2006-02-28 03:20:26

by Luke Yang

Subject: [PATCH] Replace "vmalloc_node" with "vmalloc" for no-mmu architectures in oprofile driver

Hi,

This is a fix to the oprofile driver. It calls "vmalloc_node()" but
no-mmu CPUs do not have that function. "vmalloc()" is OK for no-mmu
CPUs.

Signed-off-by: Luke Yang <[email protected]>

Index: linux-2.6/drivers/oprofile/cpu_buffer.c
===================================================================
--- linux-2.6/drivers/oprofile/cpu_buffer.c 2006-02-16
16:16:35.000000000 +0800
+++ linux-2.6/drivers/oprofile/cpu_buffer.c 2006-02-16
16:20:58.000000000 +0800
@@ -51,9 +51,13 @@

for_each_online_cpu(i) {
struct oprofile_cpu_buffer * b = &cpu_buffer[i];
-
+
+#ifdef MMU
b->buffer = vmalloc_node(sizeof(struct op_sample) * buffer_size,
cpu_to_node(i));
+#else
+ b->buffer = vmalloc(sizeof(struct op_sample) * buffer_size);
+#endif
if (!b->buffer)
goto fail;

--
Best regards,
Luke Yang
[email protected]; [email protected]

2006-02-28 03:34:25

by Andrew Morton

Subject: Re: [PATCH] Replace "vmalloc_node" with "vmalloc" for no-mmu architectures in oprofile driver

"Luke Yang" <[email protected]> wrote:
>
> This is a fix to the oprofile driver. It calls "vmalloc_node()" but
> no-mmu CPUs do not have that function. "vmalloc()" is OK for no-mmu
> CPUs.
>
> Signed-off-by: Luke Yang <[email protected]>
>
> Index: linux-2.6/drivers/oprofile/cpu_buffer.c
> ===================================================================
> --- linux-2.6/drivers/oprofile/cpu_buffer.c 2006-02-16
> 16:16:35.000000000 +0800
> +++ linux-2.6/drivers/oprofile/cpu_buffer.c 2006-02-16
> 16:20:58.000000000 +0800
> @@ -51,9 +51,13 @@
>
> for_each_online_cpu(i) {
> struct oprofile_cpu_buffer * b = &cpu_buffer[i];
> -
> +
> +#ifdef MMU
> b->buffer = vmalloc_node(sizeof(struct op_sample) * buffer_size,
> cpu_to_node(i));
> +#else
> + b->buffer = vmalloc(sizeof(struct op_sample) * buffer_size);
> +#endif
> if (!b->buffer)
> goto fail;

You wanted CONFIG_MMU there.

A better fix is to provide vmalloc_node() on nommu architectures. COuld you
compile-test this please?

--- devel/mm/nommu.c~nommu-implement-vmalloc_node 2006-02-27 19:30:47.000000000 -0800
+++ devel-akpm/mm/nommu.c 2006-02-27 19:31:53.000000000 -0800
@@ -53,7 +53,6 @@ DECLARE_RWSEM(nommu_vma_sem);
struct vm_operations_struct generic_file_vm_ops = {
};

-EXPORT_SYMBOL(vmalloc);
EXPORT_SYMBOL(vfree);
EXPORT_SYMBOL(vmalloc_to_page);
EXPORT_SYMBOL(vmalloc_32);
@@ -205,6 +204,13 @@ void *vmalloc(unsigned long size)
{
return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
}
+EXPORT_SYMBOL(vmalloc);
+
+void *vmalloc_node(unsigned long size, int node)
+{
+ return vmalloc(size);
+}
+EXPORT_SYMBOL(vmalloc_node);

/*
* vmalloc_32 - allocate virtually continguos memory (32bit addressable)
_

2006-02-28 03:45:45

by Luke Yang

Subject: Re: [PATCH] Replace "vmalloc_node" with "vmalloc" for no-mmu architectures in oprofile driver

On 2/28/06, Andrew Morton <[email protected]> wrote:

> You wanted CONFIG_MMU there.
Yes.
>
> A better fix is to provide vmalloc_node() on nommu architectures. COuld you
> compile-test this please?
Tested. Then my patch is obsolete. Thanks!

>
> --- devel/mm/nommu.c~nommu-implement-vmalloc_node 2006-02-27 19:30:47.000000000 -0800
> +++ devel-akpm/mm/nommu.c 2006-02-27 19:31:53.000000000 -0800
> @@ -53,7 +53,6 @@ DECLARE_RWSEM(nommu_vma_sem);
> struct vm_operations_struct generic_file_vm_ops = {
> };
>
> -EXPORT_SYMBOL(vmalloc);
> EXPORT_SYMBOL(vfree);
> EXPORT_SYMBOL(vmalloc_to_page);
> EXPORT_SYMBOL(vmalloc_32);
> @@ -205,6 +204,13 @@ void *vmalloc(unsigned long size)
> {
> return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
> }
> +EXPORT_SYMBOL(vmalloc);
> +
> +void *vmalloc_node(unsigned long size, int node)
> +{
> + return vmalloc(size);
> +}
> +EXPORT_SYMBOL(vmalloc_node);
>
> /*
> * vmalloc_32 - allocate virtually continguos memory (32bit addressable)
> _
>
>


--
Best regards,
Luke Yang
[email protected]; [email protected]