[Feature] SKX: Support crystall ridge / far / near memory indication in PEBS
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
intel |
Fix Released
|
Undecided
|
Unassigned | ||
linux (Ubuntu) |
Fix Released
|
Medium
|
Seth Forshee |
Bug Description
On SKX PEBS bits should report whether a memory address DLA sample is crystal ridge:
hey have two PEBS encodings for L4_HIT Local and L4_HIT Remote. Since we have no L4, we setup the PRE for Server so that the PRE encodings that map onto L4_HIT are actually crystal ridge. So PEBS 0x8 (L4 HIT Local) should mean serviced by local crystal ridge memory on a SKX part, and PEBS 0x9 (L4 HIT Remote) should mean serviced by remote crystal ridge memory on a SKX part.
That said, there are multiple steps in the chain that gets that information from the memory controller to the core, and I don't think any of those steps has been formally validated.
We need to extend perf to report these distinctions. First needs some testing if it actually works. Would require kernel changes, ABI changes, perf report changes.
HW: Purley
Upstream schedule:
kernel: TBD
information type: | Proprietary → Public |
Changed in linux (Ubuntu): | |
assignee: | nobody → Seth Forshee (sforshee) |
importance: | Undecided → Medium |
status: | New → Triaged |
Changed in linux (Ubuntu): | |
status: | Triaged → Fix Committed |
Changed in intel: | |
status: | New → Fix Released |
Any updates on the kernel target here? Did this make v4.8? Or should we defer this to 17.10?