Re: [edk2-devel] [PATCH 3/4] UefiCpuPkg/CpuTimerDxeRiscV64: Add support for Sstc

["Laszlo Ersek" <lersek@redhat.com>]

On 1/3/24 14:58, Sunil V L wrote:
> Sstc extension allows to program the timer and receive the interrupt
> without using an SBI call. This reduces the latency to generate the timer
> interrupt. So, detect whether Sstc extension is supported and use the
> stimecmp register directly to program the timer interrupt.
> 
> Cc: Gerd Hoffmann <kraxel@redhat.com>
> Cc: Rahul Kumar <rahul1.kumar@intel.com>
> Cc: Laszlo Ersek <lersek@redhat.com>
> Cc: Ray Ni <ray.ni@intel.com>
> Cc: Andrei Warkentin <andrei.warkentin@intel.com>
> Signed-off-by: Sunil V L <sunilvl@ventanamicro.com>
> ---
>  .../CpuTimerDxeRiscV64/CpuTimerDxeRiscV64.inf |  1 +
>  UefiCpuPkg/CpuTimerDxeRiscV64/Timer.h         |  2 ++
>  UefiCpuPkg/CpuTimerDxeRiscV64/Timer.c         | 30 +++++++++++++++++--
>  3 files changed, 31 insertions(+), 2 deletions(-)
> 
> diff --git a/UefiCpuPkg/CpuTimerDxeRiscV64/CpuTimerDxeRiscV64.inf b/UefiCpuPkg/CpuTimerDxeRiscV64/CpuTimerDxeRiscV64.inf
> index aba660186dc0..f2a2cf12caef 100644
> --- a/UefiCpuPkg/CpuTimerDxeRiscV64/CpuTimerDxeRiscV64.inf
> +++ b/UefiCpuPkg/CpuTimerDxeRiscV64/CpuTimerDxeRiscV64.inf
> @@ -41,6 +41,7 @@ [Sources.RISCV64]
>    Timer.c
>  
>  [Pcd]
> +  gEfiMdePkgTokenSpaceGuid.PcdRiscVFeatureOverride           ## CONSUMES
>    gUefiCpuPkgTokenSpaceGuid.PcdCpuCoreCrystalClockFrequency  ## CONSUMES
>  
>  [Protocols]
> diff --git a/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.h b/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.h
> index 9b3542230cb5..5e5071b3f0b2 100644
> --- a/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.h
> +++ b/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.h
> @@ -26,6 +26,8 @@
>  //
>  #define DEFAULT_TIMER_TICK_DURATION  100000
>  
> +#define RISCV_CPU_FEATURE_SSTC_BITMASK  0x2

(1) Not a bug by any means, but BIT1 might read more idiomatic.

> +
>  extern VOID
>  RiscvSetTimerPeriod (
>    UINT32  TimerPeriod
> diff --git a/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.c b/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.c
> index 30e48061cd06..4babfb4bfc60 100644
> --- a/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.c
> +++ b/UefiCpuPkg/CpuTimerDxeRiscV64/Timer.c
> @@ -44,6 +44,19 @@ STATIC EFI_TIMER_NOTIFY  mTimerNotifyFunction;
>  STATIC UINT64  mTimerPeriod     = 0;
>  STATIC UINT64  mLastPeriodStart = 0;
>  
> +/**
> +  Check whether Sstc is enabled in PCD.
> +
> +**/
> +STATIC
> +BOOLEAN
> +RiscVIsSstcEnabled (
> +  VOID
> +  )
> +{
> +  return ((PcdGet64 (PcdRiscVFeatureOverride) & RISCV_CPU_FEATURE_SSTC_BITMASK) != 0);
> +}
> +
>  /**
>    Timer Interrupt Handler.
>  
> @@ -94,7 +107,12 @@ TimerInterruptHandler (
>                           ),
>                         1000000u
>                         );  // convert to tick
> -  SbiSetTimer (PeriodStart);
> +  if (RiscVIsSstcEnabled ()) {

(2) Even though the PCD is currently declared as fixed or
patchable-in-module, seeing a PcdGet64() call on the call stack of the
timer interrupt handler (and at a high TPL) makes me uncomfortable. It
carries a risk that later on we relax the PCD decl to dynamic, and then
this code would become brittle.

I propose: either replace the PcdGet64 call above with FixedPcdGet64 (so
it can never land in the runtime / dynamic PCD protocol), or perform the
PCD check in the entry point function of the driver, and store the
result in a STATIC BOOLEAN variable. Then further PCD accesses (dynamic
or otherwise) will not be needed.

> +    RiscVSetSupervisorTimeCompareRegister (PeriodStart);
> +  } else {
> +    SbiSetTimer (PeriodStart);
> +  }
> +
>    RiscVEnableTimerInterrupt (); // enable SMode timer int
>    gBS->RestoreTPL (OriginalTPL);
>  }
> @@ -197,7 +215,11 @@ TimerDriverSetTimerPeriod (
>                           ),
>                         1000000u
>                         ); // convert to tick
> -  SbiSetTimer (PeriodStart);
> +  if (RiscVIsSstcEnabled ()) {
> +    RiscVSetSupervisorTimeCompareRegister (PeriodStart);
> +  } else {
> +    SbiSetTimer (PeriodStart);
> +  }
>  
>    mCpu->EnableInterrupt (mCpu);
>    RiscVEnableTimerInterrupt (); // enable SMode timer int

(3) This seems like duplicated code. How about replacing the
RiscVIsSstcEnabled() function with a more substantive function that
incorporates both the feature check *and* the "PeriodStart" setting?
Then you can easily call that function from both TimerInterruptHandler()
and TimerDriverSetTimerPeriod().

> @@ -282,6 +304,10 @@ TimerDriverInitialize (
>    //
>    mTimerNotifyFunction = NULL;
>  
> +  if (RiscVIsSstcEnabled ()) {
> +    DEBUG ((DEBUG_INFO, "%a: Timer interrupt is via Sstc extension\n", __func__));
> +  }
> +

Right, this would be the place to fetch the PCD explicitly and to store
the result (based on bit-masking) into the global boolean.

>    //
>    // Make sure the Timer Architectural Protocol is not already installed in the system
>    //

Thanks
Laszlo



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#113176): https://edk2.groups.io/g/devel/message/113176
Mute This Topic: https://groups.io/mt/103501843/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/leave/12367111/7686176/1913456212/xyzzy [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-