[xdp-hints] Re: [RFC bpf-next 03/23] ice: make RX checksum checking code more reusable

[Larysa Zaremba <larysa.zaremba@intel.com>]

On Tue, Sep 05, 2023 at 07:44:32PM +0200, Maciej Fijalkowski wrote:
> On Tue, Sep 05, 2023 at 06:53:37PM +0200, Larysa Zaremba wrote:
> > On Tue, Sep 05, 2023 at 05:37:27PM +0200, Maciej Fijalkowski wrote:
> > > On Mon, Sep 04, 2023 at 08:01:06PM +0200, Larysa Zaremba wrote:
> > > > On Mon, Sep 04, 2023 at 05:02:40PM +0200, Maciej Fijalkowski wrote:
> > > > > On Thu, Aug 24, 2023 at 09:26:42PM +0200, Larysa Zaremba wrote:
> > > > > > Previously, we only needed RX checksum flags in skb path,
> > > > > > hence all related code was written with skb in mind.
> > > > > > But with the addition of XDP hints via kfuncs to the ice driver,
> > > > > > the same logic will be needed in .xmo_() callbacks.
> > > > > > 
> > > > > > Put generic process of determining checksum status into
> > > > > > a separate function.
> > > > > > 
> > > > > > Now we cannot operate directly on skb, when deducing
> > > > > > checksum status, therefore introduce an intermediate enum for checksum
> > > > > > status. Fortunately, in ice, we have only 4 possibilities: checksum
> > > > > > validated at level 0, validated at level 1, no checksum, checksum error.
> > > > > > Use 3 bits for more convenient conversion.
> > > > > > 
> > > > > > Signed-off-by: Larysa Zaremba <larysa.zaremba@intel.com>
> > > > > > ---
> > > > > >  drivers/net/ethernet/intel/ice/ice_txrx_lib.c | 105 ++++++++++++------
> > > > > >  1 file changed, 69 insertions(+), 36 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/net/ethernet/intel/ice/ice_txrx_lib.c b/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> > > > > > index b2f241b73934..8b155a502b3b 100644
> > > > > > --- a/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> > > > > > +++ b/drivers/net/ethernet/intel/ice/ice_txrx_lib.c
> > > > > > @@ -102,18 +102,41 @@ ice_rx_hash_to_skb(const struct ice_rx_ring *rx_ring,
> > > > > >  		skb_set_hash(skb, hash, ice_ptype_to_htype(rx_ptype));
> > > > > >  }
> > > > > >  
> > > > > > +enum ice_rx_csum_status {
> > > > > > +	ICE_RX_CSUM_LVL_0	= 0,
> > > > > > +	ICE_RX_CSUM_LVL_1	= BIT(0),
> > > > > > +	ICE_RX_CSUM_NONE	= BIT(1),
> > > > > > +	ICE_RX_CSUM_ERROR	= BIT(2),
> > > > > > +	ICE_RX_CSUM_FAIL	= ICE_RX_CSUM_NONE | ICE_RX_CSUM_ERROR,
> > > > > > +};
> > > > > > +
> > > > > >  /**
> > > > > > - * ice_rx_csum - Indicate in skb if checksum is good
> > > > > > - * @ring: the ring we care about
> > > > > > - * @skb: skb currently being received and modified
> > > > > > + * ice_rx_csum_lvl - Get checksum level from status
> > > > > > + * @status: driver-specific checksum status
> > > 
> > > nit: describe retval?
> > >
> > 
> > I think that kernel-doc is already too much for a one-liner.
> > Also, checksum level is fully explained in sk_buff documentation.
> > 
> > > > > > + */
> > > > > > +static u8 ice_rx_csum_lvl(enum ice_rx_csum_status status)
> > > > > > +{
> > > > > > +	return status & ICE_RX_CSUM_LVL_1;
> > > > > > +}
> > > > > > +
> > > > > > +/**
> > > > > > + * ice_rx_csum_ip_summed - Checksum status from driver-specific to generic
> > > > > > + * @status: driver-specific checksum status
> > > 
> > > ditto
> > 
> > Same as above. Moreover, there are only 2 possible return values that anyone can 
> > easily look up. Describing them here would only balloon the file length.
> 
> You really think 5 additional lines would balloon the file length? :D
> 
> I am not sure what to say here. We have many pretty pointless kdoc retval
> descriptions like 'returns 0 on success, error otherwise' but to me this
> is following the guidelines from Documentation/doc-guide/kernel-doc.rst.
> If i generate kdoc I don't want to open up the source code to easily look
> up retvals.
> 
> Just my 0.02$, not a thing that I'd like to keep on arguing on :)
>

I have consulted with the team and we came to a conclusion that maybe removing 
kernel-doc for these functions would be the best solution. In ice, functions in 
source files are always documented, but I do not think this rule is set in 
stone.

Sorry, if I was being rude, I had been traumatized by documentaion requirements
at my previous job (automotive) :D

> > 
> > > 
> > > > > > + */
> > > > > > +static u8 ice_rx_csum_ip_summed(enum ice_rx_csum_status status)
> > > > > > +{
> > > > > > +	return status & ICE_RX_CSUM_NONE ? CHECKSUM_NONE : CHECKSUM_UNNECESSARY;
> > > > > 
> > > > > 	return !(status & ICE_RX_CSUM_NONE);
> > > > > 
> > > > > ?
> > > > 
> > > > status & ICE_RX_CSUM_NONE ? CHECKSUM_NONE : CHECKSUM_UNNECESSARY;
> > > > 
> > > > is immediately understandable and results in 3 asm operations (I have checked):
> > > > 
> > > > result = status >> 1;
> > > > result ^= 1;
> > > > result &= 1;
> > > > 
> > > > I do not think "!(status & ICE_RX_CSUM_NONE);" could produce less.
> > > 
> > > oh, nice. Just the fact that branch being added caught my eye.
> > > 
> > > (...)