vt: Do not lock request comming from terminal, only those originated by mouse. Because the terminal surrounds requests to vt(4) with locking.

[ray]

Summary

Reported by: bz, adrian
MFC after: 2 weeks

[ray]

Updating D57442: vt: Do not lock request comming from terminal, only those originated by

mouse. Because the terminal surrounds requests to vt(4) with locking.

[adrian]

ok, i see what's going on here. im' ok with this to unbrick -head, but personally:

• have xxx_locked() versions of functions which mtx_assert(mtx, MA_OWNED)
• have unlocked versions of functions that mtx_assert(mtx, MA_UNOWNED); grab lock; call _locked version; mtx_unlock();
• callers would then either call the xxx() version if they don't own the lock, or xxx_locked() if they already hold the lock.

If this fixes the crash then I'm going to approve it to unbrick head, but it does tell me we should do a pass or two around locking in vt.

[glebius]

sys/dev/vt/vt.h

248–251 ↗

(On Diff #179215)

I'd suggest to just use bool locked. This is a very common style throughout the kernel in the last years. Or use Adrian's suggestion.

sys/dev/vt/vt_buf.c

743–757

Alternatively to this whole patch you can just use mtx_owned(9) and don't need to modify the function argument list:

#define VTBUF_LOCK_OWNED(vb) mtx_owned(&(vb)->vb_lock)
...
static void
vtbuf_flush_mark(...
	bool dounlock;
...
	if (!VTBUF_LOCK_OWNED(vb)) {
		dounlock = true;
		VTBUF_LOCK(vb);
	} else
		dounlock = false;
	vtbuf_dirty(vb, &area);
	if (dounlock)
		VTBUF_UNLOCK(vb);

[ray]

Check if lock owned.

Updating D57442: vt: Do not lock request comming from terminal, only those originated by

mouse. Because the terminal surrounds requests to vt(4) with locking.