Update L0 provider to use getpidfd function

[vinser52]

Description

As we agreed, while L0 driver has not fixed IPC implementation, in UMF we will use getpidfd approach.
So in this PR does the following:

• updates L0 provider implementation.
• refactors OS memory provider and moved common functions to utils.
• refactors multi-process IPC test and created a version for L0 provider in addition to the test for OS provider.

Checklist

• Code compiles without errors locally
• New tests added, especially if they will fail without my changes
• All newly added source files have a license
• All newly added source files are referenced in CMake files
• All tests pass locally
• CI workflows execute properly

[ldorau]

The umf_example_ipc_ipcapi_anon_fd test fails

[vinser52]

The umf_example_ipc_ipcapi_anon_fd test fails

Yeah, looks like it fails because utils_duplicate_fd: __NR_pidfd_open or __NR_pidfd_getfd not available. It is strange, did this test pass before on Ubuntu 20.04?

[vinser52]

The umf_example_ipc_ipcapi_anon_fd test fails

Yeah, looks like it fails because utils_duplicate_fd: __NR_pidfd_open or __NR_pidfd_getfd not available. It is strange, did this test pass before on Ubuntu 20.04?

Ok, I understand the issue. Prior to this PR we returned Not supported if required syscalls are not available, This PR just returns UMF_RESULT_ERROR_MEMORY_PROVIDER_SPECIFIC. I will fix it to keep the current behavior.

[lplewa]

Please create errnoToUmfResult function and use it here

[vinser52]

Done

[lplewa]

Using memcpy to copy single int is an overkill

[vinser52]

Compiler optimizes it. When the size is low enough there is now an actual call to memcpy, it is replaced with move instructions.

[lplewa]

It's more code readability issue then performance. Assignment is shorter, and easier to read.

[vinser52]

regarding readability, I am not sure:

fd_remote = *(int*)&ze_ipc_handle;

vs

memcpy(&fd_remote, &ze_ipc_handle, sizeof(fd_remote));

and

*((int*)&ze_ipc_handle) = fd_local;

vs

memcpy(&ze_ipc_handle, &fd_local, sizeof(fd_local));

Are you still suggesting to get rid of memcpy? If so, I will do that. Please confirm.

[lplewa]

This IPC api in L0 is awful....

I checked definition of ze_ipc_mem_handle_t and as this is opaque struct, memcpy here make sense here.

NIT: We should consider something like this - as working on opaque structs without clear definition of it's layout is quite error prone.

union umf_ipc_mem_handle {
    ze_ipc_mem_handle_t ze;
    struct {
        int fd;
    } def;
};

[vinser52]

all this is needed just because L0 IPC implementation is broken.
According to the L0 spec all these steps not needed at all. When L0 driver fix their implementation we will get rid of these changes.

[ldorau]

@vinser52 Please rebase. There is a strange message here:
"This branch is out-of-date with the base branch
Merge the latest changes from main into this branch."

[lukaszstolarczuk]

@vinser52 Please rebase. There is a strange message here: "This branch is out-of-date with the base branch Merge the latest changes from main into this branch."

That's a new addition. I'm testing a few security options to fix the issue: #580 This precise option seem a little annoying (but perhaps maintainers can just do the update before merging) - I will keep investigating these security options

[vinser52]

@vinser52 Please rebase. There is a strange message here: "This branch is out-of-date with the base branch Merge the latest changes from main into this branch."

Done.

[ldorau]

@lplewa please respond to your issues