Re: dev driver / pci throughput

[Sebastian Heidl <heidl@zib.de>]

On Fri, Nov 09, 2001 at 10:48:58AM +0000, Matthew Clark wrote:
> #define CHUNK	512->4096 depending on implementation
> 
> static ssize_t BSL_write(..., const char *buf, size_t count..){
> char chunk[CHUNK];
> int	i,pos,
> 	for(i=0,pos=0;i<amount of data;i++,pos+=CHUNK){
> 		copy_from_user(buff,buf+pos,CHUNK);
> 		/* reorder data				*/
> 		/* not significant in throughput	*/
> 		for(k=0;k<CHUNK;k++){
>                         chunk[k]=buff[B_SM(j+k)];
> 			}
> 		memcpy_toio(MEM_reg+pos+i*CHUNK+j,chunk,CHUNK);
> 		}
> 	return count;
> 	}
> + lots of not important details-----
> 
> MEM_reg=ioremap(pci_resource_start(dev,B_SM_MEM)&PCI_BASE_ADDRESS_MEM_MASK,REG_SIZE);

Don't know if this will improve the performance much but if you checked the
area you are copying from with access_ok you can replace copy_from_user by
__copy_from_user. The second version does not check the area and so saves some
cycles.

regards,
_sh_