[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <6f4c21b5e1cb4fa38a5e0c1716658329@AcuMS.aculab.com>
Date: Sun, 20 Feb 2022 18:48:58 +0000
From: David Laight <David.Laight@...LAB.COM>
To: 'Armin Wolf' <W_Armin@....de>, "pali@...nel.org" <pali@...nel.org>
CC: "jdelvare@...e.com" <jdelvare@...e.com>,
"linux@...ck-us.net" <linux@...ck-us.net>,
"linux-hwmon@...r.kernel.org" <linux-hwmon@...r.kernel.org>,
"linux-assembly@...r.kernel.org" <linux-assembly@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH v2] hwmon: (dell-smm) Improve assembly code
From: Armin Wolf
> Sent: 19 February 2022 21:10
>
> The new assembly code works on both 32 bit and 64 bit
> cpus and allows for more compiler optimisations by not
> requiring smm_regs to be packed
I'm intrigued about the __packed..
Prior to 5.17-rc1 __packed was only applied to the fields after 'eax'.
This actually has no effect (on any architecture) because there is
no padding to remove - so all the later fields are still assumed to
be 32bit aligned.
5.17-rc1 (565210c781201) moved the __packed to the end of the
structure.
AFAICT this structure is only ever used in one file and for on-stack
items. It will always actually be aligned and is only read by the
code in the file - so why was it ever marked __packed at all!
On x86 it would make no difference anyway.
I can only guess it was to ensure that the asm code didn't go
'wrong' because of the compiler adding 'random' padding.
That isn't what __packed is for at all.
The linux kernel requires that the compiler doesn't add 'random'
padding - even if the C standard might allow it.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Powered by blists - more mailing lists