[v11,0/4] scsi: ufs: Add Host Performance Booster Support

Message ID 231786897.01599016802080.JavaMail.epsvc@epcpadp1
Headers show
Series
  • scsi: ufs: Add Host Performance Booster Support
Related show

Message

Daejun Park Sept. 2, 2020, 3:17 a.m.
Changelog:

v10 -> v11
Add a newline at end the last line on Kconfig file.

v9 -> v10
1. Fix 64-bit division error
2. Fix problems commentted in Bart's review.

v8 -> v9
1. Change sysfs initialization.
2. Change reading descriptor during HPB initialization
3. Fix problems commentted in Bart's review.
4. Change base commit from 5.9/scsi-queue to 5.10/scsi-queue.

v7 -> v8
Remove wrongly added tags.

v6 -> v7
1. Remove UFS feature layer.
2. Cleanup for sparse error.

v5 -> v6
Change base commit to b53293fa662e28ae0cdd40828dc641c09f133405

v4 -> v5
Delete unused macro define.

v3 -> v4
1. Cleanup.

v2 -> v3
1. Add checking input module parameter value.
2. Change base commit from 5.8/scsi-queue to 5.9/scsi-queue.
3. Cleanup for unused variables and label.

v1 -> v2
1. Change the full boilerplate text to SPDX style.
2. Adopt dynamic allocation for sub-region data structure.
3. Cleanup.

NAND flash memory-based storage devices use Flash Translation Layer (FTL)
to translate logical addresses of I/O requests to corresponding flash
memory addresses. Mobile storage devices typically have RAM with
constrained size, thus lack in memory to keep the whole mapping table.
Therefore, mapping tables are partially retrieved from NAND flash on
demand, causing random-read performance degradation.

To improve random read performance, JESD220-3 (HPB v1.0) proposes HPB
(Host Performance Booster) which uses host system memory as a cache for the
FTL mapping table. By using HPB, FTL data can be read from host memory
faster than from NAND flash memory. 

The current version only supports the DCM (device control mode).
This patch consists of 3 parts to support HPB feature.

1) HPB probe and initialization process
2) READ -> HPB READ using cached map information
3) L2P (logical to physical) map management

In the HPB probe and init process, the device information of the UFS is
queried. After checking supported features, the data structure for the HPB
is initialized according to the device information.

A read I/O in the active sub-region where the map is cached is changed to
HPB READ by the HPB.

The HPB manages the L2P map using information received from the
device. For active sub-region, the HPB caches through ufshpb_map
request. For the in-active region, the HPB discards the L2P map.
When a write I/O occurs in an active sub-region area, associated dirty
bitmap checked as dirty for preventing stale read.

HPB is shown to have a performance improvement of 58 - 67% for random read
workload. [1]

This series patches are based on the 5.9/scsi-queue branch.

[1]:
https://www.usenix.org/conference/hotstorage17/program/presentation/jeong

Daejun park (4):
 scsi: ufs: Add HPB feature related parameters
 scsi: ufs: Introduce HPB feature
 scsi: ufs: L2P map management for HPB read
 scsi: ufs: Prepare HPB read for cached sub-region
 
 drivers/scsi/ufs/Kconfig  |   10 +
 drivers/scsi/ufs/Makefile |    1 +
 drivers/scsi/ufs/ufs.h    |   47 +
 drivers/scsi/ufs/ufshcd.c |   60 ++
 drivers/scsi/ufs/ufshcd.h |   23 +-
 drivers/scsi/ufs/ufshpb.c | 1845 ++++++++++++++++++++++++++++++++++++++++
 drivers/scsi/ufs/ufshpb.h |  229 +++++
 7 files changed, 2214 insertions(+), 1 deletion(-)
 created mode 100644 drivers/scsi/ufs/ufshpb.c
 created mode 100644 drivers/scsi/ufs/ufshpb.h

Comments

Daejun Park Sept. 15, 2020, 11:05 p.m. | #1
Hi All,

I want to know how to improve this patch.
Could you review this patch?

Martin,

Can we move forward with this one?

Thanks,
Daejun

> Changelog:

> 

> v10 -> v11

> Add a newline at end the last line on Kconfig file.

> 

> v9 -> v10

> 1. Fix 64-bit division error

> 2. Fix problems commentted in Bart's review.

> 

> v8 -> v9

> 1. Change sysfs initialization.

> 2. Change reading descriptor during HPB initialization

> 3. Fix problems commentted in Bart's review.

> 4. Change base commit from 5.9/scsi-queue to 5.10/scsi-queue.

> 

> v7 -> v8

> Remove wrongly added tags.

> 

> v6 -> v7

> 1. Remove UFS feature layer.

> 2. Cleanup for sparse error.

> 

> v5 -> v6

> Change base commit to b53293fa662e28ae0cdd40828dc641c09f133405

> 

> v4 -> v5

> Delete unused macro define.

> 

> v3 -> v4

> 1. Cleanup.

> 

> v2 -> v3

> 1. Add checking input module parameter value.

> 2. Change base commit from 5.8/scsi-queue to 5.9/scsi-queue.

> 3. Cleanup for unused variables and label.

> 

> v1 -> v2

> 1. Change the full boilerplate text to SPDX style.

> 2. Adopt dynamic allocation for sub-region data structure.

> 3. Cleanup.

> 

> NAND flash memory-based storage devices use Flash Translation Layer (FTL)

> to translate logical addresses of I/O requests to corresponding flash

> memory addresses. Mobile storage devices typically have RAM with

> constrained size, thus lack in memory to keep the whole mapping table.

> Therefore, mapping tables are partially retrieved from NAND flash on

> demand, causing random-read performance degradation.

> 

> To improve random read performance, JESD220-3 (HPB v1.0) proposes HPB

> (Host Performance Booster) which uses host system memory as a cache for the

> FTL mapping table. By using HPB, FTL data can be read from host memory

> faster than from NAND flash memory. 

> 

> The current version only supports the DCM (device control mode).

> This patch consists of 3 parts to support HPB feature.

> 

> 1) HPB probe and initialization process

> 2) READ -> HPB READ using cached map information

> 3) L2P (logical to physical) map management

> 

> In the HPB probe and init process, the device information of the UFS is

> queried. After checking supported features, the data structure for the HPB

> is initialized according to the device information.

> 

> A read I/O in the active sub-region where the map is cached is changed to

> HPB READ by the HPB.

> 

> The HPB manages the L2P map using information received from the

> device. For active sub-region, the HPB caches through ufshpb_map

> request. For the in-active region, the HPB discards the L2P map.

> When a write I/O occurs in an active sub-region area, associated dirty

> bitmap checked as dirty for preventing stale read.

> 

> HPB is shown to have a performance improvement of 58 - 67% for random read

> workload. [1]

> 

> This series patches are based on the 5.9/scsi-queue branch.

> 

> [1]:

> https://www.usenix.org/conference/hotstorage17/program/presentation/jeong

> 

> Daejun park (4):

>  scsi: ufs: Add HPB feature related parameters

>  scsi: ufs: Introduce HPB feature

>  scsi: ufs: L2P map management for HPB read

>  scsi: ufs: Prepare HPB read for cached sub-region

>  

>  drivers/scsi/ufs/Kconfig  |   10 +

>  drivers/scsi/ufs/Makefile |    1 +

>  drivers/scsi/ufs/ufs.h    |   47 +

>  drivers/scsi/ufs/ufshcd.c |   60 ++

>  drivers/scsi/ufs/ufshcd.h |   23 +-

>  drivers/scsi/ufs/ufshpb.c | 1845 ++++++++++++++++++++++++++++++++++++++++

>  drivers/scsi/ufs/ufshpb.h |  229 +++++

>  7 files changed, 2214 insertions(+), 1 deletion(-)

>  created mode 100644 drivers/scsi/ufs/ufshpb.c

>  created mode 100644 drivers/scsi/ufs/ufshpb.h
Christoph Hellwig Sept. 16, 2020, 5:22 a.m. | #2
On Wed, Sep 16, 2020 at 08:05:17AM +0900, Daejun Park wrote:
> Hi All,

> 

> I want to know how to improve this patch.


Drop it and fix the actual UFS feature to not be so horrible?
Daejun Park Sept. 17, 2020, 12:25 a.m. | #3
> > Hi All,

> > 

> > I want to know how to improve this patch.

> 

> Drop it and fix the actual UFS feature to not be so horrible?

> 


Hello Christoph,

Thanks for the comment.

The HPB is essential feature for mobile devices.

NAND-based storage needs logical to physical mapping, which is cached in
the storage to minimize translation overhead. UFS is a NAND-based storage
device with limited resources, mainly used for mobile devices. Typically,
SSD has enough cache space for mapping information in SSD, but UFS has not
enough.
So, UFS stores L2P mapping in NAND and performs IO using demand loading 
for translation. Due to overhead of demand loading, it degrades random read
performance.

The HPB is a feature which uses host memory to relieve this problem. By 
using the HPB feature, UFS can provide improved random read performance
without mapping data thrashing problem. Therefore, the HPB is currently
already included in the Linux kernel code in android devices, however it 
is maintained as out-of-tree.

While upstreaming HPB feature to mainline kernel, we received various
comments from several reviewers (thanks!) and the HPB feature can be
improved. I think it would be good to make mainline the feature that are
mainly used.

Thanks,
Daejun
Bart Van Assche Sept. 18, 2020, 7:51 p.m. | #4
On 9/15/20 10:22 PM, Christoph Hellwig wrote:
> On Wed, Sep 16, 2020 at 08:05:17AM +0900, Daejun Park wrote:

>> Hi All,

>>

>> I want to know how to improve this patch.

> 

> Drop it and fix the actual UFS feature to not be so horrible?


A new UFS specification could be defined and could be implemented for future UFS
devices. I think this patch series is intended to support the millions of UFS
devices that are already in use, e.g. in Android smartphones.

Thanks,

Bart.
Daejun Park Sept. 21, 2020, 8:03 a.m. | #5
>> On Wed, Sep 16, 2020 at 08:05:17AM +0900, Daejun Park wrote:

>>> Hi All,

>>>

>>> I want to know how to improve this patch.

>> 

>> Drop it and fix the actual UFS feature to not be so horrible?

> 

> A new UFS specification could be defined and could be implemented for future UFS

> devices. I think this patch series is intended to support the millions of UFS

> devices that are already in use, e.g. in Android smartphones.


It is important point. 

I am not suggesting a strange feature, but want to make mainline the HPB
feature that is already widely used in Android smartphones.

Thanks,

Daejun.