Roland Dreier wrote:
I had a chance to look at this, and I don't believe it is precisely
correct for mem-free HCAs with the current FMR implementation.
> + /* on memfull HCA an FMR can be remapped 2^B - 1 times where B < 32 is
> + * the number of bits which are not used for MPT addressing, on memfree
> + * HCA B=8 so an FMR can be remapped 255 times.
> + */
> + if(!mthca_is_memfree(mdev))
> + props->max_map_per_fmr = (1 << (32 -
> + long_log2(mdev->limits.num_mpts))) - 1;
> + else
> + props->max_map_per_fmr = (1 << 8) - 1;
Look at mthca_arbel_map_phys_fmr(). The question is how often key
will repeat after being indexed, and when MTHCA_FLAG_SINAI_OPT is not
set, then the same increment is used in the mem-free case as in the
Tavor case.
So I think the code I quoted should really be:
if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
props->max_map_per_fmr = (1 << (32 -
long_log2(mdev->limits.num_mpts))) - 1;
else
props->max_map_per_fmr = (1 << 8) - 1;
Do you agree? If so I can fix this patch up myself and apply it.
Yes it makes sense, but you need the check should be
if (!(dev->mthca_flags & MTHCA_FLAG_SINAI_OPT))
instead of
if (dev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
also, what about the other patch which changes fmr_pool.c to query the
device, have you got(reviewed/accepted) it? i have modified it to
allocate the device attr struct on the heap as you have asked.
Or.
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
