netdev - Re: [PATCH 3/3] chelsio: more rx speedup

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Tue, 09 Jan 2007 18:08:07 -0800
From:	Divy Le Ray <divy@...lsio.com>
To:	Stephen Hemminger <shemminger@...l.org>
CC:	Ingo Oeser <netdev@...eo.de>, Jeff Garzik <jgarzik@...ox.com>,
	netdev@...r.kernel.org
Subject: Re: [PATCH 3/3] chelsio: more rx speedup

Stephen Hemminger wrote:
> On Tue, 9 Jan 2007 09:42:03 +0100
> Ingo Oeser <netdev@...eo.de> wrote:
>
>   
>> Hi Stephen,
>>
>> Stephen Hemminger schrieb:
>>     
>>> --- netdev-2.6.orig/drivers/net/chelsio/sge.c
>>> +++ netdev-2.6/drivers/net/chelsio/sge.c
>>>       
>> [...]
>>     
>>> @@ -1043,45 +1046,42 @@ static void recycle_fl_buf(struct freelQ
>>>   *	be copied but there is no memory for the copy.
>>>   */
>>>  static inline struct sk_buff *get_packet(struct pci_dev *pdev,
>>> -					 struct freelQ *fl, unsigned int len,
>>> -					 int dma_pad, int skb_pad,
>>> -					 unsigned int copy_thres,
>>> -					 unsigned int drop_thres)
>>> +					 struct freelQ *fl, unsigned int len)
>>>  {
>>>  	struct sk_buff *skb;
>>> -	struct freelQ_ce *ce = &fl->centries[fl->cidx];
>>> +	const struct freelQ_ce *ce = &fl->centries[fl->cidx];
>>>  
>>> -	if (len < copy_thres) {
>>> -		skb = alloc_skb(len + skb_pad, GFP_ATOMIC);
>>> -		if (likely(skb != NULL)) {
>>> -			skb_reserve(skb, skb_pad);
>>> -			skb_put(skb, len);
>>> -			pci_dma_sync_single_for_cpu(pdev,
>>> -					    pci_unmap_addr(ce, dma_addr),
>>> -					    pci_unmap_len(ce, dma_len),
>>> -					    PCI_DMA_FROMDEVICE);
>>> -			memcpy(skb->data, ce->skb->data + dma_pad, len);
>>> -			pci_dma_sync_single_for_device(pdev,
>>> +	if (len < copybreak) {
>>> +		skb = alloc_skb(len + 2, GFP_ATOMIC);
>>> +		if (!skb)
>>> +			goto use_orig_buf;
>>> +
>>> +		skb_reserve(skb, 2);	/* align IP header */
>>>       
>> Please use NET_IP_ALIGN here:
>>     
>
> Wrong, NET_IP_ALIGN is intended to deal with platforms where alignment of DMA is more
> important of alignment of structures.  Therefore if data is copied, it should
> always be 2.
>
>   
>> +		skb = alloc_skb(len + NET_IP_ALIGN, GFP_ATOMIC);
>> +		if (!skb)
>> +			goto use_orig_buf;
>> +
>> +		skb_reserve(skb, NET_IP_ALIGN);
>>
>>     
>>> +		skb_put(skb, len);
>>> +		pci_dma_sync_single_for_cpu(pdev,
>>>  					    pci_unmap_addr(ce, dma_addr),
>>>  					    pci_unmap_len(ce, dma_len),
>>>  					    PCI_DMA_FROMDEVICE);
>>> -		} else if (!drop_thres)
>>> -			goto use_orig_buf;
>>> -
>>> +		memcpy(skb->data, ce->skb->data, len);
>>> +		pci_dma_sync_single_for_device(pdev,
>>> +					       pci_unmap_addr(ce, dma_addr),
>>> +					       pci_unmap_len(ce, dma_len),
>>> +					       PCI_DMA_FROMDEVICE);
>>>  		recycle_fl_buf(fl, fl->cidx);
>>>  		return skb;
>>>  	}
>>>  
>>> -	if (fl->credits < drop_thres) {
>>> +use_orig_buf:
>>> +	if (fl->credits < 2) {
>>>       
>> Why 2? What does this magic number mean?
>>     
>
> No idea, it was there in the original. (as a parameter).
>
>   
The T2 HW behaves nicely when it is guaranteed to have 2 available 
entries in the rx free list.

Cheers,
Divy
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html