prov/shm: fi_setopt causes a segmentation fault when setting FI_OPT_MIN_MULTI_RECV

[AOA-Mohammed]

Describe the bug
fi_setopt causes a segmentation fault when setting FI_OPT_MIN_MULTI_RECV before enabling the endpoint.
According to fi_endpoint man page, it is recommended to set FI_OPT_MIN_MULTI_RECV before enabling the endpoint.
Doing so, causes a segmentation fault with the following backtrace:

AddressSanitizer:DEADLYSIGNAL
=================================================================
==2613126==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000008 (pc 0x7f128262fc1d bp 0x7f11fab08410 sp 0x7f11fab082d0 T2)
==2613126==The signal is caused by a READ memory access.
==2613126==Hint: address points to the zero page.
    #0 0x7f128262fc1d in smr_ep_setopt (/lib/libfabric.so.1+0xabc1d) (BuildId: 3f0a904075fdd37c2719c171f33bd661bbe68aeb)
    #1 0x7f128280d103 in fi_setopt /include/rdma/fi_endpoint.h:232:9
    #2 0x7f128280d103 in mstro_ep_build_from_ofi maestro-core/maestro/ofi.c:1198:14

However, setting FI_OPT_MIN_MULTI_RECV after enabling the endpoint works for the shm provider (unlike other providers and opposite to what is recommended in the man pages).

To Reproduce
set FI_OPT_MIN_MULTI_RECV before enabling the endpoint.

Expected behavior
The code should work.

Output
The application fails with a seg fault as above

Environment:
OS (if not Linux), provider, endpoint type, etc.
libfabric 1.21.0
shm provider
icx compiler 2023.2.0

[aingerson]

@AOA-Mohammed Sorry for the late response. I've opened a PR that should fix this issue. If you have a chance, could you test with PR #10618 and verify that fixes the issue? Thanks for reporting!

[AOA-Mohammed]

@aingerson I tested the fix and fi_setopt completes safely. Thanks for the fix.