gdb: Fix the cache sync in write_instr()
Needs ReviewPublic
Actions

Authored by markj on Wed, Nov 12, 9:32 PM.

Details

Reviewers

andrew

Group Reviewers

bhyve

Summary

Previously this code simply wasn't compiled since it tests arm64
instead of aarch64. It was also wrong since it tried to use the
broadcast variant of "ic", which can't be executed in EL0.

Use builtin_clear_cache() instead. It clears the virtual address
range to point of unification and invalidates the corresponding range in
the icache.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 68583
Build 65466: arc lint + arc unit

Event Timeline

markj created this revision.Wed, Nov 12, 9:32 PM

Herald added a reviewer: bhyve. · View Herald TranscriptWed, Nov 12, 9:32 PM

Herald added subscribers: bcran, rgrimes, emaste, imp. · View Herald Transcript

markj requested review of this revision.Wed, Nov 12, 9:32 PM

Harbormaster completed remote builds in B68583: Diff 166332.Wed, Nov 12, 9:32 PM

__clear_cache may not work for this use on CPUs with a VIPT i-cache. It will only invalidate the i-cache by virtual address, if the instruction is in the guests i-cache it will not be handled correctly.

Having talked to a Linux dev who is familiar with the kvm code I think we need to:

ensure all VCPUs are stopped
write the instruction
clean the d-cache to the point of unification
if we have a VIPT icache and CTR_EL0.DIC == 0 then perform the i-cache maintenance in the kernel

Revision Contents
Changeset List

Path

Size

usr.sbin/

bhyve/

gdb.c

10 lines

Diff 166332

View Options

gdb: Fix the cache sync in write_instr()Needs ReviewPublicActions