[jdk11u-dev] RFR: 8261302: NMT: Improve malloc site table hashing

[María Arias de Reyna]

On Thu, 21 Mar 2024 10:01:46 GMT, Goetz Lindenmaier <goetz at openjdk.org> wrote:

>> I have a list of potential NMT improvements (following @tstuefe steps) that could be backported here. To me it makes sense to be as similar as JDK21, but of course this is not my call, so let me know if I should proceed.
>
> @Delawen, @stuefe, can you please share the list of intended backports?

Here is the short version list @GoeLin 

To JDK11:

The one with the opened PR with its follow-up:
* JDK-8261302: NMT: Improve malloc site table hashing
* JDK-8261520

Other related issues:
* JDK-8261644: NMT: Simplifications and cleanups
* JDK-8262165: NMT report should state how many callsites had been skipped
* JDK-8256844: Make NMT late-initializable
* JDK-8275320: NMT should perform buffer overrun checks
* JDK-8277822: Remove debug-only heap overrun checks in os::malloc and friends
* JDK-8281015: Further simplify NMT backend
* JDK-8292072: repurpose Tracking overhead counter as global malloc counter
* JDK-8297958: Display peak values
* JDK-8317772: Make peak values available in release builds

TO JDK17 and below:
* JDK-8280583 (fixed in 19): Always build NMT
* JDK-8281023 (fixed in 19): NMT integration into pp debug command does not work
* JDK-8281460 (fixed in 19): Let ObjectMonitor have its own NMT category
* JDK-8289182 (fixed in 20): MemTracker::baseline should return void
* JDK-8289633 (fixed in 20): Forbid raw C-heap allocation functions in hotspot and fix findings hotspot integrated
* JDK-8290868 (fixed in 20): MallocSiteTable statistics improvements
* JDK-8290870 (fixed in 20): Increase MallocSiteTable size and allocate it only when needed
* JDK-8291878 (fixed in 20): Diagnostic malloc limits
* JDK-8292071 (fixed in 20): move MallocHeader to its own header and inline header checks
* JDK-8292073 (fixed in 20): remove unused constructor parameter from MallocHeader
* JDK-8293313: (fixed in 21): Add "Fake OOM" mode to MallocLimit
* JDK-8302820 (fixed in 21): Remove costs for NMTPreInit when NMT is off
* JDK-8292561 (fixed in 20): Make "ReplayCompiles" a diagnostic product switch
* JDK-8319437 (fixed in 22, to check JDK21): should show library names in call stacks
* JDK-8320061 (fixed in 22, to check JDK21): Multiple issues with peak accounting

Maybe I should start with the ones associated with JDK17?

-------------

PR Comment: https://git.openjdk.org/jdk11u-dev/pull/2608#issuecomment-2011879244