Re: [PATCH -next v2 2/2] riscv: extable: add new extable type EX_TYPE_KACCESS_ERR_ZERO support

[Tong Tiangen]

在 2022/9/22 4:25, Palmer Dabbelt 写道:
> On Sat, 27 Aug 2022 03:39:38 PDT (-0700), tongtiangen@xxxxxxxxxx wrote:
> > 在 2022/8/26 16:16, Andrew Jones 写道:
> > > On Fri, Aug 26, 2022 at 02:44:48PM +0800, Tong Tiangen wrote:
> > > > 在 2022/8/25 19:06, Andrew Jones 写道:
> > > > > On Mon, Aug 15, 2022 at 03:20:25AM +0000, Tong Tiangen wrote:
> > > > > > Currently, The extable type EX_TYPE_UACCESS_ERR_ZERO is used by
> > > > > > __get/put_kernel_nofault(), but those helpers are not uaccess 
> > > > > > type, so we
> > > > > > add a new extable type EX_TYPE_KACCESS_ERR_ZERO which can be used by
> > > > > > __get/put_kernel_no_fault().
> > > > > >
> > > > > > Only refactor code without any functional changes.
> > > > >
> > > > > This isn't quite true. __get/put_kernel_nofault now sets a different
> > > > > extable type (as the commit message says). But, nothing special seems
> > > > > to be done with that, so there's effectively no functional change. Can
> > > > > you please elaborate on the motivation for this change? Where will the
> > > > > KACCESS type need to be distinguished from the UACCESS type?
> > > >
> > > > The introduction of EX_TYPE_KACCESS_ERR_ZERO does not change any 
> > > > function,
> > > > but makes a correct distinction in the actual type, indicating that 
> > > > there
> > > > are indeed some kaccess entries in extable. I think this 
> > > > optimization is
> > > > more clear and reasonable.
> > >
> > > Well, creating new types, just for new type sake, just bloats code.
> > >
> > > > A few weeks ago, I did something similar on arm64[1]. I think this
> > > > optimization can also be used on riscv.
> > > >
> > > > We can do some features that are used on uaccss but not applicable on
> > > > kaccess in the future[2].
> > > >
> > > > [1]
> > > > https://lore.kernel.org/lkml/20220621072638.1273594-2-tongtiangen@xxxxxxxxxx/ 
> > > >
> > > > [2]https://lore.kernel.org/lkml/20220812070557.1028499-4-tongtiangen@xxxxxxxxxx/ 
> > >
> > > This is part of the information, but I had already found this. What's
> > > still missing to me are the riscv patches, or at least a riscv plan, for
> > > actually implementing something which requires kaccess and uaccess to 
> > > have
> > > distinct types.
> > >
> > > Thanks,
> > > drew
> >
> > At present, there is no such plan on riscv, because it is rely on
> > hardware support.
> > I think this patch can be merged as a small code optimization and
> > without any function change.
>
> Generally we need some use of the code in the upstream kernel to justify 
> its existence.  In this case I don't really see that: it's just another 
> type that's exactly the same as the existing one, having some out of 
> tree code that depends on making these types do something different 
> isn't a sufficient justification.
> .
Hi palmer:

I agree with this point very much，many thanks.

Tong.