mbox series

[v5,0/3] scsi: ufs: Allow RTT negotiation

Message ID 20240523125827.818-1-avri.altman@wdc.com
Headers show
Series scsi: ufs: Allow RTT negotiation | expand

Message

Avri Altman May 23, 2024, 12:58 p.m. UTC 
The rtt-upiu packets precede any data-out upiu packets, thus
synchronizing the data input to the device: this mostly applies to write
operations, but there are other operations that requires rtt as well.

There are several rules binding this rtt - data-out dialog, specifically
There can be at most outstanding bMaxNumOfRTT such packets.  This might
have an effect on write performance (sequential write in particular), as
each data-out upiu must wait for its rtt sibling.

UFSHCI expects bMaxNumOfRTT to be min(bDeviceRTTCap, NORTT). However,
as of today, there does not appear to be no-one who sets it: not the
host controller nor the driver.  It wasn't an issue up to now:
bMaxNumOfRTT is set to 2 after manufacturing, and wasn't limiting the
write performance.

UFS4.0, and specifically gear 5 changes this, and requires the device to
be more attentive.  This doesn't come free - the device has to allocate
more resources to that end, but the sequential write performance
improvement is significant. Early measurements shows 25% gain when
moving from rtt 2 to 9. Therefore, set bMaxNumOfRTT to be
min(bDeviceRTTCap, NORTT) as UFSHCI expects.

v4 -> v5:
Quiesce the queues before writing bMaxNumOfRTT (Bart)
Make bDeviceRTTCap available in ufshcd_device_params_init() (Bart)

v3 -> v4:
Allow bMaxNumOfRTT to be configured via sysfs (Bart)

v2 -> v3:
Allow platform vendors to take precedence having their own rtt
negotiation mechanism (Peter)

v1 -> v2:
bMaxNumOfRTT is a Persistent attribute - do not override if it was
written (Bean)

Avri Altman (3):
  scsi: ufs: Allow RTT negotiation
  scsi: ufs: Allow platform vendors to set rtt
  scsi: ufs: sysfs: Make max_number_of_rtt read-write

 Documentation/ABI/testing/sysfs-driver-ufs | 14 +++--
 drivers/ufs/core/ufs-sysfs.c               | 72 +++++++++++++++++++++-
 drivers/ufs/core/ufshcd-priv.h             | 12 ++++
 drivers/ufs/core/ufshcd.c                  | 53 ++++++++++++----
 include/ufs/ufs.h                          |  2 +
 include/ufs/ufshcd.h                       |  4 ++
 include/ufs/ufshci.h                       |  1 +
 7 files changed, 139 insertions(+), 19 deletions(-)

Comments

Avri Altman May 24, 2024, 3:54 a.m. UTC | #1 
> 
> On 5/23/24 05:58, Avri Altman wrote:
> > +     /* make sure that there are no outstanding requests when rtt is set */
> > +     ufshcd_scsi_block_requests(hba);
> > +     blk_mq_wait_quiesce_done(&hba->host->tag_set);
> > +
> > +     ret = ufshcd_query_attr(hba, UPIU_QUERY_OPCODE_WRITE_ATTR,
> > +             QUERY_ATTR_IDN_MAX_NUM_OF_RTT, 0, 0, &rtt);
> > +
> > +     ufshcd_scsi_unblock_requests(hba);
> 
> The above doesn't look correct to me. ufshcd_scsi_block_requests() does not
> guarantee that all pending commands have finished by the time it returns.
> Please blk_mq_freeze_queue() / blk_mq_unfreeze_queue() instead.
Thanks.  Will use those to drain the queues.

Thanks,
Avri

> 
> Thanks,
> 
> Bart.