linux-mtd.lists.infradead.org archive mirror
 help / color / mirror / Atom feed
From: Boris Brezillon <boris.brezillon@collabora.com>
To: Stefan Agner <stefan@agner.ch>
Cc: Sascha Hauer <s.hauer@pengutronix.de>,
	linux-mtd@lists.infradead.org,
	Miquel Raynal <miquel.raynal@bootlin.com>
Subject: Re: nand_op_parser_exec_op should use longest pattern
Date: Sat, 30 Mar 2019 10:21:34 +0100	[thread overview]
Message-ID: <20190330102134.4e85e130@collabora.com> (raw)
In-Reply-To: <bc75ec7a12c681a618ee641736141100@agner.ch>

+Miquel

On Fri, 29 Mar 2019 15:37:56 +0100
Stefan Agner <stefan@agner.ch> wrote:

> On 29.03.2019 14:35, Sascha Hauer wrote:
> > Hi All,
> > 
> > I just played with the new exec_op interface for the first time and
> > together with Boris we found a problem in the pattern table parser.
> > 
> > The vf610 driver uses this pattern table:
> > 
> > static const struct nand_op_parser vf610_nfc_op_parser = NAND_OP_PARSER(
> > 	NAND_OP_PARSER_PATTERN(vf610_nfc_cmd,
> > 		NAND_OP_PARSER_PAT_CMD_ELEM(true),
> > 		NAND_OP_PARSER_PAT_ADDR_ELEM(true, 5),
> > 		NAND_OP_PARSER_PAT_DATA_OUT_ELEM(true, PAGE_2K + OOB_MAX),
> > 		NAND_OP_PARSER_PAT_CMD_ELEM(true),
> > 		NAND_OP_PARSER_PAT_WAITRDY_ELEM(true)),
> > 	NAND_OP_PARSER_PATTERN(vf610_nfc_cmd,
> > 		NAND_OP_PARSER_PAT_CMD_ELEM(true),
> > 		NAND_OP_PARSER_PAT_ADDR_ELEM(true, 5),
> > 		NAND_OP_PARSER_PAT_CMD_ELEM(true),
> > 		NAND_OP_PARSER_PAT_WAITRDY_ELEM(true),
> > 		NAND_OP_PARSER_PAT_DATA_IN_ELEM(true, PAGE_2K + OOB_MAX)),
> > 	);
> > 
> > It has two patterns, one supposed for writing and one for reading. All elements
> > are optional. Now with a typical page read we'll get this:
> > 
> > [   33.932464] nand:     ->CMD      [0x00]
> > [   33.936338] nand:     ->ADDR     [5 cyc: 00 00 00 0a 00]
> > [   33.941755] nand:     ->CMD      [0x30]
> > [   33.945628] nand:     ->WAITRDY  [max 1 ms]
> > [   33.949909] nand:       DATA_IN  [2176 B]
> > 
> > Only the first four elements are executed in one go, the fifth is
> > exectuted separately. This is because the pattern table parser finds
> > that the first pattern (supposed for writing) already matches for the
> > first four elements and then uses it instead of realizing that the
> > second pattern matches the whole operation.  
> 
> Hm, I do not remember noticing that during development. I wonder if it
> was the case already back then.
> 
> If yes, it did not seem to have a negative impact on performance
> compared to the old interface:
> https://linux-mtd.infradead.narkive.com/qZgEnPsC/patch-v6-0-3-mtd-rawnand-vf610-nfc-make-use-of-exec-op
> 
> > 
> > I have no fix for this, just wanted to let you know. It turned out that
> > in my case for the GPMI nand driver I probably won't need any pattern
> > table.  
> 
> Thanks for bringing it up! Will try it out when I come around.

Here is a new version of the proposed fix that compiles, at
least :-). Still not tested tested on a real HW though.

--->8---
diff --git a/drivers/mtd/nand/raw/nand_base.c b/drivers/mtd/nand/raw/nand_base.c
index ddd396e93e32..7a5178b5f13e 100644
--- a/drivers/mtd/nand/raw/nand_base.c
+++ b/drivers/mtd/nand/raw/nand_base.c
@@ -2131,6 +2131,23 @@ static void nand_op_parser_trace(const struct nand_op_parser_ctx *ctx)
 }
 #endif
 
+static int nand_op_parser_cmp_ctx(const struct nand_op_parser_ctx *a,
+                                 const struct nand_op_parser_ctx *b)
+{
+
+       if (a->subop.ninstrs < b->subop.ninstrs)
+               return -1;
+       else if (a->subop.ninstrs > b->subop.ninstrs)
+               return 1;
+
+       if (a->subop.last_instr_end_off < b->subop.last_instr_end_off)
+               return -1;
+       else if (a->subop.last_instr_end_off > b->subop.last_instr_end_off)
+               return 1;
+
+       return 0;
+}
+
 /**
  * nand_op_parser_exec_op - exec_op parser
  * @chip: the NAND chip
@@ -2165,30 +2182,38 @@ int nand_op_parser_exec_op(struct nand_chip *chip,
        unsigned int i;
 
        while (ctx.subop.instrs < op->instrs + op->ninstrs) {
-               int ret;
+               const struct nand_op_parser_pattern *pattern;
+               struct nand_op_parser_ctx best_ctx;
+               int ret, best_pattern = -1;
 
                for (i = 0; i < parser->npatterns; i++) {
-                       const struct nand_op_parser_pattern *pattern;
+                       struct nand_op_parser_ctx test_ctx = ctx;
 
                        pattern = &parser->patterns[i];
-                       if (!nand_op_parser_match_pat(pattern, &ctx))
+                       if (!nand_op_parser_match_pat(pattern, &test_ctx))
                                continue;
 
-                       nand_op_parser_trace(&ctx);
+                       if (best_pattern >= 0 &&
+                           nand_op_parser_cmp_ctx(&test_ctx, &best_ctx) <= 0)
+                               continue;
 
-                       if (check_only)
-                               break;
+                       best_pattern = i;
+                       best_ctx = test_ctx;
+               }
 
+               if (best_pattern < 0) {
+                       pr_debug("->exec_op() parser: pattern not found!\n");
+                       return -ENOTSUPP;
+               }
+
+               ctx = best_ctx;
+               nand_op_parser_trace(&ctx);
+
+               if (!check_only) {
+                       pattern = &parser->patterns[best_pattern];
                        ret = pattern->exec(chip, &ctx.subop);
                        if (ret)
                                return ret;
-
-                       break;
-               }
-
-               if (i == parser->npatterns) {
-                       pr_debug("->exec_op() parser: pattern not found!\n");
-                       return -ENOTSUPP;
                }
 
                /*

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

  reply	other threads:[~2019-03-30  9:21 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-03-29 13:35 nand_op_parser_exec_op should use longest pattern Sascha Hauer
2019-03-29 13:57 ` Boris Brezillon
2019-03-29 14:37 ` Stefan Agner
2019-03-30  9:21   ` Boris Brezillon [this message]
2019-04-08 20:00     ` Stefan Agner
2019-04-08 21:30       ` Miquel Raynal
2019-04-16 21:22       ` Miquel Raynal

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=20190330102134.4e85e130@collabora.com \
    --to=boris.brezillon@collabora.com \
    --cc=linux-mtd@lists.infradead.org \
    --cc=miquel.raynal@bootlin.com \
    --cc=s.hauer@pengutronix.de \
    --cc=stefan@agner.ch \
    /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;
as well as URLs for NNTP newsgroup(s).