On Mon, Sep 02, 2024 at 03:01:28PM +0200, Eric Dumazet wrote:
> On Sat, Aug 31, 2024 at 1:32 PM Joe Damato <jdamato@xxxxxxxxxx> wrote:
> >
> > In commit 6f8b12d661d0 ("net: napi: add hard irqs deferral feature")
> > napi_defer_irqs was added to net_device and napi_defer_irqs_count was
> > added to napi_struct, both as type int.
> >
> > This value never goes below zero. Change the type for both from int to
> > u32, and add an overflow check to sysfs to limit the value to S32_MAX.
> >
> > Before this patch:
> >
> > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs'
> > $ cat /sys/class/net/eth4/napi_defer_hard_irqs
> > -2147483647
> >
> > After this patch:
> >
> > $ sudo bash -c 'echo 2147483649 > /sys/class/net/eth4/napi_defer_hard_irqs'
> > bash: line 0: echo: write error: Numerical result out of range
> >
> > Fixes: 6f8b12d661d0 ("net: napi: add hard irqs deferral feature")
> > Cc: stable@xxxxxxxxxx
> > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > Suggested-by: Jakub Kicinski <kuba@xxxxxxxxxx>
> > Signed-off-by: Joe Damato <jdamato@xxxxxxxxxx>
> > ---
>
> I do not think this deserves a change to stable trees.
>
> Signed or unsigned, what is the issue ?
>
> Do you really need one extra bit ?
>
> /sys/class/net/XXXXX/tx_queue_len has a similar behavior.
Sorry, Eric, I'm not following.
Are you asking me to allow u32_max for napi_defer_hard_irqs the same
way tx_queue_len does and avoid overflow that way:
include/linux/netdevice.h: unsigned int tx_queue_len;
dev_change_tx_queue_len(..., unsigned long new_len):
if (new_len != (unsigned int)new_len)
return -ERANGE;
