linux-kernel - Re: [PATCH] x86/mm/pat: Fix missing preemption disable for __native_flush

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] [day] [month] [year] [list]

Message-ID: <749919a4-cdb1-48a3-adb4-adb81a5fa0b5@intel.com>
Date:   Mon, 12 Nov 2018 11:07:55 -0800
From:   Dave Hansen <dave.hansen@...el.com>
To:     Dan Williams <dan.j.williams@...el.com>,
        Andy Lutomirski <luto@...capital.net>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
        Andy Lutomirski <luto@...nel.org>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Borislav Petkov <bp@...en8.de>,
        stable <stable@...r.kernel.org>, X86 ML <x86@...nel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        "Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>
Subject: Re: [PATCH] x86/mm/pat: Fix missing preemption disable for
 __native_flush_tlb()

On 11/10/18 4:31 PM, Dan Williams wrote:
>> If it indeed can run late in boot or after boot, then it sure looks
>> buggy. Either the __flush_tlb_all() should be removed or it should
>> be replaced with flush_tlb_kernel_range(). It’s unclear to me why a
>> flush is needed at all, but if it’s needed, surely all CPUs need
>> flushing.
> Yeah, I don't think __flush_tlb_all() is needed at 
> kernel_physical_mapping_init() time, and at 
> kernel_physical_mapping_remove() time we do a full flush_tlb_all().

It doesn't look strictly necessary to me.  I _think_ we're only ever
populating previously non-present entries, and those never need TLB
flushes.  I didn't look too deeply, so I'd appreciate anyone else
double-checking me on this.

The __flush_tlb_all() actually appears to predate git and it was
originally entirely intended for early-boot-only.  It probably lasted
this long because it looks really important. :)

It was even next to where we set MMU features in CR4, which is *really*
early in boot:

> +       asm volatile("movq %%cr4,%0" : "=r" (mmu_cr4_features));
> +       __flush_tlb_all();

I also totally agree with Andy that if it were needed on the local CPU,
this code would be buggy because it doesn't initiate any *remote* TLB
flushes.

So, let's remove it, but also add some comments about not being allowed
to *change* page table entries, only populate them.  We could even add
some warnings to keep this enforced.