Add comments about off-page subtleties

[kib]

Summary

Explain two issues in the pshared implementation which affect correctness.

Please help me to fix the language.

[emaste]

lib/libthr/thread/thr_pshared.c

90–92 ↗

(On Diff #13883)

Maybe more clear as:
Among all processes sharing a lock only one executes pthread_lock_destroy(). Other processes still have the hash and mapped off-page.

94–97 ↗

(On Diff #13883)

Looks good.

146 ↗

(On Diff #13883)

I'd suggest:
return either the new (just mapped) or old (hashed)

151 ↗

(On Diff #13883)

not yet stored in or not yet included in

155 ↗

(On Diff #13883)

the value from the hash, the first thread might operate on

but then this gets rather confusing.

Otherwise, if we unmap the value obtained from the hash, the first thread might operate on an unmapped off-page object.

158–161 ↗

(On Diff #13883)

How about:

There is still an issue: if the hashed key was unmapped and then another page is mapped at the same key address, the hash would return the old val.

Also clarify 'handle race' -- i.e., what happens if there's a race?

[kib]

lib/libthr/thread/thr_pshared.c

158–161 ↗

(On Diff #13883)

The race was explained in the paragraph above. I made more evolved reference to it in the revised text.

[kib]

Edits suggested by Ed.

[emaste]

LGTM with the additional clarification we discussed on IRC.