ioremap balanced with iounmap for drivers/char/epca.c
PATCH 1/5 ioremap balanced with iounmap for drivers/char/epca.c
ioremap must be balanced by an iounmap and failing to do so can result in a memory leak.
Tested (compilation only):
- using allmodconfig
- making sure the files are compiling without any warning/error due to new changes
--- epca.c | 5 ++++- 1 files changed, 4 insertions(+), 1 deletion(-)
---
drivers/char/epca.c +4 -1
@@ -1474,8 +1474,11 @@ static void post_fep_init(unsigned int cif ((bd->type ==PCXEVE || bd->type ==PCXE ) && (readw (memaddr +XEPORTS ) <3 )) shrinkmem =1 ;if (bd->type <PCIXEM ) -if (!request_region ((int )bd->port,4 , board_desc[bd->type])) +if (!request_region ((int )bd->port,4 , board_desc[bd->type])) { +iounmap (bd->re_map_membase); + bd->re_map_membase =NULL ;return ; + }memwinon (bd,0 );/* --------------------------------------------------------------------
Quoted 18 lines · On Fri, 06 Oct 2006 10:27:05 +0530 Amol Lad <[email protected]> wrote:
ioremap must be balanced by an iounmap and failing to do so can result in a memory leak.
Tested (compilation only):
- using allmodconfig
- making sure the files are compiling without any warning/error due to new changesSigned-off-by Amol Lad <[email protected]>--- epca.c | 5 ++++- 1 files changed, 4 insertions(+), 1 deletion(-)---
diff -uprN -X linux-2.6.19-rc1-orig/Documentation/dontdifflinux-2.6.19-rc1-orig/drivers/char/epca.clinux-2.6.19-rc1/drivers/char/epca.cdrivers/char/epca.c +0
-0@@ -1474,8 +1474,11 @@ static void post_fep_init(unsigned int cif ((bd->type ==PCXEVE || bd->type ==PCXE ) && (readw (memaddr +XEPORTS ) <3 ))
drivers/char/epca.c static void post_fep_init(unsigned int c · line 1483
Quoted 10 lines
drivers/char/epca.c +4
-1shrinkmem =1 ;if (bd->type <PCIXEM ) -if (!request_region ((int )bd->port,4 , board_desc[bd->type])) +if (!request_region ((int )bd->port,4 , board_desc[bd->type])) { +iounmap (bd->re_map_membase); + bd->re_map_membase =NULL ;return ; + }memwinon (bd,0 );
I think this will do the wrong thing if (bd->type >= PCIXEM). Maybe it's OK, but it's not immediately obvious from a quick reading.
I'm quite worried about changes in crufty old drivers like these - if we break them, the breakage will take a *long* time to be discovered. Too late for us to fix them.
Plus a lot of them are just plain badly coded, so extra care is needed to understand the tricks which they're playing.
Quoted 26 lines · Andrew Morton
Quoted 23 lines · On Fri, 06 Oct 2006 10:27:05 +0530 Amol Lad <[email protected]> wrote:
ioremap must be balanced by an iounmap and failing to do so can result in a memory leak.
Tested (compilation only):
- using allmodconfig
- making sure the files are compiling without any warning/error due to new changesSigned-off-by Amol Lad <[email protected]>--- epca.c | 5 ++++- 1 files changed, 4 insertions(+), 1 deletion(-)---
diff -uprN -X linux-2.6.19-rc1-orig/Documentation/dontdifflinux-2.6.19-rc1-orig/drivers/char/epca.clinux-2.6.19-rc1/drivers/char/epca.cdrivers/char/epca.c +2
-1@@ -1474,8 +1474,11 @@ static void post_fep_init(unsigned int cif ((bd->type ==PCXEVE || bd->type ==PCXE ) && (readw (memaddr +XEPORTS ) <3 )) shrinkmem =1 ;if (bd->type <PCIXEM ) -if (!request_region ((int )bd->port,4 , board_desc[bd->type])) +if (!request_region ((int )bd->port,4 , board_desc[bd->type])) { +iounmap (bd->re_map_membase);
Quoted 8 lines
drivers/char/epca.c static void post_fep_init(unsigned int c · line 1483
Quoted 5 lines
drivers/char/epca.c +2
-0+ bd->re_map_membase =NULL ;return ; + }memwinon (bd,0 );I think this will do the wrong thing if (
bd->type>= PCIXEM). Maybe it's OK, but it's not immediately obvious from a quick reading.
A laymans thought here. As ioremap was done for bd->type < PCIXEM, so iounmap should also be done for the same case.
But as you see the function is void and not handling errors, so this change can have side effects... well.. it can misbehave even without this change in the failure case..
Quoted 3 lines
Plus a lot of them are just plain badly coded, so extra care is needed to understand the tricks which they're playing.
I think module owner should take responsibility for this. One more example of badly coded driver is drivers/char/cyclades.c. I was not at all able to do ioremap balance with iounmap for this one.