[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <VI1PR0401MB259145C2DFDB5E4084EA5DFC98D20@VI1PR0401MB2591.eurprd04.prod.outlook.com>
Date: Thu, 29 Jun 2017 07:52:04 +0000
From: Horia Geantă <horia.geanta@....com>
To: Logan Gunthorpe <logang@...tatee.com>,
Arnd Bergmann <arnd@...db.de>
CC: Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
linux-arch <linux-arch@...r.kernel.org>,
"linux-ntb@...glegroups.com" <linux-ntb@...glegroups.com>,
"linux-crypto@...r.kernel.org" <linux-crypto@...r.kernel.org>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Jyri Sarha <jsarha@...com>,
Stephen Bates <sbates@...thlin.com>,
Dan Douglass <dan.douglass@....com>,
Herbert Xu <herbert@...dor.apana.org.au>,
"David S. Miller" <davem@...emloft.net>
Subject: Re: [PATCH v2 3/3] crypto: caam: cleanup CONFIG_64BIT ifdefs when
using io{read|write}64
On 6/28/2017 7:51 PM, Logan Gunthorpe wrote:
>
>
> On 28/06/17 04:20 AM, Arnd Bergmann wrote:
>> On Wed, Jun 28, 2017 at 1:02 AM, Logan Gunthorpe <logang@...tatee.com> wrote:
>>> #include <linux/types.h>
>>> #include <linux/bitops.h>
>>> -#include <linux/io.h>
>>> +#include <linux/io-64-nonatomic-hi-lo.h>
>>
>> Here you include the hi-lo variant unconditionally.
>>
Arnd, thanks for spotting this.
This was not in the patch I signed off.
The lo-hi variant should be used instead for CAAM, see further below.
>>> -#else /* CONFIG_64BIT */
>>> -static inline void wr_reg64(void __iomem *reg, u64 data)
>>> -{
>>> -#ifndef CONFIG_CRYPTO_DEV_FSL_CAAM_IMX
>>> - if (caam_little_end) {
>>> - wr_reg32((u32 __iomem *)(reg) + 1, data >> 32);
>>> - wr_reg32((u32 __iomem *)(reg), data);
>>> - } else
>>> #endif
>>> - {
>>> - wr_reg32((u32 __iomem *)(reg), data >> 32);
>>> - wr_reg32((u32 __iomem *)(reg) + 1, data);
>>> - }
>>> + iowrite64be(data, reg);
>>> }
>>
>> However, the #else path here uses lo-hi instead. I guess we have
>> to decide how to define iowrite64be_lo_hi() first: it could
>> either byteswap the 64-bit value first, then write the two halves,
>> or it could write the two halves, doing a 32-bit byte swap on
>> each.
>
> Ok, I studied this a bit more:
>
> The lo_hi/hi_lo functions seem to always refer to the data being written
> or read not to the address operated on. So, in the v3 version of this
> set, which I'm working on, I've defined:
>
> static inline void iowrite64_hi_lo(u64 val, void __iomem *addr)
> {
> iowrite32(val >> 32, addr + sizeof(u32));
> iowrite32(val, addr);
> }
>
> static inline void iowrite64be_hi_lo(u64 val, void __iomem *addr)
> {
> iowrite32be(val >> 32, addr);
> iowrite32be(val, addr + sizeof(u32));
> }
>
> So the two hi_lo functions match both paths of the #if and thus, I
> believe, the patch will be correct in v3 without changes.
>
To be consistent with CAAM engine HW spec: in case of 64-bit registers,
irrespective of device endianness, the lower address should be read from
/ written to first, followed by the upper address.
Indeed the I/O accessors in CAAM driver currently don't follow the spec,
however this is a good opportunity to fix the code.
I don't consider this requires a separate patch, as we haven't noticed
any problem. I'd say a simple note in the commit message mentioning the
change (lo-hi r/w order replacing hi-lo for little endian case) is enough.
Thanks,
Horia
Powered by blists - more mailing lists