Question: where should the SCSI driver place MODE_SENSE data ?

Question: where should the SCSI driver place MODE_SENSE data ?

[Ju, Seokmann]

In the 2.6 (2.6.9 and scsi-misc in git) kernel, MODE_SENSE SCSI command
packet (scsi_cmnd) carries following entries with unexpectedly small in
size.
- request_bufflen
- bufflen

Especially for MODE SENSE with page code 8 (caching page), driver has
minumum 12 Bytes MODE_SENSE data to deliver besides 'mode parameter
header' and 'block descriptors'.
When I dump those entries, they both are 4 Bytes in size.
To me, it seems like that SCSI mid layer allocated 512 Bytes for
MODE_SENSE data buffer, but the buffer length passed down to LLD
incorrectly.

Please guide me if there is anything missing.

Thank you,

Seokmann

Re: Question: where should the SCSI driver place MODE_SENSE data ?

[James Bottomley]

On Tue, 2006-03-21 at 09:44 -0700, Ju, Seokmann wrote:
> In the 2.6 (2.6.9 and scsi-misc in git) kernel, MODE_SENSE SCSI command
> packet (scsi_cmnd) carries following entries with unexpectedly small in
> size.
> - request_bufflen
> - bufflen
> 
> Especially for MODE SENSE with page code 8 (caching page), driver has
> minumum 12 Bytes MODE_SENSE data to deliver besides 'mode parameter
> header' and 'block descriptors'.
> When I dump those entries, they both are 4 Bytes in size.
> To me, it seems like that SCSI mid layer allocated 512 Bytes for
> MODE_SENSE data buffer, but the buffer length passed down to LLD
> incorrectly.

I don't understand the question.  Are you asking why
sd_read_write_protect_flag and sd_read_cache_type operate in the way
they do?  i.e. header first then actual data.

James

RE: Question: where should the SCSI driver place MODE_SENSE data ?

[James Bottomley]

On Wed, 2006-03-22 at 08:52 -0700, Ju, Seokmann wrote:
> On Tuesday, March 21, 2006 6:02 PM, James Bottomley wrote:
> > I don't understand the question.  Are you asking why
> > sd_read_write_protect_flag and sd_read_cache_type operate in the way
> > they do?  i.e. header first then actual data.
> For any SCSI command including MODE_SENSE, 'bufflen'in scsi_cmnd structure holds actual data buffer size in bytes if 'use_sg' flage is NOT set.
> The question is that "value of bufflen is 4 for the sd_read_cache_type operation and that is NOT sufficient to return header and page data by driver".
> If there is something that I misunderstood with the operation, please guide.

If you look at the functions in sd.c you'll see it goes about asking for
mode sense very carefully, because there are a lot of broken devices out
there.  The first request is only for enough of the headers to work out
the length of the page.

James

RE: Question: where should the SCSI driver place MODE_SENSE data ?

[Ju, Seokmann]

On Tuesday, March 21, 2006 6:02 PM, James Bottomley wrote:
> I don't understand the question.  Are you asking why
> sd_read_write_protect_flag and sd_read_cache_type operate in the way
> they do?  i.e. header first then actual data.
For any SCSI command including MODE_SENSE, 'bufflen'in scsi_cmnd structure holds actual data buffer size in bytes if 'use_sg' flage is NOT set.
The question is that "value of bufflen is 4 for the sd_read_cache_type operation and that is NOT sufficient to return header and page data by driver".
If there is something that I misunderstood with the operation, please guide.

> -----Original Message-----
> From: James Bottomley [mailto:James.Bottomley@SteelEye.com] 
> Sent: Tuesday, March 21, 2006 6:02 PM
> To: Ju, Seokmann
> Cc: linux-scsi; linux-kernel
> Subject: Re: Question: where should the SCSI driver place 
> MODE_SENSE data ?
> 
> On Tue, 2006-03-21 at 09:44 -0700, Ju, Seokmann wrote:
> > In the 2.6 (2.6.9 and scsi-misc in git) kernel, MODE_SENSE 
> SCSI command
> > packet (scsi_cmnd) carries following entries with 
> unexpectedly small in
> > size.
> > - request_bufflen
> > - bufflen
> > 
> > Especially for MODE SENSE with page code 8 (caching page), 
> driver has
> > minumum 12 Bytes MODE_SENSE data to deliver besides 'mode parameter
> > header' and 'block descriptors'.
> > When I dump those entries, they both are 4 Bytes in size.
> > To me, it seems like that SCSI mid layer allocated 512 Bytes for
> > MODE_SENSE data buffer, but the buffer length passed down to LLD
> > incorrectly.
> 
> I don't understand the question.  Are you asking why
> sd_read_write_protect_flag and sd_read_cache_type operate in the way
> they do?  i.e. header first then actual data.
> 
> James
> 
> 
> 

RE: Question: where should the SCSI driver place MODE_SENSE data ?

[Ju, Seokmann]

Any comment on this question would be appreciated.
Thanks you, 

> -----Original Message-----
> From: Ju, Seokmann 
> Sent: Tuesday, March 21, 2006 11:45 AM
> To: 'James Bottomley'
> Cc: linux-scsi; linux-kernel
> Subject: Question: where should the SCSI driver place 
> MODE_SENSE data ?
> 
> In the 2.6 (2.6.9 and scsi-misc in git) kernel, MODE_SENSE 
> SCSI command packet (scsi_cmnd) carries following entries 
> with unexpectedly small in size.
> - request_bufflen
> - bufflen
> 
> Especially for MODE SENSE with page code 8 (caching page), 
> driver has minumum 12 Bytes MODE_SENSE data to deliver 
> besides 'mode parameter header' and 'block descriptors'.
> When I dump those entries, they both are 4 Bytes in size.
> To me, it seems like that SCSI mid layer allocated 512 Bytes 
> for MODE_SENSE data buffer, but the buffer length passed down 
> to LLD incorrectly.
> 
> Please guide me if there is anything missing.
> 
> Thank you,
> 
> Seokmann