2020-03-24 16:13:32

by Peter Zijlstra

Subject: [PATCH v3 16/26] objtool: Optimize find_rela_by_dest_range()

Perf shows there is significant time in find_rela_by_dest(); this is
because we have to iterate the address space per byte, looking for
relocation entries.

Optimize this by reducing the address space granularity.

This reduces objtool on vmlinux.o runtime from 4.8 to 4.4 seconds.

Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Acked-by: Josh Poimboeuf <[email protected]>
---
tools/objtool/elf.c | 15 +++++++++++----
tools/objtool/elf.h | 16 +++++++++++++++-
2 files changed, 26 insertions(+), 5 deletions(-)

--- a/tools/objtool/elf.c
+++ b/tools/objtool/elf.c
@@ -215,7 +215,7 @@ struct symbol *find_symbol_by_name(struc
struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
unsigned long offset, unsigned int len)
{
- struct rela *rela;
+ struct rela *rela, *r = NULL;
unsigned long o;

if (!sec->rela)
@@ -223,12 +223,19 @@ struct rela *find_rela_by_dest_range(str

sec = sec->rela;

- for (o = offset; o < offset + len; o++) {
+ for_offset_range(o, offset, offset + len) {
hash_for_each_possible(elf->rela_hash, rela, hash,
sec_offset_hash(sec, o)) {
- if (rela->sec == sec && rela->offset == o)
- return rela;
+ if (rela->sec != sec)
+ continue;
+
+ if (rela->offset >= offset && rela->offset < offset + len) {
+ if (!r || rela->offset < r->offset)
+ r = rela;
+ }
}
+ if (r)
+ return r;
}

return NULL;
--- a/tools/objtool/elf.h
+++ b/tools/objtool/elf.h
@@ -83,9 +83,23 @@ struct elf {
DECLARE_HASHTABLE(rela_hash, 20);
};

+#define OFFSET_STRIDE_BITS 4
+#define OFFSET_STRIDE (1UL << OFFSET_STRIDE_BITS)
+#define OFFSET_STRIDE_MASK (~(OFFSET_STRIDE - 1))
+
+#define for_offset_range(_offset, _start, _end) \
+ for (_offset = ((_start) & OFFSET_STRIDE_MASK); \
+ _offset <= ((_end) & OFFSET_STRIDE_MASK); \
+ _offset += OFFSET_STRIDE)
+
static inline u32 sec_offset_hash(struct section *sec, unsigned long offset)
{
- u32 ol = offset, oh = offset >> 32, idx = sec->idx;
+ u32 ol, oh, idx = sec->idx;
+
+ offset &= OFFSET_STRIDE_MASK;
+
+ ol = offset;
+ oh = offset >> 32;

__jhash_mix(ol, oh, idx);

2020-03-25 12:20:40

by Miroslav Benes

Subject: Re: [PATCH v3 16/26] objtool: Optimize find_rela_by_dest_range()

On Tue, 24 Mar 2020, Peter Zijlstra wrote:

> Perf shows there is significant time in find_rela_by_dest(); this is
> because we have to iterate the address space per byte, looking for
> relocation entries.
>
> Optimize this by reducing the address space granularity.
>
> This reduces objtool on vmlinux.o runtime from 4.8 to 4.4 seconds.
>
> Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
> Acked-by: Josh Poimboeuf <[email protected]>

Reviewed-by: Miroslav Benes <[email protected]>

M

2020-03-26 10:11:05

by tip-bot2 for Alexey Makhalov

Subject: [tip: core/objtool] objtool: Optimize find_rela_by_dest_range()

The following commit has been merged into the core/objtool branch of tip:

Commit-ID: 74b873e49d92f90deb41d1a2a8fbb70328aebd67
Gitweb: https://git.kernel.org/tip/74b873e49d92f90deb41d1a2a8fbb70328aebd67
Author: Peter Zijlstra <[email protected]>
AuthorDate: Thu, 12 Mar 2020 11:30:50 +01:00
Committer: Peter Zijlstra <[email protected]>
CommitterDate: Wed, 25 Mar 2020 18:28:31 +01:00

objtool: Optimize find_rela_by_dest_range()

Perf shows there is significant time in find_rela_by_dest(); this is
because we have to iterate the address space per byte, looking for
relocation entries.

Optimize this by reducing the address space granularity.

This reduces objtool on vmlinux.o runtime from 4.8 to 4.4 seconds.

Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Reviewed-by: Miroslav Benes <[email protected]>
Acked-by: Josh Poimboeuf <[email protected]>
Link: https://lkml.kernel.org/r/[email protected]
---
tools/objtool/elf.c | 15 +++++++++++----
tools/objtool/elf.h | 16 +++++++++++++++-
2 files changed, 26 insertions(+), 5 deletions(-)

diff --git a/tools/objtool/elf.c b/tools/objtool/elf.c
index 8a0a1bc..09ddc8f 100644
--- a/tools/objtool/elf.c
+++ b/tools/objtool/elf.c
@@ -215,7 +215,7 @@ struct symbol *find_symbol_by_name(struct elf *elf, const char *name)
struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,
unsigned long offset, unsigned int len)
{
- struct rela *rela;
+ struct rela *rela, *r = NULL;
unsigned long o;

if (!sec->rela)
@@ -223,12 +223,19 @@ struct rela *find_rela_by_dest_range(struct elf *elf, struct section *sec,

sec = sec->rela;

- for (o = offset; o < offset + len; o++) {
+ for_offset_range(o, offset, offset + len) {
hash_for_each_possible(elf->rela_hash, rela, hash,
sec_offset_hash(sec, o)) {
- if (rela->sec == sec && rela->offset == o)
- return rela;
+ if (rela->sec != sec)
+ continue;
+
+ if (rela->offset >= offset && rela->offset < offset + len) {
+ if (!r || rela->offset < r->offset)
+ r = rela;
+ }
}
+ if (r)
+ return r;
}

return NULL;
diff --git a/tools/objtool/elf.h b/tools/objtool/elf.h
index dfd2431..ebbb10c 100644
--- a/tools/objtool/elf.h
+++ b/tools/objtool/elf.h
@@ -83,9 +83,23 @@ struct elf {
DECLARE_HASHTABLE(rela_hash, 20);
};

+#define OFFSET_STRIDE_BITS 4
+#define OFFSET_STRIDE (1UL << OFFSET_STRIDE_BITS)
+#define OFFSET_STRIDE_MASK (~(OFFSET_STRIDE - 1))
+
+#define for_offset_range(_offset, _start, _end) \
+ for (_offset = ((_start) & OFFSET_STRIDE_MASK); \
+ _offset <= ((_end) & OFFSET_STRIDE_MASK); \
+ _offset += OFFSET_STRIDE)
+
static inline u32 sec_offset_hash(struct section *sec, unsigned long offset)
{
- u32 ol = offset, oh = offset >> 32, idx = sec->idx;
+ u32 ol, oh, idx = sec->idx;
+
+ offset &= OFFSET_STRIDE_MASK;
+
+ ol = offset;
+ oh = offset >> 32;

__jhash_mix(ol, oh, idx);