Re: [PATCH] 2.6.6 invalid usage of GFP_DMA in drivers/scsi/pluto.c

[Andrew Morton <akpm@osdl.org>]

Yury Umanets <torque@ukrpost.net> wrote:
>
> Hello Andrew, guys,
> 
> Found this, what seems to be an invalid usage of GFP_DMA flag. Is this
> patch okay?

Nope.

GFP_DMA means "from the lower 16MB of memory".  It's needed for crufty old
eisa hardware which only does 24-bit DMA.  It's meaningless to OR this with
GFP_KERNEL.

However it's a bit odd that GFP_DMA implies !__GFP_WAIT.  It would be valid
to hunt down GFP_DMA users who should really be using GFP_DMA|__GFP_WAIT,
but this stuff is so old and crufty I'd be inclined to leave it all alone.


> 
> --- ./linux-2.6.6/drivers/scsi/pluto.c	Mon May 10 05:32:27 2004
> +++ ./linux-2.6.6-modified/drivers/scsi/pluto.c	Tue Jun  8 11:26:07 2004
> @@ -117,7 +117,8 @@ int __init pluto_detect(Scsi_Host_Templa
>  #endif
>  			return 0;
>  	}
> -	fcs = (struct ctrl_inquiry *) kmalloc (sizeof (struct ctrl_inquiry) *
> fcscount, GFP_DMA);
> +	fcs = (struct ctrl_inquiry *) kmalloc (sizeof (struct ctrl_inquiry) *
> fcscount, 
> +			GFP_KERNEL | GFP_DMA);
>  	if (!fcs) {
>  		printk ("PLUTO: Not enough memory to probe\n");
>  		return 0;
> 

Your patch is wordwrapped and uses weird headers (please omit the leading
./ from the pathnames).