From: Nadav Amit <nadav.amit@gmail.com>
To: Andy Lutomirski <luto@amacapital.net>
Cc: Linux-MM <linux-mm@kvack.org>,
LKML <linux-kernel@vger.kernel.org>,
Andy Lutomirski <luto@kernel.org>,
Andrea Arcangeli <aarcange@redhat.com>,
Andrew Morton <akpm@linux-foundation.org>,
Dave Hansen <dave.hansen@linux.intel.com>,
Peter Zijlstra <peterz@infradead.org>,
Thomas Gleixner <tglx@linutronix.de>,
Will Deacon <will@kernel.org>, Yu Zhao <yuzhao@google.com>,
X86 ML <x86@kernel.org>
Subject: Re: [RFC 15/20] mm: detect deferred TLB flushes in vma granularity
Date: Tue, 2 Feb 2021 12:51:56 -0800 [thread overview]
Message-ID: <846661D8-0963-4DCF-AE8D-1417064CB721@gmail.com> (raw)
In-Reply-To: <8F37526F-8189-483A-A16E-E0EB8662AD98@amacapital.net>
> On Feb 1, 2021, at 4:14 PM, Andy Lutomirski <luto@amacapital.net> wrote:
>
>
>> On Feb 1, 2021, at 2:04 PM, Nadav Amit <nadav.amit@gmail.com> wrote:
>>
>> Andy’s comments managed to make me realize this code is wrong. We must
>> call inc_mm_tlb_gen(mm) every time.
>>
>> Otherwise, a CPU that saw the old tlb_gen and updated it in its local
>> cpu_tlbstate on a context-switch. If the process was not running when the
>> TLB flush was issued, no IPI will be sent to the CPU. Therefore, later
>> switch_mm_irqs_off() back to the process will not flush the local TLB.
>>
>> I need to think if there is a better solution. Multiple calls to
>> inc_mm_tlb_gen() during deferred flushes would trigger a full TLB flush
>> instead of one that is specific to the ranges, once the flush actually takes
>> place. On x86 it’s practically a non-issue, since anyhow any update of more
>> than 33-entries or so would cause a full TLB flush, but this is still ugly.
>
> What if we had a per-mm ring buffer of flushes? When starting a flush, we would stick the range in the ring buffer and, when flushing, we would read the ring buffer to catch up. This would mostly replace the flush_tlb_info struct, and it would let us process multiple partial flushes together.
I wanted to sleep on it, and went back and forth on whether it is the right
direction, hence the late response.
I think that what you say make sense. I think that I even tried to do once
something similar for some reason, but my memory plays tricks on me.
So tell me what you think on this ring-based solution. As you said, you keep
per-mm ring of flush_tlb_info. When you queue an entry, you do something
like:
#define RING_ENTRY_INVALID (0)
gen = inc_mm_tlb_gen(mm);
struct flush_tlb_info *info = mm->ring[gen % RING_SIZE];
spin_lock(&mm->ring_lock);
WRITE_ONCE(info->new_tlb_gen, RING_ENTRY_INVALID);
smp_wmb();
info->start = start;
info->end = end;
info->stride_shift = stride_shift;
info->freed_tables = freed_tables;
smp_store_release(&info->new_tlb_gen, gen);
spin_unlock(&mm->ring_lock);
When you flush you use the entry generation as a sequence lock. On overflow
of the ring (i.e., sequence number mismatch) you perform a full flush:
for (gen = mm->tlb_gen_completed; gen < mm->tlb_gen; gen++) {
struct flush_tlb_info *info = &mm->ring[gen % RING_SIZE];
// detect overflow and invalid entries
if (smp_load_acquire(info->new_tlb_gen) != gen)
goto full_flush;
start = min(start, info->start);
end = max(end, info->end);
stride_shift = min(stride_shift, info->stride_shift);
freed_tables |= info.freed_tables;
smp_rmb();
// seqlock-like check that the information was not updated
if (READ_ONCE(info->new_tlb_gen) != gen)
goto full_flush;
}
On x86 I suspect that performing a full TLB flush would anyhow be the best
thing to do if there is more than a single entry. I am also not sure that it
makes sense to check the ring from flush_tlb_func_common() (i.e., in each
IPI handler) as it might cause cache thrashing.
Instead it may be better to do so from flush_tlb_mm_range(), when the
flushes are initiated, and use an aggregated flush_tlb_info for the flush.
It may also be better to have the ring arch-independent, so it would
resemble more of mmu_gather (the parts about the TLB flush information,
without the freed pages stuff).
We can detect deferred TLB flushes either by storing “deferred_gen” in the
page-tables/VMA (as I did) or by going over the ring, from tlb_gen_completed
to tlb_gen, and checking for an overlap. I think page-tables would be most
efficient/scalable, but perhaps going over the ring would be easier to
understand logic.
Makes sense? Thoughts?
next prev parent reply other threads:[~2021-02-02 20:52 UTC|newest]
Thread overview: 67+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-31 0:11 [RFC 00/20] TLB batching consolidation and enhancements Nadav Amit
2021-01-31 0:11 ` [RFC 01/20] mm/tlb: fix fullmm semantics Nadav Amit
2021-01-31 1:02 ` Andy Lutomirski
2021-01-31 1:19 ` Nadav Amit
2021-01-31 2:57 ` Andy Lutomirski
2021-02-01 7:30 ` Nadav Amit
2021-02-01 11:36 ` Peter Zijlstra
2021-02-02 9:32 ` Nadav Amit
2021-02-02 11:00 ` Peter Zijlstra
2021-02-02 21:35 ` Nadav Amit
2021-02-03 9:44 ` Will Deacon
2021-02-04 3:20 ` Nadav Amit
2021-01-31 0:11 ` [RFC 02/20] mm/mprotect: use mmu_gather Nadav Amit
2021-01-31 0:11 ` [RFC 03/20] mm/mprotect: do not flush on permission promotion Nadav Amit
2021-01-31 1:07 ` Andy Lutomirski
2021-01-31 1:17 ` Nadav Amit
2021-01-31 2:59 ` Andy Lutomirski
[not found] ` <7a6de15a-a570-31f2-14d6-a8010296e694@citrix.com>
2021-02-01 5:58 ` Nadav Amit
2021-02-01 15:38 ` Andrew Cooper
2021-01-31 0:11 ` [RFC 04/20] mm/mapping_dirty_helpers: use mmu_gather Nadav Amit
2021-01-31 0:11 ` [RFC 05/20] mm/tlb: move BATCHED_UNMAP_TLB_FLUSH to tlb.h Nadav Amit
2021-01-31 0:11 ` [RFC 06/20] fs/task_mmu: use mmu_gather interface of clear-soft-dirty Nadav Amit
2021-01-31 0:11 ` [RFC 07/20] mm: move x86 tlb_gen to generic code Nadav Amit
2021-01-31 18:26 ` Andy Lutomirski
2021-01-31 0:11 ` [RFC 08/20] mm: store completed TLB generation Nadav Amit
2021-01-31 20:32 ` Andy Lutomirski
2021-02-01 7:28 ` Nadav Amit
2021-02-01 16:53 ` Andy Lutomirski
2021-02-01 11:52 ` Peter Zijlstra
2021-01-31 0:11 ` [RFC 09/20] mm: create pte/pmd_tlb_flush_pending() Nadav Amit
2021-01-31 0:11 ` [RFC 10/20] mm: add pte_to_page() Nadav Amit
2021-01-31 0:11 ` [RFC 11/20] mm/tlb: remove arch-specific tlb_start/end_vma() Nadav Amit
2021-02-01 12:09 ` Peter Zijlstra
2021-02-02 6:41 ` Nicholas Piggin
2021-02-02 7:20 ` Nadav Amit
2021-02-02 9:31 ` Peter Zijlstra
2021-02-02 9:54 ` Nadav Amit
2021-02-02 11:04 ` Peter Zijlstra
2021-01-31 0:11 ` [RFC 12/20] mm/tlb: save the VMA that is flushed during tlb_start_vma() Nadav Amit
2021-02-01 12:28 ` Peter Zijlstra
2021-01-31 0:11 ` [RFC 13/20] mm/tlb: introduce tlb_start_ptes() and tlb_end_ptes() Nadav Amit
2021-01-31 9:57 ` Damian Tometzki
2021-01-31 10:07 ` Damian Tometzki
2021-02-01 7:29 ` Nadav Amit
2021-02-01 13:19 ` Peter Zijlstra
2021-02-01 23:00 ` Nadav Amit
2021-01-31 0:11 ` [RFC 14/20] mm: move inc/dec_tlb_flush_pending() to mmu_gather.c Nadav Amit
2021-01-31 0:11 ` [RFC 15/20] mm: detect deferred TLB flushes in vma granularity Nadav Amit
2021-02-01 22:04 ` Nadav Amit
2021-02-02 0:14 ` Andy Lutomirski
2021-02-02 20:51 ` Nadav Amit [this message]
2021-02-04 4:35 ` Andy Lutomirski
2021-01-31 0:11 ` [RFC 16/20] mm/tlb: per-page table generation tracking Nadav Amit
2021-01-31 0:11 ` [RFC 17/20] mm/tlb: updated completed deferred TLB flush conditionally Nadav Amit
2021-01-31 0:11 ` [RFC 18/20] mm: make mm_cpumask() volatile Nadav Amit
2021-01-31 0:11 ` [RFC 19/20] lib/cpumask: introduce cpumask_atomic_or() Nadav Amit
2021-01-31 0:11 ` [RFC 20/20] mm/rmap: avoid potential races Nadav Amit
2021-08-23 8:05 ` Huang, Ying
2021-08-23 15:50 ` Nadav Amit
2021-08-24 0:36 ` Huang, Ying
2021-01-31 0:39 ` [RFC 00/20] TLB batching consolidation and enhancements Andy Lutomirski
2021-01-31 1:08 ` Nadav Amit
2021-01-31 3:30 ` Nicholas Piggin
2021-01-31 7:57 ` Nadav Amit
2021-01-31 8:14 ` Nadav Amit
2021-02-01 12:44 ` Peter Zijlstra
2021-02-02 7:14 ` Nicholas Piggin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=846661D8-0963-4DCF-AE8D-1417064CB721@gmail.com \
--to=nadav.amit@gmail.com \
--cc=aarcange@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=dave.hansen@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=luto@amacapital.net \
--cc=luto@kernel.org \
--cc=peterz@infradead.org \
--cc=tglx@linutronix.de \
--cc=will@kernel.org \
--cc=x86@kernel.org \
--cc=yuzhao@google.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).