On Thu, Jan 19, 2017 at 03:34:58PM -0500, Tejun Heo wrote: > So, unless we allocate one mempool per port, we're gonna have to > synchronize around its use anyway. mempool can't guarantee allocation > to multiple users at the same time. If this is something which > affects scalability, I'm completley fine with making it per-port (or > device). Each ata_port carries 512 byte buffer anyway. Maybe we can > reuse that?
I'll play around with these things a bit more. For the slow path ops I think I can get away without any dynamic allocation at all - just use small on-stack buffers. For DSM / Write Same we rewrite into the buffer the SCSI layer provided us. This is a bit of an issue anyway as this might modify user data that is not expected to be rewritten or even mapped read-only. Maybe we need to kill off that emulation entirely and just have ATA DSM using the ATA_16 CDB as another provisioning option in sd.c. While that is a bit of a layering violation it would solve a lot of problems with the way TRIM is currently implemented. -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
