On Sat, 1 Oct 2022, Greg KH wrote:
On Fri, Sep 30, 2022 at 11:32:30AM -0400, Mikulas Patocka wrote:
Hi
Here I'm submitting backport of patches 8238b4579866b7c1bb99883cfe102a43db5506ff and d6ffe6067a54972564552ea45d320fb98db1ac5e to the stable branches.
Thanks, but you provide no information as to why these are needed.
What needs them? They are just adding new functions to the tree from what I can tell.
thanks,
greg k-h
There's a race condition in wait_on_bit. wait_on_bit tests a bit using the "test_bit" function, however this function doesn't do any memory barrier, so the memory accesses that follow wait_on_bit may be reordered before it and return invalid data.
Linus didn't want to add a memory barrier to wait_on_bit, he instead wanted to introduce a new function test_bit_acquire that performs the "acquire" memory barrier and use it in wait_on_bit.
The patch d6ffe6067a54972564552ea45d320fb98db1ac5e fixes an oversight in the patch 8238b4579866b7c1bb99883cfe102a43db5506ff where the function test_bit_acquire was not defined for some architectures and this caused compile failure.
The backport of the patch 8238b4579866b7c1bb99883cfe102a43db5506ff should be applied first and the backport of the patch d6ffe6067a54972564552ea45d320fb98db1ac5e afterwards.
Mikulas