Re: [PATCH] alsa: fix comparison of unsigned < 0

[Takashi Iwai <tiwai@suse.de>]

Hi,

thanks for checking this.
They seem actually hitting some real bugs.


At Wed, 30 May 2007 03:35:54 -0400,
Bill Nottingham wrote:
> 
> diff -ru linux-2.6.21-old/sound/pci/ac97/ac97_patch.c linux-2.6.21/sound/pci/ac97/ac97_patch.c
> --- linux-2.6.21-old/sound/pci/ac97/ac97_patch.c	2007-05-30 02:53:05.000000000 -0400
> +++ linux-2.6.21/sound/pci/ac97/ac97_patch.c	2007-05-30 02:32:41.000000000 -0400
> @@ -2086,8 +2086,7 @@
>  	struct snd_ac97 *ac97 = snd_kcontrol_chip(kcontrol);
>  	unsigned short val;
>  
> -	if (ucontrol->value.enumerated.item[0] > 3
> -	    || ucontrol->value.enumerated.item[0] < 0)
> +	if (ucontrol->value.enumerated.item[0] > 3)
>  		return -EINVAL;
>  	val = ctrl2reg[ucontrol->value.enumerated.item[0]]
>  	      << AC97_AD198X_VREF_SHIFT;

This one is fine.


> diff -ru linux-2.6.21-old/sound/pci/ali5451/ali5451.c linux-2.6.21/sound/pci/ali5451/ali5451.c
> --- linux-2.6.21-old/sound/pci/ali5451/ali5451.c	2007-05-30 02:53:05.000000000 -0400
> +++ linux-2.6.21/sound/pci/ali5451/ali5451.c	2007-05-30 02:33:27.000000000 -0400
> @@ -2057,10 +2057,8 @@
>  {
>  	if (codec->hw_initialized)
>  		snd_ali_disable_address_interrupt(codec);
> -	if (codec->irq >= 0) {
> -		synchronize_irq(codec->irq);
> -		free_irq(codec->irq, codec);
> -	}
> +	synchronize_irq(codec->irq);
> +	free_irq(codec->irq, codec);
>  	if (codec->port)
>  		pci_release_regions(codec->pci);
>  	pci_disable_device(codec->pci);

Bah, the check isn't wrong but the type of codec->irq.
It should be int instead of unsigned long.  I'll fix it.


> diff -ru linux-2.6.21-old/sound/pci/ca0106/ca0106_proc.c linux-2.6.21/sound/pci/ca0106/ca0106_proc.c
> --- linux-2.6.21-old/sound/pci/ca0106/ca0106_proc.c	2007-05-30 02:53:05.000000000 -0400
> +++ linux-2.6.21/sound/pci/ca0106/ca0106_proc.c	2007-05-30 02:34:19.000000000 -0400
> @@ -305,7 +305,7 @@
>          while (!snd_info_get_line(buffer, line, sizeof(line))) {
>                  if (sscanf(line, "%x %x", &reg, &val) != 2)
>                          continue;
> -                if ((reg < 0x40) && (reg >=0) && (val <= 0xffffffff) ) {
> +                if ((reg < 0x40) && (val <= 0xffffffff) ) {

It's a 32bit unsigned int, so the check of 0xffffffff isn't needed,
too.

>  			spin_lock_irqsave(&emu->emu_lock, flags);
>  			outl(val, emu->port + (reg & 0xfffffffc));
>  			spin_unlock_irqrestore(&emu->emu_lock, flags);
> @@ -406,7 +406,7 @@
>          while (!snd_info_get_line(buffer, line, sizeof(line))) {
>                  if (sscanf(line, "%x %x %x", &reg, &channel_id, &val) != 3)
>                          continue;
> -                if ((reg < 0x80) && (reg >=0) && (val <= 0xffffffff) && (channel_id >=0) && (channel_id <= 3) )
> +                if ((reg < 0x80) && (val <= 0xffffffff) && (channel_id <= 3) )
>                          snd_ca0106_ptr_write(emu, reg, channel_id, val);
>          }
>  }

Ditto.


> diff -ru linux-2.6.21-old/sound/pci/rme9652/rme9652.c linux-2.6.21/sound/pci/rme9652/rme9652.c
> --- linux-2.6.21-old/sound/pci/rme9652/rme9652.c	2007-05-30 02:53:05.000000000 -0400
> +++ linux-2.6.21/sound/pci/rme9652/rme9652.c	2007-05-30 02:35:16.000000000 -0400
> @@ -406,8 +406,6 @@
>  		} else if (!frag)
>  			return 0;
>  		offset -= rme9652->max_jitter;
> -		if (offset < 0)
> -			offset += period_size * 2;

I think this check is actually needed, but it must be cast to int.
Will fix this, too.


Takashi