RE: [PATCH] riscv: Support non-coherency memory model

[Gary Guo]

> -----Original Message-----
> From: Guo Ren <guoren@xxxxxxxxxx>
> Sent: Wednesday, April 24, 2019 03:08
> To: Gary Guo <gary@xxxxxxxxxxx>
> Cc: Christoph Hellwig <hch@xxxxxx>; linux-arch@xxxxxxxxxxxxxxx; Palmer
> Dabbelt <palmer@xxxxxxxxxx>; Andrew Waterman <andrew@xxxxxxxxxx>; Arnd
> Bergmann <arnd@xxxxxxxx>; Anup Patel <anup.patel@xxxxxxx>; Xiang
> Xiaoyan <xiaoyan_xiang@xxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; Mike
> Rapoport <rppt@xxxxxxxxxxxxx>; Vincent Chen <vincentc@xxxxxxxxxxxxx>;
> Greentime Hu <green.hu@xxxxxxxxx>; ren_guo@xxxxxxxxx; linux-
> riscv@xxxxxxxxxxxxxxxxxxx; Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>;
> Robin Murphy <robin.murphy@xxxxxxx>; Scott Wood <swood@xxxxxxxxxx>;
> tech-privileged@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH] riscv: Support non-coherency memory model
> 
> Hi Gary,
> 
> On Tue, Apr 23, 2019 at 03:57:30PM +0000, Gary Guo wrote:
> > >>> Another point is we could get more attribute bits by modify the riscv
> > >>> spec:
> > >>>   - Remove Global bit, I think it's duplicate with the User bit in linux.
> > >>
> > >> It is in Linux, but it is conceptually very different.
> > > Yes, but hardware could ignore one of them and in riscv linux
> > > _PAGE_GLOBAL is no use at all, see:
> > > grep _PAGE_GLOBAL arch/riscv -r
> > >
> > > In fact, the _PAGE_KERNEL for pte doesn't contain _PAGE_GLOBAL and it
> > > works on FU540 and qemu. As I've mentioned page attribute bits is very
> > > precious, define a useless bit make people confused.
> >  >
> >
> > The fact that it isn't used yet doesn't imply it is not useful. We don't
> > use ASIDs at the moment, and without using ASIDs the "global" bit is
> > indeed not useful. However with ASIDs the bit will be vital for saving
> > TLB spaces. Without the global bit, the kernel pages become synonyms to
> > themselves (i.e. they have different tags in TLB but refer to the same
> > physical page).
> >
> > The global bit also exists in many other ISAs as well. It's definitely
> > not a "useless" bits.
> >
> > Moreover, this bit is already implemented in both Rocket and Ariane. It
> > is also in the spec for quite a while. The fact that Linux doesn't use
> > it at the moment is not a reason for removing it.
> >
> 
> Look:
> linux-next git:(riscv_asid_allocator_v2)$ grep GLOBAL arch/riscv -r
> arch/riscv/include/asm/pgtable-bits.h:#define _PAGE_GLOBAL    (1 << 5)    /*
> Global */
> arch/riscv/include/asm/pgtable-bits.h:                                    _PAGE_USER |
> _PAGE_GLOBAL))
> 
> Your patch tell us _PAGE_USER and _PAGE_GLOBAL are duplicate and why we
> couldn't make _PAGE_USER implies _PAGE_GLOBAL? Can you give an example
> of a real scene in PTE about:
>   _PAGE_USER:0 + _PAGE_GLOBAL:1
> or
>   _PAGE_USER:1 + _PAGE_GLOBAL:0
> 
> Of cause I know USER & GLOBAL are conceptually very different, but
> there are only 10 attribute-bits for riscv (In fact we've wasted two bits
> to support huge RV32-pfn :P). So I think it is time to merge these two bits
> before hardware supports GLOBAL. Reserve them for future!

Two cases I can think of:
* vdso like things. They're user pages that can really be shared across address spaces (i.e. global). Kernels like L4 implement most systems calls similar to VDSO, so USER + GLOBAL is useful.
* hypervisor without H-extension: This requires shadow page tables. Supervisor pages are mapped to supervisor shadow pages. However these shadow pages cannot be GLOBAL because they can't be shared between VMs. So  !USER + !GLOBAL is useful.

Remember Linux isn't the only supervisor software that RISC-V cares! 

> 
> Best Regards
>  Guo Ren