** Description changed:
+ SRU Justification:
+
+ [ Impact ]
+
+ * A new IBM Z hardware generation (z17) was introduced. libutil
+ needs to become aware of its machine types (9175/9176) to map
+ type <-> name and CPU-MF counters correctly.
+
+ * Without patching, z17 hardware is reported as "Unknown machine
+ type", and lscpumf does not list the z17-specific extended
+ counter sets.
+
+ * This is the Jammy counterpart of LP: #2103414
+ this is already incl. in noble and above.
+
+ * Required commits:
+ - b3db52fd ("libutil: Add machine type definition for machines
+ 9175 and 9176")
+ - 120ad30f ("cpumf/lscpumf: Add support for IBM z17 counter sets")
+ - ef1799f3 ("s390-tools/libutil: Add machine type 3932"),
+ pre-requisite for b3db52fd, as already done for Noble.
+ - 7d855acd ("util_arch: Fix max HSA size for Z16 machine")
+
+ * The last commit is only req. for jammy, because jammy's
+ s390-tools v2.20.0 is still based on the pre-7d855acd numeric
+ comparison in util_arch_hsa_maxsize(), which ef1799f3 assumes
+ is already a switch statement.
+ This 'old' numeric comparison also under-reports the max HSA size
+ for IBM z16 (32 MiB instead of 512 MiB, since z16's type number
+ is numerically lower than z15's).
+ This additional potential issue is the reason why I haven't
+ backported ef1799f3 to the 'old' implementation, since it would
+ not solve the under-reporting.
+ And adding a fix for the under-reporting to a potential
+ ef1799f3 backport, would req. a similar amount of changes
+ compared to the additional commit, but would (unnecessarily)
+ diverge the backport significantly from the upstream commit,
+ which again would harm further maintenance.
+
+ [ Test Plan ]
+
+ * On an IBM z17 LPAR running Ubuntu 22.04 with this update (and
+ matching kernel CPU-MF support), enable the relevant counter sets
+ in the LPAR activation profile and confirm via 'perf list' /
+ 'perf stat' that the new counters are usable.
+
+ * Check machine identification directly via lscpumf (uses libutil):
+ # lscpumf -C | head -1
+ perf event counter list for IBM z17
+ i.e. explicitly confirm it does NOT report "Unknown machine
+ type" (this exact check was flagged as missing during LP:
+ #2103414's review and had to be added there too).
+
+ * Confirm 'lscpumf -C' lists the full z17 extended counter set (see
+ upstream commit 120ad30f), using the same command IBM used to
+ verify LP: #2103414 on real z17 hardware:
+ # lscpumf -C | grep '^10:' | wc -l
+ Expected: 105 (confirmed by upstream on both Plucky and Noble).
+ Without commit 120ad30f, this reports only 28 (extended counter
+ set missing).
+
+ * Regression check: 'lscpumf -v' still correctly identifies existing
+ hardware (z13/z14/z15/z16). On a z16 (or z16 A02, type 3932)
+ system, confirm any util_arch_hsa_maxsize() consumer (e.g.
+ zgetdump's HSA dump size) now uses 512 MiB, not 32 MiB.
+
+ [ Where problems could occur ]
+
+ * As always with such table/enum extensions: a typo in a name or
+ type number, or an incorrectly placed/terminated case statement,
+ could cause wrong identification or reintroduce "Unknown machine
+ type" - leading to wrong event exposure, invalid perf
+ measurements, or sysfs inconsistencies.
+
+ * The extra 7d855acd patch changes the *implementation* of
+ util_arch_hsa_maxsize() (comparison -> switch), not just adds
+ cases. A logic error there could misreport max HSA size for an
+ already-supported machine type. Risk is mitigated by using the
+ verbatim upstream commit, already in production since s390-tools
+ 2.31.0/Ubuntu 24.04.
+
+ * All of this is s390x-only code; no other architectures affected.
+ Patches 2-4 only add new code/cases; patch 1 only corrects the
+ Z16/Z16_A02 handling of one existing function.
+
+ [ Other Info ]
+
+ * The CPU-MF counters kernel work was already done.
+
+ * Already Fix Released for Noble/24.04 (2.31.0-0ubuntu5.2) via
+ LP: #2103414 and newer releases.
+
+ * This bug was previously marked "Won't Fix", partner requested
+ this as z17 enablement for jammy.
+ __________
+
Description:
Support for CPU-MF counters of new IBM z17 in s390-tools for Jammy (22.04)
and Noble (24.04)
This item covers the s390-tools part of the z17 support, i.e. changes to
lscpumf and libutil.
Required commit for s390-tools:
https://github.com/ibm-s390-linux/s390-tools/commit/b3db52fd165e6a26d9c73e51bd5322b55017ef94
libutil: Add machine type definition for machines 9175 and 9176
** Changed in: s390-tools (Ubuntu Jammy)
Status: Won't Fix => In Progress
** Changed in: s390-tools-signed (Ubuntu Jammy)
Status: Won't Fix => In Progress
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2121810
Title:
[UBUNTU 22.04] Support CPU-MF Counters for new IBM z17 - s390-tools
part
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu-z-systems/+bug/2121810/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
