Accepted. > >> +Otherwise, the first started performance monitoring process allocates all >> available >> +4128 KiB and the other processes will fail to proceed due to the lack of >> memory. >> + >> +RLIMIT_MEMLOCK and perf_event_mlock_kb *resource* constraints are ignored >> for >> +processes with CAP_IPC_LOCK capability. Thus, perf_events/Perf privileged >> users > > with *the* CAP_IPC_LOCK Accepted. > >> +can be provided with memory above the constraints for perf_events/Perf >> performance >> +monitoring purpose by providing the Perf executable with CAP_IPC_LOCK >> capability. >> + >> Bibliography >> ------------ >> >> @@ -94,4 +128,6 @@ Bibliography >> .. [5] `<https://www.kernel.org/doc/html/latest/security/credentials.html>`_ >> .. [6] `<http://man7.org/linux/man-pages/man7/capabilities.7.html>`_ >> .. [7] `<http://man7.org/linux/man-pages/man2/ptrace.2.html>`_ >> +.. [11] `<http://man7.org/linux/man-pages/man2/getrlimit.2.html>`_ >> +.. [12] `<http://man7.org/linux/man-pages/man5/limits.conf.5.html>`_ > > Thanks, > > jon > Thanks, Alexey
