On 2021-01-12 15:45:12 [+0000], Christoph Hellwig wrote: > What is the problem with simply adding a gfp_t argument to the existing > calls? The end result of this series looks fine, but the way we get > there looks extremely cumbersome. Maybe I don't understand you fully but if you want to avoid adding the two _gftp functions (+ remove the other & rename at the end of series) and passing the gfp_t argument right away then this what I had in my inbox at the very beginning. It was one big patch with a long description of the relevant code paths and why it is the way it is. Since the two functions are used by many drivers you had to patch all at once. So I suggested to split in smaller chunks to make it easier to review (and bisect) and at the end the old functions can be removed once all users are gone (and rename if the maintainer wishes). Once we had the individual patches for driver/folder it was easier to review them. Then we also identified the first few patches which got the Fixes: tag because in_interrupt() didn't take disabled interrupts into account. Sebastian
