codeconstruct / dbus-sensors Goto Github PK

I am officially enabling the unit test for presubmit.

https://gbmc-review.googlesource.com/c/dbus-sensors/+/12994/2

But the new interfaces introduced in the ns management failed the unit-test

Currently, the dbus-sensor enables project wise io_uring flag:

https://gerrit.openbmc.org/c/openbmc/dbus-sensors/+/64039.

However, the nvmed disabled the io_uring due to https://gerrit.openbmc.org/c/openbmc/dbus-sensors/+/62299

Consequentially, it failed the ubsan:

/usr/local/include/boost/asio/detail/deadline_timer_service.hpp:272:30: runtime error: member call on address 0x617000000080 which does not point to an object of type 'io_uring_service'
0x617000000080: note: object is of type 'boost::asio::detail::select_reactor'
 00 00 00 00  e8 a7 39 5c 4e 56 00 00  00 00 00 00 00 00 00 00  52 70 95 5e 4e 56 00 00  a0 6d 95 5e
              ^~~~~~~~~~~~~~~~~~~~~~~
              vptr for 'boost::asio::detail::select_reactor'

Detailed in https://gbmc-review.googlesource.com/c/dbus-sensors/+/14996/comment/ac6097a4_e5a96d29/

details in https://gbmc-review.googlesource.com/c/dbus-sensors/+/14995/comment/6485bd92_646ee5db/

Hi @mkj,

Jinliang get an idea to improve the createNamespace() process:

https://gbmc-review.googlesource.com/c/dbus-sensors/+/13495.

Would you like to take a look?

A diff on nsid list is not trustworthy for scanning NS changes. https://github.com/CodeConstruct/dbus-sensors/blob/794103cd4d978d4f2a5209c3d45655a06dea804d/src/NVMeSubsys.cpp#L769C10-L769C10

Should turn to NAC(Namespace Attribute Changed) via health status poll.

From the comment: 0bd3368#r129278792

Currently NVMeSubsystem uses different techniques to exit on forbidden state transaction (exception, assertion, printing&return), we would better to unify them.

https://gbmc-review.googlesource.com/c/dbus-sensors/+/14999/comment/e209fab3_6376efef/

The modification of subsystem with init() makes it incompatible with the unit test. I have to make a temporary walk around to make the unit test work (https://gbmc-review.googlesource.com/c/dbus-sensors/+/12914/3/src/NVMeSubsys.hpp).

I think the requirement of the init() function is debatable since it is similar to the start() function.

The change c6a4016 created a crash on the google release.

2023-12-01 3:49:48 systemd,Started NVMe Sensor.
2023-12-01 3:49:48 nvmesensor,error getting  SpecialMode status No route to host
2023-12-01 3:49:54 nvmesensor,"[bus: 6, addr: 48, eid: 17]fail to do nvme identify:"
2023-12-01 3:49:54 nvmesensor,adminIdentify:NVMe MI: Invalid Parameter (MI status 0x4)
2023-12-01 3:49:54 nvmesensor,fail to do nvme identify: NVMe MI: Invalid Parameter (MI status 0x4)
2023-12-01 3:49:54 nvmesensor,fail to identify secondary controller list
2023-12-01 3:49:54 nvmesensor,"Failed to identify secondary controller list. error NVMe MI: Invalid Parameter (MI status 0x4) data size 4096 expected size 4096. Fallback, using arbitrary controller as primary."  
2023-12-01 3:49:54  systemd,Started Process Core Dump (PID 4452/UID 0).
2023-12-01 3:50:00  systemd-coredump,Process 4432 (nvmesensor) of user 0 dumped core.
2023-12-01 3:50:00  systemd,"xyz.openbmc_project.nvmesensor.service: Main process exited, code=dumped, status=11/SEGV"
2023-12-01 3:50:00 systemd,xyz.openbmc_project.nvmesensor.service: Failed with result 'core-dump'.

Reverted the change in : https://gbmc-review.googlesource.com/c/dbus-sensors/+/14474

Currently the association is created in NVMeSubsys. Such association would be better to bind with the volume.

The unit test here is to manually trigger a destruction of nvme instance and check the resource recollection status. Is it designed to prevent the resource is held by some async event and never get the chance to release.

The unit test is using ObjectMapper as a observer so it require Dbus and O-M to be enabled in the docker image. We could enforce the feature for presubmit.

https://gbmc.googlesource.com/meta-gbmc-staging/+/refs/heads/gbmc-release-23.18.x/recipes-phosphor/sensors/dbus-sensors/0030-nvmesensor-add-unit-test.patch

Hello,

When the NVMe device is disconnected, the subsystem should transit into disable status with all substructure deleted. However the shared ptr of Pcontroller is held by the subsystem and thus cannot be release:

https://github.com/CodeConstruct/dbus-sensors/blob/dev/nvme-redfish/include/NVMeSubsys.hpp#L124C44-L124C61

Detailed in https://gbmc-review.googlesource.com/c/dbus-sensors/+/14999/comment/d88462cb_b06e6af6/