[ofa-general] Re: [PATCH] opensm/opensm/osm_state_mgr.c: Add check for valid physical port before using pointer.

Sasha Khapyorsky sashak at voltaire.com
Sun Nov 23 01:05:57 PST 2008


Hi Ira,

On 14:06 Tue 18 Nov     , Ira Weiny wrote:
> I am not sure this will fix my bug.
> 
> The stack trace in my bug ended with:
> 
>    #0  osm_vendor_get (h_bind=0x0, mad_size=256, p_vw=0x69bbe8) at
> 
> The h_bind was being extracted from the osm_physp_t object.  Would this fix
> ensure that the h_bind pointer was valid in the osm_physp_t object returned?

Not always :(. It will protect against port moving, but may not help in
case of PortInfo Get failure (as far as I understand now it is your case).

Finally I just removed osm_node_get_any_physp_ptr() (as well as
osm_node_get_any_dr_path() which uses similar assumption about local
port number in NodeInfo) in all places where is was used.

I think we can do the same in __osm_state_mgr_get_node_desc() and remove
osm_node_get_any_physp_ptr() completely. The patch shortly.

Sasha



More information about the general mailing list