2020-04-24 12:12:58

by Geert Uytterhoeven

Subject: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

On 32-bit architectures (e.g. m68k):

ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!

Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
Reported-by: [email protected]
Signed-off-by: Geert Uytterhoeven <[email protected]>
---
net/openvswitch/meter.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
index 915f31123f235c03..3498a5ab092ab2b8 100644
--- a/net/openvswitch/meter.c
+++ b/net/openvswitch/meter.c
@@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
* Start with a full bucket.
*/
band->bucket = (band->burst_size + band->rate) * 1000ULL;
- band_max_delta_t = band->bucket / band->rate;
+ band_max_delta_t = do_div(band->bucket, band->rate);
if (band_max_delta_t > meter->max_delta_t)
meter->max_delta_t = band_max_delta_t;
band++;
--
2.17.1

2020-04-24 15:08:25

by Eric Dumazet

Subject: Re: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

On 4/24/20 5:10 AM, Geert Uytterhoeven wrote:
> On 32-bit architectures (e.g. m68k):
>
> ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
> ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!
>
> Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
> Reported-by: [email protected]
> Signed-off-by: Geert Uytterhoeven <[email protected]>
> ---
> net/openvswitch/meter.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
> index 915f31123f235c03..3498a5ab092ab2b8 100644
> --- a/net/openvswitch/meter.c
> +++ b/net/openvswitch/meter.c
> @@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
> * Start with a full bucket.
> */
> band->bucket = (band->burst_size + band->rate) * 1000ULL;
> - band_max_delta_t = band->bucket / band->rate;
> + band_max_delta_t = do_div(band->bucket, band->rate);
> if (band_max_delta_t > meter->max_delta_t)
> meter->max_delta_t = band_max_delta_t;
> band++;
>

This is fascinating... Have you tested this patch ?

Please double check what do_div() return value is supposed to be !

Thanks.

2020-04-24 15:15:31

by Geert Uytterhoeven

Subject: Re: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

Hi Eric,

On Fri, Apr 24, 2020 at 5:05 PM Eric Dumazet <[email protected]> wrote:
> On 4/24/20 5:10 AM, Geert Uytterhoeven wrote:
> > On 32-bit architectures (e.g. m68k):
> >
> > ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
> > ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!
> >
> > Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
> > Reported-by: [email protected]
> > Signed-off-by: Geert Uytterhoeven <[email protected]>
> > ---
> > net/openvswitch/meter.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
> > index 915f31123f235c03..3498a5ab092ab2b8 100644
> > --- a/net/openvswitch/meter.c
> > +++ b/net/openvswitch/meter.c
> > @@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
> > * Start with a full bucket.
> > */
> > band->bucket = (band->burst_size + band->rate) * 1000ULL;
> > - band_max_delta_t = band->bucket / band->rate;
> > + band_max_delta_t = do_div(band->bucket, band->rate);
> > if (band_max_delta_t > meter->max_delta_t)
> > meter->max_delta_t = band_max_delta_t;
> > band++;
> >
>
> This is fascinating... Have you tested this patch ?

Sorry, I should have said this is compile-tested only.

> Please double check what do_div() return value is supposed to be !

I do not have any openvswitch setups, let alone on the poor m68k box.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

2020-04-24 16:48:31

by Toke Høiland-Jørgensen

Subject: Re: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

Geert Uytterhoeven <[email protected]> writes:

> Hi Eric,
>
> On Fri, Apr 24, 2020 at 5:05 PM Eric Dumazet <[email protected]> wrote:
>> On 4/24/20 5:10 AM, Geert Uytterhoeven wrote:
>> > On 32-bit architectures (e.g. m68k):
>> >
>> > ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
>> > ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!
>> >
>> > Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
>> > Reported-by: [email protected]
>> > Signed-off-by: Geert Uytterhoeven <[email protected]>
>> > ---
>> > net/openvswitch/meter.c | 2 +-
>> > 1 file changed, 1 insertion(+), 1 deletion(-)
>> >
>> > diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
>> > index 915f31123f235c03..3498a5ab092ab2b8 100644
>> > --- a/net/openvswitch/meter.c
>> > +++ b/net/openvswitch/meter.c
>> > @@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
>> > * Start with a full bucket.
>> > */
>> > band->bucket = (band->burst_size + band->rate) * 1000ULL;
>> > - band_max_delta_t = band->bucket / band->rate;
>> > + band_max_delta_t = do_div(band->bucket, band->rate);
>> > if (band_max_delta_t > meter->max_delta_t)
>> > meter->max_delta_t = band_max_delta_t;
>> > band++;
>> >
>>
>> This is fascinating... Have you tested this patch ?
>
> Sorry, I should have said this is compile-tested only.
>
>> Please double check what do_div() return value is supposed to be !
>
> I do not have any openvswitch setups, let alone on the poor m68k box.

I think what Eric is referring to is this, from the documentation of
do_div:

* do_div - returns 2 values: calculate remainder and update new dividend
* @n: uint64_t dividend (will be updated)
* @base: uint32_t divisor
*
* Summary:
* ``uint32_t remainder = n % base;``
* ``n = n / base;``
*
* Return: (uint32_t)remainder


Specifically that last part :)

-Toke

2020-04-24 17:59:05

by Geert Uytterhoeven

Subject: Re: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

Hi Toke,

On Fri, Apr 24, 2020 at 6:45 PM Toke Høiland-Jørgensen <[email protected]> wrote:
> Geert Uytterhoeven <[email protected]> writes:
> > On Fri, Apr 24, 2020 at 5:05 PM Eric Dumazet <[email protected]> wrote:
> >> On 4/24/20 5:10 AM, Geert Uytterhoeven wrote:
> >> > On 32-bit architectures (e.g. m68k):
> >> >
> >> > ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
> >> > ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!
> >> >
> >> > Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
> >> > Reported-by: [email protected]
> >> > Signed-off-by: Geert Uytterhoeven <[email protected]>
> >> > ---
> >> > net/openvswitch/meter.c | 2 +-
> >> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >> >
> >> > diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
> >> > index 915f31123f235c03..3498a5ab092ab2b8 100644
> >> > --- a/net/openvswitch/meter.c
> >> > +++ b/net/openvswitch/meter.c
> >> > @@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
> >> > * Start with a full bucket.
> >> > */
> >> > band->bucket = (band->burst_size + band->rate) * 1000ULL;
> >> > - band_max_delta_t = band->bucket / band->rate;
> >> > + band_max_delta_t = do_div(band->bucket, band->rate);
> >> > if (band_max_delta_t > meter->max_delta_t)
> >> > meter->max_delta_t = band_max_delta_t;
> >> > band++;
> >> >
> >>
> >> This is fascinating... Have you tested this patch ?
> >
> > Sorry, I should have said this is compile-tested only.
> >
> >> Please double check what do_div() return value is supposed to be !
> >
> > I do not have any openvswitch setups, let alone on the poor m68k box.
>
> I think what Eric is referring to is this, from the documentation of
> do_div:
>
> * do_div - returns 2 values: calculate remainder and update new dividend
> * @n: uint64_t dividend (will be updated)
> * @base: uint32_t divisor
> *
> * Summary:
> * ``uint32_t remainder = n % base;``
> * ``n = n / base;``
> *
> * Return: (uint32_t)remainder

Oops, that was a serious brain fart. Sorry for that!

That should have been div_u64() instead of do_div().
Feel free to update my patch while applying, else I'll send a v2 later.

/me hides in a brown paper bag for the whole weekend...

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

2020-04-25 03:44:56

by Tonghao Zhang

Subject: Re: [PATCH] net: openvswitch: use do_div() for 64-by-32 divisions:

On Sat, Apr 25, 2020 at 1:54 AM Geert Uytterhoeven <[email protected]> wrote:
>
> Hi Toke,
>
> On Fri, Apr 24, 2020 at 6:45 PM Toke Høiland-Jørgensen <[email protected]> wrote:
> > Geert Uytterhoeven <[email protected]> writes:
> > > On Fri, Apr 24, 2020 at 5:05 PM Eric Dumazet <[email protected]> wrote:
> > >> On 4/24/20 5:10 AM, Geert Uytterhoeven wrote:
> > >> > On 32-bit architectures (e.g. m68k):
> > >> >
> > >> > ERROR: modpost: "__udivdi3" [net/openvswitch/openvswitch.ko] undefined!
> > >> > ERROR: modpost: "__divdi3" [net/openvswitch/openvswitch.ko] undefined!
> > >> >
> > >> > Fixes: e57358873bb5d6ca ("net: openvswitch: use u64 for meter bucket")
> > >> > Reported-by: [email protected]
> > >> > Signed-off-by: Geert Uytterhoeven <[email protected]>
> > >> > ---
> > >> > net/openvswitch/meter.c | 2 +-
> > >> > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >> >
> > >> > diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
> > >> > index 915f31123f235c03..3498a5ab092ab2b8 100644
> > >> > --- a/net/openvswitch/meter.c
> > >> > +++ b/net/openvswitch/meter.c
> > >> > @@ -393,7 +393,7 @@ static struct dp_meter *dp_meter_create(struct nlattr **a)
> > >> > * Start with a full bucket.
> > >> > */
> > >> > band->bucket = (band->burst_size + band->rate) * 1000ULL;
> > >> > - band_max_delta_t = band->bucket / band->rate;
> > >> > + band_max_delta_t = do_div(band->bucket, band->rate);
> > >> > if (band_max_delta_t > meter->max_delta_t)
> > >> > meter->max_delta_t = band_max_delta_t;
> > >> > band++;
> > >> >
> > >>
> > >> This is fascinating... Have you tested this patch ?
> > >
> > > Sorry, I should have said this is compile-tested only.
> > >
> > >> Please double check what do_div() return value is supposed to be !
> > >
> > > I do not have any openvswitch setups, let alone on the poor m68k box.
> >
> > I think what Eric is referring to is this, from the documentation of
> > do_div:
> >
> > * do_div - returns 2 values: calculate remainder and update new dividend
> > * @n: uint64_t dividend (will be updated)
> > * @base: uint32_t divisor
> > *
> > * Summary:
> > * ``uint32_t remainder = n % base;``
> > * ``n = n / base;``
> > *
> > * Return: (uint32_t)remainder
>
> Oops, that was a serious brain fart. Sorry for that!
>
> That should have been div_u64() instead of do_div().
> Feel free to update my patch while applying, else I'll send a v2 later.
>
> /me hides in a brown paper bag for the whole weekend...
I sent the patch with updated, change the reported tag, test it, and
add more commit msg,
and author should be you. Thanks.
> Gr{oetje,eeting}s,
>
> Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
> -- Linus Torvalds



--
Best regards, Tonghao