Skip to contentSkip to conversation

Notes from November 5, 2026

Linux Memory Hotness and Promotion David Rientjes 1 participant 1 message Nov 6 Memory management

lore
Share & tools
lore
#0 lore raw

Hi everybody,

Here are the notes from the last Linux Memory Hotness and Promotion call that happened on Thursday, November 5. Thanks to everybody who was involved!

These notes are intended to bring people up to speed who could not attend the call as well as keep the conversation going in between meetings.

----->o-----
Bharata updated offline that there had been no progress made recently on the async promotion work after the discussion at LPC. He will follow-up in the next meeting to discuss progress and the next upstream patch series.

----->o-----
Davidlohr Bueso provided an update on integrating the CHMU driver, expressing that the generic layer might be overengineered to satisfy all users. He reported that a working prototype is functioning with testing numbers similar to pghot, and code is expected to be released by the end of the month. The driver translates DPA to HPA, providing contiguous ranges as large as possible using 4KB granularity for testing.

Jonathan and Yiannis discussed interleave details. We also briefly chatted about the ability to derive sub-4KB hotness information for offline analysis.

Jonathan shared the original CHMU driver[1] with perf aux buffers which more closely resembled a tracing driver.

Davidlohr will continue working toward releasing the code.

----->o-----
I noted that an upcoming Linux MM Alignment Session would focus on private nodes and CRAM, led by Gregory Price. The announcement will go out to the [email protected] mailing list on Monday.

----->o-----
Huan Nguyen provided an update on the non-temporal store motion posted two days prior, noting open discussions around integrating changes. Huan noted that a patch for reducing overhead within single folios has not yet landed upstream, which will require updates to their assembly code. Zi from Nvidia provided minor refactoring feedback, and Yiannis offered assistance to resolve posting and tagging confusion offline. Huan and Yiannis will coordinate offline to fix tagging issues.

----->o-----
Teja evaluated SDXI for page migration in internal Google use cases, observing significant benefits that met internal requirements. Teja noted a performance regression bug where an interrupt is generated for every single copy, but confirmed that the driver authors (Shivank and Nathan) are aware and plan to fix this in the next patch. Once the small driver fix is released, Teja indicated the series should be ready for production use and supported upstream.

----->o-----
Joshua discussed challenges with upstreaming the tier-aware memcg limits due to complex edge cases, such as handling memcg charges and reclaim during batch migrations. Reviewers identified dangerous edge cases involving tier changes within a memory control group where folio locks and reclaim clash. Joshua debated whether to post a buggy series to maintain momentum or wait until all details are resolved. I advised posting the series with a clear "do not merge" warning and a detailed cover letter to gather early feedback from the mailing list and Google engineers.

Joshua agreed to proceed with this approach for their shared heterogeneous workload use case. He will prepare and post the patch series with known deficiencies explicitly noted, and I will coordinate internal Google review and testing.

----->o-----
Next meeting will be on Thursday, November 19 at 8:30am PST (UTC-8), everybody is welcome: https://meet.google.com/jak-ytdx-hnm

Topics for the next meeting:

- update on Bharata's asynchronous promotion mechanism separated out from the rest of pghot, targeting posting the revision before the next meeting
- discuss current status of CHMU driver and its integration with DAMON or any other generic backend as needed
- update on private node support as well as CRAM to be posted now that the 7.4 merge windows closed
- update on combined patch series for supporting non-temporal stores in migrate_pages() with memory error handling (series from Yiannis + Huan)
- discuss Shivank's series for enlightening migrate_pages() for hardware assists and competition between the kernel and user use cases
- update on Bharata's support for sources of page hotness and hardware support
- Teja's update on SDXI page migration based on AMD patches and hardware issues being encountered that do not result in page migration
- update on tier-aware memcg limits status and production testing based on the latest major overhaul

Please let me know if you'd like to propose additional topics for discussion, thank you!

[1]
https://lore.kernel.org/all/[email protected]/