prov/psm3: Internal polling must use timeout to prevent infinite loop…

[tatarintsevsv]

… during resources acquiring

Due to a lack of resources, the AMSH_POLL_UNTIL() and PSMI_BLOCKUNTIL() may never complete. If resources polling was not finished within PSMI_MIN_EP_CONNECT_TIMEOUT, an interrupt will occur and error will be returned.

Fixes #11893

[j-xiong]

In addition to the comment below, note that the return value of psm3_amsh_am_short_request and psm3_amsh_generic_inner is not handled by the upper layer in its call chain. As a result, in case of resource exhaustion, the application is able to break out from the infinite loop but may silently treat this error as success.

[tatarintsevsv]

In addition to the comment below, note that the return value of psm3_amsh_am_short_request and psm3_amsh_generic_inner is not handled by the upper layer in its call chain. As a result, in case of resource exhaustion, the application is able to break out from the infinite loop but may silently treat this error as success.

Error hadling added.
In some cases i'm not sure how to handle error properly, so app will just aborted

[j-xiong]

Thanks. I do see a few other places in the call chain the error code is still ignored (e.g. psm3_amsh_short_reply, psm3_amsh_long_reply, psm3_am_reqq_drain). However, those are probably fine since they either lead to send side timeout due to not getting the reply or is already in the error processing path. @acgoldma Do you have any comment?

The review process here is different from some other repos that do auto squash on merging. Please squash the commits and do a force push. For this specific case, it is fine to keep the commits separate, but then you will need to add proper commit message and "signed-off" line.

[tatarintsevsv]

Thanks! Commits squashed

[acgoldma]

I will review the code and get back to you.
That being said, the change in PSMI_BLOCKUNTIL() does not seem right. psm3_poll_internal already uses timers and will return when then expire with PSM2_EP_NO_RESOURCES.
Can you maybe dump the psm3 stack to show more about where you hang is?

[tatarintsevsv]

I will review the code and get back to you. That being said, the change in PSMI_BLOCKUNTIL() does not seem right. psm3_poll_internal already uses timers and will return when then expire with PSM2_EP_NO_RESOURCES. Can you maybe dump the psm3 stack to show more about where you hang is?

Thanks, i will wait for your review!
My test case i'm described here
There little example described the problem.
I think problem is not that psm3_poll_internal returns some incorrect. The problem is that the loop doesn't check for a oveall timeout.
I will give you call stack as soon as possible. Thanks!

[acgoldma]

I still think there is a better solution as we should already have timers waiting for ACKs. the point of the yield to receive thread is that the rec thread should check timers.

[tatarintsevsv]

some fixes pushed

[acgoldma]

Looks good.

[j-xiong]

Please remove the second commit. Or use rebase instead of merge if you want the CI to pick up all the changes from main.

[tatarintsevsv]

Please remove the second commit. Or use rebase instead of merge if you want the CI to pick up all the changes from main.

done