Re: [PATCH] staging: vt6656: Declare a few variables as __read_mostly

From: Oscar Carter
Date: Sat Mar 07 2020 - 03:30:03 EST


On Sun, Mar 01, 2020 at 04:09:13PM +0100, Greg Kroah-Hartman wrote:
> On Sun, Mar 01, 2020 at 02:17:01PM +0100, Oscar Carter wrote:
> > On Sun, Mar 01, 2020 at 01:25:14PM +0100, Greg Kroah-Hartman wrote:
> > > On Sun, Mar 01, 2020 at 12:26:20PM +0100, Oscar Carter wrote:
> > > > These include module parameters.
> > > >
> > > > Signed-off-by: Oscar Carter <oscar.carter@xxxxxxx>
> > > > ---
> > > > drivers/staging/vt6656/main_usb.c | 4 ++--
> > > > 1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/staging/vt6656/main_usb.c b/drivers/staging/vt6656/main_usb.c
> > > > index 5e48b3ddb94c..701300202b21 100644
> > > > --- a/drivers/staging/vt6656/main_usb.c
> > > > +++ b/drivers/staging/vt6656/main_usb.c
> > > > @@ -49,12 +49,12 @@ MODULE_LICENSE("GPL");
> > > > MODULE_DESCRIPTION(DEVICE_FULL_DRV_NAM);
> > > >
> > > > #define RX_DESC_DEF0 64
> > > > -static int vnt_rx_buffers = RX_DESC_DEF0;
> > > > +static int __read_mostly vnt_rx_buffers = RX_DESC_DEF0;
> > > > module_param_named(rx_buffers, vnt_rx_buffers, int, 0644);
> > > > MODULE_PARM_DESC(rx_buffers, "Number of receive usb rx buffers");
> > > >
> > > > #define TX_DESC_DEF0 64
> > > > -static int vnt_tx_buffers = TX_DESC_DEF0;
> > > > +static int __read_mostly vnt_tx_buffers = TX_DESC_DEF0;
> > > > module_param_named(tx_buffers, vnt_tx_buffers, int, 0644);
> > > > MODULE_PARM_DESC(tx_buffers, "Number of receive usb tx buffers");
> > > >
> > >
> > > Why? What does this help with?
> >
> > If we declare these variables __read_mostly we can improve the performance. If
> > these variables are read many more times than written, each core of a multicore
> > system can maintain a copy in a local cache and the time to access is less than
> > if they use the shared-cache.
>
> This is a USB driver, performance is always limited to the hardware, not
> the CPU location of variables.

Thank you for the explanation.

>
> Please always benchmark things to see if it actually makes sense to make
> changes like this, before proposing them.

I'm sorry.

>
> thanks,
>
> greg k-h

thanks,

Oscar