diff options
| author | Ali Mashtizadeh <ali@mashtizadeh.com> | 2026-04-07 23:33:03 +0000 |
|---|---|---|
| committer | Mitchell Horne <mhorne@FreeBSD.org> | 2026-04-09 16:35:26 +0000 |
| commit | f86ba3eec5961c0217b5458663e0eae5bbad314e (patch) | |
| tree | 4f8b0dabd553f344c91401ba95e88bcce324b339 | |
| parent | 7a1aaca06c3384f90202dafa60440081d67d00fd (diff) | |
hwpmc: prevent IBS fetch from getting stuck
Both fetch and op IBS sampling have the same problem where we need to
rewrite the control MSR to ensure sampling continues at the correct
rate. I also like this because it resets the counter reducing the
chances that we collect a sample inside the NMI handler.
Reported by: Aalok Agarwal
Reviewed by: mhorne
Sponsored by: Netflix
Fixes: e51ef8ae490f ("hwpmc: Initial support for AMD IBS")
Pull Request: https://github.com/freebsd/freebsd-src/pull/2130
| -rw-r--r-- | sys/dev/hwpmc/hwpmc_ibs.c | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/sys/dev/hwpmc/hwpmc_ibs.c b/sys/dev/hwpmc/hwpmc_ibs.c index ec1afcb47666..a230288f157e 100644 --- a/sys/dev/hwpmc/hwpmc_ibs.c +++ b/sys/dev/hwpmc/hwpmc_ibs.c @@ -342,6 +342,8 @@ pmc_ibs_process_fetch(struct pmc *pm, struct trapframe *tf, uint64_t config) } pmc_process_interrupt_mp(PMC_HR, pm, tf, &mpd); + + wrmsr(IBS_FETCH_CTL, pm->pm_md.pm_ibs.ibs_ctl | IBS_FETCH_CTL_ENABLE); } static void |