Re: [PATCH 2/3] tools: testing: Use existing atomic.h for vma/radix-tree tests

[Pedro Falcato <pfalcato@suse.de>]

On Wed, Aug 27, 2025 at 11:04:42AM +0000, Brendan Jackman wrote:
> The shared userspace logic used for unit-testing radix-tree and VMA code
> currently has its own replacements for atomics helpers. This is not
> needed as the necessary APIs already have userspace implementations in
> the tools tree. Switching over to that allows deleting a bit of code.
> 
> Note that the implementation is different; while the version being
> deleted here is implemented using liburcu, the existing version in tools
> uses either x86 asm or compiler builtins. It's assumed that both are
> equally likely to be correct.
> 
> The tools tree's version of atomic_t is a struct type while the version
> being deleted was just a typedef of an integer. This means it's no
> longer valid to call __sync_bool_compare_and_swap() directly on it. One
> option would be to just peek into the struct and call it on the field,
> but it seems a little cleaner to just use the corresponding atomic.h
> API. On non-x86 archs this is implemented using
> __sync_val_compare_and_swap(). It's not clear why the old version uses
> the bool variant instead of the generic "val" one, for now it's assumed
> that this was a mistake.
>

I don't think it's a mistake. Namely we're checking if the cmpxchg occured.
So in the new version you'll have trouble incrementing i_mmap_writeable from
0 to 1, where in practice you should (AIUI) see 0 -> 1 (old val = 0, retry) -> 2,
which is obviously not correct here.

At the very least you'll need some:

do {
} while(atomic_cmpxchg(&mapping->i_mmap_writeable, c, c+1) != c);

to keep the same semantics.

-- 
Pedro