On 7/16/21 5:39 AM, Avri Altman wrote:
But does your platform make use of REG_UTP_TRANSFER_REQ_LIST_COMPL? With 60k IOPS I suspect it doesn't, and the comparison is irrelevant.
Yes, my test setup supports the UTRLCNR register (let's use the acronyms from the UFS specification). However, it is not clear to me why that register has been introduced? Using that register in the completion path requires one register read + one register write while using the doorbell register in the completion path involves a single register read. My understanding is that register reads and writes are much slower than DRAM reads or writes. Does this mean that using the doorbell register in the compleiton path is always faster than using the UTRLCNR register?
Thanks, Bart.
