From: Kent Overstreet <kent.overstreet@linux.dev>
We're introducing alloc tagging, which tracks memory allocations by
callsite. Converting alloc_inode_sb() to a macro means allocations will
be tracked by its caller, which is a bit more useful.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
---
include/linux/fs.h | 6 +-----
1 file changed, 1 insertion(+), 5 deletions(-)
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 00fc429b0af0..034f0c918eea 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -3083,11 +3083,7 @@ int setattr_should_drop_sgid(struct mnt_idmap *idmap,
* This must be used for allocating filesystems specific inodes to set
* up the inode reclaim context correctly.
*/
-static inline void *
-alloc_inode_sb(struct super_block *sb, struct kmem_cache *cache, gfp_t gfp)
-{
- return kmem_cache_alloc_lru(cache, &sb->s_inode_lru, gfp);
-}
+#define alloc_inode_sb(_sb, _cache, _gfp) kmem_cache_alloc_lru(_cache, &_sb->s_inode_lru, _gfp)
extern void __insert_inode_hash(struct inode *, unsigned long hashval);
static inline void insert_inode_hash(struct inode *inode)
--
2.44.0.291.gc1ea87d7ee-googOn Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote:
> From: Kent Overstreet <kent.overstreet@linux.dev>
>
> We're introducing alloc tagging, which tracks memory allocations by
> callsite. Converting alloc_inode_sb() to a macro means allocations will
> be tracked by its caller, which is a bit more useful.
I'd have thought that there would be many similar
inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode().
Do we have to go converting things to macros as people report
misleading or less useful results, or is there some more general
solution to this?
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -3083,11 +3083,7 @@ int setattr_should_drop_sgid(struct mnt_idmap *idmap,
> * This must be used for allocating filesystems specific inodes to set
> * up the inode reclaim context correctly.
> */
> -static inline void *
> -alloc_inode_sb(struct super_block *sb, struct kmem_cache *cache, gfp_t gfp)
> -{
> - return kmem_cache_alloc_lru(cache, &sb->s_inode_lru, gfp);
> -}
> +#define alloc_inode_sb(_sb, _cache, _gfp) kmem_cache_alloc_lru(_cache, &_sb->s_inode_lru, _gfp)
Parenthesizing __sb seems sensible here?
On Thu, Mar 21, 2024 at 01:31:47PM -0700, Andrew Morton wrote:
> On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote:
>
> > From: Kent Overstreet <kent.overstreet@linux.dev>
> >
> > We're introducing alloc tagging, which tracks memory allocations by
> > callsite. Converting alloc_inode_sb() to a macro means allocations will
> > be tracked by its caller, which is a bit more useful.
>
> I'd have thought that there would be many similar
> inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode().
> Do we have to go converting things to macros as people report
> misleading or less useful results, or is there some more general
> solution to this?
No, this is just what we have to do.
But a fair number of these helpers shouldn't exist - jbd2_alloc_inode()
is one of those, it looks like it predates kmalloc() being able to use
the page allocator for large allocations.
>
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -3083,11 +3083,7 @@ int setattr_should_drop_sgid(struct mnt_idmap *idmap,
> > * This must be used for allocating filesystems specific inodes to set
> > * up the inode reclaim context correctly.
> > */
> > -static inline void *
> > -alloc_inode_sb(struct super_block *sb, struct kmem_cache *cache, gfp_t gfp)
> > -{
> > - return kmem_cache_alloc_lru(cache, &sb->s_inode_lru, gfp);
> > -}
> > +#define alloc_inode_sb(_sb, _cache, _gfp) kmem_cache_alloc_lru(_cache, &_sb->s_inode_lru, _gfp)
>
> Parenthesizing __sb seems sensible here?
yeah, we can do that
On Thu, 21 Mar 2024 17:15:39 -0400 Kent Overstreet <kent.overstreet@linux.dev> wrote: > On Thu, Mar 21, 2024 at 01:31:47PM -0700, Andrew Morton wrote: > > On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote: > > > > > From: Kent Overstreet <kent.overstreet@linux.dev> > > > > > > We're introducing alloc tagging, which tracks memory allocations by > > > callsite. Converting alloc_inode_sb() to a macro means allocations will > > > be tracked by its caller, which is a bit more useful. > > > > I'd have thought that there would be many similar > > inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode(). > > Do we have to go converting things to macros as people report > > misleading or less useful results, or is there some more general > > solution to this? > > No, this is just what we have to do. Well, this is something we strike in other contexts - kallsyms gives us an inlined function and it's rarely what we wanted. I think kallsyms has all the data which is needed to fix this - how hard can it be to figure out that a particular function address lies within an outer function? I haven't looked...
On Thu, Mar 21, 2024 at 03:09:08PM -0700, Andrew Morton wrote: > On Thu, 21 Mar 2024 17:15:39 -0400 Kent Overstreet <kent.overstreet@linux.dev> wrote: > > > On Thu, Mar 21, 2024 at 01:31:47PM -0700, Andrew Morton wrote: > > > On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote: > > > > > > > From: Kent Overstreet <kent.overstreet@linux.dev> > > > > > > > > We're introducing alloc tagging, which tracks memory allocations by > > > > callsite. Converting alloc_inode_sb() to a macro means allocations will > > > > be tracked by its caller, which is a bit more useful. > > > > > > I'd have thought that there would be many similar > > > inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode(). > > > Do we have to go converting things to macros as people report > > > misleading or less useful results, or is there some more general > > > solution to this? > > > > No, this is just what we have to do. > > Well, this is something we strike in other contexts - kallsyms gives us > an inlined function and it's rarely what we wanted. > > I think kallsyms has all the data which is needed to fix this - how > hard can it be to figure out that a particular function address lies > within an outer function? I haven't looked... This is different, though - even if a function is inlined in multiple places there's only going to be one instance of a static var defined within that function.
On Thu, Mar 21, 2024 at 3:17 PM Kent Overstreet
<kent.overstreet@linux.dev> wrote:
>
> On Thu, Mar 21, 2024 at 03:09:08PM -0700, Andrew Morton wrote:
> > On Thu, 21 Mar 2024 17:15:39 -0400 Kent Overstreet <kent.overstreet@linux.dev> wrote:
> >
> > > On Thu, Mar 21, 2024 at 01:31:47PM -0700, Andrew Morton wrote:
> > > > On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote:
> > > >
> > > > > From: Kent Overstreet <kent.overstreet@linux.dev>
> > > > >
> > > > > We're introducing alloc tagging, which tracks memory allocations by
> > > > > callsite. Converting alloc_inode_sb() to a macro means allocations will
> > > > > be tracked by its caller, which is a bit more useful.
> > > >
> > > > I'd have thought that there would be many similar
> > > > inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode().
> > > > Do we have to go converting things to macros as people report
> > > > misleading or less useful results, or is there some more general
> > > > solution to this?
> > >
> > > No, this is just what we have to do.
> >
> > Well, this is something we strike in other contexts - kallsyms gives us
> > an inlined function and it's rarely what we wanted.
> >
> > I think kallsyms has all the data which is needed to fix this - how
> > hard can it be to figure out that a particular function address lies
> > within an outer function? I haven't looked...
>
> This is different, though - even if a function is inlined in multiple
> places there's only going to be one instance of a static var defined
> within that function.
I guess one simple way to detect the majority of these helpers would
be to filter all entries from /proc/allocinfo which originate from
header files.
~# grep ".*\.h:." /proc/allocinfo
933888 228 include/linux/mm.h:2863 func:pagetable_alloc
848 53 include/linux/mm_types.h:1175 func:mm_alloc_cid
0 0 include/linux/bpfptr.h:70 func:kvmemdup_bpfptr
0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node
0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu
0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu
0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node
0 0 include/linux/bpf.h:2249 func:bpf_map_kvcalloc
0 0 include/linux/bpf.h:2243 func:bpf_map_kzalloc
0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node
0 0 include/linux/ptr_ring.h:471
func:__ptr_ring_init_queue_alloc
0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu
0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node
0 0 include/net/tcx.h:80 func:tcx_entry_create
0 0 arch/x86/include/asm/pgalloc.h:156 func:p4d_alloc_one
487424 119 include/linux/mm.h:2863 func:pagetable_alloc
0 0 include/linux/mm.h:2863 func:pagetable_alloc
832 13 include/linux/jbd2.h:1607 func:jbd2_alloc_inode
0 0 include/linux/jbd2.h:1591 func:jbd2_alloc_handle
0 0 fs/nfs/iostat.h:51 func:nfs_alloc_iostats
0 0 include/net/netlabel.h:281 func:netlbl_secattr_cache_alloc
0 0 include/net/netlabel.h:381 func:netlbl_secattr_alloc
0 0 include/crypto/internal/acompress.h:76
func:__acomp_request_alloc
8064 84 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
1016 74 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
384 4 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
704 3 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
32 1 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
64 1 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
40 2 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
32 1 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
30000 625 include/acpi/platform/aclinuxex.h:67
func:acpi_os_acquire_object
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:67
func:acpi_os_acquire_object
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
512 1 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
192 6 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
192 3 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate
61992 861 include/acpi/platform/aclinuxex.h:67
func:acpi_os_acquire_object
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 include/acpi/platform/aclinuxex.h:67
func:acpi_os_acquire_object
0 0 include/acpi/platform/aclinuxex.h:57
func:acpi_os_allocate_zeroed
0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page
0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page
0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/hid_bpf.h:154 func:call_hid_bpf_rdesc_fixup
0 0 include/linux/skbuff.h:3392 func:__dev_alloc_pages
114688 56 include/linux/ptr_ring.h:471
func:__ptr_ring_init_queue_alloc
0 0 include/linux/skmsg.h:415 func:sk_psock_init_link
0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node
0 0 include/linux/ptr_ring.h:628 func:ptr_ring_resize_multiple
24576 3 include/linux/ptr_ring.h:471
func:__ptr_ring_init_queue_alloc
0 0 include/net/netlink.h:1896 func:nla_memdup
0 0 include/linux/sockptr.h:97 func:memdup_sockptr
0 0 include/net/request_sock.h:131 func:reqsk_alloc
0 0 include/net/tcp.h:2456 func:tcp_v4_save_options
0 0 include/net/tcp.h:2456 func:tcp_v4_save_options
0 0 include/crypto/hash.h:586 func:ahash_request_alloc
0 0 include/linux/sockptr.h:97 func:memdup_sockptr
0 0 include/linux/sockptr.h:97 func:memdup_sockptr
0 0 net/sunrpc/auth_gss/auth_gss_internal.h:38
func:simple_get_netobj
0 0 include/crypto/hash.h:586 func:ahash_request_alloc
0 0 include/net/netlink.h:1896 func:nla_memdup
0 0 include/crypto/skcipher.h:869 func:skcipher_request_alloc
0 0 include/net/fq_impl.h:361 func:fq_init
0 0 include/net/netlabel.h:316 func:netlbl_catmap_alloc
and it finds our example:
832 13 include/linux/jbd2.h:1607 func:jbd2_alloc_inode
Interestingly the inlined functions which are called from multiple
places will have multiple entries with the same file+line:
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
0 0 include/linux/dma-fence-chain.h:91
func:dma_fence_chain_alloc
So, duplicate entries can be also used as an indication of an inlined allocator.
I'll go chase these down and will post a separate patch converting them.
On Thu, Mar 21, 2024 at 3:47 PM Suren Baghdasaryan <surenb@google.com> wrote: > > On Thu, Mar 21, 2024 at 3:17 PM Kent Overstreet > <kent.overstreet@linux.dev> wrote: > > > > On Thu, Mar 21, 2024 at 03:09:08PM -0700, Andrew Morton wrote: > > > On Thu, 21 Mar 2024 17:15:39 -0400 Kent Overstreet <kent.overstreet@linux.dev> wrote: > > > > > > > On Thu, Mar 21, 2024 at 01:31:47PM -0700, Andrew Morton wrote: > > > > > On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote: > > > > > > > > > > > From: Kent Overstreet <kent.overstreet@linux.dev> > > > > > > > > > > > > We're introducing alloc tagging, which tracks memory allocations by > > > > > > callsite. Converting alloc_inode_sb() to a macro means allocations will > > > > > > be tracked by its caller, which is a bit more useful. > > > > > > > > > > I'd have thought that there would be many similar > > > > > inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode(). > > > > > Do we have to go converting things to macros as people report > > > > > misleading or less useful results, or is there some more general > > > > > solution to this? > > > > > > > > No, this is just what we have to do. > > > > > > Well, this is something we strike in other contexts - kallsyms gives us > > > an inlined function and it's rarely what we wanted. > > > > > > I think kallsyms has all the data which is needed to fix this - how > > > hard can it be to figure out that a particular function address lies > > > within an outer function? I haven't looked... > > > > This is different, though - even if a function is inlined in multiple > > places there's only going to be one instance of a static var defined > > within that function. > > I guess one simple way to detect the majority of these helpers would > be to filter all entries from /proc/allocinfo which originate from > header files. > > ~# grep ".*\.h:." /proc/allocinfo > 933888 228 include/linux/mm.h:2863 func:pagetable_alloc > 848 53 include/linux/mm_types.h:1175 func:mm_alloc_cid > 0 0 include/linux/bpfptr.h:70 func:kvmemdup_bpfptr > 0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node > 0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu > 0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu > 0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node > 0 0 include/linux/bpf.h:2249 func:bpf_map_kvcalloc > 0 0 include/linux/bpf.h:2243 func:bpf_map_kzalloc > 0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node > 0 0 include/linux/ptr_ring.h:471 > func:__ptr_ring_init_queue_alloc > 0 0 include/linux/bpf.h:2256 func:bpf_map_alloc_percpu > 0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node > 0 0 include/net/tcx.h:80 func:tcx_entry_create > 0 0 arch/x86/include/asm/pgalloc.h:156 func:p4d_alloc_one > 487424 119 include/linux/mm.h:2863 func:pagetable_alloc > 0 0 include/linux/mm.h:2863 func:pagetable_alloc > 832 13 include/linux/jbd2.h:1607 func:jbd2_alloc_inode > 0 0 include/linux/jbd2.h:1591 func:jbd2_alloc_handle > 0 0 fs/nfs/iostat.h:51 func:nfs_alloc_iostats > 0 0 include/net/netlabel.h:281 func:netlbl_secattr_cache_alloc > 0 0 include/net/netlabel.h:381 func:netlbl_secattr_alloc > 0 0 include/crypto/internal/acompress.h:76 > func:__acomp_request_alloc > 8064 84 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 1016 74 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 384 4 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 704 3 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 32 1 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 64 1 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 40 2 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 32 1 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 30000 625 include/acpi/platform/aclinuxex.h:67 > func:acpi_os_acquire_object > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:67 > func:acpi_os_acquire_object > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 512 1 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 192 6 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 192 3 include/acpi/platform/aclinuxex.h:52 func:acpi_os_allocate > 61992 861 include/acpi/platform/aclinuxex.h:67 > func:acpi_os_acquire_object > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 include/acpi/platform/aclinuxex.h:67 > func:acpi_os_acquire_object > 0 0 include/acpi/platform/aclinuxex.h:57 > func:acpi_os_allocate_zeroed > 0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page > 0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page > 0 0 drivers/iommu/amd/amd_iommu.h:141 func:alloc_pgtable_page > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/hid_bpf.h:154 func:call_hid_bpf_rdesc_fixup > 0 0 include/linux/skbuff.h:3392 func:__dev_alloc_pages > 114688 56 include/linux/ptr_ring.h:471 > func:__ptr_ring_init_queue_alloc > 0 0 include/linux/skmsg.h:415 func:sk_psock_init_link > 0 0 include/linux/bpf.h:2237 func:bpf_map_kmalloc_node > 0 0 include/linux/ptr_ring.h:628 func:ptr_ring_resize_multiple > 24576 3 include/linux/ptr_ring.h:471 > func:__ptr_ring_init_queue_alloc > 0 0 include/net/netlink.h:1896 func:nla_memdup > 0 0 include/linux/sockptr.h:97 func:memdup_sockptr > 0 0 include/net/request_sock.h:131 func:reqsk_alloc > 0 0 include/net/tcp.h:2456 func:tcp_v4_save_options > 0 0 include/net/tcp.h:2456 func:tcp_v4_save_options > 0 0 include/crypto/hash.h:586 func:ahash_request_alloc > 0 0 include/linux/sockptr.h:97 func:memdup_sockptr > 0 0 include/linux/sockptr.h:97 func:memdup_sockptr > 0 0 net/sunrpc/auth_gss/auth_gss_internal.h:38 > func:simple_get_netobj > 0 0 include/crypto/hash.h:586 func:ahash_request_alloc > 0 0 include/net/netlink.h:1896 func:nla_memdup > 0 0 include/crypto/skcipher.h:869 func:skcipher_request_alloc > 0 0 include/net/fq_impl.h:361 func:fq_init > 0 0 include/net/netlabel.h:316 func:netlbl_catmap_alloc > > and it finds our example: > > 832 13 include/linux/jbd2.h:1607 func:jbd2_alloc_inode > > Interestingly the inlined functions which are called from multiple > places will have multiple entries with the same file+line: > > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > 0 0 include/linux/dma-fence-chain.h:91 > func:dma_fence_chain_alloc > > So, duplicate entries can be also used as an indication of an inlined allocator. > I'll go chase these down and will post a separate patch converting them. I just posted https://lore.kernel.org/all/20240404165404.3805498-1-surenb@google.com/ to report allocations done from the inlined functions in the headers to their callers.
On Thu, Mar 21, 2024 at 1:31 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Thu, 21 Mar 2024 09:36:27 -0700 Suren Baghdasaryan <surenb@google.com> wrote:
>
> > From: Kent Overstreet <kent.overstreet@linux.dev>
> >
> > We're introducing alloc tagging, which tracks memory allocations by
> > callsite. Converting alloc_inode_sb() to a macro means allocations will
> > be tracked by its caller, which is a bit more useful.
>
> I'd have thought that there would be many similar
> inlines-which-allocate-memory. Such as, I dunno, jbd2_alloc_inode().
> Do we have to go converting things to macros as people report
> misleading or less useful results, or is there some more general
> solution to this?
Yeah, that's unfortunately inevitable. Even if we had compiler support
we would have to add annotations for such inlined functions.
For the given example of jbd2_alloc_inode() it's not so bad since it's
used only from one location but in general yes, that's something we
will have to improve as we find more such cases.
>
> > --- a/include/linux/fs.h
> > +++ b/include/linux/fs.h
> > @@ -3083,11 +3083,7 @@ int setattr_should_drop_sgid(struct mnt_idmap *idmap,
> > * This must be used for allocating filesystems specific inodes to set
> > * up the inode reclaim context correctly.
> > */
> > -static inline void *
> > -alloc_inode_sb(struct super_block *sb, struct kmem_cache *cache, gfp_t gfp)
> > -{
> > - return kmem_cache_alloc_lru(cache, &sb->s_inode_lru, gfp);
> > -}
> > +#define alloc_inode_sb(_sb, _cache, _gfp) kmem_cache_alloc_lru(_cache, &_sb->s_inode_lru, _gfp)
>
> Parenthesizing __sb seems sensible here?
Ack.
Let's wait for more comments and then I'll post fixes.
Thanks!
© 2016 - 2026 Red Hat, Inc.