nommu patch against 2.6.12-rc3-mm3, for mm/nommu.c :
- need to fix the vmalloc definition.
- most malloc functions in nommu does not provide the 'len'. should consider
"zero" len value.
If it does return with error when the 'len' is zero, the whole exist
applications are unable to run, because of the std-c libraries for uClinux.
Signed-off-by : Hyok S. Choi <[email protected]>
Index: linux-2.6.12-rc3-mm3/mm/nommu.c
================================================================
--- linux-2.6.12-rc3-mm3/mm/nommu.c 2005-05-06 09:58:44.000000000 +0900
+++ linux-2.6.12-rc3-mm3-hsc0/mm/nommu.c 2005-05-06
13:53:54.000000000 +0900
@@ -6,6 +6,7 @@
*
* See Documentation/nommu-mmap.txt
*
+ * Copyright (c) 2005 Hyok S. Choi <[email protected]>
* Copyright (c) 2004-2005 David Howells <[email protected]>
* Copyright (c) 2000-2003 David McCullough <[email protected]>
* Copyright (c) 2000-2001 D Jeff Dionne <[email protected]>
@@ -146,7 +147,7 @@
kfree(addr);
}
-void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
+void *__vmalloc(unsigned long size, unsigned int gfp_mask, pgprot_t prot)
{
/*
* kmalloc doesn't like __GFP_HIGHMEM for some reason
@@ -901,10 +902,12 @@
#ifdef DEBUG
printk("do_munmap:\n");
#endif
-
+ /* if the len is zero(as some simple-malloc libs does),
+ * treat it with the start addr only.
+ */
for (parent = &mm->context.vmlist; *parent; parent =
&(*parent)->next)
if ((*parent)->vma->vm_start == addr &&
- (*parent)->vma->vm_end == end)
+ (!len || (*parent)->vma->vm_end == end))
goto found;
printk("munmap of non-mmaped memory by process %d (%s): %p\n",
@@ -914,6 +917,9 @@
found:
vml = *parent;
+ if(!len)
+ len = vml->vma->vm_end - addr;
+
put_vma(vml->vma);
*parent = vml->next;
Hyok S. Choi <[email protected]> wrote:
> - (*parent)->vma->vm_end == end)
> + (!len || (*parent)->vma->vm_end == end))
Please make this configurable. It's bypassing an error case check.
David
David Howells wrote:
> Hyok S. Choi <[email protected]> wrote:
>
>
>>- (*parent)->vma->vm_end == end)
>>+ (!len || (*parent)->vma->vm_end == end))
>
>
> Please make this configurable. It's bypassing an error case check.
I disagree, I don't think it warrents yet another configuration option.
Most uClinux archiectures don't want to do this size error check - at
least any that rely on the current implementation of light weight uClibc
mmap() based malloc(). And up until recently this was _every_ uClinux
architecture.
Regards
Greg
------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: [email protected]
SnapGear -- a CyberGuard Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com