In D54639#1248394, @imp wrote:

You could combine the two lines too by

OFFSET=$((0x$(hd...) ))

You could combine the two lines too by

OFFSET=$((0X$(hd...) ))

option #2, thanks warner! this is cleaner!

Impressive

Is this behavior desirable? It seems very ambiguous to me...

(no changes planned, please review!)

In D54572#1248346, @markj wrote:

In D54572#1247713, @kib wrote:

Or give up on the slight over-accounting and charge the whole object unconditionally, as it is done in the patch.

How do we know it is slight? Is there some way to express a bound on the amount of swap that might be wasted due to the over-accounting?

s/size1/oldsize/
add comment

Updated mpifileutils to version 0.12

let's land this and then figure out how to expand the spin mutexes a bit to cover features a bit cleaner.
(And then see i they really NEED to be spin mutexes..)

yeah, looking at asmc again with spin mutex use i really don't want to clean that up whilst removing GIANT here.
it should be ok-ish for now.

That diff doesn't render too well. Isn't deprecating older macros usually done, particularly when major functionality of the older macros has become rotten? I don't know whether the functionality which has become broken due to changing standards is the main reason why the macros are only used 50% of the time, but I suspect it is a factor.

I had a chat with seuros on discord about it. I think its fine as an intermediary step, but I do think it'll be worth revisiting the driver locking later and pushing more stuff into functionality locks rather than per device IO locks.

In D54572#1247713, @kib wrote:

In D54572#1247459, @markj wrote:

Also I suspect that if OBJ_ONEMAPPING is set, then the original object's ref_count will be 1, and vm_object_split() is a no-op. If this isn't true, I'd want to understand how that can happen.

This is not necessarily true. For instance, if you create an anon mapping of three pages ABC, and then unmap B, I believe that the object behind the original vm_map get the ref count of 2 (first incremented to 3 by split, then decremented by actual delete), but the OBJ_ONEMAPPING flag is kept.