[Skiboot] [PATCH v2 2/2] phb4: Disable nodal scoped DMA accesses when PB pump mode is enabled
stewart at linux.vnet.ibm.com
Wed Jul 18 16:14:02 AEST 2018
Alistair Popple <alistair at popple.id.au> writes:
> By default when a PCIe device issues a read request via the PHB it is first
> issued with nodal scope. When accessing GPU memory the NPU does not know at the
> time of response if the requested memory page is off node or not. Therefore
> every read of GPU memory by a PHB is retried with larger scope which introduces
> bandwidth and latency issues.
> On smaller boxes which have pump mode enabled nodal and group scoped reads are
> treated the same and both types of request are broadcast to one chip. Therefore
> we can avoid the retry by disabling nodal scope on the PHB for these boxes. On
> larger boxes nodal (single chip) and group (multiple chip) scoped reads are
> treated differently. Therefore we avoid disabling nodal scope on large boxes
> which have pump mode disabled to avoid all PHB requests being broadcast to
> multiple chips.
> Signed-off-by: Alistair Popple <alistair at popple.id.au>
> Changes from v1:
> - Added more accurate details to the commit message
Thanks. Series merged to
master as of 68518e542e6f7adfe4e97ac22024970ac2400872 and to
6.0.x as of 85a5e14748d3947aca768882ce10a6beb95bcbbc
OPAL Architect, IBM.
More information about the Skiboot