i applied the latest napi patch for eepro100, but i still met the problem.
when i tested the 64bytes frame with smartbits, in the beginning both the RX and TX of the system are normal. but after a while,the network card stop to receive and transmit packets .
i checked the proc/ params, and found that when soft_reset_count increased one, the system would stop to receive packets for a while. the phenomenon is just like the congestion problem of the interrupt driver as metioned in the NAPI article.
the code after patched is as follows. at what condition will soft_reset_count increase? what did the speedo_rx_soft_reset() do? and is there something incorrect when dealing with that? according to the comment,maybe the sp->cur_rx should be dealt.
if (!(status & RxComplete)) {
int intr_status;
unsigned long ioaddr = dev->base_addr;
unsigned long flags;
spin_lock_irqsave(&sp->lock,flags);
intr_status = inw(ioaddr + SCBStatus);
/* We check receiver state here because if
* we have to do soft reset,sp->cur_rx should
* point to an empty entry or something
* unexpected will happen
*/
if ((intr_status & 0x3c) != 0x10) {
if (speedo_debug > 4)
printk("No resource,reset\n");
speedo_rx_soft_reset(dev);
sp->soft_reset_count++;
}
spin_unlock_irqrestore(&sp->lock,flags);
break;
}
please cc. thanks.
chuanbo zheng
[email protected]
> I don't know which version you get.The ealier versions do have
>serious problems. The latest one(6.19) on NAPI website works
>well for me,but someone report problem of it too.Since i get no
>environment and time to investigate it,the problem is pending now.
>I will send you the latest patch in case you can't find it in other mail.
>
>zhengchuanbo wrote:
>
>>i tried ehe eepro100 NAPI driver on linux2.4.19. the kernel was compiled successfully. but when i tested the throughput of the system,i met some problem.
>>i tested the system with smartbits. when the frame size is 64bytes, in the beginning the system can receive and transmit packets. but after a while, the network card would not receive and transmit packets any more.
>>then with frame size bigger than 128bytes, it worked well. the throughput was improved. (but sometimes it also has some problem just like 64bytes frames).
>>so what's the problem? is there something wrong with the driver?
>>please cc. thanks.
>>
>>
>>zhengchuanbo
>>
>>
>>