From: Paul Barker <paul.barker.ct@bp.renesas.com>

The HW checksum value is used as a 16-bit flag, it is zero when the
checksum has been validated and non-zero otherwise. Therefore we don't
need to treat this as an actual __wsum type or call csum_unfold(), we
can just use a u16 pointer.

Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
---
 drivers/net/ethernet/renesas/ravb_main.c | 30 +++++++++++-------------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
index 1dd2152734b0..9350ca10ab22 100644
--- a/drivers/net/ethernet/renesas/ravb_main.c
+++ b/drivers/net/ethernet/renesas/ravb_main.c
@@ -749,12 +749,11 @@ static void ravb_get_tx_tstamp(struct net_device *ndev)
 static void ravb_rx_csum_gbeth(struct sk_buff *skb)
 {
 	struct skb_shared_info *shinfo = skb_shinfo(skb);
-	__wsum csum_proto;
-	skb_frag_t *last_frag;
-	u8 *hw_csum;
+	size_t csum_len;
+	u16 *hw_csum;
 
-	/* The hardware checksum status is contained in sizeof(__sum16) * 2 = 4
-	 * bytes appended to packet data.
+	/* The hardware checksum status is contained in 4 bytes appended to
+	 * packet data.
 	 *
 	 * For ipv4, the first 2 bytes are the ip header checksum status. We can
 	 * ignore this as it will always be re-checked in inet_gro_receive().
@@ -762,23 +761,22 @@ static void ravb_rx_csum_gbeth(struct sk_buff *skb)
 	 * The last 2 bytes are the protocol checksum status which will be zero
 	 * if the checksum has been validated.
 	 */
-	if (unlikely(skb->len < sizeof(__sum16) * 2))
+	csum_len = sizeof(*hw_csum) * 2;
+	if (unlikely(skb->len < csum_len))
 		return;
 
 	if (skb_is_nonlinear(skb)) {
-		last_frag = &shinfo->frags[shinfo->nr_frags - 1];
-		hw_csum = skb_frag_address(last_frag) +
-			  skb_frag_size(last_frag);
-		skb_frag_size_sub(last_frag, 2 * sizeof(__sum16));
+		skb_frag_t *last_frag = &shinfo->frags[shinfo->nr_frags - 1];
+
+		hw_csum = (u16 *)(skb_frag_address(last_frag) +
+				  skb_frag_size(last_frag));
+		skb_frag_size_sub(last_frag, csum_len);
 	} else {
-		hw_csum = skb_tail_pointer(skb);
-		skb_trim(skb, skb->len - 2 * sizeof(__sum16));
+		hw_csum = (u16 *)skb_tail_pointer(skb);
+		skb_trim(skb, skb->len - csum_len);
 	}
 
-	hw_csum -= sizeof(__sum16);
-	csum_proto = csum_unfold((__force __sum16)get_unaligned_le16(hw_csum));
-
-	if (!csum_proto)
+	if (!*--hw_csum)
 		skb->ip_summed = CHECKSUM_UNNECESSARY;
 }
 
-- 
2.43.0

On 9/30/24 19:08, Paul Barker wrote:

> From: Paul Barker <paul.barker.ct@bp.renesas.com>
> 
> The HW checksum value is used as a 16-bit flag, it is zero when the

   I think I prefer s/HW/hardware/ but there's no hard feelings... :-)

> checksum has been validated and non-zero otherwise. Therefore we don't
> need to treat this as an actual __wsum type or call csum_unfold(), we
> can just use a u16 pointer.
> 
> Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
[...]
> diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
> index 1dd2152734b0..9350ca10ab22 100644
> --- a/drivers/net/ethernet/renesas/ravb_main.c
> +++ b/drivers/net/ethernet/renesas/ravb_main.c
[...]
> @@ -762,23 +761,22 @@ static void ravb_rx_csum_gbeth(struct sk_buff *skb)
>  	 * The last 2 bytes are the protocol checksum status which will be zero
>  	 * if the checksum has been validated.
>  	 */
> -	if (unlikely(skb->len < sizeof(__sum16) * 2))
> +	csum_len = sizeof(*hw_csum) * 2;

   Could've been done by an initializer instead?

> +	if (unlikely(skb->len < csum_len))
>  		return;
>  
>  	if (skb_is_nonlinear(skb)) {
> -		last_frag = &shinfo->frags[shinfo->nr_frags - 1];
> -		hw_csum = skb_frag_address(last_frag) +
> -			  skb_frag_size(last_frag);
> -		skb_frag_size_sub(last_frag, 2 * sizeof(__sum16));
> +		skb_frag_t *last_frag = &shinfo->frags[shinfo->nr_frags - 1];

   Could've been done in the previous patch...

> +
> +		hw_csum = (u16 *)(skb_frag_address(last_frag) +
> +				  skb_frag_size(last_frag));
> +		skb_frag_size_sub(last_frag, csum_len);
>  	} else {
> -		hw_csum = skb_tail_pointer(skb);
> -		skb_trim(skb, skb->len - 2 * sizeof(__sum16));
> +		hw_csum = (u16 *)skb_tail_pointer(skb);
> +		skb_trim(skb, skb->len - csum_len);
>  	}
>  
> -	hw_csum -= sizeof(__sum16);
> -	csum_proto = csum_unfold((__force __sum16)get_unaligned_le16(hw_csum));
> -
> -	if (!csum_proto)
> +	if (!*--hw_csum)

   Hm, you lost get_unaligned_le16() here. The checksum can be anywhere,
unaligned too...

[...]

MBR, Sergey

On 30/09/2024 20:11, Sergey Shtylyov wrote:
> On 9/30/24 19:08, Paul Barker wrote:
> 
>> From: Paul Barker <paul.barker.ct@bp.renesas.com>
>>
>> The HW checksum value is used as a 16-bit flag, it is zero when the
> 
>    I think I prefer s/HW/hardware/ but there's no hard feelings... :-)
> 
>> checksum has been validated and non-zero otherwise. Therefore we don't
>> need to treat this as an actual __wsum type or call csum_unfold(), we
>> can just use a u16 pointer.
>>
>> Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
> [...]
>> diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
>> index 1dd2152734b0..9350ca10ab22 100644
>> --- a/drivers/net/ethernet/renesas/ravb_main.c
>> +++ b/drivers/net/ethernet/renesas/ravb_main.c
> [...]
>> @@ -762,23 +761,22 @@ static void ravb_rx_csum_gbeth(struct sk_buff *skb)
>>  	 * The last 2 bytes are the protocol checksum status which will be zero
>>  	 * if the checksum has been validated.
>>  	 */
>> -	if (unlikely(skb->len < sizeof(__sum16) * 2))
>> +	csum_len = sizeof(*hw_csum) * 2;
> 
>    Could've been done by an initializer instead?

So, if I move this to the initializers at the start of the function,
csum_len must be declared after hw_csum which breaks reverse Christmas
tree ordering:

    struct skb_shared_info *shinfo = skb_shinfo(skb);
    u16 *hw_csum;
    size_t csum_len = sizeof(*hw_csum) * 2;

Thanks,

-- 
Paul Barker

On 10/3/24 12:23, Paul Barker wrote:
[...]

>>> From: Paul Barker <paul.barker.ct@bp.renesas.com>
>>>
>>> The HW checksum value is used as a 16-bit flag, it is zero when the
>>
>>    I think I prefer s/HW/hardware/ but there's no hard feelings... :-)
>>
>>> checksum has been validated and non-zero otherwise. Therefore we don't
>>> need to treat this as an actual __wsum type or call csum_unfold(), we
>>> can just use a u16 pointer.
>>>
>>> Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
>> [...]
>>> diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
>>> index 1dd2152734b0..9350ca10ab22 100644
>>> --- a/drivers/net/ethernet/renesas/ravb_main.c
>>> +++ b/drivers/net/ethernet/renesas/ravb_main.c
>> [...]
>>> @@ -762,23 +761,22 @@ static void ravb_rx_csum_gbeth(struct sk_buff *skb)
>>>  	 * The last 2 bytes are the protocol checksum status which will be zero
>>>  	 * if the checksum has been validated.
>>>  	 */
>>> -	if (unlikely(skb->len < sizeof(__sum16) * 2))
>>> +	csum_len = sizeof(*hw_csum) * 2;
>>
>>    Could've been done by an initializer instead?
> 
> So, if I move this to the initializers at the start of the function,
> csum_len must be declared after hw_csum which breaks reverse Christmas
> tree ordering:
> 
>     struct skb_shared_info *shinfo = skb_shinfo(skb);
>     u16 *hw_csum;
>     size_t csum_len = sizeof(*hw_csum) * 2;

   Could use sizeof(u16) instead, but it's OK to ignore me on this
matter. :-)

> Thanks,

MBR, Sergey

On 9/30/24 22:11, Sergey Shtylyov wrote:
[...]

>> From: Paul Barker <paul.barker.ct@bp.renesas.com>
>>
>> The HW checksum value is used as a 16-bit flag, it is zero when the
> 
>    I think I prefer s/HW/hardware/ but there's no hard feelings... :-)
> 
>> checksum has been validated and non-zero otherwise. Therefore we don't
>> need to treat this as an actual __wsum type or call csum_unfold(), we
>> can just use a u16 pointer.
>>
>> Signed-off-by: Paul Barker <paul.barker.ct@bp.renesas.com>
> [...]
>> diff --git a/drivers/net/ethernet/renesas/ravb_main.c b/drivers/net/ethernet/renesas/ravb_main.c
>> index 1dd2152734b0..9350ca10ab22 100644
>> --- a/drivers/net/ethernet/renesas/ravb_main.c
>> +++ b/drivers/net/ethernet/renesas/ravb_main.c
> [...]
>> @@ -762,23 +761,22 @@ static void ravb_rx_csum_gbeth(struct sk_buff *skb)
[...]
>> +	if (unlikely(skb->len < csum_len))
>>  		return;
>>  
>>  	if (skb_is_nonlinear(skb)) {
>> -		last_frag = &shinfo->frags[shinfo->nr_frags - 1];
>> -		hw_csum = skb_frag_address(last_frag) +
>> -			  skb_frag_size(last_frag);
>> -		skb_frag_size_sub(last_frag, 2 * sizeof(__sum16));
>> +		skb_frag_t *last_frag = &shinfo->frags[shinfo->nr_frags - 1];
> 
>    Could've been done in the previous patch...

   Even fit better there, I think...

>> +
>> +		hw_csum = (u16 *)(skb_frag_address(last_frag) +
>> +				  skb_frag_size(last_frag));
>> +		skb_frag_size_sub(last_frag, csum_len);
>>  	} else {
>> -		hw_csum = skb_tail_pointer(skb);
>> -		skb_trim(skb, skb->len - 2 * sizeof(__sum16));
>> +		hw_csum = (u16 *)skb_tail_pointer(skb);
>> +		skb_trim(skb, skb->len - csum_len);
>>  	}
>>  
>> -	hw_csum -= sizeof(__sum16);
>> -	csum_proto = csum_unfold((__force __sum16)get_unaligned_le16(hw_csum));
>> -
>> -	if (!csum_proto)
>> +	if (!*--hw_csum)
> 
>    Hm, you lost get_unaligned_le16() here. The checksum can be anywhere,
> unaligned too...

   No need to keep using get_unaligned_le16() itself but you should then switch to 
using get_unaligned().

[...]

MBR, Sergey