Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sun, 30 Jan 2022 11:13:26 +0200
From:   Mike Rapoport <rppt@kernel.org>
To:     Michel Lespinasse <michel@lespinasse.org>
Cc:     Linux-MM <linux-mm@kvack.org>, linux-kernel@vger.kernel.org,
        Andrew Morton <akpm@linux-foundation.org>, kernel-team@fb.com,
        Laurent Dufour <ldufour@linux.ibm.com>,
        Jerome Glisse <jglisse@google.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Michal Hocko <mhocko@suse.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Davidlohr Bueso <dave@stgolabs.net>,
        Matthew Wilcox <willy@infradead.org>,
        Liam Howlett <liam.howlett@oracle.com>,
        Rik van Riel <riel@surriel.com>,
        Paul McKenney <paulmck@kernel.org>,
        Song Liu <songliubraving@fb.com>,
        Suren Baghdasaryan <surenb@google.com>,
        Minchan Kim <minchan@google.com>,
        Joel Fernandes <joelaf@google.com>,
        David Rientjes <rientjes@google.com>,
        Axel Rasmussen <axelrasmussen@google.com>,
        Andy Lutomirski <luto@kernel.org>
Subject: Re: [PATCH v2 33/35] arm64/mm: attempt speculative mm faults first
Message-ID: <YfZWtsC3QIzXR0LI@kernel.org>
References: <20220128131006.67712-1-michel@lespinasse.org>
 <20220128131006.67712-34-michel@lespinasse.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220128131006.67712-34-michel@lespinasse.org>
Precedence: bulk

On Fri, Jan 28, 2022 at 05:10:04AM -0800, Michel Lespinasse wrote:
> Attempt speculative mm fault handling first, and fall back to the
> existing (non-speculative) code if that fails.
> 
> This follows the lines of the x86 speculative fault handling code,
> but with some minor arch differences such as the way that the
> VM_FAULT_BADACCESS case is handled.
> 
> Signed-off-by: Michel Lespinasse <michel@lespinasse.org>
> ---
>  arch/arm64/mm/fault.c | 62 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 62 insertions(+)
> 
> diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> index 77341b160aca..2598795f4e70 100644
> --- a/arch/arm64/mm/fault.c
> +++ b/arch/arm64/mm/fault.c
> @@ -25,6 +25,7 @@
>  #include <linux/perf_event.h>
>  #include <linux/preempt.h>
>  #include <linux/hugetlb.h>
> +#include <linux/vm_event_item.h>
>  
>  #include <asm/acpi.h>
>  #include <asm/bug.h>
> @@ -524,6 +525,11 @@ static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
>  	unsigned long vm_flags;
>  	unsigned int mm_flags = FAULT_FLAG_DEFAULT;
>  	unsigned long addr = untagged_addr(far);
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	struct vm_area_struct *vma;
> +	struct vm_area_struct pvma;
> +	unsigned long seq;
> +#endif
>  
>  	if (kprobe_page_fault(regs, esr))
>  		return 0;
> @@ -574,6 +580,59 @@ static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
>  
>  	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
>  
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	/*
> +	 * No need to try speculative faults for kernel or
> +	 * single threaded user space.
> +	 */
> +	if (!(mm_flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
> +		goto no_spf;
> +
> +	count_vm_event(SPF_ATTEMPT);
> +	seq = mmap_seq_read_start(mm);
> +	if (seq & 1) {
> +		count_vm_spf_event(SPF_ABORT_ODD);
> +		goto spf_abort;
> +	}
> +	rcu_read_lock();
> +	vma = __find_vma(mm, addr);
> +	if (!vma || vma->vm_start > addr) {
> +		rcu_read_unlock();
> +		count_vm_spf_event(SPF_ABORT_UNMAPPED);
> +		goto spf_abort;
> +	}
> +	if (!vma_is_anonymous(vma)) {
> +		rcu_read_unlock();
> +		count_vm_spf_event(SPF_ABORT_NO_SPECULATE);
> +		goto spf_abort;
> +	}
> +	pvma = *vma;
> +	rcu_read_unlock();
> +	if (!mmap_seq_read_check(mm, seq, SPF_ABORT_VMA_COPY))
> +		goto spf_abort;
> +	vma = &pvma;
> +	if (!(vma->vm_flags & vm_flags)) {
> +		count_vm_spf_event(SPF_ABORT_ACCESS_ERROR);
> +		goto spf_abort;
> +	}
> +	fault = do_handle_mm_fault(vma, addr & PAGE_MASK,
> +			mm_flags | FAULT_FLAG_SPECULATIVE, seq, regs);
> +
> +	/* Quick path to respond to signals */
> +	if (fault_signal_pending(fault, regs)) {
> +		if (!user_mode(regs))
> +			goto no_context;
> +		return 0;
> +	}
> +	if (!(fault & VM_FAULT_RETRY))
> +		goto done;
> +
> +spf_abort:
> +	count_vm_event(SPF_ABORT);
> +no_spf:
> +
> +#endif	/* CONFIG_SPECULATIVE_PAGE_FAULT */

The speculative page fault implementation here (and for PowerPC as well)
looks very similar to x86. Can we factor it our rather than copy 3 (or
more) times?

> +
>  	/*
>  	 * As per x86, we may deadlock here. However, since the kernel only
>  	 * validly references user space from well defined areas of the code,
> @@ -612,6 +671,9 @@ static int __kprobes do_page_fault(unsigned long far, unsigned int esr,
>  		goto retry;
>  	}
>  	mmap_read_unlock(mm);
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +done:
> +#endif
>  
>  	/*
>  	 * Handle the "normal" (no error) case first.
> -- 
> 2.20.1
> 
> 

-- 
Sincerely yours,
Mike.