Message-ID: <Yxmd6ItV05ynoFar@kernel.org>
Date:   Thu, 8 Sep 2022 10:46:48 +0300
From:   Jarkko Sakkinen <jarkko@...nel.org>
To:     "Kalra, Ashish" <Ashish.Kalra@....com>
Cc:     Marc Orr <marcorr@...gle.com>, Borislav Petkov <bp@...en8.de>,
        x86 <x86@...nel.org>, LKML <linux-kernel@...r.kernel.org>,
        kvm list <kvm@...r.kernel.org>,
        "linux-coco@...ts.linux.dev" <linux-coco@...ts.linux.dev>,
        Linux Memory Management List <linux-mm@...ck.org>,
        Linux Crypto Mailing List <linux-crypto@...r.kernel.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Joerg Roedel <jroedel@...e.de>,
        "Lendacky, Thomas" <Thomas.Lendacky@....com>,
        "H. Peter Anvin" <hpa@...or.com>, Ard Biesheuvel <ardb@...nel.org>,
        Paolo Bonzini <pbonzini@...hat.com>,
        Sean Christopherson <seanjc@...gle.com>,
        Vitaly Kuznetsov <vkuznets@...hat.com>,
        Jim Mattson <jmattson@...gle.com>,
        Andy Lutomirski <luto@...nel.org>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Sergio Lopez <slp@...hat.com>, Peter Gonda <pgonda@...gle.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>,
        David Rientjes <rientjes@...gle.com>,
        Dov Murik <dovmurik@...ux.ibm.com>,
        Tobin Feldman-Fitzthum <tobin@....com>,
        "Roth, Michael" <Michael.Roth@....com>,
        Vlastimil Babka <vbabka@...e.cz>,
        "Kirill A . Shutemov" <kirill@...temov.name>,
        Andi Kleen <ak@...ux.intel.com>,
        Tony Luck <tony.luck@...el.com>,
        Sathyanarayanan Kuppuswamy 
        <sathyanarayanan.kuppuswamy@...ux.intel.com>,
        Alper Gun <alpergun@...gle.com>,
        "Dr . David Alan Gilbert" <dgilbert@...hat.com>
Subject: Re: [PATCH Part2 v6 09/49] x86/fault: Add support to handle the RMP
 fault for user address

On Tue, Sep 06, 2022 at 06:44:23PM +0300, Jarkko Sakkinen wrote:
> On Tue, Sep 06, 2022 at 02:17:15PM +0000, Kalra, Ashish wrote:
> > [AMD Official Use Only - General]
> > 
> > >> On Tue, Aug 09, 2022 at 06:55:43PM +0200, Borislav Petkov wrote:
> > >> > On Mon, Jun 20, 2022 at 11:03:43PM +0000, Ashish Kalra wrote:
> > >> > > +   pfn = pte_pfn(*pte);
> > >> > > +
> > >> > > +   /* If its large page then calculte the fault pfn */
> > >> > > +   if (level > PG_LEVEL_4K) {
> > >> > > +           unsigned long mask;
> > >> > > +
> > >> > > +           mask = pages_per_hpage(level) - pages_per_hpage(level - 1);
> > >> > > +           pfn |= (address >> PAGE_SHIFT) & mask;
> > >> >
> > >> > Oh boy, this is unnecessarily complicated. Isn't this
> > >> >
> > >> >       pfn |= pud_index(address);
> > >> >
> > >> > or
> > >> >       pfn |= pmd_index(address);
> > >>
> > >> I played with this a bit and ended up with
> > >>
> > >>         pfn = pte_pfn(*pte) | PFN_DOWN(address & page_level_mask(level 
> > >> - 1));
> > >>
> > >> Unless I got something terribly wrong, this should do the same (see 
> > >> the attached patch) as the existing calculations.
> > 
> > >Actually, I don't think they're the same. I think Jarkko's version is correct. Specifically:
> > >- For level = PG_LEVEL_2M they're the same.
> > >- For level = PG_LEVEL_1G:
> > >The current code calculates a garbage mask:
> > >mask = pages_per_hpage(level) - pages_per_hpage(level - 1); translates to:
> > >>> hex(262144 - 512)
> > >'0x3fe00'
> > 
> > No actually this is not a garbage mask, as I explained in earlier responses we need to capture the address bits 
> > to get to the correct 4K index into the RMP table.
> > Therefore, for level = PG_LEVEL_1G:
> > mask = pages_per_hpage(level) - pages_per_hpage(level - 1) => 0x3fe00 (which is the correct mask).
> > 
> > >But I believe Jarkko's version calculates the correct mask (below), incorporating all 18 offset bits into the 1G page.
> > >>> hex(262144 -1)
> > >'0x3ffff'
> > 
> > We can get this simply by doing (page_per_hpage(level)-1), but as I mentioned above this is not what we need.
> 
> I think you're correct, so I'll retry:
> 
> (address / PAGE_SIZE) & (pages_per_hpage(level) - pages_per_hpage(level - 1)) =
> 
> (address / PAGE_SIZE) & ((page_level_size(level) / PAGE_SIZE) - (page_level_size(level - 1) / PAGE_SIZE)) =
> 
> [ factor out 1 / PAGE_SIZE ]
> 
> (address & (page_level_size(level) - page_level_size(level - 1))) / PAGE_SIZE  =
> 
> [ Substitute with PFN_DOWN() ] 
> 
> PFN_DOWN(address & (page_level_size(level) - page_level_size(level - 1)))
> 
> So you can just:
> 
> pfn = pte_pfn(*pte) | PFN_DOWN(address & (page_level_size(level) - page_level_size(level - 1)));
> 
> Which is IMHO way better still what it is now because no branching
> and no ad-hoc helpers (the current is essentially just page_level_size
> wrapper).

I created a small test program:

$ cat test.c
#include <stdio.h>
int main(void)
{
        unsigned long arr[] = {0x8, 0x1000, 0x200000, 0x40000000, 0x8000000000};
        int i;

        for (i = 1; i < sizeof(arr)/sizeof(unsigned long); i++) {
                printf("%048b\n", arr[i] - arr[i - 1]);
                printf("%048b\n", (arr[i] - 1) ^ (arr[i - 1] - 1));
        }
}

kultaheltta in linux on  host-snp-v7 [?]
$ gcc -o test test.c

kultaheltta in linux on  host-snp-v7 [?]
$ ./test
000000000000000000000000000000000000111111111000
000000000000000000000000000000000000111111111000
000000000000000000000000000111111111000000000000
000000000000000000000000000111111111000000000000
000000000000000000111111111000000000000000000000
000000000000000000111111111000000000000000000000
000000000000000011000000000000000000000000000000
000000000000000011000000000000000000000000000000

So the operation could be described as:

        pfn = PFN_DOWN(address & (~page_level_mask(level) ^ ~page_level_mask(level - 1)));

Which IMHO already documents itself quite well: index
with the granularity of PGD by removing bits used for
PGD's below it.

BR, Jarkko

Powered by blists - more mailing lists