From: Ignat Korchagin <ignat@cloudflare.com>
To: Eric Biggers <ebiggers@kernel.org>
Cc: agk@redhat.com, Mike Snitzer <snitzer@redhat.com>,
dm-devel@redhat.com, dm-crypt@saout.de,
linux-kernel <linux-kernel@vger.kernel.org>,
kernel-team <kernel-team@cloudflare.com>
Subject: Re: [dm-crypt] [RFC PATCH 1/1] Add DM_CRYPT_FORCE_INLINE flag to dm-crypt target
Date: Wed, 24 Jun 2020 09:24:07 +0100 [thread overview]
Message-ID: <CALrw=nFduv_X83V1Dfz+bt4bZqT19OSi3q5f7umhty1-DQ2SPg@mail.gmail.com> (raw)
In-Reply-To: <20200624050452.GB844@sol.localdomain>
On Wed, Jun 24, 2020 at 6:04 AM Eric Biggers <ebiggers@kernel.org> wrote:
>
> On Fri, Jun 19, 2020 at 05:41:32PM +0100, Ignat Korchagin wrote:
> > Sometimes extra thread offloading imposed by dm-crypt hurts IO latency. This is
> > especially visible on busy systems with many processes/threads. Moreover, most
> > Crypto API implementaions are async, that is they offload crypto operations on
> > their own, so this dm-crypt offloading is excessive.
>
> This really should say "some Crypto API implementations are async" instead of
> "most Crypto API implementations are async".
The most accurate would probably be: most hardware-accelerated Crypto
API implementations are async
> Notably, the AES-NI implementation of AES-XTS is synchronous if you call it in a
> context where SIMD instructions are usable. It's only asynchronous when SIMD is
> not usable. (This seems to have been missed in your blog post.)
No, it was not. This is exactly why we made xts-proxy Crypto API
module as a second patch. But it seems now it does not make a big
difference if a used Crypto API implementation is synchronous as well
(based on some benchmarks outlined in the cover letter to this patch).
I think the v2 of this patch will not require a synchronous Crypto
API. This is probably a right thing to do, as the "inline" flag should
control the way how dm-crypt itself handles requests, not how Crypto
API handles requests. If a user wants to ensure a particular
synchronous Crypto API implementation, they can already reconfigure
dm-crypt and specify the implementation with a "capi:" prefix in the
the dm table description.
> > This adds a new flag, which directs dm-crypt not to offload crypto operations
> > and process everything inline. For cases, where crypto operations cannot happen
> > inline (hard interrupt context, for example the read path of the NVME driver),
> > we offload the work to a tasklet rather than a workqueue.
>
> This patch both removes some dm-crypt specific queueing, and changes decryption
> to use softIRQ context instead of a workqueue. It would be useful to know how
> much of a difference the workqueue => softIRQ change makes by itself. Such a
> change could be useful for fscrypt as well. (fscrypt uses a workqueue for
> decryption, but besides that doesn't use any other queueing.)
>
> > @@ -127,7 +128,7 @@ struct iv_elephant_private {
> > * and encrypts / decrypts at the same time.
> > */
> > enum flags { DM_CRYPT_SUSPENDED, DM_CRYPT_KEY_VALID,
> > - DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD };
> > + DM_CRYPT_SAME_CPU, DM_CRYPT_NO_OFFLOAD, DM_CRYPT_FORCE_INLINE = (sizeof(unsigned long) * 8 - 1) };
>
> Assigning a specific enum value isn't necessary.
Yes, this is a leftover from our "internal" patch which I wanted to
make "future proof" in case future iterations of dm-crypt will add
some flags to avoid flag collisions. Will remove in v2.
>
> > @@ -1458,13 +1459,18 @@ static void crypt_alloc_req_skcipher(struct crypt_config *cc,
> >
> > skcipher_request_set_tfm(ctx->r.req, cc->cipher_tfm.tfms[key_index]);
> >
> > - /*
> > - * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
> > - * requests if driver request queue is full.
> > - */
> > - skcipher_request_set_callback(ctx->r.req,
> > - CRYPTO_TFM_REQ_MAY_BACKLOG,
> > - kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
> > + if (test_bit(DM_CRYPT_FORCE_INLINE, &cc->flags))
> > + /* make sure we zero important fields of the request */
> > + skcipher_request_set_callback(ctx->r.req,
> > + 0, NULL, NULL);
> > + else
> > + /*
> > + * Use REQ_MAY_BACKLOG so a cipher driver internally backlogs
> > + * requests if driver request queue is full.
> > + */
> > + skcipher_request_set_callback(ctx->r.req,
> > + CRYPTO_TFM_REQ_MAY_BACKLOG,
> > + kcryptd_async_done, dmreq_of_req(cc, ctx->r.req));
> > }
>
> This looks wrong. Unless type=0 and mask=CRYPTO_ALG_ASYNC are passed to
> crypto_alloc_skcipher(), the skcipher implementation can still be asynchronous,
> in which case providing a callback is required.
>
> Do you intend that the "force_inline" option forces the use of a synchronous
> skcipher (alongside the other things it does)? Or should it still allow
> asynchronous ones?
As mentioned above, I don't think we should require synchronous crypto
with the "force_inline" flag anymore. Although we may remove
CRYPTO_TFM_REQ_MAY_BACKLOG with the inline flag.
> We may not actually have a choice in that matter, since xts-aes-aesni has the
> CRYPTO_ALG_ASYNC bit set (as I mentioned) despite being synchronous in most
> cases; thus, the crypto API won't give you it if you ask for a synchronous
> cipher. So I think you still need to allow async skciphers? That means a
> callback is still always required.
>
> - Eric
next prev parent reply other threads:[~2020-06-24 8:24 UTC|newest]
Thread overview: 31+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-06-19 16:41 [RFC PATCH 0/1] dm-crypt excessive overhead Ignat Korchagin
2020-06-19 16:41 ` [RFC PATCH 1/1] Add DM_CRYPT_FORCE_INLINE flag to dm-crypt target Ignat Korchagin
2020-06-24 5:04 ` [dm-crypt] " Eric Biggers
2020-06-24 5:21 ` [dm-devel] " Damien Le Moal
2020-06-24 5:27 ` Eric Biggers
2020-06-24 6:46 ` Damien Le Moal
2020-06-24 7:24 ` Damien Le Moal
2020-06-24 7:49 ` Damien Le Moal
2020-06-24 8:24 ` Ignat Korchagin [this message]
2020-06-24 16:24 ` Eric Biggers
2020-06-24 17:00 ` Ignat Korchagin
2020-06-24 5:12 ` [dm-devel] " Damien Le Moal
2020-06-19 16:55 ` [RFC PATCH 0/1] dm-crypt excessive overhead Mike Snitzer
2020-06-19 18:39 ` Mikulas Patocka
2020-06-19 19:44 ` Ignat Korchagin
2020-06-20 1:23 ` Herbert Xu
2020-06-20 19:36 ` Mikulas Patocka
2020-06-20 21:02 ` Ignat Korchagin
2020-06-23 15:33 ` Mike Snitzer
2020-06-23 16:24 ` Ignat Korchagin
2020-06-24 0:23 ` Herbert Xu
2020-06-22 0:45 ` [dm-devel] " Damien Le Moal
2020-06-22 7:55 ` Ignat Korchagin
2020-06-22 8:08 ` Damien Le Moal
2020-06-23 15:01 ` Mike Snitzer
2020-06-23 15:07 ` Ignat Korchagin
2020-06-23 15:22 ` Mike Snitzer
2020-06-24 4:54 ` [dm-devel] " Damien Le Moal
2020-06-24 5:22 ` Mike Snitzer
2020-06-24 8:02 ` Ignat Korchagin
2020-06-24 4:28 ` Damien Le Moal
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='CALrw=nFduv_X83V1Dfz+bt4bZqT19OSi3q5f7umhty1-DQ2SPg@mail.gmail.com' \
--to=ignat@cloudflare.com \
--cc=agk@redhat.com \
--cc=dm-crypt@saout.de \
--cc=dm-devel@redhat.com \
--cc=ebiggers@kernel.org \
--cc=kernel-team@cloudflare.com \
--cc=linux-kernel@vger.kernel.org \
--cc=snitzer@redhat.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).