Re: [edk2-devel] [PATCH] UefiCpuPkg: add volatile qualifier to page table related variable

["Zhou, Jianfeng" <jianfeng.zhou@intel.com>]

Thank all for review.

> Volatile-qualifying the local variables does not seem useful for 
> anything. It's fine -- actually: it's beneficial -- if the compiler 
> optimizes accesses to those locals -- being on the stack -- as heavily 
> as it can. In other words, those parts of the patch look like a small 
> performance regression.

Obviously, local variables like LocalPleB just a temporary variable. 
From optimization perspective, it may be replaced by function parameter PleB.
Qualifying local variable is for preventing it is optimized. In this case, code like "LocalPleB.Bits.Present = Attribute->Bits.Present" may lead to unexpected result, as it is not atomic.
It might lead to performance degradation, as it can't be optimized to register. I think such code will not be called in large quantities and slight drop in performance is acceptable.


>   IN OUT volatile IA32_PAGE_NON_LEAF_ENTRY  *Pnle
> and similar. Then the existent assignment expressions
>   Pnle->Uint64 = LocalPnle.Uint64;
> don't have to be changed.

agree that such improvement would have the same effect, and the code would be a little more graceful.
Will update this patch.

Thanks & Regards,
Zhou Jianfeng

-----Original Message-----
From: Pedro Falcato <pedro.falcato@gmail.com> 
Sent: Thursday, February 22, 2024 5:44 AM
To: devel@edk2.groups.io; lersek@redhat.com
Cc: Zhou, Jianfeng <jianfeng.zhou@intel.com>; Ni, Ray <ray.ni@intel.com>; Kumar, Rahul R <rahul.r.kumar@intel.com>; Gerd Hoffmann <kraxel@redhat.com>
Subject: Re: [edk2-devel] [PATCH] UefiCpuPkg: add volatile qualifier to page table related variable

On Wed, Feb 21, 2024 at 8:36 PM Laszlo Ersek <lersek@redhat.com> wrote:
>
> On 2/21/24 02:25, Zhou Jianfeng wrote:
> > Add volatile qualifier to page table related variable to prevent 
> > compiler from optimizing away the variables which may lead to 
> > unexpected result.
> >
> > Signed-off-by: Zhou Jianfeng <jianfeng.zhou@intel.com>
> > Cc: Ray Ni <ray.ni@intel.com>
> > Cc: Laszlo Ersek <lersek@redhat.com>
> > Cc: Rahul Kumar <rahul1.kumar@intel.com>
> > Cc: Gerd Hoffmann <kraxel@redhat.com>

I'd appreciate getting CC'd on my own suggestion....

> > ---
> >  UefiCpuPkg/Library/CpuPageTableLib/CpuPageTableMap.c | 12 
> > ++++++------
> >  1 file changed, 6 insertions(+), 6 deletions(-)
>
> (1) subject should be something like:
>
>   UefiCpuPkg/CpuPageTableLib: qualify page table accesses as volatile
>
> >
> > diff --git a/UefiCpuPkg/Library/CpuPageTableLib/CpuPageTableMap.c 
> > b/UefiCpuPkg/Library/CpuPageTableLib/CpuPageTableMap.c
> > index 2ea40666cc..5cf6e8fea0 100644
> > --- a/UefiCpuPkg/Library/CpuPageTableLib/CpuPageTableMap.c
> > +++ b/UefiCpuPkg/Library/CpuPageTableLib/CpuPageTableMap.c
> > @@ -26,7 +26,7 @@ PageTableLibSetPte4K (
> >    IN IA32_MAP_ATTRIBUTE      *Mask
> >    )
> >  {
> > -  IA32_PTE_4K  LocalPte4K;
> > +  volatile IA32_PTE_4K  LocalPte4K;
> >
> >    LocalPte4K.Uint64 = Pte4K->Uint64;
> >    if (Mask->Bits.PageTableBaseAddressLow || 
> > Mask->Bits.PageTableBaseAddressHigh) { @@ -78,7 +78,7 @@ PageTableLibSetPte4K (
> >    }
> >
> >    if (Pte4K->Uint64 != LocalPte4K.Uint64) {
> > -    Pte4K->Uint64 = LocalPte4K.Uint64;
> > +    *(volatile UINT64 *)&(Pte4K->Uint64) = LocalPte4K.Uint64;
> >    }
> >  }
> >
> > @@ -100,7 +100,7 @@ PageTableLibSetPleB (
> >    IN IA32_MAP_ATTRIBUTE                     *Mask
> >    )
> >  {
> > -  IA32_PAGE_LEAF_ENTRY_BIG_PAGESIZE  LocalPleB;
> > +  volatile IA32_PAGE_LEAF_ENTRY_BIG_PAGESIZE  LocalPleB;
> >
> >    LocalPleB.Uint64 = PleB->Uint64;
> >    if (Mask->Bits.PageTableBaseAddressLow || 
> > Mask->Bits.PageTableBaseAddressHigh) { @@ -154,7 +154,7 @@ PageTableLibSetPleB (
> >    }
> >
> >    if (PleB->Uint64 != LocalPleB.Uint64) {
> > -    PleB->Uint64 = LocalPleB.Uint64;
> > +    *(volatile UINT64 *)&(PleB->Uint64) = LocalPleB.Uint64;
> >    }
> >  }
> >
> > @@ -200,7 +200,7 @@ PageTableLibSetPnle (
> >    IN IA32_MAP_ATTRIBUTE            *Mask
> >    )
> >  {
> > -  IA32_PAGE_NON_LEAF_ENTRY  LocalPnle;
> > +  volatile IA32_PAGE_NON_LEAF_ENTRY  LocalPnle;
> >
> >    LocalPnle.Uint64 = Pnle->Uint64;
> >    if (Mask->Bits.Present) {
> > @@ -231,7 +231,7 @@ PageTableLibSetPnle (
> >    LocalPnle.Bits.WriteThrough  = 0;
> >    LocalPnle.Bits.CacheDisabled = 0;
> >    if (Pnle->Uint64 != LocalPnle.Uint64) {
> > -    Pnle->Uint64 = LocalPnle.Uint64;
> > +    *(volatile UINT64 *)&(Pnle->Uint64) = LocalPnle.Uint64;
> >    }
> >  }
>
> I agree with the idea (I think it's a necessary change, or put 
> differently, an improvement, even though I may not be convinced that 
> it is a *sufficient* improvement; but let's not rehash all that here 
> again); however, I think the implementation is not the greatest.
>
> Volatile-qualifying the local variables does not seem useful for 
> anything. It's fine -- actually: it's beneficial -- if the compiler 
> optimizes accesses to those locals -- being on the stack -- as heavily 
> as it can. In other words, those parts of the patch look like a small 
> performance regression.
>
> (2) What we want to qualify as volatile here are the *targets* of the 
> Pte4K, PleB and Pnle pointers. Your other patch ("UefiCpuPkg: Fix IN 
> OUT parameters marked as IN") correctly marks those as "IN OUT", so in 
> this patch, we should update them to:
>
>   IN OUT volatile IA32_PAGE_NON_LEAF_ENTRY  *Pnle
>
> and similar. Then the existent assignment expressions
>
>   Pnle->Uint64 = LocalPnle.Uint64;
>
> don't have to be changed.

I echo these comments :)

>
> Note that call sites will not have to be updated either; see C99 
> 6.3.2.3 Pointers, paragraph 2:
>
>     For any qualifier q, a pointer to a non-q-qualified type may be
>     converted to a pointer to the q-qualified version of the type; the
>     values stored in the original and converted pointers shall compare
>     equal.

Ugh, honestly converting to volatile implicitly is kind-of yucky, but I guess it works; personally I'd rather have explicit conversion, but it's just a matter of taste.
What I *really* prefer in these cases (when we're not dealing with
MMIO) is something like READ_ONCE and WRITE_ONCE, where the "volatility points" are very well annotated, but oh well :)

--
Pedro


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#115872): https://edk2.groups.io/g/devel/message/115872
Mute This Topic: https://groups.io/mt/104483610/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-