Re: [PATCH RFC 00/23] resctrl: MBA control emulation and ARM MPAM MB_NODE support
Reinette Chatre <[email protected]>
| Newsgroups | org.infradead.lists.linux-arm-kernel,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
Hi Fenghua, On 7/16/26 2:02 PM, Fenghua Yu wrote: > This series enables MBA bandwidth control emulation on ARM MPAM when the > default MB control is disabled, using a node-scoped MB_NODE control as > the native backend. It extends memory-bandwidth monitoring and > control to memory-level MSCs or CPU-less NUMA nodes, and adds > documentation and tests. Just to confirm, I believe that [1] simplifies support for CPU-less NUMA nodes. Some highlights of that related to this series: - There was agreement that resctrl will treat each resource at a particular scope as a separate resource. Just like L2 and L3 are two resources that represent cache allocation at different scope, MB and MB_NODE will be two resources representing memory bandwidth allocation at different scope with the MB resource implying L3 scope. - [1] contains support for switching control mode between legacy and native. From the MB_NODE examples in this cover letter it looks to me as though MB_NODE will take the shape of a percentage based control. As I understand the hardware has multiple controls with finer grained capabilities and how these will be used may still need emulation if MB_NODE is planned to be percentage based. As a sidenote, since it is introduced after the new control properties it could possibly just start out as a finer grained control? - There is no need to change the info/ hierarchy on switch between legacy and native. Regarding the monitoring support, I still believe that the monitoring support needs to be separated from the resources to accurately reflect that the monitoring is associated with a scope, not a resource. This change does not form part of [1] though. It is fine with me if we continue discussion at [1]. This message is just intended as a summary of changes in the works that should simplify the support for this hardware. Reinette [1] https://lore.kernel.org/lkml/[email protected]/