Re: [PATCH] scsi: Let scsi_execute_cmd() mark args->sshdr as invalid

[Juergen Gross <jgross@suse.com>]

[-- Attachment #1.1.1: Type: text/plain, Size: 1030 bytes --]

On 11.05.23 15:23, Martin Wilck wrote:
> On Thu, 2023-05-11 at 15:17 +0200, Juergen Gross wrote:
>>>
>>> We know for certain that sizeof(*sshdr) is 8 bytes, and will most
>>> probably remain so. Thus
>>>
>>>       memset(sshdr, 0, sizeof(*sshdr))
>>>
>>> would result in more efficient code.
>>
>> I fail to see why zeroing a single byte would be less efficient than
>> zeroing
>> a possibly unaligned 8-byte area.
> 
> I don't think it can be unaligned. gcc seems to think the same. It
> compiles the memset(sshdr, ...) in scsi_normalize_sense() into a single
> instruction on x86_64.
> 
> 0xffffffff8177e9d0 <scsi_normalize_sense>:      nopl   0x0(%rax,%rax,1) [FTRACE NOP]
> 0xffffffff8177e9d5 <scsi_normalize_sense+5>:    test   %rdi,%rdi
> 0xffffffff8177e9d8 <scsi_normalize_sense+8>:    movq   $0x0,(%rdx)

A struct with 8 "u8" fields can be unaligned.

x86_64 can do unaligned 8-byte stores.

Other architectures can't (e.g. MIPS). And 32-bit architectures might need
2 stores.


Juergen

[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 3149 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]