On 7/31/19 8:37 AM, Michal Privoznik wrote:
On 6/11/19 5:31 AM, Wang Huaqiang wrote:
> Patches submitted for purpose of refactoring existing 'resctrl' related
> source code, including some code cleanups as well as some fixes. This is
> also a preparation for memory bandwidth monitor codes.
>
> Plan to support Resctrl Control Monitors, which is a feature introduced
> by kernel 'resctrl' sub-model. Submit some cleanup and refactoring
> patches
> for upcoming memory bandwidth resource monitoring (MBM) monitors.
>
> Related MBM RFC is
>
https://www.redhat.com/archives/libvir-list/2019-April/msg01409.html.
> This RFC is not actively discussed since libvirt already implemented
> similar
> resctrl cache monitoring (CMT), and lots details have been discussed
> and implemented during the work of CMT.
>
> The cleanups and refactoring includes:
> v2 changes:
> 1. Addressed comments of v1.
> 2. Introduce a new algorithm for verifying new monitor vcpus and existing
> monitors and allocations.
> 3. Fixes for creating default-allocation-monitor in 'resctrl' file
> system.(patch 0001).
>
> v1 changes:
> 1. Removing some reluctant lines and white spaces that is existing
> in current code and not meet the libvirt coding style.
> 2. Replace 'virResctrlAllocIsEmpty' with @n==0 for indicating no
> resctrl allocation in configuration file.
> 3. Private API changes, removed 'virResctrlMonitorGetCacheOccupancy'
> and exported a new API named 'virResctrlMonitorGetStats' with similar
> functionality, but with capability to be used for retrieving MBM
> statistical information.
> 4. Refactoring 'virResctrlMonitorFreeStats' for more reusing in code.
> 5. Extend data structure 'virResctrlMonitorStats' with the capability
> to carry multiple statistical information from monitor.
>
> Wang Huaqiang (11):
> util,conf: Handle default monitor group of an allocation properly
> conf: code cleanup, remove empty line and one space
> conf: code cleanup for return error code directly
> conf: some code cleanup
> conf: refactor 'virDomainResctrlVcpuMatch' and some code cleanup
I think this patch introduced an issue. resctrl returned from VcpuMatch
should not be free'd, but that can happen in virDomainMemorytuneDefParse
Use the attached XML and do:
./tools/virsh --connect test:///default define bad.xml
> conf: Append 'resctrl' object according to number of
monitor group
> directly
> util: Refactor and rename 'virResctrlMonitorFreeStats'
> util: Refactor 'virResctrlMonitorStats'
> util: Extend virresctl API to retrieve multiple monitor statistics
> util: Remove unused virResctrlMonitorGetCacheOccupancy
> conf: Refactor and rename the function to validate a new resctrl
> monitor
>
> src/conf/domain_conf.c | 145
> ++++++++++++++++++++++++-----------------------
> src/libvirt_private.syms | 5 +-
> src/qemu/qemu_driver.c | 41 ++++++++++----
> src/qemu/qemu_process.c | 3 +-
> src/util/virresctrl.c | 75 ++++++++++--------------
> src/util/virresctrl.h | 23 +++++---
> 6 files changed, 156 insertions(+), 136 deletions(-)
>
Reviewed-by: Michal Privoznik <mprivozn(a)redhat.com>
I'll push these once the freeze is over.
- Cole