From: Song Liu <song@kernel.org>
To: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Mike Rapoport <rppt@kernel.org>,
"bpf@vger.kernel.org" <bpf@vger.kernel.org>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
"akpm@linux-foundation.org" <akpm@linux-foundation.org>,
"x86@kernel.org" <x86@kernel.org>,
"peterz@infradead.org" <peterz@infradead.org>,
"hch@lst.de" <hch@lst.de>,
"rick.p.edgecombe@intel.com" <rick.p.edgecombe@intel.com>,
"aaron.lu@intel.com" <aaron.lu@intel.com>,
"mcgrof@kernel.org" <mcgrof@kernel.org>
Subject: Re: [PATCH bpf-next v2 0/5] execmem_alloc for BPF programs
Date: Tue, 8 Nov 2022 13:40:04 -0800 [thread overview]
Message-ID: <CAPhsuW6w04_zgX=aXFVrXVYX1nnie1KN4oZZBrBNdL32-L1-qg@mail.gmail.com> (raw)
In-Reply-To: <d0c60ab6-e618-425a-4279-454901a60235@csgroup.eu>
On Tue, Nov 8, 2022 at 11:43 AM Christophe Leroy
<christophe.leroy@csgroup.eu> wrote:
>
>
>
> Le 08/11/2022 à 19:41, Song Liu a écrit :
> > On Tue, Nov 8, 2022 at 3:27 AM Mike Rapoport <rppt@kernel.org> wrote:
> >>
> >> Hi Song,
> >>
> >> On Mon, Nov 07, 2022 at 02:39:16PM -0800, Song Liu wrote:
> >>> This patchset tries to address the following issues:
> >>>
> >>> 1. Direct map fragmentation
> >>>
> >>> On x86, STRICT_*_RWX requires the direct map of any RO+X memory to be also
> >>> RO+X. These set_memory_* calls cause 1GB page table entries to be split
> >>> into 2MB and 4kB ones. This fragmentation in direct map results in bigger
> >>> and slower page table, and pressure for both instruction and data TLB.
> >>>
> >>> Our previous work in bpf_prog_pack tries to address this issue from BPF
> >>> program side. Based on the experiments by Aaron Lu [4], bpf_prog_pack has
> >>> greatly reduced direct map fragmentation from BPF programs.
> >>
> >> Usage of set_memory_* APIs with memory allocated from vmalloc/modules
> >> virtual range does not change the direct map, but only updates the
> >> permissions in vmalloc range. The direct map splits occur in
> >> vm_remove_mappings() when the memory is *freed*.
> >>
> >> That said, both bpf_prog_pack and these patches do reduce the
> >> fragmentation, but this happens because the memory is freed to the system
> >> in 2M chunks and there are no splits of 2M pages. Besides, since the same
> >> 2M page used for many BPF programs there should be way less vfree() calls.
> >>
> >>> 2. iTLB pressure from BPF program
> >>>
> >>> Dynamic kernel text such as modules and BPF programs (even with current
> >>> bpf_prog_pack) use 4kB pages on x86, when the total size of modules and
> >>> BPF program is big, we can see visible performance drop caused by high
> >>> iTLB miss rate.
> >>
> >> Like Luis mentioned several times already, it would be nice to see numbers.
> >>
> >>> 3. TLB shootdown for short-living BPF programs
> >>>
> >>> Before bpf_prog_pack loading and unloading BPF programs requires global
> >>> TLB shootdown. This patchset (and bpf_prog_pack) replaces it with a local
> >>> TLB flush.
> >>>
> >>> 4. Reduce memory usage by BPF programs (in some cases)
> >>>
> >>> Most BPF programs and various trampolines are small, and they often
> >>> occupies a whole page. From a random server in our fleet, 50% of the
> >>> loaded BPF programs are less than 500 byte in size, and 75% of them are
> >>> less than 2kB in size. Allowing these BPF programs to share 2MB pages
> >>> would yield some memory saving for systems with many BPF programs. For
> >>> systems with only small number of BPF programs, this patch may waste a
> >>> little memory by allocating one 2MB page, but using only part of it.
> >>
> >> I'm not convinced there are memory savings here. Unless you have hundreds
> >> of BPF programs, most of 2M page will be wasted, won't it?
> >> So for systems that have moderate use of BPF most of the 2M page will be
> >> unused, right?
> >
> > There will be some memory waste in such cases. But it will get better with:
> > 1) With 4/5 and 5/5, BPF programs will share this 2MB page with kernel .text
> > section (_stext to _etext);
> > 2) modules, ftrace, kprobe will also share this 2MB page;
> > 3) There are bigger BPF programs in many use cases.
>
> And what I love with this series (for powerpc/32) is that we will likely
> now be able to have bpf, ftrace, kprobe without the performance cost of
> CONFIG_MODULES.
Yeah, I remember reading emails about using tracing tools without
CONFIG_MODULES. We still need more work (beyond this set) to make it
happen for powerpc/32. For example, current powerpc bpf_jit doesn't
support jitting into ROX memory.
Song
>
> Today, CONFIG_MODULES means page mapping, which means handling of kernel
> page in ITLB miss handlers.
>
> By using some of the space between end of rodata and start of inittext,
> we are able to use ROX linear memory which is mapped by blocks. It means
> there is no need to handle kernel text in ITLB handler (You can look at
> https://elixir.bootlin.com/linux/v6.1-rc3/source/arch/powerpc/kernel/head_8xx.S#L191
> to better understand what I'm talking about).
>
> Thanks
> Christophe
next prev parent reply other threads:[~2022-11-08 21:41 UTC|newest]
Thread overview: 91+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-07 22:39 [PATCH bpf-next v2 0/5] execmem_alloc for BPF programs Song Liu
2022-11-07 22:39 ` [PATCH bpf-next v2 1/5] vmalloc: introduce execmem_alloc, execmem_free, and execmem_fill Song Liu
2022-11-07 22:39 ` [PATCH bpf-next v2 2/5] x86/alternative: support execmem_alloc() and execmem_free() Song Liu
2022-11-07 22:39 ` [PATCH bpf-next v2 3/5] bpf: use execmem_alloc for bpf program and bpf dispatcher Song Liu
2022-11-07 22:39 ` [PATCH bpf-next v2 4/5] vmalloc: introduce register_text_tail_vm() Song Liu
2022-11-07 22:39 ` [PATCH bpf-next v2 5/5] x86: use register_text_tail_vm Song Liu
2022-11-08 19:04 ` Edgecombe, Rick P
2022-11-08 22:15 ` Song Liu
2022-11-15 17:28 ` Edgecombe, Rick P
2022-11-07 22:55 ` [PATCH bpf-next v2 0/5] execmem_alloc for BPF programs Luis Chamberlain
2022-11-07 23:13 ` Song Liu
2022-11-07 23:39 ` Luis Chamberlain
2022-11-08 0:13 ` Edgecombe, Rick P
2022-11-08 2:45 ` Luis Chamberlain
2022-11-08 18:20 ` Song Liu
2022-11-08 18:12 ` Song Liu
2022-11-08 11:27 ` Mike Rapoport
2022-11-08 12:38 ` Aaron Lu
2022-11-09 6:55 ` Christoph Hellwig
2022-11-09 11:05 ` Peter Zijlstra
2022-11-08 16:51 ` Edgecombe, Rick P
2022-11-08 18:50 ` Song Liu
2022-11-09 11:17 ` Mike Rapoport
2022-11-09 17:04 ` Edgecombe, Rick P
2022-11-09 17:53 ` Song Liu
2022-11-13 10:34 ` Mike Rapoport
2022-11-14 20:30 ` Song Liu
2022-11-15 21:18 ` Luis Chamberlain
2022-11-15 21:39 ` Edgecombe, Rick P
2022-11-16 22:34 ` Luis Chamberlain
2022-11-17 8:50 ` Mike Rapoport
2022-11-17 18:36 ` Song Liu
2022-11-20 10:41 ` Mike Rapoport
2022-11-21 14:52 ` Song Liu
2022-11-30 9:39 ` Mike Rapoport
2022-11-09 17:43 ` Song Liu
2022-11-09 21:23 ` Christophe Leroy
2022-11-10 1:50 ` Song Liu
2022-11-13 10:42 ` Mike Rapoport
2022-11-14 20:45 ` Song Liu
2022-11-15 20:51 ` Luis Chamberlain
2022-11-20 10:44 ` Mike Rapoport
2022-11-08 18:41 ` Song Liu
2022-11-08 19:43 ` Christophe Leroy
2022-11-08 21:40 ` Song Liu [this message]
2022-11-13 9:58 ` Mike Rapoport
2022-11-14 20:13 ` Song Liu
2022-11-08 11:44 ` Christophe Leroy
2022-11-08 18:47 ` Song Liu
2022-11-08 19:32 ` Christophe Leroy
2022-11-08 11:48 ` Mike Rapoport
2022-11-15 1:30 ` Song Liu
2022-11-15 17:34 ` Edgecombe, Rick P
2022-11-15 21:54 ` Song Liu
2022-11-15 22:14 ` Edgecombe, Rick P
2022-11-15 22:32 ` Song Liu
2022-11-16 1:20 ` Song Liu
2022-11-16 21:22 ` Edgecombe, Rick P
2022-11-16 22:03 ` Song Liu
2022-11-15 21:09 ` Luis Chamberlain
2022-11-15 21:32 ` Luis Chamberlain
2022-11-15 22:48 ` Song Liu
2022-11-16 22:33 ` Luis Chamberlain
2022-11-16 22:47 ` Edgecombe, Rick P
2022-11-16 23:53 ` Luis Chamberlain
2022-11-17 1:17 ` Song Liu
2022-11-17 9:37 ` Mike Rapoport
2022-11-29 10:23 ` Thomas Gleixner
2022-11-29 17:26 ` Song Liu
2022-11-29 23:56 ` Thomas Gleixner
2022-11-30 16:18 ` Song Liu
2022-12-01 9:08 ` Thomas Gleixner
2022-12-01 19:31 ` Song Liu
2022-12-02 1:38 ` Thomas Gleixner
2022-12-02 8:38 ` Song Liu
2022-12-02 9:22 ` Thomas Gleixner
2022-12-06 20:25 ` Song Liu
2022-12-07 15:36 ` Thomas Gleixner
2022-12-07 16:53 ` Christophe Leroy
2022-12-07 19:29 ` Song Liu
2022-12-07 21:04 ` Thomas Gleixner
2022-12-07 21:48 ` Christophe Leroy
2022-12-07 19:26 ` Song Liu
2022-12-07 20:57 ` Thomas Gleixner
2022-12-07 23:17 ` Song Liu
2022-12-02 10:46 ` Christophe Leroy
2022-12-02 17:43 ` Thomas Gleixner
2022-12-01 20:23 ` Mike Rapoport
2022-12-01 22:34 ` Thomas Gleixner
2022-12-03 14:46 ` Mike Rapoport
2022-12-03 20:58 ` Thomas Gleixner
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='CAPhsuW6w04_zgX=aXFVrXVYX1nnie1KN4oZZBrBNdL32-L1-qg@mail.gmail.com' \
--to=song@kernel.org \
--cc=aaron.lu@intel.com \
--cc=akpm@linux-foundation.org \
--cc=bpf@vger.kernel.org \
--cc=christophe.leroy@csgroup.eu \
--cc=hch@lst.de \
--cc=linux-mm@kvack.org \
--cc=mcgrof@kernel.org \
--cc=peterz@infradead.org \
--cc=rick.p.edgecombe@intel.com \
--cc=rppt@kernel.org \
--cc=x86@kernel.org \
/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).