[ofa-general] page mask calculation in mthca mthca_reg_phys_mr

[Liang Zhen]

Hi,
I think there is a similar bug in mthca as in cxgb3 
(http://lists.openfabrics.org/pipermail/general/2008-January/045246.html),
in mthca_reg_phys_mr(), mask is built like this way:

        for (i = 0; i < num_phys_buf; ++i) {
                if (i != 0)
                        mask |= buffer_list[i].addr;
                if (i != num_phys_buf - 1)
                        mask |= buffer_list[i].addr + buffer_list[i].size;

                total_size += buffer_list[i].size;
        }

It can get wrong mask as start address of the first fragment and end 
address of the last fragment are ignored, here is some example of this:

INPUT:
        ipb[0].addr = 0x10002000;
        ipb[0].size = 0x4000;
        ipb[1].addr = 0x10006000;
        ipb[1].size = 0x1000;
OUTPUT:
        shift 13 npages 3
        page_list[0] = 10002000
        page_list[1] = 10004000
        page_list[2] = 10006000

2) --------------------------------
INPUT:
        ipb[0].addr = 0x10001000;
        ipb[0].size = 0x4000;
        ipb[1].addr = 0x10006000;
        ipb[1].size = 0x1000;
OUTPUT:
        shift 12 npages 5
        page_list[0] = 10001000
        page_list[1] = 10002000
        page_list[2] = 10003000
        page_list[3] = 10004000
        page_list[4] = 10006000

Regards
Liang