linux-kernel - Re: [RFC PATCH] kvm: Use huge pages for DAX-backed files

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 6 Nov 2018 22:16:30 +0100
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Barret Rhoden <brho@...gle.com>,
        Dan Williams <dan.j.williams@...el.com>,
        Dave Jiang <dave.jiang@...el.com>,
        Ross Zwisler <zwisler@...nel.org>,
        Vishal Verma <vishal.l.verma@...el.com>,
        Radim Krčmář <rkrcmar@...hat.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>
Cc:     linux-nvdimm@...ts.01.org, linux-kernel@...r.kernel.org,
        "H. Peter Anvin" <hpa@...or.com>, x86@...nel.org,
        kvm@...r.kernel.org, yu.c.zhang@...el.com, yi.z.zhang@...el.com
Subject: Re: [RFC PATCH] kvm: Use huge pages for DAX-backed files

On 06/11/2018 22:05, Barret Rhoden wrote:
> On 2018-10-29 at 17:07 Barret Rhoden <brho@...gle.com> wrote:
>> Another issue is that kvm_mmu_zap_collapsible_spte() also uses
>> PageTransCompoundMap() to detect huge pages, but we don't have a way to
>> get the HVA easily.  Can we just aggressively zap DAX pages there?
> 
> Any thoughts about this?  Is there a way to determine the HVA or GFN in
> this function:

Yes, iter.gfn is the gfn inside the loop and iter.level is the level
(1=PTE, 2=PDE, ...).  iter.level of course is unusable here, similar to
*levelp in transparent_hugepage_adjust, but you can use iter.gfn and
gfn_to_hva.

Paolo

> static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm,                        
>                                          struct kvm_rmap_head *rmap_head)        
> {       
>         u64 *sptep;                                                              
>         struct rmap_iterator iter;                                               
>         int need_tlb_flush = 0;
>         kvm_pfn_t pfn;
>         struct kvm_mmu_page *sp;
>                                                                                  
> restart:        
>         for_each_rmap_spte(rmap_head, &iter, sptep) {
>                 sp = page_header(__pa(sptep));
>                 pfn = spte_to_pfn(*sptep);
> 
>                 /*
>                  * We cannot do huge page mapping for indirect shadow pages,     
>                  * which are found on the last rmap (level = 1) when not using   
>                  * tdp; such shadow pages are synced with the page table in      
>                  * the guest, and the guest page table is using 4K page size     
>                  * mapping if the indirect sp has level = 1.                     
>                  */     
>                 if (sp->role.direct &&                                           
>                         !kvm_is_reserved_pfn(pfn) &&                             
>                         PageTransCompoundMap(pfn_to_page(pfn))) {                
>                         pte_list_remove(rmap_head, sptep);                       
>                         need_tlb_flush = 1;                                      
>                         goto restart;                                            
>                 }                                                                
>         }
>                                    
>         return need_tlb_flush;                                                   
> }    
> 
> If not, I was thinking of changing that loop to always remove PTEs for
> DAX mappings, with the understanding that they'll get faulted back in
> later.  Ideally, we'd like to check if the page is huge, but DAX can't
> use the PageTransCompoundMap check.
> 
> Thanks,
> 
> Barret
> 
> 
>