> On 31 Jan 2019, at 23:45, Jason Gunthorpe <jgg@xxxxxxxx> wrote: > > On Wed, Jan 30, 2019 at 01:59:37PM +0100, Håkon Bugge wrote: >> Hey there, >> >> From IBTA: >> >> C17-19: For deadlock prevention, the CA shall not continuously and >> permanently assert backpressure (i.e. fail to grant link credits). > > 'permanently' here really does mean permanently. A filled EQ should be > a transient condition.. So it isn't a spec violation, although it is > probably bad. If the OS freezes, the HCA is not reset, the sys admin is on vacation, then pretty permanent I would say ;-) One could say to Mellanox' excuse that said clause does not apply for RoCE devices. But, CX-5 is supposed to work (well) with IB link-layer as well. Thxs, Håkon
