Hello, Christoph. On Sat, Jan 14, 2017 at 05:50:01PM +0100, Christoph Hellwig wrote: > Of course we can't just do a blind GFP_NOIO from ->queuecommand, > mea culpa. For most of the commands GFP_ATOMIC should be absolutely > fine as they are only called from the probe path. If you're > paranoid TRIM might want a sector-sized mempool, but other common > drivers like NVMe rely on plain GFP_ATOMIC allocations for those > as well. Let me know, and I'll send a mempool patch on top of it. Ugh... I don't know. What we had previously is always guaranteed to work. I'm not really liking the fact that we're adding a possibility of failure here. Even if we do mempool, we would still have to protect it with a spinlock as mempool only guarantees one allocation at a time. Until we have a better solution, can we please revert back to where we were at least for the buffers needed from atomic context? Thanks. -- tejun -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
