Re: [BUG] 2.6.24-git usb reset problems

[Jens Axboe <jens.axboe@oracle.com>]

On Tue, Jan 29 2008, James Bottomley wrote:
> On Tue, 2008-01-29 at 21:03 +0100, Jens Axboe wrote:
> > On Tue, Jan 29 2008, Boaz Harrosh wrote:
> > > On Tue, Jan 29 2008 at 21:45 +0200, Jens Axboe <jens.axboe@oracle.com> wrote:
> > > > On Tue, Jan 29 2008, Jens Axboe wrote:
> > > >> On Tue, Jan 29 2008, James Bottomley wrote:
> > > >>> On Tue, 2008-01-29 at 11:10 -0800, Matthew Dharm wrote:
> > > >>>> On Tue, Jan 29, 2008 at 07:39:11PM +0100, Jens Axboe wrote:
> > > >>>>> On Tue, Jan 29 2008, Jens Axboe wrote:
> > > >>>>>> On Tue, Jan 29 2008, Oliver Neukum wrote:
> > > >>>>>>> Am Dienstag, 29. Januar 2008 15:11:08 schrieb Jens Axboe:
> > > >>>>>>>> On Tue, Jan 29 2008, Boaz Harrosh wrote:
> > > >>>>>>>>> On Tue, Jan 29 2008 at 15:54 +0200, Jens Axboe <jens.axboe@oracle.com> wrote:
> > > >>>>>>>>>> On Tue, Jan 29 2008, Boaz Harrosh wrote:
> > > >>>>>>>>>>> Greg KH wrote:
> > > >>>>>>>  
> > > >>>>>>>>>> No difference, still just a lot of resets.
> > > >>>>>>>>>>
> > > >>>>>>>>> Where you able to figure out which usb storage transport is used?
> > > >>>>>>>>>
> > > >>>>>>>>> in drivers/usb/storage/usb.c you have get_protocol() and get_transport()
> > > >>>>>>>>> functions. I'm not sure if these get stored in sysfs perhaps. This will
> > > >>>>>>>>> pinpoint better where to look. Let me research a bit. 
> > > >>>>>>>> Did the quick'n easy and dumped it. Protocol is 'Transparent SCSI' and
> > > >>>>>>>> transport is 'Bulk'
> > > >>>>>>> You can recompile your kernel with CONFIG_USB_DEBUG and CONFIG_STORAGE_DEBUG
> > > >>>>>>> That should tell the reason for the resets.
> > > >>>>>> Sure, I'll do that. Will post the results tonight.
> > > >>>>> OK, fresh boot with CONFIG_USB_DEBUG and CONFIG_STORAGE_DEBUG. Plugged
> > > >>>>> in the device, waited 10 seconds or so and pulled it out. These are the
> > > >>>>> messages.
> > > >>>>>
> > > >>>>> It all looks good until the MODE_SENSE command, where it only transfers
> > > >>>>> 4 of 192 bytes.
> > > >>>> No, that's not the problem.  This is the problem:
> > > >>>>  
> > > >>>>> usb-storage: *** thread awakened.
> > > >>>>> usb-storage: Command TEST_UNIT_READY (6 bytes)
> > > >>>>> usb-storage:  00 00 00 00 00 00
> > > >>>>> usb-storage: Bulk Command S 0x43425355 T 0xd L 0 F 0 Trg 0 LUN 1 CL 6
> > > >>>>> usb-storage: usb_stor_bulk_transfer_buf: xfer 31 bytes
> > > >>>>> usb-storage: Status code 0; transferred 31/31
> > > >>>>> usb-storage: -- transfer complete
> > > >>>>> usb-storage: Bulk command transfer result=0
> > > >>>>> usb-storage: Attempting to get CSW...
> > > >>>>> usb-storage: usb_stor_bulk_transfer_buf: xfer 13 bytes
> > > >>>>> usb-storage: Status code 0; transferred 13/13
> > > >>>>> usb-storage: -- transfer complete
> > > >>>>> usb-storage: Bulk status result = 0
> > > >>>>> usb-storage: Bulk Status S 0x53425355 T 0xd R 0 Stat 0x1
> > > >>>>> usb-storage: -- transport indicates command failure
> > > >>>>> usb-storage: Issuing auto-REQUEST_SENSE
> > > >>>>> usb-storage: Bulk Command S 0x43425355 T 0xe L 18 F 128 Trg 0 LUN 1 CL 6
> > > >>>>> usb-storage: usb_stor_bulk_transfer_buf: xfer 31 bytes
> > > >>>>> usb-storage: Status code 0; transferred 31/31
> > > >>>>> usb-storage: -- transfer complete
> > > >>>>> usb-storage: Bulk command transfer result=0
> > > >>>>> usb-storage: usb_stor_bulk_transfer_sglist: xfer 18 bytes, 1 entries
> > > >>>>> usb-storage: usb_sg_init returned -22
> > > >>>>> usb-storage: Bulk data transfer result 0x4
> > > >>>>> usb-storage: -- auto-sense failure
> > > >>>>> usb-storage: storage_pre_reset
> > > >>>>> ehci_hcd 0000:00:1d.7: port 1 high speed
> > > >>>>> ehci_hcd 0000:00:1d.7: GetStatus port 1 status 001005 POWER sig=se0 PE CONNECT
> > > >>>>> usb 5-1: reset high speed USB device using ehci_hcd and address 2
> > > >>>>> ehci_hcd 0000:00:1d.7: port 1 high speed
> > > >>>>> ehci_hcd 0000:00:1d.7: GetStatus port 1 status 001005 POWER sig=se0 PE CONNECT
> > > >>>>> usb-storage: storage_post_reset
> > > >>>>> usb-storage: usb_reset_composite_device returns 0
> > > >>>>> usb-storage: scsi cmd done, result=0x70000
> > > >>>>> usb-storage: *** thread sleeping.
> > > >>>> For some reason, usb_sg_init is boned during auto-sense.
> > > >>> OK, that's implicating the scsi_eh_prep_cmnd() in the auto sense
> > > >>> code ... that was also an update in 2.6.24
> > > >> yeah, already found the bug - it's assuming ->request_buffer holds the
> > > >> sglist, oops. Preparing a fix.
> > > > 
> > > > ok here goes, this saves and restores the sg table correctly. it also
> > > > fixes the usb bug for me.
> > > > 
> > > > diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
> > > > index 547e85a..12770ef 100644
> > > > --- a/drivers/scsi/scsi_error.c
> > > > +++ b/drivers/scsi/scsi_error.c
> > > > @@ -622,13 +622,15 @@ void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
> > > >  	ses->use_sg = scmd->use_sg;
> > > >  	ses->resid = scmd->resid;
> > > >  	ses->result = scmd->result;
> > > > +	memcpy(&ses->sense_sgl, &scmd->sg_table, sizeof(ses->sense_sgl));
> > > >  
> > > >  	if (sense_bytes) {
> > > >  		scmd->request_bufflen = min_t(unsigned,
> > > >  		                       SCSI_SENSE_BUFFERSIZE, sense_bytes);
> > > >  		sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
> > > >  		                                       scmd->request_bufflen);
> > > > -		scmd->request_buffer = &ses->sense_sgl;
> > > > +		scmd->sg_table.sgl = &ses->sense_sgl;
> > > > +		scmd->sg_table.nents = scmd->sg_table.orig_nents = 1;
> > > >  		scmd->sc_data_direction = DMA_FROM_DEVICE;
> > > >  		scmd->use_sg = 1;
> > > >  		memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
> > > > @@ -679,6 +681,7 @@ void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
> > > >  	scmd->request_bufflen = ses->bufflen;
> > > >  	scmd->request_buffer = ses->buffer;
> > > >  	scmd->use_sg = ses->use_sg;
> > > > +	memcpy(&scmd->sg_table, &ses->sg_table, sizeof(scmd->sg_table));
> > > >  	scmd->resid = ses->resid;
> > > >  	scmd->result = ses->result;
> > > >  }
> > > > diff --git a/include/scsi/scsi_eh.h b/include/scsi/scsi_eh.h
> > > > index d21b891..d43dc83 100644
> > > > --- a/include/scsi/scsi_eh.h
> > > > +++ b/include/scsi/scsi_eh.h
> > > > @@ -75,6 +75,7 @@ struct scsi_eh_save {
> > > >  
> > > >  	void *buffer;
> > > >  	unsigned bufflen;
> > > > +	struct sg_table sg_table;
> > > >  	unsigned short use_sg;
> > > >  	int resid;
> > > >  
> > > > 
> > > Ok this is not in Linus tree is it? Hence I did not have that failure.
> > 
> > Yes it is, it's in current -git! James, comments on the patch? Will you
> > push it or shall I?
> 
> I will ... but it will cause an explosion in the bidirectional tree
> again.  I think the bidi updates also fix this.  However, give me time
> to rebase and verify.

OK thanks, just wanted to make sure that we didn't both expect each
other to handle it :-)

-- 
Jens Axboe