split_page function to split higher order pages?
Subsystems (6)
RFC 1/1 split_page function to split higher order pages?
Hi,
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
The patch isn't actually for merging (because it relies on some xtensa fixes which I'm waiting for the maintainer to comment on).
Nick
--
Have an explicit mm call to split higher order pages into individual pages. Should help to avoid bugs and be more explicit about the code's intention.
include/linux/mm.h +2 -0
@@ -338,6 +338,8 @@ static inline void get_page(struct pagevoid put_page (struct page *page); +void split_page (struct page *page,unsigned int order); +/* * Multiple processes may "see" the same page. E.g. for untouched * mappings of /dev/null, all processes see the same page full of
mm/memory.c +1 -3
@@ -1204,9 +1204,7 @@ out:* The page has to be a nice clean _individual_ kernel allocation. * If you allocate a compound page, you need to have marked it as * such (__GFP_COMP), or manually just split the page up yourself -* (which is mainly an issue of doing "set_page_count(page, 1)" for -* each sub-page, and then freeing them one by one when you free -* them rather than freeing it as a compound page). +* (see split_page()). * * NOTE! Traditionally this was done with "remap_pfn_range()" which * took an arbitrary page protection parameter. This doesn't allow
mm/page_alloc.c +17 -0
@@ -748,6 +748,23 @@ static inline void prep_zero_page(struct }/* +* split_page takes a non-compound higher-order page, and splits it into +* n (1<<order) sub-pages: page[0..n] +* Each sub-page must be freed individually. +*/ +void split_page (struct page *page,unsigned int order) +{ +int i; + +BUG_ON (PageCompound (page)); +BUG_ON (!page_count (page)); +for (i =1 ; i < (1 << order); i++) { +BUG_ON (page_count (page + i)); +set_page_count (page + i,1 ); + } +} + +/* * Really, prep_compound_page() should be called from __rmqueue_bulk(). But * we cheat by calling it from here, in the order > 0 path. Saves a branch * or two.
arch/ppc/kernel/dma-mapping.c +2 -2
@@ -223,6 +223,8 @@ __dma_alloc_coherent(size_t size, dma_adpte_t *pte = consistent_pte +CONSISTENT_OFFSET (vaddr);struct page *end = page + (1 << order); +split_page (page, order); +/* * Set the "dma handle" */
@@ -231,7 +233,6 @@ __dma_alloc_coherent(size_t size, dma_ad do { BUG_ON(!pte_none(*pte)); - set_page_count(page, 1); SetPageReserved(page); set_pte_at(&init_mm, vaddr, pte, mk_pte(page, pgprot_noncached(PAGE_KERNEL)));
@@ -244,7 +245,6 @@ __dma_alloc_coherent(size_t size, dma_ad* Free the otherwise unused pages. */ while (page < end) { -set_page_count (page,1 );__free_page (page); page++; }
arch/xtensa/mm/pgtable.c +3 -7
@@ -21,13 +21,9 @@ pte_t* pte_alloc_one_kernel(struct mm_st p = (pte_t *)__get_free_pages (GFP_KERNEL |__GFP_REPEAT,COLOR_ORDER );if (likely (p)) { -struct page *page; +split_page (virt_to_page (p),COLOR_ORDER );for (i =0 ; i <COLOR_SIZE ; i++) { - page =virt_to_page (p); - -set_page_count (page,1 ); -if (ADDR_COLOR (p) == color) pte = p;else
@@ -55,9 +51,9 @@ struct page* pte_alloc_one(struct mm_str p =alloc_pages (GFP_KERNEL | __GFP_REPEAT,PTE_ORDER );if (likely (p)) { -for (i =0 ; i <PAGE_ORDER ; i++) { -set_page_count (p,1 ); +split_page (p,COLOR_ORDER ); +for (i =0 ; i <PAGE_ORDER ; i++) {if (PADDR_COLOR (page_address (p)) == color) page = p;else
arch/arm/mm/consistent.c +2 -2
@@ -223,6 +223,8 @@ __dma_alloc(struct device *dev, size_t s pte = consistent_pte[idx] + off; c->vm_pages = page; +split_page (page, order); +/* * Set the "dma handle" */
@@ -231,7 +233,6 @@ __dma_alloc(struct device *dev, size_t s do { BUG_ON(!pte_none(*pte)); - set_page_count(page, 1); /* * x86 does not mark the pages reserved... */
@@ -250,7 +251,6 @@ __dma_alloc(struct device *dev, size_t s* Free the otherwise unused pages. */ while (page < end) { -set_page_count (page,1 );__free_page (page); page++; }
arch/frv/mm/dma-alloc.c +1 -3
@@ -115,9 +115,7 @@ void *consistent_alloc(gfp_t gfp, size_t*/ if (order >0 ) {struct page *rpage =virt_to_page (page); - -for (i =1 ; i < (1 << order); i++) -set_page_count (rpage + i,1 ); +split_page (rpage, order); } err =0 ;
arch/sh/mm/consistent.c +1 -2
@@ -23,6 +23,7 @@ void *consistent_alloc(gfp_t gfp, size_t page =alloc_pages (gfp, order);if (!page)return NULL ; +split_page (page, order); ret =page_address (page); *handle =virt_to_phys (ret);
@@ -37,8 +38,6 @@ void *consistent_alloc(gfp_t gfp, size_t end = page + (1 << order);while (++page < end) { -set_page_count (page,1 ); -/* Free any unused pages */ if (page >= free) {__free_page (page);
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 4 lines · Nick Piggin
Hi,
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
Maybe it's worth documenting that this is for kernel (or even architecture) internal use only and that drivers really shouldn't be doing this..
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 9 lines · Arjan van de Ven
Quoted 4 lines · Nick Piggin
Hi,
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
Maybe it's worth documenting that this is for kernel (or even architecture) internal use only and that drivers really shouldn't be doing this..
I guess it doesn't seem like something drivers would need to use (and none appear to do anything like it).
I'll add a note.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 9 lines
Quoted 6 lines
Quoted 2 lines
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
Maybe it's worth documenting that this is for kernel (or even architecture) internal use only and that drivers really shouldn't be doing this..
I guess it doesn't seem like something drivers would need to use (and none appear to do anything like it).
And I thought this could/should be used together with vm_insert_page() that drivers are supposed to use nowadays instead of remap_pfn_range(). Why shouldn't drivers use this?
Thanks,
Heiko
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 7 lines · Heiko Carstens
Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
Maybe it's worth documenting that this is for kernel (or even architecture) internal use only and that drivers really shouldn't be doing this..
Quoted 6 lines
Quoted 2 lines
I guess it doesn't seem like something drivers would need to use (and none appear to do anything like it).
And I thought this could/should be used together with
vm_insert_page()that drivers are supposed to use nowadays instead ofremap_pfn_range(). Why shouldn't drivers use this?
because in that case the drivers shouldn't allocate higher order pages in the first place...
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
Quoted 7 lines · Heiko Carstens
Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
Just wondering what people think of the idea of using a helper function to split higher order pages instead of doing it manually?
Maybe it's worth documenting that this is for kernel (or even architecture) internal use only and that drivers really shouldn't be doing this..
Quoted 7 lines
Quoted 2 lines
I guess it doesn't seem like something drivers would need to use (and none appear to do anything like it).
And I thought this could/should be used together with
vm_insert_page()that drivers are supposed to use nowadays instead ofremap_pfn_range(). Why shouldn't drivers use this?
Actually fs/ramfs/file-nommu.c does use this (but it need not because !CONFIG_MMU page allocator already does something similar).
But largely, no in-tree drivers use the facility so I think it would be good for the mm folk to consider the first use case if and when it comes up. So I've added a comment in split_page to that effect.
Thanks,
Nick
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>