[ofa-general] [PATCH/RFC] Add support for "send with invalidate" to libibverbs

[Roland Dreier]

 > drivers/infiniband/hw/ehca/ehca_hca.c 376:
 > 	props->max_mw = min_t(unsigned, rblock->max_mw, INT_MAX);

 > Note, ehca may set it to huge negative values,

I think the code is OK as it stands... it takes the minimum (as unsigned
int values) of rblock->max_mw and INT_MAX, and returns that.

This should be working OK, at least since 76dea3bc ("IB/ehca: Fix
clipping of device limits to INT_MAX").

 > drivers/infiniband/hw/nes/nes_verbs.c 3915:
 > 	props->max_mw = nesibdev->max_mr;

 > nes puts the wrong value in the attribute field! (typo?)

I'm not positive but it's plausible that the nes limit on the number of
memory windows is the same as its limit on MRs.  And nes has an
implementation of bind_mw, so it is at least possible that it works.

Actually now that I think of it, I have a nes setup where I could test
your MW code... what is the sysctl to set?

 - R.