Page MenuHomeFreeBSD
Differential D22374 Diff 64393 sys/amd64/amd64/machdep.c

Changeset View

Standalone View

View Options

sys/amd64/amd64/machdep.c

Show First 20 Lines • Show All 1,773 Lines • ▼ Show 20 Lines if (preload_search_info(kmdp, MODINFO_METADATA | MODINFOMD_EFI_MAP)
!= NULL) != NULL)
vty_set_preferred(VTY_VT); vty_set_preferred(VTY_VT);
TUNABLE_INT_FETCH("hw.ibrs_disable", &hw_ibrs_disable); TUNABLE_INT_FETCH("hw.ibrs_disable", &hw_ibrs_disable);
TUNABLE_INT_FETCH("hw.spec_store_bypass_disable", &hw_ssb_disable); TUNABLE_INT_FETCH("hw.spec_store_bypass_disable", &hw_ssb_disable);
TUNABLE_INT_FETCH("machdep.syscall_ret_l1d_flush", TUNABLE_INT_FETCH("machdep.syscall_ret_l1d_flush",
&syscall_ret_l1d_flush_mode); &syscall_ret_l1d_flush_mode);
TUNABLE_INT_FETCH("hw.mds_disable", &hw_mds_disable); TUNABLE_INT_FETCH("hw.mds_disable", &hw_mds_disable);
TUNABLE_INT_FETCH("hw.tsx_disable", &hw_tsx_disable);
kibUnsubmitted
Done
Inline Actions

This fetch must be done early enough so it must occur in hammer_time().

kib: This fetch must be done early enough so it must occur in hammer_time().
emasteUnsubmitted
Not Done
Inline Actions

we should really use positive sense sysctls

emaste: we should really use positive sense sysctls
scottlAuthorUnsubmitted
Done
Inline Actions

Agreed, but I propose that we do a sweep and rename all of the nearby and related knobs. For now, hw.tsx_disable is consistent with the others, which I think it important.

scottl: Agreed, but I propose that we do a sweep and rename all of the nearby and related knobs. For…
emasteUnsubmitted
Not Done
Inline Actions

It's not exactly, because mds_disable is actually enabling a workaround - we could rename it mds_workaround_enable and the values the user sets it to do not change.

If tsx_disable becomes tsx_enable the setting also has to change.

emaste: It's not exactly, because mds_disable is actually enabling a workaround - we could rename it…
scottphUnsubmitted
Not Done
Inline Actions

I think the name here needs to be changed to taa_disable. The intention is to mitigate the TSX Async Abort vulnerability, not necessarily just general purpose TSX feature control.

scottph: I think the name here needs to be changed to `taa_disable`. The intention is to mitigate the…
emasteUnsubmitted
Not Done
Inline Actions

Indeed, but is it not usable as a general TSX disable control? I.e., if in the future another problem is found with TSX we could still use this to turn it off?

emaste: Indeed, but is it not usable as a general TSX disable control? I.e., if in the future another…
scottlAuthorUnsubmitted
Done
Inline Actions

hw.x86.mitigations.taa is my proposal. 0 = no mitigation, 1 = TSX-disable, 2 = VERW, 3 = auto

scottl: `hw.x86.mitigations.taa` is my proposal. 0 = no mitigation, 1 = TSX-disable, 2 = VERW, 3 = auto
finishidentcpu(); /* Final stage of CPU initialization */ finishidentcpu(); /* Final stage of CPU initialization */
initializecpu(); /* Initialize CPU registers */ initializecpu(); /* Initialize CPU registers */
amd64_bsp_ist_init(pc); amd64_bsp_ist_init(pc);
/* Set the IO permission bitmap (empty due to tss seg limit) */ /* Set the IO permission bitmap (empty due to tss seg limit) */
pc->pc_common_tss.tss_iobase = sizeof(struct amd64tss) + pc->pc_common_tss.tss_iobase = sizeof(struct amd64tss) +
▲ Show 20 Lines • Show All 970 Lines • Show Last 20 Lines