Re: [PATCH] scsi_debug [was: scsi_debug issues]

[Jens Axboe <axboe@suse.de>]

On Tue, Oct 19 2004, Douglas Gilbert wrote:
> Nishanth Aravamudan wrote:
> >On Mon, Oct 18, 2004 at 11:44:59PM +1000, Douglas Gilbert wrote:
> >
> >>James Bottomley wrote:
> >>
> >>>On Sat, 2004-10-16 at 01:51, Douglas Gilbert wrote:
> >>>
> >>>
> >>>>So this problem seems related to highmem.
> >
> >
> ><snip>
> >
> >>The above required a fair few changes to scsi_debug.
> >>Attached is a patch that rolls "kmap" changes with
> >>patches I have sent recently for scsi_debug.
> >>Attachment is gzipped (due to size) and applies against
> >>lk 2.6.8.1 -> lk 2.6.9-rc4 .
> >>
> >>Nishanth, could you test this with highmem?
> >
> >
> >Doug,
> >
> >I ran 2.6.9-rc4 with your patch applied and had several dump_stack()s
> >occur (complaining about sleeping in an invalid context) and a final
> >panic during mkfs :) I think something is still wrong... ;) Output is
> >below and attached.
> 
> Ok, it looks like kmap_atomic() is needed.
> Could you try this additional patch.
> 
> Doug Gilbert

> --- linux/drivers/scsi/scsi_debug.c	2004-10-19 07:59:14.966673272 +1000
> +++ linux/drivers/scsi/scsi_debug.c174b	2004-10-19 07:58:46.044070176 +1000
> @@ -56,7 +56,7 @@
>  #include "scsi_debug.h"
>  
>  #define SCSI_DEBUG_VERSION "1.74"
> -static const char * scsi_debug_version_date = "20041018";
> +static const char * scsi_debug_version_date = "20041019";
>  
>  /* Additional Sense Code (ASC) used */
>  #define NO_ADDED_SENSE 0x0
> @@ -505,7 +505,8 @@
>  	active = 1;
>  	for (k = 0, req_len = 0, act_len = 0; k < scp->use_sg; ++k, ++sgpnt) {
>  		if (active) {
> -			kaddr = (unsigned char *)kmap(sgpnt->page);
> +			kaddr = (unsigned char *)
> +				kmap_atomic(sgpnt->page, KM_USER0);
>  			if (NULL == kaddr)
>  				return (DID_ERROR << 16);
>  			kaddr = (unsigned char *)kaddr + sgpnt->offset;
> @@ -515,7 +516,7 @@
>  				len = arr_len - req_len;
>  			}
>  			memcpy(kaddr, arr + req_len, len);
> -			kunmap(sgpnt->page);
> +			kunmap_atomic(sgpnt->page, KM_USER0);

BUG, kunmap_atomic(kaddr) (without the offset).

Also do make sure that you stay on the same cpu from calling kmap_atomic
to kunmap_atomic(). If you are under the queue_lock you are fine, I
didn't check if this is the case for all your calls.

-- 
Jens Axboe