On 3/18/25 3:58 PM, Uday Shankar wrote: > On Tue, Mar 18, 2025 at 12:48:44PM -0600, Jens Axboe wrote: >>> though I don't see any examples of drivers using it. Would it be a bad >>> idea to try and reuse that? >> >> We can't reuse that one, and it's not for driver use - purely internal. >> But I _think_ you could easily grab space in the union that has the hash >> and ipi_list for it. And then you could dump needing this extra data per >> request. > > Another idea is to union the refcount with end_io_data, since that's > purely for the driver. But it might be a bit weird to tell drivers that > they can use either end_io/end_io_data or the refcount but not both... > not to mention it could be a nice exploit vector if drivers get it > wrong. > > Either way, I think this work is follow-on material and shouldn't be > rolled into this patch. For sure, that's what I said in my first reply too. It's just an idea for a further improvement, not gating this patch at all. -- Jens Axboe
