[ofa-general] [PATCH/Resend] Fixed capability mask problem in ibstat introduec by commit 722b6c6428c9e4921a81f4a6db2838bcee660bb7

[Sean Hefty]

>diff --git a/infiniband-diags/src/ibstat.c b/infiniband-diags/src/ibstat.c
>index 7985be1..99af9a8 100644
>--- a/infiniband-diags/src/ibstat.c
>+++ b/infiniband-diags/src/ibstat.c
>@@ -111,7 +111,7 @@ port_dump(umad_port_t *port, int alone)
> 	printf("%sBase lid: %d\n", pre, port->base_lid);
> 	printf("%sLMC: %d\n", pre, port->lmc);
> 	printf("%sSM lid: %d\n", pre, port->sm_lid);
>-	printf("%sCapability mask: 0x%08x\n", pre, (unsigned)ntohll(port-
>>capmask));
>+	printf("%sCapability mask: 0x%08x\n", pre,
>(unsigned)(ntohl((uint32_t)(port->capmask))));

Casting from 64-bit to 32-bit, then byte swapping doesn't look right.

I think the problem may be in libibumad, umad.c, line 166:

        if (sys_read_uint64(port_dir, SYS_PORT_CAPMASK, &port->capmask) < 0)
                goto clean;

        port->capmask = htonl(port->capmask);

capmask is read as a 64-bit value, but only 32-bit swap is used.  (libibumad is
not shared between Linux and Windows, so this problem doesn't show up on
Windows.)

- Sean