On Mon, Nov 20, 2017 at 03:29:59PM -0600, Shiraz Saleem wrote: > On Mon, Nov 20, 2017 at 11:38:00AM -0700, Jason Gunthorpe wrote: > > On Fri, Nov 17, 2017 at 10:46:51AM -0600, Henry Orosco wrote: > > > From: Chien Tin Tung <chien.tin.tung@xxxxxxxxx> > > > > > > Currently there is only one sdbuf per Control QP(CQP) for programming > > > Segment Descriptor(SD) command thus limiting the number of SD work > > > requests that can be posted to one. Allocate enough memory for one > > > sdbuf per CQP SQ WQE to allow more than one SD command at a time. > > > When a SD command is posted, it will use the corresponding sdbuf for > > > the WQE. > > > > It is not clear to me from this description why this would be a rc > > candidate? > > > > Is this is a bug? or is it adding a feature of increasing the number > > of 'SD work requests'? > > > > Its a bug fix as we are reusing the same SD buffer > for all WQEs posted. This is incorrect and will become > a problem if we have multiple simultaneous SD WQEs > being posted. OK, can you send me a revised commit description making it clear what the bug is and how it could be hit? Maybe something like: The driver only allocated one Segment Descriptor (SD) per ???, but there is nothing preventing multiple WQEs from trying to use the same SD. In this case new WQEs can corrupt past SD's resulting in mis-execution of the WQE. Fix this by allocating a SD for every possible WQE and blah blah Jason -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
