[Xenomai-core] [PATCH] set VM_RESERVED for remapped mem

[Xenomai-core] [PATCH] set VM_RESERVED for remapped mem

[Jan Kiszka]

[-- Attachment #1.1: Type: text/plain, Size: 508 bytes --]

Hi,

while reviewing my own rtdm_mmap-code again, I noticed that the wrapped
remap_pfn_range() sets VM_RESERVED (+ some other bits) on the passed vma
while the compatibility function remap_page_range() doesn't do this.
This flag seems to exclude the vma from swapping considerations.

Adding the bit to the wrapper appears to me as the best way to catch
this on older kernels. And this should also make the VM_LOCKED setting
obsolete in nucleus/heap.c.

Feel free to correct/enlighten me.

Jan

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1.2: set-VM_RESERVED.patch --]
[-- Type: text/x-patch; name="set-VM_RESERVED.patch", Size: 1772 bytes --]

Index: include/asm-generic/wrappers.h
===================================================================
--- include/asm-generic/wrappers.h	(Revision 600)
+++ include/asm-generic/wrappers.h	(Arbeitskopie)
@@ -46,8 +46,10 @@
 #define module_param_named(name,var,type,mode)  module_param(var,type,mode)
 
 /* VM */
-#define wrap_remap_page_range(vma,from,to,size,prot) \
-    remap_page_range(from,to,size,prot)
+#define wrap_remap_page_range(vma,from,to,size,prot) do { \
+    vma->flags |= VM_RESERVED; \
+    remap_page_range(from,to,size,prot); \
+} while (0)
 #define wrap_switch_mm(prev,next,task)	\
     switch_mm(prev,next,task,(task)->processor)
 #define wrap_enter_lazy_tlb(mm,task)	\
@@ -145,8 +147,10 @@
 #define wrap_remap_page_range(vma,from,to,size,prot)  \
     remap_pfn_range(vma,from,(to) >> PAGE_SHIFT,size,prot)
 #else /* LINUX_VERSION_CODE < KERNEL_VERSION(2,6,10) */
-#define wrap_remap_page_range(vma,from,to,size,prot)  \
-    remap_page_range(vma,from,to,size,prot)
+#define wrap_remap_page_range(vma,from,to,size,prot) do { \
+    vma->flags |= VM_RESERVED; \
+    remap_page_range(vma,from,to,size,prot); \
+} while (0)
 #endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10) */
 #define wrap_switch_mm(prev,next,task)	\
     switch_mm(prev,next,task)
Index: ksrc/nucleus/heap.c
===================================================================
--- ksrc/nucleus/heap.c	(Revision 600)
+++ ksrc/nucleus/heap.c	(Arbeitskopie)
@@ -948,7 +948,6 @@
 	return -ENXIO;	/* Doesn't match the heap size. */
         
     vma->vm_ops = &xnheap_vmops;
-    vma->vm_flags |= VM_LOCKED;	/* Don't swap this out. */
     vma->vm_private_data = file->private_data;
 
     vaddr = (unsigned long)heap->archdep.heapbase;

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 252 bytes --]

Re: [Xenomai-core] [PATCH] set VM_RESERVED for remapped mem

[Philippe Gerum]

Jan Kiszka wrote:
> Hi,
> 
> while reviewing my own rtdm_mmap-code again, I noticed that the wrapped
> remap_pfn_range() sets VM_RESERVED (+ some other bits) on the passed vma
> while the compatibility function remap_page_range() doesn't do this.
> This flag seems to exclude the vma from swapping considerations.
> 
> Adding the bit to the wrapper appears to me as the best way to catch
> this on older kernels. And this should also make the VM_LOCKED setting
> obsolete in nucleus/heap.c.
>

Sounds good, even if the deprecated remap_page_range() did not change the VMA 
flags, but I cannot come with a better approach, and after all, we always align on 
the behaviour of the most recent version supported, i.e. remap_pfn_range(), so 
that's definitely ok. Fixed, thanks.

> Feel free to correct/enlighten me.
> 
> Jan
> 
> 
> ------------------------------------------------------------------------
> 
> Index: include/asm-generic/wrappers.h
> ===================================================================
> --- include/asm-generic/wrappers.h	(Revision 600)
> +++ include/asm-generic/wrappers.h	(Arbeitskopie)
> @@ -46,8 +46,10 @@
>  #define module_param_named(name,var,type,mode)  module_param(var,type,mode)
>  
>  /* VM */
> -#define wrap_remap_page_range(vma,from,to,size,prot) \
> -    remap_page_range(from,to,size,prot)
> +#define wrap_remap_page_range(vma,from,to,size,prot) do { \
> +    vma->flags |= VM_RESERVED; \
> +    remap_page_range(from,to,size,prot); \
> +} while (0)
>  #define wrap_switch_mm(prev,next,task)	\
>      switch_mm(prev,next,task,(task)->processor)
>  #define wrap_enter_lazy_tlb(mm,task)	\
> @@ -145,8 +147,10 @@
>  #define wrap_remap_page_range(vma,from,to,size,prot)  \
>      remap_pfn_range(vma,from,(to) >> PAGE_SHIFT,size,prot)
>  #else /* LINUX_VERSION_CODE < KERNEL_VERSION(2,6,10) */
> -#define wrap_remap_page_range(vma,from,to,size,prot)  \
> -    remap_page_range(vma,from,to,size,prot)
> +#define wrap_remap_page_range(vma,from,to,size,prot) do { \
> +    vma->flags |= VM_RESERVED; \
> +    remap_page_range(vma,from,to,size,prot); \
> +} while (0)
>  #endif /* LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10) */
>  #define wrap_switch_mm(prev,next,task)	\
>      switch_mm(prev,next,task)
> Index: ksrc/nucleus/heap.c
> ===================================================================
> --- ksrc/nucleus/heap.c	(Revision 600)
> +++ ksrc/nucleus/heap.c	(Arbeitskopie)
> @@ -948,7 +948,6 @@
>  	return -ENXIO;	/* Doesn't match the heap size. */
>          
>      vma->vm_ops = &xnheap_vmops;
> -    vma->vm_flags |= VM_LOCKED;	/* Don't swap this out. */
>      vma->vm_private_data = file->private_data;
>  
>      vaddr = (unsigned long)heap->archdep.heapbase;
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Xenomai-core mailing list
> Xenomai-core@domain.hid
> https://mail.gna.org/listinfo/xenomai-core


-- 

Philippe.