Accelerate Storage Resume (2x or more)
Get series with b4
b4 am [email protected]Versions newest first · linked by author and title
- v6 Latest3 patches 3 repliesIn review
- v53 patches 24 repliesSuperseded
Changes in v6 from the cover letter
* Per Tejun, fixed up the naming of the ata_port helper routines
* Cleaned up the return value of ata_port_pm helper routines to be 'void'
* Per Tejun, added a check of whether async scanning is disabled to gate whether async resume will be performed out of caution for systems that do not handle staggered spin-up outside of the kernel. Clarified this in the changelog and added commentary in the code.
Patches in v6 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/3 | libata, libsas: kill pm_result and related cleanupdrivers/ drivers/ drivers/ +2 more | +71 | 3 replies | |
| 2/3 | libata: async resumedrivers/ | +1 | 0 replies | |
| 3/3 | scsi: async sd resumedrivers/ drivers/ drivers/ +3 more | +109 | 0 replies |
Reviews trailers per patch
1 of 3 patches has a Reviewed-by or Acked-by.
| Patch | Alan Stern |
|---|---|
| 1/3 | |
| 2/3 | |
| 3/3 | A-b |
Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.
Cover letter [PATCH v6 0/3] Accelerate Storage Resume (2x or more)
Changes since v5:
* Per Tejun, fixed up the naming of the ata_port helper routines
* Cleaned up the return value of ata_port_pm helper routines to be 'void'
* Per Tejun, added a check of whether async scanning is disabled to gate whether async resume will be performed out of caution for systems that do not handle staggered spin-up outside of the kernel. Clarified this in the changelog and added commentary in the code.
===
It is not everyday that a kernel operation gets > 2x faster! Great find by Todd and his AnalyzeSuspend tool [1].
Todd is on vacation so I am taking care of these patches to make sure they get in the queue for 3.15.
The significant changes from Todd's last submission [2]:
1/ Split out the pure cleanup into its own patch, and expand it to clean
up libsas as well
2/ Move the entirety of scsi_device resume to an async context. As
written, v4 of the patchset overlapped the scsi-start-stop command with the scsi_device_resume(). Now in v5 the queue restart is properly ordered with the completion of the scsi-start-stop command.
Resume completion time as reported by:
"echo devices > /sys/power/pm_test && echo mem > /sys/power/state"
BEFORE: PM: resume of devices complete after 2271.097 msecs AFTER: PM: resume of devices complete after 1057.404 msecs
On a system with two SSDs attached via AHCI.
[1]: Todd's testing showing up to 12x resume latency improvement in some cases:
https://01.org/suspendresume/blogs/tebrandt/2013/hard-disk-resume-optimization-simpler-approach
[2]: Todd's v4 patchset:
[PATCH v4 0/2] http://marc.info/?l=linux-ide&m=138984698103487&w=2
[PATCH v4 1/2] http://marc.info/?l=linux-ide&m=138984713203515&w=2
[PATCH v4 2/2] http://marc.info/?l=linux-ide&m=138984722003539&w=2
---
Dan Williams (2):
libata, libsas: kill pm_result and related cleanup
scsi: async sd resume
Todd Brandt (1):
libata: async resume
drivers/ata/libata-core.c | 135 ++++++++++++++++++----------------------- drivers/ata/libata-eh.c | 13 +--- drivers/scsi/Kconfig | 3 + drivers/scsi/libsas/sas_ata.c | 35 ++--------- drivers/scsi/scsi.c | 3 + drivers/scsi/scsi_pm.c | 128 ++++++++++++++++++++++++++++++--------- drivers/scsi/scsi_priv.h | 2 + drivers/scsi/scsi_scan.c | 2 - drivers/scsi/sd.c | 1 include/linux/libata.h | 11 +-- include/scsi/libsas.h | 1 11 files changed, 180 insertions(+), 154 deletions(-)