From: Liam Howlett <liam.howlett@oracle.com>
To: Vasily Gorbik <gor@linux.ibm.com>
Cc: "maple-tree@lists.infradead.org" <maple-tree@lists.infradead.org>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
Andrew Morton <akpm@linux-foundation.org>,
Heiko Carstens <hca@linux.ibm.com>,
Alexander Gordeev <agordeev@linux.ibm.com>
Subject: Re: [PATCH v6 08/71] Maple Tree: Add new data structure
Date: Mon, 28 Feb 2022 14:36:40 +0000 [thread overview]
Message-ID: <20220228143633.r4zoemgtmrq4uzvb@revolver> (raw)
In-Reply-To: <your-ad-here.call-01645924312-ext-0398@work.hours>
* Vasily Gorbik <gor@linux.ibm.com> [220226 20:12]:
> Hi Liam,
>
> there is an endianness issue with maple_metadata. This is broken on
> all big endian architectures. Tests are crashing. See potential fixup
> below. Feel free to apply it or fix the issue in your own way. This does
> not resolve all the issues with the patch series though.
The endianness isn't what's causing the issues. The slots hold a
pointer. Either we can use the entire slot or none of the slot
space. You are just moving the metatdata around in the last slot.
>
> With current definition of struct maple_range_64 on big endian systems
> metadata end and gap fields are aligned with the most significant bytes
> of slot[15], rather than least significant.
>
> (gdb) ptype /o struct maple_range_64
> /* offset | size */ type = struct maple_range_64 {
> /* 0 | 8 */ struct maple_pnode *parent;
> /* 8 | 120 */ unsigned long pivot[15];
> /* 128 | 128 */ union {
> /* 128 */ void *slot[16];
> /* 128 */ struct {
> /* 128 | 120 */ void *pad[15];
> /* 248 | 2 */ struct maple_metadata {
> /* 248 | 1 */ unsigned char end;
> /* 249 | 1 */ unsigned char gap;
>
> /* total size (bytes): 2 */
> } meta;
> /* XXX 6-byte padding */
>
> /* total size (bytes): 128 */
> };
>
> /* total size (bytes): 128 */
> };
>
> /* total size (bytes): 256 */
> }
>
You may have been confused about my comment, which I believe is
outdated, that talks about maple_arange_64, not maple_range_64. I added
maple_range_64 metadata and use the previous pivot to see if the slot
contains data or not. If piv[14] == 0 or mas->max means we can use the
slot for metadata.
> Assuming we don't want to end up with smth like this in a code which
> otherwise relies on shifts and endianness independent.
>
> #if defined(__BYTE_ORDER) ? __BYTE_ORDER == __LITTLE_ENDIAN : defined(__LITTLE_ENDIAN)
>
> struct maple_metadata {
> unsigned char end;
> unsigned char gap;
> };
>
> #else
>
> struct maple_metadata {
> unsigned char unused[6];
> unsigned char gap;
> unsigned char end;
> };
>
> #endif
>
> rewrite node matadata access in endianness independent shifts based
> approach as well.
>
> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
> ---
> include/linux/maple_tree.h | 33 ++++++++++++++--------------
> lib/maple_tree.c | 45 +++++++++++++++++++++-----------------
> 2 files changed, 41 insertions(+), 37 deletions(-)
>
> diff --git a/include/linux/maple_tree.h b/include/linux/maple_tree.h
> index e808794e06a5..e0c9f356ba4e 100644
> --- a/include/linux/maple_tree.h
> +++ b/include/linux/maple_tree.h
> @@ -76,21 +76,6 @@ typedef struct maple_enode *maple_enode; /* encoded node */
> typedef struct maple_pnode *maple_pnode; /* parent node */
>
>
> -/*
> - * The node->meta is currently only supported in allocation range 64 (arange_64)
> - * node type. As a result of tracking gaps, there is a small area that is not
> - * used for data storage in this node type. This area is reused to store
> - * metadata related to the node itself including the data end and the largest
> - * gap location. This metadata is used to optimize the gap updating code and in
> - * reverse searching for gaps or any other code that needs to find the end of
> - * the data.
> - */
> -struct maple_metadata {
> - unsigned char end;
> - unsigned char gap;
> -
> -};
> -
> /*
> * Leaf nodes do not store pointers to nodes, they store user data. Users may
> * store almost any bit pattern. As noted above, the optimisation of storing an
> @@ -110,8 +95,22 @@ struct maple_metadata {
> * subtree with an entry attached to the value whereas keys are unique to a
> * specific position of a B-tree. Pivot values are inclusive of the slot with
> * the same index.
> + *
> + * The node->meta is currently only supported in allocation range 64 (arange_64)
> + * node type. As a result of tracking gaps, there is a small area that is not
> + * used for data storage in this node type. This area is reused to store
> + * metadata related to the node itself including the data end and the largest
> + * gap location. This metadata is used to optimize the gap updating code and in
> + * reverse searching for gaps or any other code that needs to find the end of
> + * the data.
> */
>
> +#define MN_META_MASK 0xFFFF
> +#define MN_META_GAP_MASK 0xFF00
> +#define MN_META_END_MASK 0x00FF
> +#define MN_META_GAP_SHIFT 8
> +#define MN_META_END_SHIFT 0
> +
> struct maple_range_64 {
> struct maple_pnode *parent;
> unsigned long pivot[MAPLE_RANGE64_SLOTS - 1];
> @@ -119,7 +118,7 @@ struct maple_range_64 {
> void __rcu *slot[MAPLE_RANGE64_SLOTS];
> struct {
> void __rcu *pad[MAPLE_RANGE64_SLOTS - 1];
> - struct maple_metadata meta;
> + unsigned long meta;
> };
> };
> };
> @@ -138,7 +137,7 @@ struct maple_arange_64 {
> unsigned long pivot[MAPLE_ARANGE64_SLOTS - 1];
> void __rcu *slot[MAPLE_ARANGE64_SLOTS];
> unsigned long gap[MAPLE_ARANGE64_SLOTS];
> - struct maple_metadata meta;
> + unsigned long meta;
> };
>
> struct maple_alloc {
> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
> index 7ebb34964c68..02be5a5314de 100644
> --- a/lib/maple_tree.c
> +++ b/lib/maple_tree.c
> @@ -868,15 +868,22 @@ static inline void *mas_root_locked(struct ma_state *mas)
> return mt_root_locked(mas->tree);
> }
>
> -static inline struct maple_metadata *ma_meta(struct maple_node *mn,
> - enum maple_type mt)
> +static inline unsigned long ma_get_meta_raw(struct maple_node *mn,
> + enum maple_type mt)
> {
> - switch (mt) {
> - case maple_arange_64:
> - return &mn->ma64.meta;
> - default:
> - return &mn->mr64.meta;
> - }
> + if (mt == maple_arange_64)
> + return mn->ma64.meta;
> + else
> + return mn->mr64.meta;
> +}
> +
> +static inline void ma_set_meta_raw(struct maple_node *mn, enum maple_type mt,
> + unsigned long meta)
> +{
> + if (mt == maple_arange_64)
> + mn->ma64.meta = meta;
> + else
> + mn->mr64.meta = meta;
> }
>
> /*
> @@ -889,10 +896,10 @@ static inline struct maple_metadata *ma_meta(struct maple_node *mn,
> static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt,
> unsigned char offset, unsigned char end)
> {
> - struct maple_metadata *meta = ma_meta(mn, mt);
> + unsigned long mnm = ma_get_meta_raw(mn, mt) & ~MN_META_MASK;
>
> - meta->gap = offset;
> - meta->end = end;
> + mnm |= offset << MN_META_GAP_SHIFT | end << MN_META_END_SHIFT;
> + ma_set_meta_raw(mn, mt, mnm);
> }
>
> /*
> @@ -903,9 +910,7 @@ static inline void ma_set_meta(struct maple_node *mn, enum maple_type mt,
> static inline unsigned char ma_meta_end(struct maple_node *mn,
> enum maple_type mt)
> {
> - struct maple_metadata *meta = ma_meta(mn, mt);
> -
> - return meta->end;
> + return (ma_get_meta_raw(mn, mt) & MN_META_END_MASK) >> MN_META_END_SHIFT;
> }
>
> /*
> @@ -916,8 +921,7 @@ static inline unsigned char ma_meta_end(struct maple_node *mn,
> static inline unsigned char ma_meta_gap(struct maple_node *mn,
> enum maple_type mt)
> {
> -
> - return mn->ma64.meta.gap;
> + return (ma_get_meta_raw(mn, mt) & MN_META_GAP_MASK) >> MN_META_GAP_SHIFT;
> }
>
> /*
> @@ -929,10 +933,9 @@ static inline unsigned char ma_meta_gap(struct maple_node *mn,
> static inline void ma_set_meta_gap(struct maple_node *mn, enum maple_type mt,
> unsigned char offset)
> {
> + unsigned long mnm = ma_get_meta_raw(mn, mt) & ~MN_META_GAP_MASK;
>
> - struct maple_metadata *meta = ma_meta(mn, mt);
> -
> - meta->gap = offset;
> + ma_set_meta_raw(mn, mt, mnm | offset << MN_META_GAP_SHIFT);
> }
>
> /*
> @@ -6590,7 +6593,9 @@ void mt_dump_arange64(const struct maple_tree *mt, void *entry,
> pr_cont(" contents: ");
> for (i = 0; i < MAPLE_ARANGE64_SLOTS; i++)
> pr_cont("%lu ", node->gap[i]);
> - pr_cont("| %02X %02X| ", node->meta.end, node->meta.gap);
> + pr_cont("| %02lX %02lX| ",
> + (node->meta & MN_META_END_MASK) >> MN_META_END_SHIFT,
> + (node->meta & MN_META_GAP_MASK) >> MN_META_GAP_SHIFT);
> for (i = 0; i < MAPLE_ARANGE64_SLOTS - 1; i++)
> pr_cont(MA_PTR" %lu ", node->slot[i], node->pivot[i]);
> pr_cont(MA_PTR"\n", node->slot[i]);
> --
> 2.35.1
>
>
next prev parent reply other threads:[~2022-02-28 14:36 UTC|newest]
Thread overview: 100+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-02-15 14:37 [PATCH v6 00/71] Introducing the Maple Tree Liam Howlett
2022-02-15 14:42 ` [PATCH v6 01/71] binfmt_elf: Take the mmap lock when walking the VMA list Liam Howlett
2022-02-15 14:42 ` [PATCH v6 02/71] xarray: Fix bitmap breakage Liam Howlett
2022-02-15 14:42 ` [PATCH v6 03/71] radix tree test suite: Add pr_err define Liam Howlett
2022-02-15 14:42 ` [PATCH v6 05/71] radix tree test suite: Add allocation counts and size to kmem_cache Liam Howlett
2022-02-15 14:42 ` [PATCH v6 04/71] radix tree test suite: Add kmem_cache_set_non_kernel() Liam Howlett
2022-02-15 14:42 ` [PATCH v6 06/71] radix tree test suite: Add support for slab bulk APIs Liam Howlett
2022-02-15 14:42 ` [PATCH v6 07/71] radix tree test suite: Add lockdep_is_held to header Liam Howlett
2022-02-15 14:43 ` [PATCH v6 08/71] Maple Tree: Add new data structure Liam Howlett
2022-02-16 10:11 ` Mark Hemment
2022-02-16 18:25 ` Liam Howlett
2022-02-27 1:11 ` Vasily Gorbik
2022-02-27 12:46 ` Vasily Gorbik
2022-02-28 14:36 ` Liam Howlett [this message]
2022-03-01 2:01 ` Vasily Gorbik
2022-03-01 20:39 ` Liam Howlett
2022-03-01 22:50 ` Vasily Gorbik
2022-03-01 22:56 ` Vasily Gorbik
2022-03-02 14:08 ` Liam Howlett
2022-02-15 14:43 ` [PATCH v6 09/71] lib/test_maple_tree: Add testing for maple tree Liam Howlett
2022-02-15 14:43 ` [PATCH v6 10/71] mm: Start tracking VMAs with " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 11/71] mm: Add VMA iterator Liam Howlett
2022-02-16 10:50 ` Mark Hemment
2022-02-16 18:32 ` Liam Howlett
2022-02-15 14:43 ` [PATCH v6 12/71] mmap: Use the VMA iterator in count_vma_pages_range() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 13/71] mm/mmap: Use the maple tree in find_vma() instead of the rbtree Liam Howlett
2022-02-15 14:43 ` [PATCH v6 16/71] kernel/fork: Use maple tree for dup_mmap() during forking Liam Howlett
2022-02-15 14:43 ` [PATCH v6 14/71] mm/mmap: Use the maple tree for find_vma_prev() instead of the rbtree Liam Howlett
2022-02-15 14:43 ` [PATCH v6 15/71] mm/mmap: Use maple tree for unmapped_area{_topdown} Liam Howlett
2022-02-15 14:43 ` [PATCH v6 18/71] proc: Remove VMA rbtree use from nommu Liam Howlett
2022-02-15 14:43 ` [PATCH v6 17/71] damon: Convert __damon_va_three_regions to use the VMA iterator Liam Howlett
2022-02-15 14:43 ` [PATCH v6 19/71] mm: Remove rb tree Liam Howlett
2022-02-15 14:43 ` [PATCH v6 20/71] mmap: Change zeroing of maple tree in __vma_adjust() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 22/71] mm: Optimize find_exact_vma() to use vma_lookup() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 21/71] xen: Use vma_lookup() in privcmd_ioctl_mmap() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 23/71] mm/khugepaged: Optimize collapse_pte_mapped_thp() by using vma_lookup() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 25/71] mm: Use maple tree operations for find_vma_intersection() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 24/71] mm/mmap: Change do_brk_flags() to expand existing VMA and add do_brk_munmap() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 26/71] mm/mmap: Use advanced maple tree API for mmap_region() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 27/71] mm: Remove vmacache Liam Howlett
2022-02-15 14:43 ` [PATCH v6 28/71] mm: Convert vma_lookup() to use mtree_load() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 29/71] mm/mmap: Move mmap_region() below do_munmap() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 30/71] mm/mmap: Reorganize munmap to use maple states Liam Howlett
2022-02-15 14:43 ` [PATCH v6 32/71] arm64: Remove mmap linked list from vdso Liam Howlett
2022-02-15 14:43 ` [PATCH v6 31/71] mm/mmap: Change do_brk_munmap() to use do_mas_align_munmap() Liam Howlett
2022-02-15 14:43 ` [PATCH v6 33/71] parisc: Remove mmap linked list from cache handling Liam Howlett
2022-02-15 14:43 ` [PATCH v6 34/71] powerpc: Remove mmap linked list walks Liam Howlett
2022-02-15 14:43 ` [PATCH v6 35/71] s390: Remove vma " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 38/71] cxl: Remove vma linked list walk Liam Howlett
2022-02-15 14:43 ` [PATCH v6 36/71] x86: Remove vma linked list walks Liam Howlett
2022-02-15 14:43 ` [PATCH v6 37/71] xtensa: " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 39/71] optee: Remove vma linked list walk Liam Howlett
2022-02-15 14:43 ` [PATCH v6 40/71] um: " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 42/71] coredump: " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 43/71] exec: Use VMA iterator instead of linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 41/71] binfmt_elf: Remove vma linked list walk Liam Howlett
2022-02-15 14:43 ` [PATCH v6 45/71] fs/proc/task_mmu: Stop using linked list and highest_vm_end Liam Howlett
2022-02-15 14:43 ` [PATCH v6 44/71] fs/proc/base: Use maple tree iterators in place of linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 47/71] ipc/shm: Use VMA iterator instead " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 46/71] userfaultfd: Use maple tree iterator to iterate VMAs Liam Howlett
2022-02-15 14:43 ` [PATCH v6 49/71] perf: Use VMA iterator Liam Howlett
2022-02-15 14:43 ` [PATCH v6 50/71] sched: Use maple tree iterator to walk VMAs Liam Howlett
2022-02-15 14:43 ` [PATCH v6 48/71] acct: Use VMA iterator instead of linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 51/71] fork: Use VMA iterator Liam Howlett
2022-02-15 14:43 ` [PATCH v6 52/71] bpf: Remove VMA linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 53/71] mm/gup: Use maple tree navigation instead of " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 54/71] mm/khugepaged: Stop using vma " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 55/71] mm/ksm: Use vma iterators instead of " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 58/71] mm/mempolicy: Use vma iterator & maple state " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 57/71] mm/memcontrol: Stop using mm->highest_vm_end Liam Howlett
2022-02-15 14:43 ` [PATCH v6 56/71] mm/madvise: Use vma_find() instead of vma linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 59/71] mm/mlock: Use vma iterator and " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 61/71] mm/mremap: Use vma_find_intersection() " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 60/71] mm/mprotect: Use maple tree navigation " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 63/71] mm/oom_kill: Use maple tree iterators " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 62/71] mm/msync: Use vma_find() " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 64/71] mm/pagewalk: " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 65/71] mm/swapfile: Use vma iterator " Liam Howlett
2022-02-15 14:43 ` [PATCH v6 66/71] i915: Use the VMA iterator Liam Howlett
2022-02-15 14:43 ` [PATCH v6 67/71] nommu: Remove uses of VMA linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 68/71] riscv: Use vma iterator for vdso Liam Howlett
2022-02-15 14:43 ` [PATCH v6 69/71] mm: Remove the vma linked list Liam Howlett
2022-02-15 14:43 ` [PATCH v6 70/71] mm/mmap: Drop range_has_overlap() function Liam Howlett
2022-02-15 14:43 ` [PATCH v6 71/71] mm/mmap.c: Pass in mapping to __vma_link_file() Liam Howlett
2022-02-16 19:47 ` [PATCH v6 00/71] Introducing the Maple Tree Andrew Morton
2022-02-16 20:24 ` Matthew Wilcox
2022-02-23 16:35 ` Mel Gorman
2022-02-23 16:45 ` Matthew Wilcox
2022-02-25 3:49 ` Qian Cai
2022-02-25 19:08 ` Liam Howlett
2022-02-25 20:23 ` Liam Howlett
2022-02-25 20:46 ` Qian Cai
2022-02-25 23:00 ` Nathan Chancellor
2022-02-26 1:58 ` Liam Howlett
2022-02-26 23:19 ` Hugh Dickins
2022-02-27 18:32 ` Hugh Dickins
2022-02-28 14:26 ` Liam Howlett
2022-02-28 11:56 ` Qian Cai
2022-02-27 2:22 ` Vasily Gorbik
2022-02-28 14:56 ` Liam Howlett
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220228143633.r4zoemgtmrq4uzvb@revolver \
--to=liam.howlett@oracle.com \
--cc=agordeev@linux.ibm.com \
--cc=akpm@linux-foundation.org \
--cc=gor@linux.ibm.com \
--cc=hca@linux.ibm.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=maple-tree@lists.infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox