make vmalloc gfp flags usage more apparent
Subsystems (2)
cover make vmalloc gfp flags usage more apparent
We should do a better job at enforcing gfp flags for vmalloc. Right now, we have a kernel-doc for __vmalloc_node_range(), and hope callers pass in supported flags. If a caller were to pass in an unsupported flag, we may BUG, silently clear it, or completely ignore it.
If we are more proactive about enforcing gfp flags, we can making sure callers know when they may be asking for unsupported behavior.
This patchset lets vmalloc control the incoming gfp flags, and cleans up some confusing gfp code.
----------------
Based on mm-new
I did some digging and am not entirely sure what flags vmalloc does NOT support. Is a better idea is to have explicitly supported flags and drop all others?
__GFP_COMP is an obvious one due to a BUG call in split_page(). ~GFP_BITS_MASK is also obvious.
Then I started following the kernel doc and added NORETRY and RETRY_MAYFAIL, and after forking a couple hundred times, it turns out some per-cpu allocations pass in the NORETRY flag right now.
Does anyone have a handy-dandy list of supported/unsupported vmalloc flags that we should reject/clear? Ulad?
Vishal Moola (Oracle) (4):
mm/vmalloc: warn on invalid vmalloc gfp flags mm/vmalloc: Add a helper to optimize vmalloc allocation gfps mm/vmalloc: cleanup large_gfp in vm_area_alloc_pages() mm/vmalloc: cleanup gfp flag use in new_vmap_block()
mm/vmalloc.c | 41 +++++++++++++++++++++++++++++++++-------- 1 file changed, 33 insertions(+), 8 deletions(-)
Signature · 1 line
2.51.1
RFC PATCH 1/4 mm/vmalloc: warn on invalid vmalloc gfp flags
There are some gfp flags that are not supported by vmalloc. vmalloc has been trying to handle these by clearing and setting flags wherever necessary. This is messy and makes the code harder to understand, when we could simply check for these unsupported flags immediately.
Define a helper mask and function telling callers they have passed in invalid flags, and clear those unsupported vmalloc flags.
--- mm/vmalloc.c | 17 +++++++++++++++++ 1 file changed, 17 insertions(+)
mm/vmalloc.c +17 -0
@@ -3911,6 +3911,20 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,return NULL ; } +#define GFP_VMALLOC_BUG_MASK (__GFP_COMP|__GFP_RETRY_MAYFAIL|\ + ~__GFP_BITS_MASK) +static gfp_t vmalloc_fix_flags (gfp_t flags) +{ +gfp_t invalid_mask = flags &GFP_VMALLOC_BUG_MASK ; + + flags &= ~GFP_VMALLOC_BUG_MASK ; +pr_warn ("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n" , + invalid_mask, &invalid_mask, flags, &flags); +dump_stack (); + +return flags; +} +/** * __vmalloc_node_range - allocate virtually contiguous memory * @size: allocation size
@@ -3961,6 +3975,9 @@ void *__vmalloc_node_range_noprof(unsigned long size, unsigned long align,return NULL ; } +if (unlikely (gfp_mask &GFP_VMALLOC_BUG_MASK )) + gfp_mask =vmalloc_fix_flags (gfp_mask); +if (vmap_allow_huge && (vm_flags &VM_ALLOW_HUGE_VMAP )) {/* * Try huge pages. Only try for PAGE_KERNEL allocations,
Signature · 1 line
2.51.1
RFC PATCH 2/4 mm/vmalloc: Add a helper to optimize vmalloc allocation gfps
In reply to Vishal Moola (Oracle)vm_area_alloc_pages() attempts to use different gfp flags as a way to optimize allocations. This has been done inline which makes things harder to read.
Add a helper function to make the code more readable.
--- mm/vmalloc.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-)
mm/vmalloc.c +14 -3
@@ -3614,6 +3614,17 @@ void *vmap_pfn(unsigned long *pfns, unsigned int count, pgprot_t prot)EXPORT_SYMBOL_GPL (vmap_pfn);#endif /* CONFIG_VMAP_PFN */ +/* +* Helper for vmalloc to adjust the gfp flags for certain allocations. +*/ +static inline gfp_t vmalloc_gfp_adjust (gfp_t flags,const bool large) +{ + flags |= __GFP_NOWARN; +if (large) + flags &= ~__GFP_NOFAIL; +return flags; +} +static inline unsigned int vm_area_alloc_pages (gfp_t gfp,int nid,unsigned int order,unsigned int nr_pages,struct page **pages)
@@ -3852,9 +3863,9 @@ static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,* Please note, the __vmalloc_node_range_noprof() falls-back * to order-0 pages if high-order attempt is unsuccessful. */ - area->nr_pages =vm_area_alloc_pages ((page_order ? - gfp_mask & ~__GFP_NOFAIL : gfp_mask) | __GFP_NOWARN, - node, page_order, nr_small_pages, area->pages); + area->nr_pages =vm_area_alloc_pages ( +vmalloc_gfp_adjust (gfp_mask, page_order), node, + page_order, nr_small_pages, area->pages);atomic_long_add (area->nr_pages, &nr_vmalloc_pages);/* All pages of vm should be charged to same memcg, so use first one. */
Signature · 1 line
2.51.1
RFC PATCH 3/4 mm/vmalloc: cleanup large_gfp in vm_area_alloc_pages()
In reply to Vishal Moola (Oracle)Now that we have already checked for unsupported flags, we can use the helper function to set the necessary gfp flags for the large order allocation optimization.
--- mm/vmalloc.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)
mm/vmalloc.c +1 -3
@@ -3634,10 +3634,8 @@ vm_area_alloc_pages(gfp_t gfp, int nid,unsigned int max_attempt_order =MAX_PAGE_ORDER ;struct page *page;int i; -gfp_t large_gfp = (gfp & - ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL | __GFP_COMP)) - | __GFP_NOWARN;unsigned int large_order =ilog2 (nr_remaining); +gfp_t large_gfp =vmalloc_gfp_adjust (gfp, large_order) & ~__GFP_DIRECT_RECLAIM; large_order =min (max_attempt_order, large_order);
Signature · 1 line
2.51.1
RFC PATCH 4/4 mm/vmalloc: cleanup gfp flag use in new_vmap_block()
In reply to Vishal Moola (Oracle)The only caller, vb_alloc(), passes GFP_KERNEL into new_vmap_block() which is a subset of GFP_RECLAIM_MASK. Since there's no reason to use this mask here, remove it.
--- mm/vmalloc.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
mm/vmalloc.c +1 -2
@@ -2699,8 +2699,7 @@ static void *new_vmap_block(unsigned int order, gfp_t gfp_mask) node =numa_node_id (); - vb =kmalloc_node (sizeof (struct vmap_block ), - gfp_mask &GFP_RECLAIM_MASK , node); + vb =kmalloc_node (sizeof (struct vmap_block ), gfp_mask, node);if (unlikely (!vb))return ERR_PTR (-ENOMEM );
Signature · 1 line
2.51.1
Uladzislau Rezki
maintainerQuoted 21 lines · Vishal Moola (Oracle)
We should do a better job at enforcing gfp flags for vmalloc. Right now, we have a kernel-doc for
__vmalloc_node_range(), and hope callers pass in supported flags. If a caller were to pass in an unsupported flag, we may BUG, silently clear it, or completely ignore it.If we are more proactive about enforcing gfp flags, we can making sure callers know when they may be asking for unsupported behavior.
This patchset lets vmalloc control the incoming gfp flags, and cleans up some confusing gfp code.
----------------
Based on mm-newI did some digging and am not entirely sure what flags vmalloc does NOT support. Is a better idea is to have explicitly supported flags and drop all others?
__GFP_COMPis an obvious one due to a BUG call insplit_page(). ~GFP_BITS_MASK is also obvious.
Quoted 7 lines
Then I started following the kernel doc and added NORETRY and
RETRY_MAYFAIL, and after forking a couple hundred times, it turns out some per-cpu allocations pass in the NORETRY flag right now.Does anyone have a handy-dandy list of supported/unsupported vmalloc flags that we should reject/clear? Ulad?
We recently have added GFP_ATOMIC, GFP_NOWAIT support. Both are handled based on gfpflags_allow_blocking() checking:
<snip>
* Supported GFP classes: %GFP_KERNEL, %GFP_ATOMIC, %GFP_NOWAIT,
* %GFP_NOFS and %GFP_NOIO. Zone modifiers are not supported.
* Please note %GFP_ATOMIC and %GFP_NOWAIT are supported only
* by __vmalloc().
<snip>
Quoted 5 lines
I did some digging and am not entirely sure what flags vmalloc does NOT support. Is a better idea is to have explicitly supported flags and drop all others?
Maybe we should look at it vice versa. Focus on supported flags. In the slab there is an adjust function which modifies the gfp and emits the warning if passed GFP is part of buggy mask.
--
Uladzislau Rezki
Quoted 7 lines · Uladzislau Rezki
Quoted 4 lines
I did some digging and am not entirely sure what flags vmalloc does NOT support. Is a better idea is to have explicitly supported flags and drop all others?
Maybe we should look at it vice versa. Focus on supported flags. In the slab there is an adjust function which modifies the gfp and emits the warning if passed GFP is part of buggy mask.
Yes, explicitly whitelisting the (component)flags supported seems like a much more maintainable approach.