Re: [PATCH v3] add new Cobalt LCD framebuffer driver

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 15 Jul 2008 22:27:51 +0900
Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp> wrote:

> I update cobalt_lcdfb driver.
> 
> v3
> - fix read/write count boundary check.
> - add <include/uaccess.h>.
> - fix MODULE_AUTHOR.
> 
> v2
> - add dpends MIPS_COBALT in Kconfig.
> - add handling of signal_pending().
> - check overflow of read/write count.
> 
> v1
> - first release.

This driver has been merged into the subsystem tree for a long time and
has been reviewed and has been perhaps tested by others.  Sending a
complete new version of the entire thing is really quite an unfriendly
act.  It basically invalidates all the review and test work which
everyone else has done.

This is why I will almost always turn replacement patches into
incremental patches - so I can see what changed.  But that's only good
for me.  Everyone else who is reading your new version won't bother
doing that.


So I generated the incremental patch and:

- The driver I have already includes linux/uaccess.h, so that change
  was unneeded.

- The driver I have has already fixed the reject in
  drivers/video/Makefile, so I get to fix it again, ho hum.

Here's what I ended up committing:

 drivers/video/Kconfig        |    2 +-
 drivers/video/cobalt_lcdfb.c |   24 ++++++++++++++++++------
 2 files changed, 19 insertions(+), 7 deletions(-)

diff -puN drivers/video/Kconfig~add-new-cobalt-lcd-framebuffer-driver 
drivers/video/Kconfig
--- a/drivers/video/Kconfig~add-new-cobalt-lcd-framebuffer-driver
+++ a/drivers/video/Kconfig
@@ -1989,7 +1989,7 @@ config FB_AM200EPD
 
 config FB_COBALT
        tristate "Cobalt server LCD frame buffer support"
-       depends on FB
+       depends on FB && MIPS_COBALT
 
 config FB_VIRTUAL
        tristate "Virtual Frame Buffer support (ONLY FOR TESTING!)"
diff -puN drivers/video/Makefile~add-new-cobalt-lcd-framebuffer-driver 
drivers/video/Makefile
diff -puN drivers/video/cobalt_lcdfb.c~add-new-cobalt-lcd-framebuffer-driver 
drivers/video/cobalt_lcdfb.c
--- a/drivers/video/cobalt_lcdfb.c~add-new-cobalt-lcd-framebuffer-driver
+++ a/drivers/video/cobalt_lcdfb.c
@@ -137,13 +137,16 @@ static ssize_t cobalt_lcdfb_read(struct 
 {
        char src[LCD_CHARS_MAX];
        unsigned long pos;
-       int len, retval;
+       int len, retval = 0;
 
        pos = *ppos;
-       if (pos >= LCD_CHARS_MAX)
+       if (pos >= LCD_CHARS_MAX || count == 0)
                return 0;
 
-       if (pos + count >= LCD_CHARS_MAX)
+       if (count > LCD_CHARS_MAX)
+               count = LCD_CHARS_MAX;
+
+       if (pos + count > LCD_CHARS_MAX)
                count = LCD_CHARS_MAX - pos;
 
        for (len = 0; len < count; len++) {
@@ -164,6 +167,9 @@ static ssize_t cobalt_lcdfb_read(struct 
                        pos++;
        }
 
+       if (retval < 0 && signal_pending(current))
+               return -ERESTARTSYS;
+
        if (copy_to_user(buf, src, len))
                return -EFAULT;
 
@@ -177,13 +183,16 @@ static ssize_t cobalt_lcdfb_write(struct
 {
        char dst[LCD_CHARS_MAX];
        unsigned long pos;
-       int len, retval;
+       int len, retval = 0;
 
        pos = *ppos;
-       if (pos >= LCD_CHARS_MAX)
+       if (pos >= LCD_CHARS_MAX || count == 0)
                return 0;
 
-       if (pos + count >= LCD_CHARS_MAX)
+       if (count > LCD_CHARS_MAX)
+               count = LCD_CHARS_MAX;
+
+       if (pos + count > LCD_CHARS_MAX)
                count = LCD_CHARS_MAX - pos;
 
        if (copy_from_user(dst, buf, count))
@@ -207,6 +216,9 @@ static ssize_t cobalt_lcdfb_write(struct
                        pos++;
        }
 
+       if (retval < 0 && signal_pending(current))
+               return -ERESTARTSYS;
+
        *ppos += len;
 
        return len;
_


and I'm a bit uncertain about it.

- the test for (count > LCD_CHARS_MAX) appear to be unneeded, given
  the follwing test for (pos + count > LCD_CHARS_MAX).

  Or perhaps it is an obscure test for very large tests of `count',
  which would cause overflows in `count+pos'?  Can it all be simplified?

- Do we need to test for `count == 0'?  I have a feeling that this
  case is handled at the higher levels of read(), but I cannot find
  that piece of code at present.  Perhaps I dreamed it.