2023-03-08 16:27:05

by Leesoo Ahn

[permalink] [raw]
Subject: [PATCH net-next] net: stmmac: call stmmac_finalize_xdp_rx() on a condition

The current codebase calls the function no matter net device has XDP
programs or not. So the finalize function is being called everytime when RX
bottom-half in progress. It needs a few machine instructions for nothing
in the case that XDP programs are not attached at all.

Lets it call the function on a condition that if xdp_status variable has
not zero value. That means XDP programs are attached to the net device
and it should be finalized based on the variable.

The following instructions show that it's better than calling the function
unconditionally.

0.31 │6b8: ldr w0, [sp, #196]
│ ┌──cbz w0, 6cc
│ │ mov x1, x0
│ │ mov x0, x27
│ │→ bl stmmac_finalize_xdp_rx
│6cc:└─→ldr x1, [sp, #176]

with 'if (xdp_status)' statement, jump to '6cc' label if xdp_status has
zero value.

Signed-off-by: Leesoo Ahn <[email protected]>
---
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index e4902a7bb61e..53c6e9b3a0c2 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -5145,7 +5145,8 @@ static int stmmac_rx_zc(struct stmmac_priv *priv, int limit, u32 queue)
rx_q->state.len = len;
}

- stmmac_finalize_xdp_rx(priv, xdp_status);
+ if (xdp_status)
+ stmmac_finalize_xdp_rx(priv, xdp_status);

priv->xstats.rx_pkt_n += count;
priv->xstats.rxq_stats[queue].rx_pkt_n += count;
@@ -5425,7 +5426,8 @@ static int stmmac_rx(struct stmmac_priv *priv, int limit, u32 queue)
rx_q->state.len = len;
}

- stmmac_finalize_xdp_rx(priv, xdp_status);
+ if (xdp_status)
+ stmmac_finalize_xdp_rx(priv, xdp_status);

stmmac_rx_refill(priv, queue);

--
2.34.1



2023-03-09 13:40:58

by Simon Horman

[permalink] [raw]
Subject: Re: [PATCH net-next] net: stmmac: call stmmac_finalize_xdp_rx() on a condition

On Thu, Mar 09, 2023 at 01:26:18AM +0900, Leesoo Ahn wrote:
> The current codebase calls the function no matter net device has XDP
> programs or not. So the finalize function is being called everytime when RX
> bottom-half in progress. It needs a few machine instructions for nothing
> in the case that XDP programs are not attached at all.
>
> Lets it call the function on a condition that if xdp_status variable has
> not zero value. That means XDP programs are attached to the net device
> and it should be finalized based on the variable.
>
> The following instructions show that it's better than calling the function
> unconditionally.
>
> 0.31 │6b8: ldr w0, [sp, #196]
> │ ┌──cbz w0, 6cc
> │ │ mov x1, x0
> │ │ mov x0, x27
> │ │→ bl stmmac_finalize_xdp_rx
> │6cc:└─→ldr x1, [sp, #176]
>
> with 'if (xdp_status)' statement, jump to '6cc' label if xdp_status has
> zero value.
>
> Signed-off-by: Leesoo Ahn <[email protected]>

Hi Leesoo,

I am curious to know if you considered going a step further and using
a static key.

Link: https://www.kernel.org/doc/html/latest/staging/static-keys.html

2023-03-09 15:12:30

by Leesoo Ahn

[permalink] [raw]
Subject: Re: [PATCH net-next] net: stmmac: call stmmac_finalize_xdp_rx() on a condition



On 23. 3. 9. 22:40, Simon Horman wrote:
> On Thu, Mar 09, 2023 at 01:26:18AM +0900, Leesoo Ahn wrote:
>> The current codebase calls the function no matter net device has XDP
>> programs or not. So the finalize function is being called everytime when RX
>> bottom-half in progress. It needs a few machine instructions for nothing
>> in the case that XDP programs are not attached at all.
>>
>> Lets it call the function on a condition that if xdp_status variable has
>> not zero value. That means XDP programs are attached to the net device
>> and it should be finalized based on the variable.
>>
>> The following instructions show that it's better than calling the function
>> unconditionally.
>>
>> 0.31 │6b8: ldr w0, [sp, #196]
>> │ ┌──cbz w0, 6cc
>> │ │ mov x1, x0
>> │ │ mov x0, x27
>> │ │→ bl stmmac_finalize_xdp_rx
>> │6cc:└─→ldr x1, [sp, #176]
>>
>> with 'if (xdp_status)' statement, jump to '6cc' label if xdp_status has
>> zero value.
>>
>> Signed-off-by: Leesoo Ahn <[email protected]>
> Hi Leesoo,
>
> I am curious to know if you considered going a step further and using
> a static key.
>
> Link: https://www.kernel.org/doc/html/latest/staging/static-keys.html

Thank you for the review.

The function must be called for only XDP_TX or XDP_REDIRECT cases.
So using a static key doesn't look good and the commit message is not
clear for 'why' as well.
I think that's why you suggested for using 'static key' by the latter
reason.

I will edit the message and post v2 soon.

Best regards,
Leesoo

2023-03-09 16:47:01

by Simon Horman

[permalink] [raw]
Subject: Re: [PATCH net-next] net: stmmac: call stmmac_finalize_xdp_rx() on a condition

On Fri, Mar 10, 2023 at 12:08:29AM +0900, Leesoo Ahn wrote:
>
>
> On 23. 3. 9. 22:40, Simon Horman wrote:
> > On Thu, Mar 09, 2023 at 01:26:18AM +0900, Leesoo Ahn wrote:
> > > The current codebase calls the function no matter net device has XDP
> > > programs or not. So the finalize function is being called everytime when RX
> > > bottom-half in progress. It needs a few machine instructions for nothing
> > > in the case that XDP programs are not attached at all.
> > >
> > > Lets it call the function on a condition that if xdp_status variable has
> > > not zero value. That means XDP programs are attached to the net device
> > > and it should be finalized based on the variable.
> > >
> > > The following instructions show that it's better than calling the function
> > > unconditionally.
> > >
> > > 0.31 │6b8: ldr w0, [sp, #196]
> > > │ ┌──cbz w0, 6cc
> > > │ │ mov x1, x0
> > > │ │ mov x0, x27
> > > │ │→ bl stmmac_finalize_xdp_rx
> > > │6cc:└─→ldr x1, [sp, #176]
> > >
> > > with 'if (xdp_status)' statement, jump to '6cc' label if xdp_status has
> > > zero value.
> > >
> > > Signed-off-by: Leesoo Ahn <[email protected]>
> > Hi Leesoo,
> >
> > I am curious to know if you considered going a step further and using
> > a static key.
> >
> > Link: https://www.kernel.org/doc/html/latest/staging/static-keys.html
>
> Thank you for the review.
>
> The function must be called for only XDP_TX or XDP_REDIRECT cases.
> So using a static key doesn't look good and the commit message is not clear
> for 'why' as well.
> I think that's why you suggested for using 'static key' by the latter
> reason.

Yes, my suggestion was based on the performance optimisation
aspect of your patch.

> I will edit the message and post v2 soon.

Thanks