diff options
author | Jessica Clarke <jrtc27@FreeBSD.org> | 2023-06-07 14:21:18 +0000 |
---|---|---|
committer | Jessica Clarke <jrtc27@FreeBSD.org> | 2023-06-07 14:24:29 +0000 |
commit | 21f7397a61f7bff61a1221cc6340cd980a922540 (patch) | |
tree | 1b2960e78f22a93c849fe84ff6449dde4e355ec4 | |
parent | 296a0987be590e73784e4c313d28a61ff310f1a3 (diff) | |
download | src-21f7397a61f7bff61a1221cc6340cd980a922540.tar.gz src-21f7397a61f7bff61a1221cc6340cd980a922540.zip |
libpmc: Handle PMCALLOCATE log with PMC code on PMU event system
On an arm64 system that reports as a Cortex A72 r0p3, running
pmcstat -P CPU_CYCLES command
works, but
pmcstat -P cpu-cycles command
does not. This is because the former uses the PMU event from the JSON
source, resulting in pl_event in the log event being a small index
(here, 5) into the generated events table, whilst the latter does not
match any of the JSON events and falls back on PMC's own tables, mapping
it to the PMC event 0x14111, i.e. PMC_EV_ARMV8_EVENT_11H. Then, when
libpmc gets the PMCALLOCATE event, it tries to use the event as an index
into the JSON-derived table, but doing so only makes sense for the
former, whilst for the latter it will go way out of bounds and either
read junk (which may trigger the != NULL assertion) or segfault. As far
as I can tell we don't have anything lying around to tell us which of
the two cases we're in, but we can exploit the fact that the first
0x1000 PMC event codes are reserved, and that none of our PMU events
tables reach that number of entries yet.
PR: 268857
Reviewed by: mhorne
MFC after: 1 month
Differential Revision: https://reviews.freebsd.org/D39592
-rw-r--r-- | lib/libpmc/libpmc.c | 9 | ||||
-rw-r--r-- | lib/libpmc/pmclog.c | 27 |
2 files changed, 29 insertions, 7 deletions
diff --git a/lib/libpmc/libpmc.c b/lib/libpmc/libpmc.c index 57b9c90de256..003d9615a44a 100644 --- a/lib/libpmc/libpmc.c +++ b/lib/libpmc/libpmc.c @@ -35,6 +35,7 @@ __FBSDID("$FreeBSD$"); #include <sys/pmc.h> #include <sys/syscall.h> +#include <assert.h> #include <ctype.h> #include <errno.h> #include <err.h> @@ -1084,8 +1085,14 @@ pmc_allocate(const char *ctrspec, enum pmc_mode mode, r = spec_copy = strdup(ctrspec); ctrname = strsep(&r, ","); if (pmc_pmu_enabled()) { - if (pmc_pmu_pmcallocate(ctrname, &pmc_config) == 0) + if (pmc_pmu_pmcallocate(ctrname, &pmc_config) == 0) { + /* + * XXX: pmclog_get_event exploits this to disambiguate + * PMU from PMC event codes in PMCALLOCATE events. + */ + assert(pmc_config.pm_ev < PMC_EVENT_FIRST); goto found; + } /* Otherwise, reset any changes */ pmc_config.pm_ev = 0; diff --git a/lib/libpmc/pmclog.c b/lib/libpmc/pmclog.c index 0db91cf51bc2..92bcb1c2b161 100644 --- a/lib/libpmc/pmclog.c +++ b/lib/libpmc/pmclog.c @@ -356,12 +356,27 @@ pmclog_get_event(void *cookie, char **data, ssize_t *len, PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags); PMCLOG_READ32(le,noop); PMCLOG_READ64(le,ev->pl_u.pl_a.pl_rate); - ev->pl_u.pl_a.pl_evname = pmc_pmu_event_get_by_idx(ps->ps_cpuid, ev->pl_u.pl_a.pl_event); - if (ev->pl_u.pl_a.pl_evname != NULL) - break; - else if ((ev->pl_u.pl_a.pl_evname = - _pmc_name_of_event(ev->pl_u.pl_a.pl_event, ps->ps_arch)) - == NULL) { + + /* + * Could be either a PMC event code or a PMU event index; + * assume that their encodings don't overlap (i.e. no PMU event + * table is more than 0x1000 entries) to distinguish them here. + * Otherwise pmc_pmu_event_get_by_idx will go out of bounds if + * given a PMC event code when it knows about that CPU. + * + * XXX: Ideally we'd have user flags to give us that context. + */ + if (ev->pl_u.pl_a.pl_event < PMC_EVENT_FIRST) + ev->pl_u.pl_a.pl_evname = + pmc_pmu_event_get_by_idx(ps->ps_cpuid, + ev->pl_u.pl_a.pl_event); + else if (ev->pl_u.pl_a.pl_event <= PMC_EVENT_LAST) + ev->pl_u.pl_a.pl_evname = + _pmc_name_of_event(ev->pl_u.pl_a.pl_event, + ps->ps_arch); + else + ev->pl_u.pl_a.pl_evname = NULL; + if (ev->pl_u.pl_a.pl_evname == NULL) { printf("unknown event\n"); goto error; } |