Linux SCSI subsystem development
 help / color / mirror / Atom feed
From: Ming Lei <ming.lei@redhat.com>
To: Finn Thain <fthain@telegraphics.com.au>
Cc: Michael Schmitz <schmitzmic@gmail.com>,
	devel@driverdev.osuosl.org, Hannes Reinecke <hare@suse.com>,
	Benjamin Block <bblock@linux.ibm.com>,
	linux-scsi@vger.kernel.org,
	"Martin K . Petersen" <martin.petersen@oracle.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	linux-usb@vger.kernel.org, James Smart <james.smart@broadcom.com>,
	"Ewan D . Milne" <emilne@redhat.com>, Jim Gill <jgill@vmware.com>,
	James Bottomley <James.Bottomley@hansenpartnership.com>,
	Brian King <brking@us.ibm.com>,
	"Juergen E . Fischer" <fischer@norbit.de>,
	Christoph Hellwig <hch@lst.de>,
	Dan Carpenter <dan.carpenter@oracle.com>,
	Bart Van Assche <bvanassche@acm.org>
Subject: Re: [PATCH V3 10/15] scsi: aha152x: use sg helper to operate scatterlist
Date: Fri, 14 Jun 2019 16:17:33 +0800	[thread overview]
Message-ID: <20190614081732.GC24393@ming.t460p> (raw)
In-Reply-To: <alpine.LNX.2.21.1906141404270.33@nippy.intranet>

On Fri, Jun 14, 2019 at 03:27:36PM +1000, Finn Thain wrote:
> Hi Ming,
> 
> On Fri, 14 Jun 2019, Ming Lei wrote:
> 
> > Use the scatterlist iterators and remove direct indexing of the
> > scatterlist array.
> > 
> > This way allows us to pre-allocate one small scatterlist, which can be
> > chained with one runtime allocated scatterlist if the pre-allocated one
> > isn't enough for the whole request.
> > 
> > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > ---
> >  drivers/scsi/aha152x.c | 29 +++++++++++++++++++----------
> >  1 file changed, 19 insertions(+), 10 deletions(-)
> > 
> > diff --git a/drivers/scsi/aha152x.c b/drivers/scsi/aha152x.c
> > index 97872838b983..bc9d12aa7880 100644
> > --- a/drivers/scsi/aha152x.c
> > +++ b/drivers/scsi/aha152x.c
> > @@ -2033,7 +2033,7 @@ static void datai_run(struct Scsi_Host *shpnt)
> >  				    CURRENT_SC->SCp.buffers_residual > 0) {
> >  					/* advance to next buffer */
> >  					CURRENT_SC->SCp.buffers_residual--;
> > -					CURRENT_SC->SCp.buffer++;
> > +					CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer);
> >  					CURRENT_SC->SCp.ptr           = SG_ADDRESS(CURRENT_SC->SCp.buffer);
> >  					CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length;
> >  				}
> > @@ -2139,7 +2139,7 @@ static void datao_run(struct Scsi_Host *shpnt)
> >  		if(CURRENT_SC->SCp.this_residual==0 && CURRENT_SC->SCp.buffers_residual>0) {
> >  			/* advance to next buffer */
> >  			CURRENT_SC->SCp.buffers_residual--;
> > -			CURRENT_SC->SCp.buffer++;
> > +			CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer);
> >  			CURRENT_SC->SCp.ptr           = SG_ADDRESS(CURRENT_SC->SCp.buffer);
> >  			CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length;
> >  		}
> > @@ -2160,20 +2160,29 @@ static void datao_end(struct Scsi_Host *shpnt)
> >  	if(TESTLO(DMASTAT, DFIFOEMP)) {
> >  		int data_count = (DATA_LEN - scsi_get_resid(CURRENT_SC)) -
> >  			GETSTCNT();
> 
> data_count appears to be the number of bytes remaining in the FIFO. (I 
> have to infer that much from the surrounding code. I don't have 
> documentation for this controller.)

Yeah, it should be so, that is the data which need to transfer again.

> 
> Some comments would be nice.
> 
> > +		struct scatterlist *sg = scsi_sglist(CURRENT_SC);
> > +		int left, i = 0;
> >  
> >  		CMD_INC_RESID(CURRENT_SC, data_count);
> >  
> 
> Apparently the aim is to increase the residual by the number of bytes that 
> will never leave the FIFO. Above we can see that increase performed by 
> scsi_set_resid() and now the same has to be done to the SCp struct.

Agree.

> 
> >  		data_count -= CURRENT_SC->SCp.ptr -
> >  			SG_ADDRESS(CURRENT_SC->SCp.buffer);
> 
> Here, data_count effectively has SCp.this_residual subtracted from it.
> 
> > -		while(data_count>0) {
> > -			CURRENT_SC->SCp.buffer--;
> > -			CURRENT_SC->SCp.buffers_residual++;
> > -			data_count -= CURRENT_SC->SCp.buffer->length;
> > -		}
> 
> So far, so good.
> 
> > -		CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) -
> > -			data_count;
> > -		CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length +
> > -			data_count;
> 
> This is like saying ptr = buffer + residual, which is bogus. This must be 
> an old bug, but we never hit it because the FIFO is always empty when we 
> get a DISCONNECT message. Probably because every SG segment has a length 
> that is a multiple of 128 bytes. (Juergen?)
> 
> > +
> > +		left = CURRENT_SC->transfersize - data_count;
> 
> Are you sure about that? Perhaps you meant to write,
> 		left = scsi_bufflen(CURRENT_SC) - scsi_get_resid(CURRENT_SC);
> 
> Is there a better name for this variable? Maybe 'sent' or 'bytes_sent'?

'left' can be renamed as 'done', and

	done =  GETSTCNT - (CURRENT_SC->SCp.ptr - SG_ADDRESS(CURRENT_SC->SCp.buffer))

> 
> > +		for (i = 0; left > 0 && !sg_is_last(sg); i++, sg = sg_next(sg)) {
> > +			if (left < sg->length)
> > +				break;
> > +			left -= sg->length;
> > +		}
> > +
> > +		if (data_count > 0) {
> > +			CURRENT_SC->SCp.buffers_residual += i;
> 
> Shouldn't that be,
> 			CURRENT_SC->SCp.buffers_residual = i;

Good catch.

> 
> > +			CURRENT_SC->SCp.buffer = sg;
> > +
> > +			CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) + left;
> > +			CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length -
> > +				left;
> > +		}
> >  	}
> >  
> >  	SETPORT(SXFRCTL0, CH1|CLRCH1|CLRSTCNT);
> > 
> 
> BTW, datao_run() seems to guarantee that the FIFO will never contain more 
> than min(128, SCp.this_residual) so I suspect that this code can be 
> simplified. That's just BTW. I wouldn't attempt to optimize this branch as 
> it will only run when the FIFO is not empty, if ever. So I'd prefer 
> clarity. Besides, I don't have the hardware to test it on.

I'd suggest to simply translate into code which uses current sg helper.

Follows the fixed version, could you review again?

>From f03484d4bac083c39d70665cfbadb641093b63de Mon Sep 17 00:00:00 2001
From: Ming Lei <ming.lei@redhat.com>
Date: Wed, 12 Jun 2019 20:37:35 +0800
Subject: [PATCH] scsi: aha152x: use sg helper to operate scatterlist

Use the scatterlist iterators and remove direct indexing of the
scatterlist array.

This way allows us to pre-allocate one small scatterlist, which can be
chained with one runtime allocated scatterlist if the pre-allocated one
isn't enough for the whole request.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 drivers/scsi/aha152x.c | 34 ++++++++++++++++++++--------------
 1 file changed, 20 insertions(+), 14 deletions(-)

diff --git a/drivers/scsi/aha152x.c b/drivers/scsi/aha152x.c
index 97872838b983..7faecdefda56 100644
--- a/drivers/scsi/aha152x.c
+++ b/drivers/scsi/aha152x.c
@@ -2033,7 +2033,7 @@ static void datai_run(struct Scsi_Host *shpnt)
 				    CURRENT_SC->SCp.buffers_residual > 0) {
 					/* advance to next buffer */
 					CURRENT_SC->SCp.buffers_residual--;
-					CURRENT_SC->SCp.buffer++;
+					CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer);
 					CURRENT_SC->SCp.ptr           = SG_ADDRESS(CURRENT_SC->SCp.buffer);
 					CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length;
 				}
@@ -2139,7 +2139,7 @@ static void datao_run(struct Scsi_Host *shpnt)
 		if(CURRENT_SC->SCp.this_residual==0 && CURRENT_SC->SCp.buffers_residual>0) {
 			/* advance to next buffer */
 			CURRENT_SC->SCp.buffers_residual--;
-			CURRENT_SC->SCp.buffer++;
+			CURRENT_SC->SCp.buffer = sg_next(CURRENT_SC->SCp.buffer);
 			CURRENT_SC->SCp.ptr           = SG_ADDRESS(CURRENT_SC->SCp.buffer);
 			CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length;
 		}
@@ -2158,22 +2158,28 @@ static void datao_run(struct Scsi_Host *shpnt)
 static void datao_end(struct Scsi_Host *shpnt)
 {
 	if(TESTLO(DMASTAT, DFIFOEMP)) {
-		int data_count = (DATA_LEN - scsi_get_resid(CURRENT_SC)) -
-			GETSTCNT();
+		int done = GETSTCNT();
+		int data_count = (DATA_LEN - scsi_get_resid(CURRENT_SC)) - done;
+		struct scatterlist *sg = scsi_sglist(CURRENT_SC);
+		int i;
 
 		CMD_INC_RESID(CURRENT_SC, data_count);
 
-		data_count -= CURRENT_SC->SCp.ptr -
-			SG_ADDRESS(CURRENT_SC->SCp.buffer);
-		while(data_count>0) {
-			CURRENT_SC->SCp.buffer--;
-			CURRENT_SC->SCp.buffers_residual++;
-			data_count -= CURRENT_SC->SCp.buffer->length;
+		/*
+		 * rewind where we have done, and we have to start from
+		 * the beginning
+		 */
+		for (i = 0; done > 0 && !sg_is_last(sg); i++, sg = sg_next(sg)) {
+			if (done < sg->length)
+				break;
+			done -= sg->length;
 		}
-		CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) -
-			data_count;
-		CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length +
-			data_count;
+
+		CURRENT_SC->SCp.buffers_residual = i;
+		CURRENT_SC->SCp.buffer = sg;
+		CURRENT_SC->SCp.ptr = SG_ADDRESS(CURRENT_SC->SCp.buffer) + done;
+		CURRENT_SC->SCp.this_residual = CURRENT_SC->SCp.buffer->length -
+			done;
 	}
 
 	SETPORT(SXFRCTL0, CH1|CLRCH1|CLRSTCNT);
-- 
2.20.1


thanks,
Ming

  reply	other threads:[~2019-06-14  8:17 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-06-14  2:53 [PATCH V3 00/15] use sg helper to operate scatterlist Ming Lei
2019-06-14  2:53 ` [PATCH V3 01/15] scsi: vmw_pscsi: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 02/15] scsi: advansys: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 03/15] scsi: lpfc: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 04/15] scsi: mvumi: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 05/15] scsi: ipr: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 06/15] scsi: pmcraid: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 07/15] usb: image: microtek: " Ming Lei
2019-06-14  5:39   ` Finn Thain
2019-06-14  6:57     ` Ming Lei
2019-06-14  2:53 ` [PATCH V3 08/15] staging: unisys: visorhba: " Ming Lei
2019-06-14  5:48   ` Greg Kroah-Hartman
2019-06-14  2:53 ` [PATCH V3 09/15] s390: zfcp_fc: " Ming Lei
2019-06-16 16:25   ` Benjamin Block
2019-06-14  2:53 ` [PATCH V3 10/15] scsi: aha152x: " Ming Lei
2019-06-14  5:27   ` Finn Thain
2019-06-14  8:17     ` Ming Lei [this message]
2019-06-14 10:36       ` Finn Thain
2019-06-17  1:14         ` Ming Lei
2019-06-14  2:53 ` [PATCH V3 11/15] scsi: imm: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 12/15] scsi: pcmcia: nsp_cs: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 13/15] scsi: ppa: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 14/15] scsi: wd33c93: " Ming Lei
2019-06-14  2:53 ` [PATCH V3 15/15] NCR5380: Support chained sg lists Ming Lei

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20190614081732.GC24393@ming.t460p \
    --to=ming.lei@redhat.com \
    --cc=James.Bottomley@hansenpartnership.com \
    --cc=bblock@linux.ibm.com \
    --cc=brking@us.ibm.com \
    --cc=bvanassche@acm.org \
    --cc=dan.carpenter@oracle.com \
    --cc=devel@driverdev.osuosl.org \
    --cc=emilne@redhat.com \
    --cc=fischer@norbit.de \
    --cc=fthain@telegraphics.com.au \
    --cc=gregkh@linuxfoundation.org \
    --cc=hare@suse.com \
    --cc=hch@lst.de \
    --cc=james.smart@broadcom.com \
    --cc=jgill@vmware.com \
    --cc=linux-scsi@vger.kernel.org \
    --cc=linux-usb@vger.kernel.org \
    --cc=martin.petersen@oracle.com \
    --cc=schmitzmic@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox