Re: smr inp breaks some jail use cases and panics with i915kms don't switch to the console anymore

On 12/13/21 9:35 AM, Gleb Smirnoff wrote:
>    Hi John,
> 
> On Mon, Dec 13, 2021 at 07:45:07AM -0800, John Baldwin wrote:
> J> So there are two things here.  The root issue is that the devel/apr1 port
> J> runs a configure test for TCP_NDELAY being inherited by accepted sockets.
> J> This test panics because prison_check_ip4() tries to lock a prison mutex
> J> to walk the IPs assigned to a jail, but the caller (in_pcblookup_hash()) has
> J> done an smr_enter() which is a critical_enter():
> 
> The first one is known, and I got a patch to fix it:
> 
> https://reviews.freebsd.org/D33340
> 
> However, a pre-requisite to this simple patch is more complex:
> 
> https://reviews.freebsd.org/D33339
> 
> There is some discussion on how to improve that, and I decided to do that
> rather than stick to original version. So I takes a few extra days.
> 
> We could push D33340 into main, if the negative effects (raciness of
> the prison check) is considered lesser evil then potentially contested
> mtx_lock in smr section.

I think raciness is probably better than always panicking as it does today.
  
> J> However, it was a bit harder to see this originally as the 915kms driver
> J> tries to do a malloc(M_WAITOK) from cn_grab() when entering DDB which
> J> recursively panics (even a malloc(M_NOWAIT) from cn_grab() is probably a
> J> bad idea).  When it panicked in X the result was that the screen just froze
> J> on whatever it had most recently drawn and the machine looked hung.  (The
> J> fact that that sysbeep is off so I couldn't tell if typing in commands was
> J> doing anything vs emitting errors probably didn't improve trying to diagnose
> J> the hang as "sitting in ddb" initially, though I don't know if DDB itself
> J> emits a beep for invalid commands, etc.)
> 
> Didn't know about this one. Is this isolated to actually entering DDB or
> there is some path that in a normal inpcb lookup we would M_WAITOK?

This is in the drm(4) driver, nothing to do with in_pcb, just made it harder
to see the in_pcb issue.

-- 
John Baldwin