thanks for the answers, lactose & danmcd_

obscure question of the afternoon: i've been looking at /proc/<pid>/* files today for reasons and i notice that in usr/src/uts/common/fs/proc/prsubr.c we ASSERT(AS_WRITE_HELD(as)) for a lot of things that *seem* to only need to read things from the as. prpdsize() feels like a particularly good example. does anyone see an obvious reason we should *need* exclusive access, or is this maybe just a bit

excessive out of caution?

as they are they date all the way back to trusty OpenSolaris Launch so... beyond the horizon of time for me :)

Just from a quick scan, it can cause side effects.

Though that may just be history for how it's dealing with the impl of pr_getprot() and others.

So I'm not exactly sure, but definitely some of the bits in prpdread() make me ask questions.

Writers do starve readers, so you would think that that would give it a consistent enough snapshot while walking if it just had a reader lock.

The block comment about the transient nature of page_exists() in prpdread above the next check is definitely suggestive of stuff going on.

But I'm not personally sure.

oh, interesting, that's one i hadn't noticed

some to think about

iximeow: I might've linked two more recent procfs fixes as related to your newly filed bug... maybe I shouldn't have. Feel free to remove them if it's too intrusive. Some of what you and rmustacc are discussing does indeed go WAY back.

uggh. s/might've/did/g for the above.

danmcd: seems good to me - i wouldn't have seen those otherwise. honestly i was/am pretty focused on the `/{x,}map` stuff but there's a lot in prsubr.c and i assume it supports other procfs paths too

There's some mess in there that did get better. pmooney did some work in there as well.

wow, so many guys on line, like a big famaly, an illumos famaly <3 (y)

[illumos-gate] 17319 want define for in-band PD disable support bit in PCIe slot capabilities 2 -- Dan Cross <cross⊙oc>