Using the real macro is usually more intuitive and readable,
When the original file is guaranteed to contain the minmax.h header file 
and compile correctly.

Signed-off-by: Yan Zhen <yanzhen@vivo.com>
---
 drivers/net/ethernet/marvell/mvneta.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
index d72b2d5f96db..415d2b9e63f9 100644
--- a/drivers/net/ethernet/marvell/mvneta.c
+++ b/drivers/net/ethernet/marvell/mvneta.c
@@ -4750,8 +4750,7 @@ mvneta_ethtool_set_ringparam(struct net_device *dev,
 
 	if ((ring->rx_pending == 0) || (ring->tx_pending == 0))
 		return -EINVAL;
-	pp->rx_ring_size = ring->rx_pending < MVNETA_MAX_RXD ?
-		ring->rx_pending : MVNETA_MAX_RXD;
+	pp->rx_ring_size = min(ring->rx_pending, MVNETA_MAX_RXD);
 
 	pp->tx_ring_size = clamp_t(u16, ring->tx_pending,
 				   MVNETA_MAX_SKB_DESCS * 2, MVNETA_MAX_TXD);
-- 
2.34.1

On Tue, Aug 27, 2024 at 07:58:48PM +0800, Yan Zhen wrote:
> Using the real macro is usually more intuitive and readable,
> When the original file is guaranteed to contain the minmax.h header file 
> and compile correctly.
> 
> Signed-off-by: Yan Zhen <yanzhen@vivo.com>
> ---
>  drivers/net/ethernet/marvell/mvneta.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> index d72b2d5f96db..415d2b9e63f9 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -4750,8 +4750,7 @@ mvneta_ethtool_set_ringparam(struct net_device *dev,
>  
>  	if ((ring->rx_pending == 0) || (ring->tx_pending == 0))
>  		return -EINVAL;
> -	pp->rx_ring_size = ring->rx_pending < MVNETA_MAX_RXD ?
> -		ring->rx_pending : MVNETA_MAX_RXD;
> +	pp->rx_ring_size = min(ring->rx_pending, MVNETA_MAX_RXD);

Given that the type of ring->rx_pending is __32, and MVNETA_MAX_RXD is
a positive value.

See: 80fcac55385c ("minmax: add umin(a, b) and umax(a, b)")
     https://git.kernel.org/torvalds/c/80fcac55385c

>  
>  	pp->tx_ring_size = clamp_t(u16, ring->tx_pending,
>  				   MVNETA_MAX_SKB_DESCS * 2, MVNETA_MAX_TXD);
> -- 
> 2.34.1
> 
>

On Tue, Aug 27, 2024 at 06:54:08PM +0100, Simon Horman wrote:
> On Tue, Aug 27, 2024 at 07:58:48PM +0800, Yan Zhen wrote:
> > Using the real macro is usually more intuitive and readable,
> > When the original file is guaranteed to contain the minmax.h header file 
> > and compile correctly.
> > 
> > Signed-off-by: Yan Zhen <yanzhen@vivo.com>
> > ---
> >  drivers/net/ethernet/marvell/mvneta.c | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> > 
> > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > index d72b2d5f96db..415d2b9e63f9 100644
> > --- a/drivers/net/ethernet/marvell/mvneta.c
> > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > @@ -4750,8 +4750,7 @@ mvneta_ethtool_set_ringparam(struct net_device *dev,
> >  
> >  	if ((ring->rx_pending == 0) || (ring->tx_pending == 0))
> >  		return -EINVAL;
> > -	pp->rx_ring_size = ring->rx_pending < MVNETA_MAX_RXD ?
> > -		ring->rx_pending : MVNETA_MAX_RXD;
> > +	pp->rx_ring_size = min(ring->rx_pending, MVNETA_MAX_RXD);
> 
> Given that the type of ring->rx_pending is __32, and MVNETA_MAX_RXD is
> a positive value.

Sorry, I hit send to soon. What I wanted to say is:

I think that it is appropriate to use umin() here.
Because:
1) As I understand things, the type of MVNETA_MAX_RXD is signed,
   but it always holds a positive value
2) ring->rx_pending is unsigned

> See: 80fcac55385c ("minmax: add umin(a, b) and umax(a, b)")
>      https://git.kernel.org/torvalds/c/80fcac55385c

From: Simon Horman
> Sent: 27 August 2024 18:58
> 
> On Tue, Aug 27, 2024 at 06:54:08PM +0100, Simon Horman wrote:
> > On Tue, Aug 27, 2024 at 07:58:48PM +0800, Yan Zhen wrote:
> > > Using the real macro is usually more intuitive and readable,
> > > When the original file is guaranteed to contain the minmax.h header file
> > > and compile correctly.
> > >
> > > Signed-off-by: Yan Zhen <yanzhen@vivo.com>
> > > ---
> > >  drivers/net/ethernet/marvell/mvneta.c | 3 +--
> > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > > index d72b2d5f96db..415d2b9e63f9 100644
> > > --- a/drivers/net/ethernet/marvell/mvneta.c
> > > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > > @@ -4750,8 +4750,7 @@ mvneta_ethtool_set_ringparam(struct net_device *dev,
> > >
> > >  	if ((ring->rx_pending == 0) || (ring->tx_pending == 0))
> > >  		return -EINVAL;
> > > -	pp->rx_ring_size = ring->rx_pending < MVNETA_MAX_RXD ?
> > > -		ring->rx_pending : MVNETA_MAX_RXD;
> > > +	pp->rx_ring_size = min(ring->rx_pending, MVNETA_MAX_RXD);
> >
> > Given that the type of ring->rx_pending is __32, and MVNETA_MAX_RXD is
> > a positive value.
> 
> Sorry, I hit send to soon. What I wanted to say is:
> 
> I think that it is appropriate to use umin() here.
> Because:
> 1) As I understand things, the type of MVNETA_MAX_RXD is signed,
>    but it always holds a positive value
> 2) ring->rx_pending is unsigned

Provided MVNETA_MAX_RXD is constant it is fine.
umin() is only needed for signed variables that can only contain
non-negative values.

You only need to use it is the compiler bleats...

umin(x, y) is safer than min_t(unsigned_type, x, y) because you can't
get the type wrong.
If will also generate better code since it never sign extends a
32bit value to 64bits (expensive on 32bit).

	David

> 
> > See: 80fcac55385c ("minmax: add umin(a, b) and umax(a, b)")
> >      https://git.kernel.org/torvalds/c/80fcac55385c

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

On Sat, Aug 31, 2024 at 12:39:28PM +0000, David Laight wrote:
> From: Simon Horman
> > Sent: 27 August 2024 18:58
> > 
> > On Tue, Aug 27, 2024 at 06:54:08PM +0100, Simon Horman wrote:
> > > On Tue, Aug 27, 2024 at 07:58:48PM +0800, Yan Zhen wrote:
> > > > Using the real macro is usually more intuitive and readable,
> > > > When the original file is guaranteed to contain the minmax.h header file
> > > > and compile correctly.
> > > >
> > > > Signed-off-by: Yan Zhen <yanzhen@vivo.com>
> > > > ---
> > > >  drivers/net/ethernet/marvell/mvneta.c | 3 +--
> > > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> > > > index d72b2d5f96db..415d2b9e63f9 100644
> > > > --- a/drivers/net/ethernet/marvell/mvneta.c
> > > > +++ b/drivers/net/ethernet/marvell/mvneta.c
> > > > @@ -4750,8 +4750,7 @@ mvneta_ethtool_set_ringparam(struct net_device *dev,
> > > >
> > > >  	if ((ring->rx_pending == 0) || (ring->tx_pending == 0))
> > > >  		return -EINVAL;
> > > > -	pp->rx_ring_size = ring->rx_pending < MVNETA_MAX_RXD ?
> > > > -		ring->rx_pending : MVNETA_MAX_RXD;
> > > > +	pp->rx_ring_size = min(ring->rx_pending, MVNETA_MAX_RXD);
> > >
> > > Given that the type of ring->rx_pending is __32, and MVNETA_MAX_RXD is
> > > a positive value.
> > 
> > Sorry, I hit send to soon. What I wanted to say is:
> > 
> > I think that it is appropriate to use umin() here.
> > Because:
> > 1) As I understand things, the type of MVNETA_MAX_RXD is signed,
> >    but it always holds a positive value
> > 2) ring->rx_pending is unsigned
> 
> Provided MVNETA_MAX_RXD is constant it is fine.
> umin() is only needed for signed variables that can only contain
> non-negative values.
> 
> You only need to use it is the compiler bleats...
> 
> umin(x, y) is safer than min_t(unsigned_type, x, y) because you can't
> get the type wrong.
> If will also generate better code since it never sign extends a
> 32bit value to 64bits (expensive on 32bit).

Hi David,

My understanding of umin() was a bit off - I thought it was
also relevant when one of the arguments is a constant.

Thanks for setting me straight.