Compatible Ubuntu and kernel versions for building/loading snvme driver?

[WWWzq-01]

Hi team,

I’m trying to build and load the custom snvme/snvme-core modules. On my setup (Ubuntu 20.04 + kernel 5.15.20-051520-generic), the modules compile but fail to load with symbol version mismatches against the in-tree nvme_core (dmesg shows “snvme: disagrees about version of symbol … / Unknown symbol … (err -22)” and “snvme_core: exports duplicate symbol __nvme_check_ready (owned by nvme_core)”). Because the root filesystem is on NVMe, I also can’t rmmod the stock nvme/nvme_core while the system is running.

Could you clarify:

• Which Ubuntu release do you recommend?
• Which Linux kernel version (and nvme driver baseline) should we use to build and load the custom modules successfully?

Any additional setup notes (e.g., Module.symvers expectations or tested configurations) would be greatly appreciated.

[WWWzq-01]

Additional info: I previously tried compiling snvme/snvme-core against several other kernels. On those, the build failed because many NVMe functions were missing (API changed). With kernel 5.15.20-051520-generic the modules compile, but insmod still fails. dmesg shows symbol version mismatches vs. the in-tree nvme_core (“disagrees about version of symbol … / Unknown symbol … (err -22)”) and “snvme_core: exports duplicate symbol __nvme_check_ready (owned by nvme_core)”.

[qq502233945]

Hi ziqiang:
we don't use a ubuntu release kernel, we download the 5.15.0 on github and replace the orignal kernel, and it can handle the problem "Unknown symbol ". And you need an independent nvme device, because you need unbind and bind before run the program.

[WWWzq-01]

I installed the OS on a SATA SSD, and the previous “Unknown symbol” problem has been resolved. Now I’m trying to test GeminiFS, but I encountered the following issue:

sudo ./TestForNvmeBacking.exe   
mmap is 7f83e19df000
dstrd  is 0, ctrl->timeout is 30000, max qs is  4096
SQs: 64 CQs: 64 n_qps: 64
[ioctl_map] Page mapping kernel request failed (ptr=0x7f83b7a10000, n_pages=1): Bad address
terminate called after throwing an instance of 'std::runtime_error'
  what():  Failed to map device memory: Bad address

After launching the test program, the kernel module reports a page-mapping failure when trying to map the NVMe device memory. The process terminates with a std::runtime_error indicating Failed to map device memory: Bad address.