Received-SPF: pass (google.com: domain of linux-crypto-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   David Laight <David.Laight@ACULAB.COM>
To:     'Robert Elliott' <elliott@hpe.com>,
        "herbert@gondor.apana.org.au" <herbert@gondor.apana.org.au>,
        "davem@davemloft.net" <davem@davemloft.net>,
        "tim.c.chen@linux.intel.com" <tim.c.chen@linux.intel.com>,
        "ap420073@gmail.com" <ap420073@gmail.com>,
        "ardb@kernel.org" <ardb@kernel.org>,
        "linux-crypto@vger.kernel.org" <linux-crypto@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: RE: [PATCH v2 05/19] crypto: x86/crc - limit FPU preemption
Thread-Topic: [PATCH v2 05/19] crypto: x86/crc - limit FPU preemption
Thread-Index: AQHY3oYZ7LB/MaiZeEeC/a2xezfdB64NQsbA
Date:   Fri, 14 Oct 2022 04:02:03 +0000
Message-ID: <85a029f30048495b963ae0ee0308853f@AcuMS.aculab.com>
References: <20221006223151.22159-1-elliott@hpe.com>
 <20221012215931.3896-1-elliott@hpe.com>
 <20221012215931.3896-6-elliott@hpe.com>
In-Reply-To: <20221012215931.3896-6-elliott@hpe.com>
Accept-Language: en-GB, en-US
MIME-Version: 1.0
Content-Language: en-US
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8BIT
Precedence: bulk

From: Robert Elliott
> Sent: 12 October 2022 22:59
> 
> As done by the ECB and CBC helpers in arch/x86/crypt/ecb_cbc_helpers.h,
> limit the number of bytes processed between kernel_fpu_begin() and
> kernel_fpu_end() calls.
> 
> Those functions call preempt_disable() and preempt_enable(), so
> the CPU core is unavailable for scheduling while running, leading to:
>     rcu: INFO: rcu_preempt detected expedited stalls on CPUs/tasks: ...

How long were the buffers being processed when the rcu stall was reported?
It looks like you are adding kernel_fpu_end(); kernel_fpu_begin()
pairs every 4096 bytes.
I'd guess the crc instruction runs at 4 bytes/clock
(or at least gets somewhere near that).
So you are talking of few thousand clocks at most.
A pci read from a device can easily take much longer than that.
So I'm surprised you need to do such small buffers to avoid
rcu stalls.

The kernel_fpu_end(); kernel_fpu_begin() pair pair will also cost.
(Maybe not as much as the first kernel_fpu_begin() ?)

Some performance figures might be enlightening.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)