Re: [PATCH] MdeModulePkg/UfsPassThruDxe: Improve Error handling of Ufs Pass Thru driver


Wu, Hao A
 

Hello Liming,

I have a patch that would like to confirm with you that whether it can be merged into the upcoming edk2-stable202102 tag.

This is a feature request: https://bugzilla.tianocore.org/show_bug.cgi?id=3217
In the BZ tracker, there are 3 improvements mentioned for UfsPassThruDxe.
According to Purna, he would like to have 1 of the improvements (improvement #3 in BZ-3217) be merged and catch the stable tag.
I have given the 'R-b' tag for improvement #3 already (https://edk2.groups.io/g/devel/message/72121)

My thought is that we can break BZ-3217 into multiple feature requests:
1. BZ-3217: Updated its title and description to only cover improvement #3
2. File new BZ feature requests to cover improvement #1 & #2

What is your suggestion for this case? Thanks in advance.

Best Regards,
Hao Wu

-----Original Message-----
From: Bandaru, Purna Chandra Rao <purna.chandra.rao.bandaru@intel.com>
Sent: Tuesday, February 23, 2021 10:36 PM
To: Wu, Hao A <hao.a.wu@intel.com>; devel@edk2.groups.io
Cc: Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>
Subject: RE: [edk2-devel] [PATCH] MdeModulePkg/UfsPassThruDxe:
Improve Error handling of Ufs Pass Thru driver

Hi Wu, Hao A

I am trying to focus on merging patch#1 for now to unblock boot issues.
March 6th might be too late, May I request you to expedite any other
alternatives like exceptions/overrides?
For the remaining two patches I will get back to you with the plan after
discussing with WSIV and MVE teams on the protocol analyzer tools etc.

Thanks
~Purna

-----Original Message-----
From: Wu, Hao A <hao.a.wu@intel.com>
Sent: Tuesday, February 23, 2021 6:46 AM
To: Bandaru, Purna Chandra Rao <purna.chandra.rao.bandaru@intel.com>;
devel@edk2.groups.io
Cc: Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>
Subject: RE: [edk2-devel] [PATCH] MdeModulePkg/UfsPassThruDxe:
Improve Error handling of Ufs Pass Thru driver

-----Original Message-----
From: Bandaru, Purna Chandra Rao
<purna.chandra.rao.bandaru@intel.com>
Sent: Tuesday, February 23, 2021 1:11 AM
To: Wu, Hao A <hao.a.wu@intel.com>; devel@edk2.groups.io
Cc: Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>
Subject: RE: [edk2-devel] [PATCH] MdeModulePkg/UfsPassThruDxe:
Improve Error handling of Ufs Pass Thru driver

Thank you Hai Bu for the response.

I have broken this into three separate patches. There were no specific
recommendation in the speciation for seen multiple issues on all the
UFS platforms like LKF, ADP-P and EHK.

Hello,

After quickly going through the new series sent, I do not see my previous
inline comments and questions get addressed.
Could you please help to provide your feedbacks and update the patches?


And these changes worked on all the three with various UFS cards.
Can you please review and help to get this changes at the earliest in
master as well as Downstream/master.

Sorry, since there is an upcoming stable tag approaching, at this moment, I
prefer to hold this feature after the stable tag (March 6th).

Best Regards,
Hao Wu



Thanks
~Purna

-----Original Message-----
From: Wu, Hao A <hao.a.wu@intel.com>
Sent: Monday, February 22, 2021 2:10 PM
To: devel@edk2.groups.io; Wu, Hao A <hao.a.wu@intel.com>; Bandaru,
Purna Chandra Rao <purna.chandra.rao.bandaru@intel.com>
Cc: Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>
Subject: RE: [edk2-devel] [PATCH] MdeModulePkg/UfsPassThruDxe:
Improve Error handling of Ufs Pass Thru driver

-----Original Message-----
From: devel@edk2.groups.io <devel@edk2.groups.io> On Behalf Of Wu,
Hao
A
Sent: Monday, February 22, 2021 4:38 PM
To: Bandaru, Purna Chandra Rao
<purna.chandra.rao.bandaru@intel.com>;
devel@edk2.groups.io
Cc: Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>
Subject: Re: [edk2-devel] [PATCH] MdeModulePkg/UfsPassThruDxe:
Improve Error handling of Ufs Pass Thru driver

-----Original Message-----
From: Bandaru, Purna Chandra Rao
<purna.chandra.rao.bandaru@intel.com>
Sent: Wednesday, February 17, 2021 5:02 PM
To: devel@edk2.groups.io
Cc: Bandaru, Purna Chandra Rao
<purna.chandra.rao.bandaru@intel.com>;
Albecki, Mateusz <mateusz.albecki@intel.com>; Ni, Ray
<ray.ni@intel.com>; Wu, Hao A <hao.a.wu@intel.com>
Subject: [PATCH] MdeModulePkg/UfsPassThruDxe: Improve Error
handling
of Ufs Pass Thru driver

From: Bandaru <purna.chandra.rao.bandaru@intel.com>

https://bugzilla.tianocore.org/show_bug.cgi?id=3217

Following is the brief description of the changes
1) There are cards that can take upto 600ms for Init and hence increase
the time out for fDeviceInit polling loop.
2) Add UFS host conctroller reset in the last retry of Link start up.
3) Retry sending NOP OUT command upto 10 times

Hello Bandaru,

Could you help to break this patch into a 3-patch series in V2?
With each patch handling just one of the above 3 improvements
mentioned.

For improvement 2) above, I do not see such UFS host controller
re-enabling process being mentioned in UFSHCI 3.0 spec section 7.1.1.
Is this process being documented somewhere else in the spec or
suggested by device vender?

Sorry for missing one comment.
Could you help to add the information on what kind of tests have been
performed for the code changes?

Thanks in advance.

Best Regards,
Hao Wu



More inline comments below:



Signed-off-by: Bandaru <purna.chandra.rao.bandaru@intel.com>
Cc: Mateusz Albecki <mateusz.albecki@intel.com>
Cc: Ray Ni <ray.ni@intel.com>
Cc: Hao A Wu <hao.a.wu@intel.com>

Change-Id: I6c0dbc1c147487e51f0ed5f2425957ae089b0160
---
MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThru.c | 26
+++++++++++++++++++++-----
MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThruHci.c | 18
++++++++++++------
2 files changed, 33 insertions(+), 11 deletions(-)

diff --git a/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThru.c
b/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThru.c
index 9768c2e6fb..89048745be 100644
--- a/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThru.c
+++ b/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThru.c
@@ -1,6 +1,6 @@
/** @file

- Copyright (c) 2014 - 2019, Intel Corporation. All rights
reserved.<BR>
+ Copyright (c) 2014 - 2021, Intel Corporation. All rights
+ reserved.<BR>
Copyright (c) Microsoft Corporation.<BR>
SPDX-License-Identifier: BSD-2-Clause-Patent

@@ -749,7 +749,7 @@ UfsFinishDeviceInitialization ( {
EFI_STATUS Status;
UINT8 DeviceInitStatus;
- UINT8 Timeout;
+ UINT16 Timeout;

DeviceInitStatus = 0xFF;

@@ -761,17 +761,23 @@ UfsFinishDeviceInitialization (
return Status;
}

- Timeout = 5;
+ Timeout = 6000; //There are cards that can take upto 600ms.

Please help to add a macro in file UfsPassThru.h:
#define UFS_INIT_COMPLETION_TIMEOUT 6000 And use the macro here.

Also a minor comment, could you help to use the below comment format?
//
// There are UFS devices that can take up to 600ms to clear the
fDeviceInit flag // Timeout = UFS_INIT_COMPLETION_TIMEOUT;


do {
+ MicroSecondDelay (100); //Give 100 us and then start polling.

For the above delay movement, do you observe any side effect for the
origin code?
If not, I prefer to leave the origin behavior:
do {
UfsReadFlag();
...
MicroSecondDelay (1);
} while (...)
since doing so will have the least performance penalty for devices
that respond fast.


Status = UfsReadFlag (Private, UfsFlagDevInit, &DeviceInitStatus);
if (EFI_ERROR (Status)) {
return Status;
}
- MicroSecondDelay (1);
Timeout--;
} while (DeviceInitStatus != 0 && Timeout != 0);

+ if (Timeout == 0) {
+ DEBUG ((DEBUG_ERROR, "UfsFinishDeviceInitialization
DeviceInitStatus=%x EFI_TIMEOUT \n", DeviceInitStatus));
+ return EFI_TIMEOUT;
+ } else {
+ DEBUG ((DEBUG_INFO, "UfsFinishDeviceInitialization Timeout
+ left=%x EFI_SUCCESS \n", Timeout));
return EFI_SUCCESS;

Please help to add two spaces for text alignment in the above line.


+ }
}

/**
@@ -905,9 +911,19 @@ UfsPassThruDriverBindingStart (
// At the end of the UFS Interconnect Layer initialization on
both host and device side,
// the host shall send a NOP OUT UPIU to verify that the device
UTP Layer is ready.
//

For the NOP OUT - NOP IN improvement, could you help to provide more
information on what is the current issue for some devices?
Is it a timeout happened for:
Status = UfsWaitMemSet (Private, UFS_HC_UTRLDBR_OFFSET, BIT0 <<
Slot, 0, UFS_TIMEOUT); (If so, have you tried increasing the last
parameter like
'10*UFS_TIMEOUT'?) Or the case is that NopInUpiu->Resp has a
non-zero value?

I found that in the UFS 3.0 spec:
|> For some implementations, the device UTP layer may not be
|> initialized yet, therefore the device may not respond promptly to
|> NOP OUT UPIU sending NOP IN UPIU.
|> The host waits until it receives the NOP IN UPIU from the device...
And there is no mention for the retry scheme.


+ for (Index = 10; Index > 0; Index--) {
Status = UfsExecNopCmds (Private);
if (EFI_ERROR (Status)) {
- DEBUG ((DEBUG_ERROR, "Ufs Sending NOP IN command Error,
Status
= %r\n", Status));
+ DEBUG ((DEBUG_ERROR, "Ufs Sending NOP IN command Error,
Index
= %x Status = %r\n", Index, Status));
+ MicroSecondDelay (100); //100 us
+ continue;
+ } else {
+ DEBUG ((DEBUG_INFO, "Ufs Sent NOP OUT successfully and
+ received
NOP IN, Status = %r\n", Status));
+ break;
+ }
+ }
+ if (!Index) {
+ DEBUG ((DEBUG_INFO, "NOP OUT failed all the 10 times Status =
+ %r\n", Status));
goto Error;
}

diff --git a/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThruHci.c
b/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThruHci.c
index 0b1030ab47..4fa5689196 100644
--- a/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThruHci.c
+++ b/MdeModulePkg/Bus/Ufs/UfsPassThruDxe/UfsPassThruHci.c
@@ -2,7 +2,7 @@
UfsPassThruDxe driver is used to produce EFI_EXT_SCSI_PASS_THRU
protocol interface
for upper layer application to execute UFS-supported SCSI cmds.

- Copyright (c) 2014 - 2019, Intel Corporation. All rights
reserved.<BR>
+ Copyright (c) 2014 - 2021, Intel Corporation. All rights
+ reserved.<BR>
Copyright (c) Microsoft Corporation.<BR>
SPDX-License-Identifier: BSD-2-Clause-Patent

@@ -1929,17 +1929,15 @@ UfsDeviceDetection (

//
// Start UFS device detection.
- // Try up to 3 times for establishing data link with device.
+ // Try up to 4 times for establishing data link with device.
//
- for (Retry = 0; Retry < 3; Retry++) {
+ for (Retry = 0; Retry < 4; Retry++) {

Please introduce a macro in file UfsPassThru.h:
#define UFS_LINK_STARTUP_RETRIES 4
And use the macro here.

Also, is it necessary to increase the retry number by 1?
Or the device can be successfully brought up by adding a host
controller re- enabling?


LinkStartupCommand.Opcode = UfsUicDmeLinkStartup;
LinkStartupCommand.Arg1 = 0;
LinkStartupCommand.Arg2 = 0;
LinkStartupCommand.Arg3 = 0;
Status = UfsExecUicCommands (Private, &LinkStartupCommand);
- if (EFI_ERROR (Status)) {
- return EFI_DEVICE_ERROR;
- }

Will the DME_LINKSTARTUP command execution fail at first and then
succeed after retry?
If not, I prefer to keep the origin code logic to return error status directly.


+ if (!EFI_ERROR (Status)) {

Status = UfsMmioRead32 (Private, UFS_HC_STATUS_OFFSET, &Data);
if (EFI_ERROR (Status)) {
@@ -1960,6 +1958,14 @@ UfsDeviceDetection (
}
}
return EFI_SUCCESS;
+ }
+ }
+ if (Retry == 2) {

Please help to update to:
if (Retry == UFS_LINK_STARTUP_RETRIES - 1) {

And add comments like:
//
// Try re-enabling the UFS host controller in the last retry attempt
//


Best Regards,
Hao Wu


+ Status = UfsEnableHostController (Private);
+ if (EFI_ERROR (Status)) {
+ DEBUG ((DEBUG_ERROR, "UfsDeviceDetection: Enable Host
Controller
Fails, Status = %r\n", Status));
+ return Status;
+ }
}
}

--
2.16.2.windows.1



Join devel@edk2.groups.io to automatically receive all group messages.