Skip to contentSkip to conversation

auxdisplay: charlcd: Fix and clean up handling of x/y commands

RFC PATCH Superseded Miguel Ojeda 3 participants 5 messages Feb 27, 2018 Auxiliary display drivers

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 auxdisplay: charlcd: Fix and clean up handling of x/y commands

The current version is not parsing multiple x/y commands as the code originally intended. On top of that, kstrtoul() expects NULL-terminated strings. Finally, the code had to do two passes over the string, while now only one is done.

Some explanations about the supported syntax are added as well.

Cc Willy Tarreau <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Andy Shevchenko <[email protected]>
Cc Robert Abel <[email protected]>

---
Only the parsing functions are tested, please try on real HW.

 drivers/auxdisplay/charlcd.c | 97 ++++++++++++++++++++++++++++++++++++--------
 1 file changed, 79 insertions(+), 18 deletions(-)
drivers/auxdisplay/charlcd.c +79 -18
index 642afd88870b..aa9a6abc2ada 100644
@@ -11,6 +11,7 @@
  */
 
 #include <linux/atomic.h>
+#include <linux/ctype.h>
 #include <linux/delay.h>
 #include <linux/fs.h>
 #include <linux/miscdevice.h>
@@ -292,13 +293,86 @@ static int charlcd_init_display(struct charlcd *lcd)
 	return 0;
 }
 
+/*
+ * Parses a base 10 number from a string, until a non-digit number is found or
+ * until PARSE_N_MAX_SIZE digits.
+ *
+ * See kstrtoul() for the meaning of the return value.
+ * It also returns the next character in the string, i.e. the first non-digit.
+ */
+#define PARSE_N_MAX_SIZE (3)
+static unsigned long parse_n(const char *s, unsigned long *res,
+	const char **next_s)
+{
+	char tmp[PARSE_N_MAX_SIZE + 1];
+	int i;
+
+	for (i = 0; ; ++i) {
+		if (!isdigit(s[i]))
+			break;
+		if (i >= PARSE_N_MAX_SIZE)
+			return -EINVAL;
+		tmp[i] = s[i];
+	}
+
+	tmp[i] = 0;
+	*next_s = &s[i];
+	return kstrtoul(tmp, 10, res);
+}
+
+/*
+ * Parses a movement command of the form "(.*);", where the group can be
+ * any number of subcommands of the form "(x|y)[0-9]+".
+ *
+ * Returns whether the command is valid. The position arguments are
+ * only written if the parsing was successful.
+ *
+ * For instance:
+ *   - ";"          returns (<original x>, <original y>).
+ *   - "x1;"        returns (1, <original y>).
+ *   - "y2x1;"      returns (1, 2).
+ *   - "x12y34x56;" returns (56, 34).
+ *   - "x1"         fails.
+ *   - "xy12;"      fails.
+ *   - "x12yy12;"   fails.
+ */
+static bool parse_xy(const char *s, unsigned long *x, unsigned long *y)
+{
+	unsigned long new_x = *x;
+	unsigned long new_y = *y;
+	const char *next_s;
+
+	for (;;) {
+		if (!*s)
+			return false;
+
+		if (*s == ';')
+			break;
+
+		if (*s == 'x') {
+			if (parse_n(s + 1, &new_x, &next_s) != 0)
+				return false;
+			s = next_s;
+		} else if (*s == 'y') {
+			if (parse_n(s + 1, &new_y, &next_s) != 0)
+				return false;
+			s = next_s;
+		} else {
+			return false;
+		}
+	}
+
+	*x = new_x;
+	*y = new_y;
+	return true;
+}
+
 /*
  * These are the file operation function for user access to /dev/lcd
  * This function can also be called from inside the kernel, by
  * setting file and ppos to NULL.
  *
  */
-
 static inline int handle_lcd_special_code(struct charlcd *lcd)
 {
 	struct charlcd_priv *priv = to_priv(lcd);
@@ -469,24 +543,11 @@ static inline int handle_lcd_special_code(struct charlcd *lcd)
 	}
 	case 'x':	/* gotoxy : LxXXX[yYYY]; */
 	case 'y':	/* gotoxy : LyYYY[xXXX]; */
-		if (!strchr(esc, ';'))
-			break;
-
-		while (*esc) {
-			if (*esc == 'x') {
-				esc++;
-				if (kstrtoul(esc, 10, &priv->addr.x) < 0)
-					break;
-			} else if (*esc == 'y') {
-				esc++;
-				if (kstrtoul(esc, 10, &priv->addr.y) < 0)
-					break;
-			} else {
-				break;
-			}
-		}
+		/* If the command is valid, move to the new address */
+		if (parse_xy(esc, &priv->addr.x, &priv->addr.y))
+			charlcd_gotoxy(lcd);
 
-		charlcd_gotoxy(lcd);
+		/* Regardless of its validity, mark as processed */
 		processed = 1;
 		break;
 	}
Signature · 1 line
2.14.1
#1 lore raw
Quoted 6 lines · Miguel Ojeda

The current version is not parsing multiple x/y commands as the code originally intended. On top of that, kstrtoul() expects NULL-terminated strings. Finally, the code had to do two passes over the string, while now only one is done.

Some explanations about the supported syntax are added as well.

Thanks Miguel for doing this. To be fair, I personally think that the code was significantly complexified compared to the original version, just for the purpose of using one library function designed to multiply by ten and add a number. Not to mention that it now requires a copy before parsing. For me it simply means that the initial code change was a wrong idea in the end. But at least with your version, the bugs introduced by the previous "fix" should now be gone, which is a good thing.

Thanks,
Willy

#2 lore raw
Quoted 9 lines · Willy Tarreau
Quoted 6 lines · Miguel Ojeda

The current version is not parsing multiple x/y commands as the code originally intended. On top of that, kstrtoul() expects NULL-terminated strings. Finally, the code had to do two passes over the string, while now only one is done.

Some explanations about the supported syntax are added as well.

Thanks Miguel for doing this. To be fair, I personally think that the code

You're welcome!

Quoted 3 lines

was significantly complexified compared to the original version, just for the purpose of using one library function designed to multiply by ten and add a number. Not to mention that it now requires a copy before parsing.

I agree. I wrote this version like this so that parse_n() can be removed in the future if we have access to an in-place alternative to the kstrtoul*() functions (cf. the discussion in the other thread with Andy); which would take the copy away as well.

Quoted 3 lines

For me it simply means that the initial code change was a wrong idea in the end. But at least with your version, the bugs introduced by the previous "fix" should now be gone, which is a good thing.

Thanks! :-) This one has the "advantage" that it only goes by the string once (i.e. no initial strchr()) and that gotoxy() is not called unless the full command is valid (but I am not sure whether this is OK or not for the HW, though).

Cheers,
Miguel

Quoted 3 lines

Thanks,
Willy

#3 lore raw
In reply to Miguel Ojeda
Quoted 6 lines · Miguel Ojeda

The current version is not parsing multiple x/y commands as the code originally intended. On top of that, kstrtoul() expects NULL-terminated strings. Finally, the code had to do two passes over the string, while now only one is done.

Some explanations about the supported syntax are added as well.

Something like this would work, but see my comments below.

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 305

Quoted 10 lines
drivers/auxdisplay/charlcd.c +10 -0
+/*
+ * Parses a base 10 number from a string, until a non-digit number is found or
+ * until PARSE_N_MAX_SIZE digits.
+ *
+ * See kstrtoul() for the meaning of the return value.
+ * It also returns the next character in the string, i.e. the first non-digit.
+ */
+#define PARSE_N_MAX_SIZE (3)
+static unsigned long parse_n(const char *s, unsigned long *res,
+       const char **next_s)

static int parse_n(const char *s, unsigned long *res, const char **next_s)

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 309

Quoted 4 lines
drivers/auxdisplay/charlcd.c +4 -0
+{
+       char tmp[PARSE_N_MAX_SIZE + 1];
+       int i;
+

First of all you need

unsigned int i = 0;

while (s[i] == '0')
i++;

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 316

Quoted 7 lines
drivers/auxdisplay/charlcd.c +7 -0
+       for (i = 0; ; ++i) {
+               if (!isdigit(s[i]))
+                       break;
+               if (i >= PARSE_N_MAX_SIZE)
+                       return -EINVAL;
+               tmp[i] = s[i];
+       }

And here you partially open coded what simple_strto*() does.

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 321

Quoted 5 lines
drivers/auxdisplay/charlcd.c +5 -0
+
+       tmp[i] = 0;
+       *next_s = &s[i];
+       return kstrtoul(tmp, 10, res);
+}

It's over engineered. Just simple use simple_strto*() and that's all. Do you really care about overflows? Is those numbers somehow critical? What will go wrong?

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#4 lore raw
Quoted 11 lines · Andy Shevchenko
Quoted 6 lines · Miguel Ojeda

The current version is not parsing multiple x/y commands as the code originally intended. On top of that, kstrtoul() expects NULL-terminated strings. Finally, the code had to do two passes over the string, while now only one is done.

Some explanations about the supported syntax are added as well.

Something like this would work, but see my comments below.

Quoted 12 lines

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 305

Quoted 10 lines
drivers/auxdisplay/charlcd.c +10 -0
+/*
+ * Parses a base 10 number from a string, until a non-digit number is found or
+ * until PARSE_N_MAX_SIZE digits.
+ *
+ * See kstrtoul() for the meaning of the return value.
+ * It also returns the next character in the string, i.e. the first non-digit.
+ */
+#define PARSE_N_MAX_SIZE (3)
+static unsigned long parse_n(const char *s, unsigned long *res,
+       const char **next_s)

static int parse_n(const char *s, unsigned long *res, const char **next_s)

Oops, thanks!

Quoted 8 lines

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 309

Quoted 5 lines
drivers/auxdisplay/charlcd.c +4 -0
+{
+       char tmp[PARSE_N_MAX_SIZE + 1];
+       int i;
+

First of all you need

Quoted 4 lines

unsigned int i = 0;

while (s[i] == '0')
i++;

Why?

Quoted 11 lines

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 316

Quoted 8 lines
drivers/auxdisplay/charlcd.c +7 -0
+       for (i = 0; ; ++i) {
+               if (!isdigit(s[i]))
+                       break;
+               if (i >= PARSE_N_MAX_SIZE)
+                       return -EINVAL;
+               tmp[i] = s[i];
+       }

And here you partially open coded what simple_strto*() does.

Quoted 9 lines

drivers/auxdisplay/charlcd.c static int charlcd_init_display(struct charlcd *lcd) · line 321

Quoted 5 lines
drivers/auxdisplay/charlcd.c +5 -0
+
+       tmp[i] = 0;
+       *next_s = &s[i];
+       return kstrtoul(tmp, 10, res);
+}

It's over engineered. Just simple use simple_strto*() and that's all. Do you really care about overflows? Is those numbers somehow critical? What will go wrong?

We shouldn't use simple_strto*() since it is deprecated. So parse_n() can be either this wrapper to call kstrto*() (which indeed is overengineered, but easier to understand than Robert's patch that modified the original sequence, and can be replaced by an inplace version whenever we have one) or we just put a simple loop for the moment (like the one Willy coded in 2008). Both are fine with me.

Cheers,
Miguel

Quoted 4 lines

--
With Best Regards,
Andy Shevchenko