Re: [PATCH edk2-platforms v1 07/38] Silicon/Hisilicon/D06: Wait for all disk ready

[Leif Lindholm <leif.lindholm@linaro.org>]

On Wed, Aug 08, 2018 at 07:44:36PM +0800, Ming wrote:
> 
> 
> 在 8/8/2018 5:59 PM, Leif Lindholm 写道:
> > On Wed, Aug 08, 2018 at 05:02:15PM +0800, Ming wrote:
> >> 在 8/3/2018 1:36 AM, Leif Lindholm 写道:
> >>> On Tue, Jul 24, 2018 at 03:08:51PM +0800, Ming Huang wrote:
> >>>> This patch is relative to D06 SasDxe driver. The SasDxe set a
> >>>> variable to notice this libray. Here Wait for all disk ready
> >>>> for 30S at most.
> >>>>
> >>>> Contributed-under: TianoCore Contribution Agreement 1.1
> >>>> Signed-off-by: Ming Huang <ming.huang@linaro.org>
> >>>> Signed-off-by: Heyi Guo <heyi.guo@linaro.org>
> >>>> ---
> >>>>  Silicon/Hisilicon/HisiPkg.dec                                               |  1 +
> >>>>  Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBm.c               | 43 ++++++++++++++++++++
> >>>>  Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBootManagerLib.inf |  2 +
> >>>>  3 files changed, 46 insertions(+)
> >>>>
> >>>> diff --git a/Silicon/Hisilicon/HisiPkg.dec b/Silicon/Hisilicon/HisiPkg.dec
> >>>> index 35bea970ec..b56a6a6af7 100644
> >>>> --- a/Silicon/Hisilicon/HisiPkg.dec
> >>>> +++ b/Silicon/Hisilicon/HisiPkg.dec
> >>>> @@ -45,6 +45,7 @@
> >>>>  
> >>>>    gHisiEfiMemoryMapGuid  = {0xf8870015, 0x6994, 0x4b98, {0x95, 0xa2, 0xbd, 0x56, 0xda, 0x91, 0xc0, 0x7f}}
> >>>>    gVersionInfoHobGuid = {0xe13a14c, 0x859c, 0x4f22, {0x82, 0xbd, 0x18, 0xe, 0xe1, 0x42, 0x12, 0xbf}}
> >>>> +  gHisiOemVariableGuid = {0xac62b9a5, 0x9939, 0x41d3, {0xff, 0x5c, 0xc5, 0x80, 0x32, 0x7d, 0x9b, 0x29}}
> >>>>    gOemBootVariableGuid = {0xb7784577, 0x5aaf, 0x4557, {0xa1, 0x99, 0xd4, 0xa4, 0x2f, 0x45, 0x06, 0xf8}}
> >>>
> >>> What is the difference between gHisiOemVariableGuid and gOemBootVariableGuid?
> >>>
> >>
> >> Just a guid using to a variable.
> >> Do you suggest to use the old one?
> > 
> > Well, I am unsure about the intended scope for either. So I don't
> > know. How is HisiOem different from Oem? Can you explain the structure?
> 
> HisiOem and Oem are the same scope, just need a guid for SetVariable.

The only purpose for the GUID is so that you can have variables of the
same name in different modules without risking clashes. Sharing the
same variable GUID across multiple modules is perfectly fine (and
often preferable) when those modules are closely related.

But as I think this was only used for the communication with the SAS
controller driver, I think maybe this one can just go.

> >>>>    gEfiHisiSocControllerGuid = {0xee369cc3, 0xa743, 0x5382, {0x75, 0x64, 0x53, 0xe4, 0x31, 0x19, 0x38, 0x35}}
> >>>>  
> >>>> diff --git a/Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBm.c b/Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBm.c
> >>>> index 7dd5ba615c..f7536bfea3 100644
> >>>> --- a/Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBm.c
> >>>> +++ b/Silicon/Hisilicon/Library/PlatformBootManagerLib/PlatformBm.c
> >>>> @@ -20,6 +20,7 @@
> >>>>  #include <Library/BmcConfigBootLib.h>
> >>>>  #include <Library/DevicePathLib.h>
> >>>>  #include <Library/PcdLib.h>
> >>>> +#include <Library/TimerLib.h>
> >>>>  #include <Library/UefiBootManagerLib.h>
> >>>>  #include <Library/UefiLib.h>
> >>>>  #include <Protocol/DevicePath.h>
> >>>> @@ -554,6 +555,47 @@ PlatformBootManagerBeforeConsole (
> >>>>    PlatformRegisterOptionsAndKeys ();
> >>>>  }
> >>>>  
> >>>> +STATIC
> >>>> +VOID
> >>>> +WaitForDiskReady (
> >>>> +  )
> >>>> +{
> >>>> +  EFI_STATUS                Status;
> >>>> +  UINT32                    Index;
> >>>> +  UINTN                     DataSize;
> >>>> +  UINT32                    DiskInfo;
> >>>> +  UINT8                     IsFinished;
> >>>> +
> >>>> +  Status = EFI_NOT_FOUND;
> >>>> +  DataSize = sizeof (UINT32);
> >>>> +  // Wait for 30 seconds at most.
> >>>> +  for (Index=0; Index<30; Index++) {
> >>>
> >>> Spaces around '=' and '<'.
> >>>
> >>>> +    Status = gRT->GetVariable (
> >>>> +                    L"SASDiskInfo",
> >>>> +                    &gHisiOemVariableGuid,
> >>>> +                    NULL,
> >>>> +                    &DataSize,
> >>>> +                    &DiskInfo
> >>>> +                    );
> >>>
> >>> Wait...
> >>> Are we synchronizing against the storage controller driver using an
> >>> environment variable and looping over it for 30 seconds?
> >>>
> >> D06:
> >> For using hard disk backboard, some disk need 15 seconds to ready.
> >> Actually, wait less 15 seconds here(minus the time from end of SAS
> >> driver to here).
> >> For using expander, wait less 6 seconds here(minus the time from end of SAS
> >> driver to here).
> >>
> >> D03/D05:
> >> no waiting here.
> >>
> >> Maybe I should change 30 to 15, it will never loop over for 30 seconds.
> >>
> >>> That can't go in.
> >>> Please look into implementing an event in the SAS driver which you can
> >>> wait for here.
> >>
> >> I don't really understand what differece between implementing with variable
> >> and implementing with event.
> > 
> > One is using the mechanism explicitly designed for this sort of
> > thing. (And hence making the code a lot more clear and avoiding
> > unintended side effects.)
> > It also mean you would wait exactly as long as you needed, and not
> > need to worry about how long.
> 
> In this case, it don't want to wait exactly time, the time depand on
> SAS driver and which backboard is used. After optimizing SAS driver,
> wait time is reduced. Here,30 just for avoiding dead loop.

Which is exactly what an event is for.
At the point where the SAS controller driver is currently updating the
variable, it can instead call SignalEvent ().

The WaitForDiskReady () function can be changed to wait/check for that event.

> >>> Why not call WaitForDiskReady() before EfiBootManagerConnectAll () in
> >>> PlatformBootManagerAfterConsole ()?
> >>>
> >>
> >> The Sas controller of D06 is a pci device, SAS driver (Start functio) run after
> >> pci enumeration. WaitForDiskReady should be called after SAS driver and before
> >> creating boot options.
> > 
> > OK, but EfiBootManagerConnectAll () is a bit of a sledge hammer, and
> > will affect boot times.
> > 
> > The SAS controller is onboard (so always present), right?
> > If so, you could connect it manually.
>
> Yes, it's always present.
> I intend to optimize SAS driver in future. It's not enouth time to do
> this before 18.08.
> Thanks.

OK.

/
    Leif