Re: [PATCH rdma-next v6] RDMA: Change capability fields in ib_device_attr from int to u32
David Laight <[email protected]> Tue, 2 Jun 2026 12:21:04 +0100
| Newsgroups | gmane.linux.kernel,gmane.linux.drivers.rdma,gmane.linux.scsi.target.devel,gmane.linux.kernel.cifs,gmane.network.samba.internals,gmane.linux.nfs,gmane.linux.network |
|---|---|
| Message-ID | <20260602122104.20afa8b4@pumpkin> |
On Tue, 2 Jun 2026 12:21:58 +0300 Andy Shevchenko <[email protected]> wrote: > On Mon, Jun 01, 2026 at 08:51:40AM -0700, Bart Van Assche wrote: > > On 6/1/26 2:25 AM, Erni Sri Satya Vennela wrote: > > ... > > > > - sdev->srq_size = min(srpt_srq_size, sdev->device->attrs.max_srq_wr); > > > + sdev->srq_size = min_t(u32, srpt_srq_size, sdev->device->attrs.max_srq_wr); > > > > min_t() shouldn't be used if there is an alternative available. For the > > SRP drivers, please make sure that both arguments of min() are unsigned > > instead of using min_t(). > > Ah, I just answered in similar way against v5. I also mentioned clamp() there. > IMHO it is also best to do min(value, 255) not min(255, value). Like an 'if' put the value you are comparing against second. The min_t(u8, x, y) you've removed are usually broken. Maybe I should change clamp() to allow clamp(int_var, 0, unsigned_var). That will need the order of the compares swapping (to do the low bound first). I think they used to be that way around, got changed by a commit that said it didn't change it! Correct code shouldn't care. -- David