Replace cpu_to_be64(le64_to_cpu()) with swab64() to simplify
ionic_prep_reg().  No functional changes.

Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
---
 drivers/infiniband/hw/ionic/ionic_datapath.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/infiniband/hw/ionic/ionic_datapath.c b/drivers/infiniband/hw/ionic/ionic_datapath.c
index aa2944887f23..1a1cf82d1745 100644
--- a/drivers/infiniband/hw/ionic/ionic_datapath.c
+++ b/drivers/infiniband/hw/ionic/ionic_datapath.c
@@ -1105,7 +1105,7 @@ static int ionic_prep_reg(struct ionic_qp *qp,
 	wqe->reg_mr.length = cpu_to_be64(mr->ibmr.length);
 	wqe->reg_mr.offset = ionic_pgtbl_off(&mr->buf, mr->ibmr.iova);
 	dma_addr = ionic_pgtbl_dma(&mr->buf, mr->ibmr.iova);
-	wqe->reg_mr.dma_addr = cpu_to_be64(le64_to_cpu(dma_addr));
+	wqe->reg_mr.dma_addr = swab64(dma_addr);
 
 	wqe->reg_mr.map_count = cpu_to_be32(mr->buf.tbl_pages);
 	wqe->reg_mr.flags = cpu_to_be16(flags);
-- 
Thorsten Blum <thorsten.blum@linux.dev>
GPG: 1D60 735E 8AEF 3BE4 73B6  9D84 7336 78FD 8DFE EAD4

On Wed, Dec 10, 2025 at 02:14:29PM +0100, Thorsten Blum wrote:
> Replace cpu_to_be64(le64_to_cpu()) with swab64() to simplify
> ionic_prep_reg().  No functional changes.
> 
> Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
> ---
>  drivers/infiniband/hw/ionic/ionic_datapath.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/infiniband/hw/ionic/ionic_datapath.c b/drivers/infiniband/hw/ionic/ionic_datapath.c
> index aa2944887f23..1a1cf82d1745 100644
> --- a/drivers/infiniband/hw/ionic/ionic_datapath.c
> +++ b/drivers/infiniband/hw/ionic/ionic_datapath.c
> @@ -1105,7 +1105,7 @@ static int ionic_prep_reg(struct ionic_qp *qp,
>  	wqe->reg_mr.length = cpu_to_be64(mr->ibmr.length);
>  	wqe->reg_mr.offset = ionic_pgtbl_off(&mr->buf, mr->ibmr.iova);
>  	dma_addr = ionic_pgtbl_dma(&mr->buf, mr->ibmr.iova);
> -	wqe->reg_mr.dma_addr = cpu_to_be64(le64_to_cpu(dma_addr));
> +	wqe->reg_mr.dma_addr = swab64(dma_addr);

This doesn't make any sense to me. The original code looks wrong and
would fail sparse, switching to swab just highlights how nonsense it
is, there is no way that is right on BE and LE.

Pensando guys what is the right thing to do here??

Jason

On 12/12/25 12:19, Jason Gunthorpe wrote:
> On Wed, Dec 10, 2025 at 02:14:29PM +0100, Thorsten Blum wrote:
>> Replace cpu_to_be64(le64_to_cpu()) with swab64() to simplify
>> ionic_prep_reg().  No functional changes.
>>
>> Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
>> ---
>>   drivers/infiniband/hw/ionic/ionic_datapath.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/infiniband/hw/ionic/ionic_datapath.c b/drivers/infiniband/hw/ionic/ionic_datapath.c
>> index aa2944887f23..1a1cf82d1745 100644
>> --- a/drivers/infiniband/hw/ionic/ionic_datapath.c
>> +++ b/drivers/infiniband/hw/ionic/ionic_datapath.c
>> @@ -1105,7 +1105,7 @@ static int ionic_prep_reg(struct ionic_qp *qp,
>>   	wqe->reg_mr.length = cpu_to_be64(mr->ibmr.length);
>>   	wqe->reg_mr.offset = ionic_pgtbl_off(&mr->buf, mr->ibmr.iova);
>>   	dma_addr = ionic_pgtbl_dma(&mr->buf, mr->ibmr.iova);
>> -	wqe->reg_mr.dma_addr = cpu_to_be64(le64_to_cpu(dma_addr));
>> +	wqe->reg_mr.dma_addr = swab64(dma_addr);
> This doesn't make any sense to me. The original code looks wrong and
> would fail sparse, switching to swab just highlights how nonsense it
> is, there is no way that is right on BE and LE.
>
> Pensando guys what is the right thing to do here??
>
> Jason

The original code does not have sparse failure. ionic_pgtbl_dma() is 
returning __le64, which is what swapped to __be64 for 
wqe->reg_me.dma_addr. However the proposed fix is definitely going to 
throw sparse warning.

Thanks,
Abhijit

On Fri, Dec 12, 2025 at 01:54:17PM +0530, Abhijit Gangurde wrote:
> 
> On 12/12/25 12:19, Jason Gunthorpe wrote:
> > On Wed, Dec 10, 2025 at 02:14:29PM +0100, Thorsten Blum wrote:
> > > Replace cpu_to_be64(le64_to_cpu()) with swab64() to simplify
> > > ionic_prep_reg().  No functional changes.
> > > 
> > > Signed-off-by: Thorsten Blum <thorsten.blum@linux.dev>
> > > ---
> > >   drivers/infiniband/hw/ionic/ionic_datapath.c | 2 +-
> > >   1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/infiniband/hw/ionic/ionic_datapath.c b/drivers/infiniband/hw/ionic/ionic_datapath.c
> > > index aa2944887f23..1a1cf82d1745 100644
> > > --- a/drivers/infiniband/hw/ionic/ionic_datapath.c
> > > +++ b/drivers/infiniband/hw/ionic/ionic_datapath.c
> > > @@ -1105,7 +1105,7 @@ static int ionic_prep_reg(struct ionic_qp *qp,
> > >   	wqe->reg_mr.length = cpu_to_be64(mr->ibmr.length);
> > >   	wqe->reg_mr.offset = ionic_pgtbl_off(&mr->buf, mr->ibmr.iova);
> > >   	dma_addr = ionic_pgtbl_dma(&mr->buf, mr->ibmr.iova);
> > > -	wqe->reg_mr.dma_addr = cpu_to_be64(le64_to_cpu(dma_addr));
> > > +	wqe->reg_mr.dma_addr = swab64(dma_addr);
> > This doesn't make any sense to me. The original code looks wrong and
> > would fail sparse, switching to swab just highlights how nonsense it
> > is, there is no way that is right on BE and LE.
> > 
> > Pensando guys what is the right thing to do here??
> 
> The original code does not have sparse failure. ionic_pgtbl_dma() is
> returning __le64, which is what swapped to __be64 for wqe->reg_me.dma_addr.
> However the proposed fix is definitely going to throw sparse warning.

Okay, so Throsten, please don't send patches for changing to swab.

If you want to improve it then the primitive should be

 le64_to_be64(x)

(and maybe vice versa) With proper sparse annotations.

Jason

On 16. Dec 2025, at 01:51, Jason Gunthorpe wrote:
> Okay, so Throsten, please don't send patches for changing to swab.

Yeah sorry, I didn't know about the sparse warnings before.

> If you want to improve it then the primitive should be
> 
> le64_to_be64(x)

le64_to_be64() or a similar helper doesn't seem to exist.

Thanks,
Thorsten

On Tue, Dec 16, 2025 at 01:35:02PM +0100, Thorsten Blum wrote:
> On 16. Dec 2025, at 01:51, Jason Gunthorpe wrote:
> > Okay, so Throsten, please don't send patches for changing to swab.
> 
> Yeah sorry, I didn't know about the sparse warnings before.
> 
> > If you want to improve it then the primitive should be
> > 
> > le64_to_be64(x)
> 
> le64_to_be64() or a similar helper doesn't seem to exist.

You can add it.

Thanks

> 
> Thanks,
> Thorsten
>