Advantages of io_uring_queue_init_mem

[AomaYple]

The io_uring_queue_init_mem seems to allocate memory by the user and then put multiple rings under the same block of memory.

What is the advantage of this? Suppose it's meant to be cache friendly, which means that an extra-large block of memory must be allocated beforehand to hold multiple rings.

How does it scale if more rings are subsequently needed (e.g., based on the number of threads created)? (Memory reallocation would invalidate the previous rings)

[isilence]

The io_uring_queue_init_mem seems to allocate memory by the user and then put multiple rings under the same block of memory.

CQ/SQ rings of different io_uring instances are not shared. Passing the same chunk of memory to multiple io_uring_queue_init_mem (without killing previous instances) would be a bug. However, you can chunk, let's say, a huge page into non-overlapping buffers and pass them to different rings.

What is the advantage of this? Suppose it's meant to be cache friendly, which means that an extra-large block of memory must be allocated beforehand to hold multiple rings.

Right, the manual tells about it.

# man io_uring_queue_init_mem.3

... Typically, the caller should allocate a huge page and pass that in to io_uring_queue_init_mem(3) ...

How does it scale if more rings are subsequently needed (e.g., based on the number of threads created)? (Memory reallocation would invalidate the previous rings)

Not sure what you mean here, but it's up to the user how it allocates the memory and manages it.