sound/oss/msnd_pinnacle.c: ioremap balanced with iounmap
PATCH 3/9 sound/oss/msnd_pinnacle.c: ioremap balanced with iounmap
Signed-off-by Amol Lad <[email protected]>
--- msnd_pinnacle.c | 10 ++++++++++ 1 files changed, 10 insertions(+)
---
sound/oss/msnd_pinnacle.c +10 -0
@@ -1441,6 +1441,8 @@ static int __init attach_multisound(void static void __exit unload_multisound(void) { +iounmap (dev.base); + dev.base =NULL ;release_region (dev.io, dev.numio);free_irq (dev.irq, &dev);unregister_sound_mixer (dev.mixer_minor);
@@ -1884,12 +1886,16 @@ static int __init msnd_init(void) printk(KERN_INFO LOGNAME ": %u byte audio FIFOs (x2)\n", dev.fifosize);if ((err =msnd_fifo_alloc (&dev.DAPF, dev.fifosize)) <0 ) { printk(KERN_ERR LOGNAME ": Couldn't allocate write FIFO\n"); +iounmap (dev.base); + dev.base =NULL ;return err; }if ((err =msnd_fifo_alloc (&dev.DARF, dev.fifosize)) <0 ) { printk(KERN_ERR LOGNAME ": Couldn't allocate read FIFO\n");msnd_fifo_free (&dev.DAPF); +iounmap (dev.base); + dev.base =NULL ;return err; }
@@ -1897,6 +1903,8 @@ static int __init msnd_init(void) printk(KERN_ERR LOGNAME ": Probe failed\n");msnd_fifo_free (&dev.DAPF);msnd_fifo_free (&dev.DARF); +iounmap (dev.base); + dev.base =NULL ;return err; }
@@ -1904,6 +1912,8 @@ static int __init msnd_init(void) printk(KERN_ERR LOGNAME ": Attach failed\n");msnd_fifo_free (&dev.DAPF);msnd_fifo_free (&dev.DARF); +iounmap (dev.base); + dev.base =NULL ;return err; }
Quoted 1 line · On Fri, 06 Oct 2006 11:08:57 +0530 Amol Lad <[email protected]> wrote:
msnd_pinnacle.c | 10 ++++++++++
This driver fails to check that ioremap() actually succeeded. Hence with this patch we can end up doing iounmap(NULL).
Now it _could_ be that the kernel permits iounmap(NULL). But I don't recall that being the rule, and from my reading the powerpc 64-bit iounmap (at least) will go splat if we do this to it. Most other implementations will permit iounmap(NULL) by accident.
Quoted 8 lines · Andrew Morton
Quoted 1 line · On Fri, 06 Oct 2006 11:08:57 +0530 Amol Lad <[email protected]> wrote:
msnd_pinnacle.c | 10 ++++++++++This driver fails to check that
ioremap()actually succeeded. Hence with this patch we can end up doing iounmap(NULL).
Is this better ?
Signed-off-by Amol Lad <[email protected]>
--- msnd_pinnacle.c | 30 +++++++++++++++++++++--------- 1 files changed, 21 insertions(+), 9 deletions(-)
---
sound/oss/msnd_pinnacle.c +21 -9
@@ -1441,6 +1441,10 @@ static int __init attach_multisound(void static void __exit unload_multisound(void) { +if (dev.base) { +iounmap (dev.base); + dev.base =NULL ; + }release_region (dev.io, dev.numio);free_irq (dev.irq, &dev);unregister_sound_mixer (dev.mixer_minor);
@@ -1884,30 +1888,38 @@ static int __init msnd_init(void) printk(KERN_INFO LOGNAME ": %u byte audio FIFOs (x2)\n", dev.fifosize);if ((err =msnd_fifo_alloc (&dev.DAPF, dev.fifosize)) <0 ) { printk(KERN_ERR LOGNAME ": Couldn't allocate write FIFO\n"); -return err; + goto fail1; }if ((err =msnd_fifo_alloc (&dev.DARF, dev.fifosize)) <0 ) { printk(KERN_ERR LOGNAME ": Couldn't allocate read FIFO\n"); -msnd_fifo_free (&dev.DAPF); -return err; + goto fail2; }if ((err =probe_multisound ()) <0 ) { printk(KERN_ERR LOGNAME ": Probe failed\n"); -msnd_fifo_free (&dev.DAPF); -msnd_fifo_free (&dev.DARF); -return err; + goto fail3; }if ((err =attach_multisound ()) <0 ) { printk(KERN_ERR LOGNAME ": Attach failed\n"); -msnd_fifo_free (&dev.DAPF); -msnd_fifo_free (&dev.DARF); -return err; + goto fail3; }return 0 ; + +fail3: +msnd_fifo_free (&dev.DARF); +fail2: +msnd_fifo_free (&dev.DAPF); +fail1: +if (dev.base) { +iounmap (dev.base); + dev.base =NULL ; + } + +return err; + } static void __exit msdn_cleanup(void)