Re: [PATCH v4 1/1] scsi: ufs: core: add device level exception support

[manivannan.sadhasivam@xxxxxxxxxx]

On Wed, Apr 02, 2025 at 12:00:26PM -0700, Bao D. Nguyen wrote:
> On 4/2/2025 12:49 AM, manivannan.sadhasivam@xxxxxxxxxx wrote:
> > Yeah, we should be cautious in changing the uAPI header as it can break the
> > userspace applications. Annotating the members that need packed attribute seems
> > like the way forward to me.
> 
> Yes, I realized potential issue when Bart raised a concern.
> 
> > 
> > Though, I'd like to understand which architecture has the alignment constraint
> > in this structure. Only if an architecture requires 8 byte alignment for __be32
> > would be a problem. That too only for osf7 and reserved. But I'm not aware of
> > such architectures in use.
> When using "__u64 value;" in place of osf3-6, I saw the compiler padded 4
> bytes, so __packed was needed for me to get correct __u64 value. I thought
> even the existing structure utp_upiu_query_v4_0 may need __packed on some
> fields where the driver reads the returned data in order to be safe across
> all architectures. However, without evidence of an actual failure, I didn't
> touch the existing structure. Only raised potential issue for discussion.
> 

If you change members to be 64bit, then for sure compiler will add padding to
avoid holes. But I don't see any issue with the unchanged utp_upiu_query_v4_0
structure.

- Mani

-- 
மணிவண்ணன் சதாசிவம்