Move building of fast lookup table from boot to sorttable tool. This saves us
6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
Huaixin Chang (2):
scripts/sorttable: Build orc fast lookup table via sorttable tool
x86/unwind/orc: Remove unwind_init() from x86 boot
arch/x86/include/asm/unwind.h | 2 -
arch/x86/kernel/setup.c | 2 -
arch/x86/kernel/unwind_orc.c | 51 ----------------------
scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++++++---
4 files changed, 92 insertions(+), 62 deletions(-)
--
2.14.4.44.g2045bb6
On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
> Move building of fast lookup table from boot to sorttable tool. This saves us
> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
And what does it add to the build time?
> On Apr 29, 2020, at 4:49 PM, Peter Zijlstra <[email protected]> wrote:
>
> On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
>> Move building of fast lookup table from boot to sorttable tool. This saves us
>> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
>
> And what does it add to the build time?
It takes a little more than 7ms to build fast lookup table in sorttable on the same CPU. And it is on the critical path.
On Thu, Apr 30, 2020 at 10:32:17AM +0800, changhuaixin wrote:
>
>
> > On Apr 29, 2020, at 4:49 PM, Peter Zijlstra <[email protected]> wrote:
> >
> > On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
> >> Move building of fast lookup table from boot to sorttable tool. This saves us
> >> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
> >
> > And what does it add to the build time?
>
> It takes a little more than 7ms to build fast lookup table in
> sorttable on the same CPU. And it is on the critical path.
Thanks, I like it. It will help make the in-kernel unwinder even
simpler. And it will enable unwinding from early boot.
Maybe someday we can move all the table sorting code into objtool, once
we have objtool running on vmlinux.o.
I'll try to review the patches soon.
--
Josh
On Wed, Apr 29, 2020 at 11:06:58PM -0500, Josh Poimboeuf wrote:
> On Thu, Apr 30, 2020 at 10:32:17AM +0800, changhuaixin wrote:
> >
> >
> > > On Apr 29, 2020, at 4:49 PM, Peter Zijlstra <[email protected]> wrote:
> > >
> > > On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
> > >> Move building of fast lookup table from boot to sorttable tool. This saves us
> > >> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
> > >
> > > And what does it add to the build time?
> >
> > It takes a little more than 7ms to build fast lookup table in
> > sorttable on the same CPU. And it is on the critical path.
>
> Thanks, I like it. It will help make the in-kernel unwinder even
> simpler. And it will enable unwinding from early boot.
>
> Maybe someday we can move all the table sorting code into objtool, once
> we have objtool running on vmlinux.o.
>
> I'll try to review the patches soon.
BTW, another cool feature would be for sorttable to run on modules
during the module linking phase.
--
Josh
On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
> Move building of fast lookup table from boot to sorttable tool. This saves us
> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
>
> Huaixin Chang (2):
> scripts/sorttable: Build orc fast lookup table via sorttable tool
> x86/unwind/orc: Remove unwind_init() from x86 boot
>
> arch/x86/include/asm/unwind.h | 2 -
> arch/x86/kernel/setup.c | 2 -
> arch/x86/kernel/unwind_orc.c | 51 ----------------------
> scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++++++---
> 4 files changed, 92 insertions(+), 62 deletions(-)
I tested this (rebased on tip/master), it seems to break ORC
completely... e.g. /proc/self/stack is empty.
--
Josh
Thanks for your kindly reply. Let me have a check.
> On May 23, 2020, at 2:28 AM, Josh Poimboeuf <[email protected]> wrote:
>
> On Wed, Apr 29, 2020 at 02:46:24PM +0800, Huaixin Chang wrote:
>> Move building of fast lookup table from boot to sorttable tool. This saves us
>> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
>>
>> Huaixin Chang (2):
>> scripts/sorttable: Build orc fast lookup table via sorttable tool
>> x86/unwind/orc: Remove unwind_init() from x86 boot
>>
>> arch/x86/include/asm/unwind.h | 2 -
>> arch/x86/kernel/setup.c | 2 -
>> arch/x86/kernel/unwind_orc.c | 51 ----------------------
>> scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++++++---
>> 4 files changed, 92 insertions(+), 62 deletions(-)
>
> I tested this (rebased on tip/master), it seems to break ORC
> completely... e.g. /proc/self/stack is empty.
>
> --
> Josh
On Sun, May 31, 2020 at 01:26:54PM +0800, changhuaixin wrote:
> It turned out to be an alignment problem. If sh_size of previous section
> orc_unwind is not 4-byte aligned, sh_offset of the following orc_lookup
> section is not 4-byte aligned too. However, the VMA of section orc_lookup
> is aligned to the nearest 4-byte. Thus, the orc_lookup section means two
> different ares for scripts/sorttable tool and kernel.
>
> Sections headers look like this when it happens:
>
> 12 .orc_unwind_ip 00172124 ffffffff82573b28 0000000002573b28 01773b28
> 2**0
> CONTENTS, ALLOC, LOAD, RELOC, READONLY, DATA
> 13 .orc_unwind 0022b1b6 ffffffff826e5c4c 00000000026e5c4c 018e5c4c
> 2**0
> CONTENTS, ALLOC, LOAD, READONLY, DATA
> 14 .orc_lookup 0003003c ffffffff82910e04 0000000002910e04 01b10e02
> 2**0
> ALLOC
> 15 .vvar 00001000 ffffffff82941000 0000000002941000 01b41000
> 2**4
> CONTENTS, ALLOC, LOAD, DATA
>
> Sorttable tool uses the are starting with offset 0x01b10e02 for 0x0003003c
> bytes. While kernel use the area starting with VMA at 0xffffffff82910e04
> for 0x0003003c bytes, meaning that each entry in this table used by kernel
> is actually 2 bytes behind the corresponding entry set from sorttable
> tool.
>
> Any suggestion on fixing this？
The VMA and LMA are both 4-byte aligned. The file offset alignment
(0x01b10e02) shouldn't matter.
Actually it looks like the problem is that the section doesn't have
CONTENTS, so it's just loaded as a BSS section (all zeros). The section
needs to be type SHT_PROGBITS instead of SHT_NOBITS.
$ readelf -S vmlinux |grep orc_lookup
[16] .orc_lookup NOBITS ffffffff82b68418 01d68418
I tried to fix it with
diff --git a/scripts/sorttable.h b/scripts/sorttable.h
index a36c76c17be4..76adb1fb88f8 100644
--- a/scripts/sorttable.h
+++ b/scripts/sorttable.h
@@ -341,6 +341,7 @@ static int do_sort(Elf_Ehdr *ehdr,
param.lookup_table_size = s->sh_size;
param.orc_lookup_table = (unsigned int *)
((void *)ehdr + s->sh_offset);
+ w(SHT_PROGBITS, &s->sh_type);
}
if (!strcmp(secstrings + idx, ".text")) {
param.text_size = s->sh_size;
But that makes kallsyms unhappy, so I guess we need to do it from the
linker script where .orc_lookup is created.
Linker script doesn't seem to allow manual specification of the section
type, so this is the best I could come up with:
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index db600ef218d7..49f4f5bc6165 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -826,6 +826,8 @@
. += (((SIZEOF(.text) + LOOKUP_BLOCK_SIZE - 1) / \
LOOKUP_BLOCK_SIZE) + 1) * 4; \
orc_lookup_end = .; \
+ /* HACK: force SHT_PROGBITS so sorttable can edit: */ \
+ BYTE(1); \
}
#else
#define ORC_UNWIND_TABLE
> On Jun 2, 2020, at 1:38 AM, Josh Poimboeuf <[email protected]> wrote:
>
> On Sun, May 31, 2020 at 01:26:54PM +0800, changhuaixin wrote:
>> It turned out to be an alignment problem. If sh_size of previous section
>> orc_unwind is not 4-byte aligned, sh_offset of the following orc_lookup
>> section is not 4-byte aligned too. However, the VMA of section orc_lookup
>> is aligned to the nearest 4-byte. Thus, the orc_lookup section means two
>> different ares for scripts/sorttable tool and kernel.
>>
>> Sections headers look like this when it happens:
>>
>> 12 .orc_unwind_ip 00172124 ffffffff82573b28 0000000002573b28 01773b28
>> 2**0
>> CONTENTS, ALLOC, LOAD, RELOC, READONLY, DATA
>> 13 .orc_unwind 0022b1b6 ffffffff826e5c4c 00000000026e5c4c 018e5c4c
>> 2**0
>> CONTENTS, ALLOC, LOAD, READONLY, DATA
>> 14 .orc_lookup 0003003c ffffffff82910e04 0000000002910e04 01b10e02
>> 2**0
>> ALLOC
>> 15 .vvar 00001000 ffffffff82941000 0000000002941000 01b41000
>> 2**4
>> CONTENTS, ALLOC, LOAD, DATA
>>
>> Sorttable tool uses the are starting with offset 0x01b10e02 for 0x0003003c
>> bytes. While kernel use the area starting with VMA at 0xffffffff82910e04
>> for 0x0003003c bytes, meaning that each entry in this table used by kernel
>> is actually 2 bytes behind the corresponding entry set from sorttable
>> tool.
>>
>> Any suggestion on fixing this？
>
> The VMA and LMA are both 4-byte aligned. The file offset alignment
> (0x01b10e02) shouldn't matter.
>
> Actually it looks like the problem is that the section doesn't have
> CONTENTS, so it's just loaded as a BSS section (all zeros). The section
> needs to be type SHT_PROGBITS instead of SHT_NOBITS.
>
> $ readelf -S vmlinux |grep orc_lookup
> [16] .orc_lookup NOBITS ffffffff82b68418 01d68418
>
> I tried to fix it with
>
> diff --git a/scripts/sorttable.h b/scripts/sorttable.h
> index a36c76c17be4..76adb1fb88f8 100644
> --- a/scripts/sorttable.h
> +++ b/scripts/sorttable.h
> @@ -341,6 +341,7 @@ static int do_sort(Elf_Ehdr *ehdr,
> param.lookup_table_size = s->sh_size;
> param.orc_lookup_table = (unsigned int *)
> ((void *)ehdr + s->sh_offset);
> + w(SHT_PROGBITS, &s->sh_type);
> }
> if (!strcmp(secstrings + idx, ".text")) {
> param.text_size = s->sh_size;
>
>
> But that makes kallsyms unhappy, so I guess we need to do it from the
> linker script where .orc_lookup is created.
>
> Linker script doesn't seem to allow manual specification of the section
> type, so this is the best I could come up with:
>
> diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
> index db600ef218d7..49f4f5bc6165 100644
> --- a/include/asm-generic/vmlinux.lds.h
> +++ b/include/asm-generic/vmlinux.lds.h
> @@ -826,6 +826,8 @@
> . += (((SIZEOF(.text) + LOOKUP_BLOCK_SIZE - 1) / \
> LOOKUP_BLOCK_SIZE) + 1) * 4; \
> orc_lookup_end = .; \
> + /* HACK: force SHT_PROGBITS so sorttable can edit: */ \
> + BYTE(1); \
> }
> #else
> #define ORC_UNWIND_TABLE
Thanks! It works.
Move building of fast lookup table from boot to sorttable tool. This saves us
6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores.
Changelog v2:
1. Type of section orc_lookup needs to be SHT_PROGBITS.
2. unwind_init() cannot be removed totally as setting lookup_num_blocks is needed.
Huaixin Chang (3):
scripts/sorttable: Change section type of orc_lookup to SHT_PROGBITS
scripts/sorttable: Build orc fast lookup table via sorttable tool
x86/unwind/orc: Simplify unwind_init() for x86 boot
arch/x86/kernel/unwind_orc.c | 40 ----------------
include/asm-generic/vmlinux.lds.h | 2 +
scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++---
3 files changed, 94 insertions(+), 47 deletions(-)
--
2.14.4.44.g2045bb6
In order to edit orc_lookup table via sorttable, type of section
orc_lookup needs to be SHT_PROGBITS instead of SHT_NOBITS.
Linker script doesn't seem to allow manual specification of the section
type, so just write a byte into the section instead.
Signed-off-by: Josh Poimboeuf <[email protected]>
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
include/asm-generic/vmlinux.lds.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index db600ef218d7..49f4f5bc6165 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -826,6 +826,8 @@
. += (((SIZEOF(.text) + LOOKUP_BLOCK_SIZE - 1) / \
LOOKUP_BLOCK_SIZE) + 1) * 4; \
orc_lookup_end = .; \
+ /* HACK: force SHT_PROGBITS so sorttable can edit: */ \
+ BYTE(1); \
}
#else
#define ORC_UNWIND_TABLE
--
2.14.4.44.g2045bb6
Since orc tables are already sorted by sorttable tool, let us move
building of fast lookup table into sorttable tool too. This saves us
6380us from boot time under Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz
with 64 cores.
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
scripts/sorttable.h | 99 +++++++++++++++++++++++++++++++++++++++++++++++++----
1 file changed, 92 insertions(+), 7 deletions(-)
diff --git a/scripts/sorttable.h b/scripts/sorttable.h
index a2baa2fefb13..a36c76c17be4 100644
--- a/scripts/sorttable.h
+++ b/scripts/sorttable.h
@@ -93,12 +93,50 @@
char g_err[ERRSTR_MAXSZ];
int *g_orc_ip_table;
struct orc_entry *g_orc_table;
+static unsigned long orc_ip_table_offset;
pthread_t orc_sort_thread;
+struct orc_sort_param {
+ size_t lookup_table_size;
+ unsigned int *orc_lookup_table;
+ unsigned long start_ip;
+ size_t text_size;
+ unsigned int orc_num_entries;
+};
+
static inline unsigned long orc_ip(const int *ip)
{
- return (unsigned long)ip + *ip;
+ return (unsigned long)ip + *ip + orc_ip_table_offset;
+}
+
+static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
+ unsigned int num_entries, unsigned long ip)
+{
+ int *first = ip_table;
+ int *last = ip_table + num_entries - 1;
+ int *mid = first, *found = first;
+
+ if (!num_entries)
+ return NULL;
+
+ /*
+ * Do a binary range search to find the rightmost duplicate of a given
+ * starting address. Some entries are section terminators which are
+ * "weak" entries for ensuring there are no gaps. They should be
+ * ignored when they conflict with a real entry.
+ */
+ while (first <= last) {
+ mid = first + ((last - first) / 2);
+
+ if (orc_ip(mid) <= ip) {
+ found = mid;
+ first = mid + 1;
+ } else
+ last = mid - 1;
+ }
+
+ return u_table + (found - ip_table);
}
static int orc_sort_cmp(const void *_a, const void *_b)
@@ -130,18 +168,24 @@ static void *sort_orctable(void *arg)
int *idxs = NULL;
int *tmp_orc_ip_table = NULL;
struct orc_entry *tmp_orc_table = NULL;
- unsigned int *orc_ip_size = (unsigned int *)arg;
- unsigned int num_entries = *orc_ip_size / sizeof(int);
+ struct orc_sort_param *param = (struct orc_sort_param *)arg;
+ unsigned int num_entries = param->orc_num_entries;
+ unsigned int orc_ip_size = num_entries * sizeof(int);
unsigned int orc_size = num_entries * sizeof(struct orc_entry);
+ unsigned int lookup_num_blocks = param->lookup_table_size / sizeof(int);
+ unsigned int *orc_lookup = param->orc_lookup_table;
+ unsigned long lookup_start_ip = param->start_ip;
+ unsigned long lookup_stop_ip = param->start_ip + param->text_size;
+ struct orc_entry *orc;
- idxs = (int *)malloc(*orc_ip_size);
+ idxs = (int *)malloc(orc_ip_size);
if (!idxs) {
snprintf(g_err, ERRSTR_MAXSZ, "malloc idxs: %s",
strerror(errno));
pthread_exit(g_err);
}
- tmp_orc_ip_table = (int *)malloc(*orc_ip_size);
+ tmp_orc_ip_table = (int *)malloc(orc_ip_size);
if (!tmp_orc_ip_table) {
snprintf(g_err, ERRSTR_MAXSZ, "malloc tmp_orc_ip_table: %s",
strerror(errno));
@@ -173,6 +217,31 @@ static void *sort_orctable(void *arg)
g_orc_table[i] = tmp_orc_table[idxs[i]];
}
+#define LOOKUP_BLOCK_ORDER 8
+#define LOOKUP_BLOCK_SIZE (1 << LOOKUP_BLOCK_ORDER)
+
+ for (i = 0; i < lookup_num_blocks-1; i++) {
+ orc = __orc_find(g_orc_ip_table, g_orc_table,
+ num_entries,
+ lookup_start_ip + (LOOKUP_BLOCK_SIZE * i));
+ if (!orc) {
+ snprintf(g_err, ERRSTR_MAXSZ,
+ "Corrupt .orc_unwind table\n");
+ pthread_exit(g_err);
+ }
+
+ orc_lookup[i] = orc - g_orc_table;
+ }
+
+ /* Initialize the ending block: */
+ orc = __orc_find(g_orc_ip_table, g_orc_table, num_entries,
+ lookup_stop_ip);
+ if (!orc) {
+ snprintf(g_err, ERRSTR_MAXSZ, "Corrupt .orc_unwind table\n");
+ pthread_exit(g_err);
+ }
+ orc_lookup[lookup_num_blocks-1] = orc - g_orc_table;
+
free(idxs);
free(tmp_orc_ip_table);
free(tmp_orc_table);
@@ -221,6 +290,8 @@ static int do_sort(Elf_Ehdr *ehdr,
unsigned int orc_ip_size = 0;
unsigned int orc_size = 0;
unsigned int orc_num_entries = 0;
+ unsigned long orc_ip_addr = 0;
+ struct orc_sort_param param;
#endif
shstrndx = r2(&ehdr->e_shstrndx);
@@ -259,17 +330,27 @@ static int do_sort(Elf_Ehdr *ehdr,
orc_ip_size = s->sh_size;
g_orc_ip_table = (int *)((void *)ehdr +
s->sh_offset);
+ orc_ip_addr = s->sh_addr;
}
if (!strcmp(secstrings + idx, ".orc_unwind")) {
orc_size = s->sh_size;
g_orc_table = (struct orc_entry *)((void *)ehdr +
s->sh_offset);
}
+ if (!strcmp(secstrings + idx, ".orc_lookup")) {
+ param.lookup_table_size = s->sh_size;
+ param.orc_lookup_table = (unsigned int *)
+ ((void *)ehdr + s->sh_offset);
+ }
+ if (!strcmp(secstrings + idx, ".text")) {
+ param.text_size = s->sh_size;
+ param.start_ip = s->sh_addr;
+ }
#endif
} /* for loop */
#if defined(SORTTABLE_64) && defined(UNWINDER_ORC_ENABLED)
- if (!g_orc_ip_table || !g_orc_table) {
+ if (!g_orc_ip_table || !g_orc_table || !param.orc_lookup_table) {
fprintf(stderr,
"incomplete ORC unwind tables in file: %s\n", fname);
goto out;
@@ -285,9 +366,13 @@ static int do_sort(Elf_Ehdr *ehdr,
goto out;
}
+ /* Make orc_ip return virtual address at execution. */
+ orc_ip_table_offset = orc_ip_addr - (unsigned long)g_orc_ip_table;
+
/* create thread to sort ORC unwind tables concurrently */
+ param.orc_num_entries = orc_num_entries;
if (pthread_create(&orc_sort_thread, NULL,
- sort_orctable, &orc_ip_size)) {
+ sort_orctable, ¶m)) {
fprintf(stderr,
"pthread_create orc_sort_thread failed '%s': %s\n",
strerror(errno), fname);
--
2.14.4.44.g2045bb6
The orc fast lookup table is built by scripts/sorttable tool. All that
is left is setting lookup_num_blocks.
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
arch/x86/kernel/unwind_orc.c | 40 ----------------------------------------
1 file changed, 40 deletions(-)
diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
index 7f969b2d240f..11b87ecf8919 100644
--- a/arch/x86/kernel/unwind_orc.c
+++ b/arch/x86/kernel/unwind_orc.c
@@ -264,48 +264,8 @@ void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
void __init unwind_init(void)
{
- size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
- size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
- size_t num_entries = orc_ip_size / sizeof(int);
- struct orc_entry *orc;
- int i;
-
- if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
- orc_size % sizeof(struct orc_entry) != 0 ||
- num_entries != orc_size / sizeof(struct orc_entry)) {
- orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
- return;
- }
-
- /*
- * Note, the orc_unwind and orc_unwind_ip tables were already
- * sorted at build time via the 'sorttable' tool.
- * It's ready for binary search straight away, no need to sort it.
- */
-
/* Initialize the fast lookup table: */
lookup_num_blocks = orc_lookup_end - orc_lookup;
- for (i = 0; i < lookup_num_blocks-1; i++) {
- orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
- num_entries,
- LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
- if (!orc) {
- orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
- return;
- }
-
- orc_lookup[i] = orc - __start_orc_unwind;
- }
-
- /* Initialize the ending block: */
- orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
- LOOKUP_STOP_IP);
- if (!orc) {
- orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
- return;
- }
- orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
-
orc_init = true;
}
--
2.14.4.44.g2045bb6
Move building of fast lookup table from boot to sorttable tool. This saves us
6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores. It
adds a little more than 7ms to boot time when testing on the same CPU.
Changelog v3:
1. Modify annotation of unwind_init().
Changelog v2:
1. Type of section orc_lookup needs to be SHT_PROGBITS.
2. unwind_init() cannot be removed totally as setting lookup_num_blocks is needed.
Huaixin Chang (3):
scripts/sorttable: Change section type of orc_lookup to SHT_PROGBITS
scripts/sorttable: Build orc fast lookup table via sorttable tool
x86/unwind/orc: Simplify unwind_init() for x86 boot
arch/x86/kernel/unwind_orc.c | 41 +---------------
include/asm-generic/vmlinux.lds.h | 2 +
scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++---
3 files changed, 96 insertions(+), 46 deletions(-)
--
2.14.4.44.g2045bb6
Since orc tables are already sorted by sorttable tool, let us move
building of fast lookup table into sorttable tool too. This saves us
6380us from boot time under Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz
with 64 cores.
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
scripts/sorttable.h | 99 +++++++++++++++++++++++++++++++++++++++++++++++++----
1 file changed, 92 insertions(+), 7 deletions(-)
diff --git a/scripts/sorttable.h b/scripts/sorttable.h
index a2baa2fefb13..a36c76c17be4 100644
--- a/scripts/sorttable.h
+++ b/scripts/sorttable.h
@@ -93,12 +93,50 @@
char g_err[ERRSTR_MAXSZ];
int *g_orc_ip_table;
struct orc_entry *g_orc_table;
+static unsigned long orc_ip_table_offset;
pthread_t orc_sort_thread;
+struct orc_sort_param {
+ size_t lookup_table_size;
+ unsigned int *orc_lookup_table;
+ unsigned long start_ip;
+ size_t text_size;
+ unsigned int orc_num_entries;
+};
+
static inline unsigned long orc_ip(const int *ip)
{
- return (unsigned long)ip + *ip;
+ return (unsigned long)ip + *ip + orc_ip_table_offset;
+}
+
+static struct orc_entry *__orc_find(int *ip_table, struct orc_entry *u_table,
+ unsigned int num_entries, unsigned long ip)
+{
+ int *first = ip_table;
+ int *last = ip_table + num_entries - 1;
+ int *mid = first, *found = first;
+
+ if (!num_entries)
+ return NULL;
+
+ /*
+ * Do a binary range search to find the rightmost duplicate of a given
+ * starting address. Some entries are section terminators which are
+ * "weak" entries for ensuring there are no gaps. They should be
+ * ignored when they conflict with a real entry.
+ */
+ while (first <= last) {
+ mid = first + ((last - first) / 2);
+
+ if (orc_ip(mid) <= ip) {
+ found = mid;
+ first = mid + 1;
+ } else
+ last = mid - 1;
+ }
+
+ return u_table + (found - ip_table);
}
static int orc_sort_cmp(const void *_a, const void *_b)
@@ -130,18 +168,24 @@ static void *sort_orctable(void *arg)
int *idxs = NULL;
int *tmp_orc_ip_table = NULL;
struct orc_entry *tmp_orc_table = NULL;
- unsigned int *orc_ip_size = (unsigned int *)arg;
- unsigned int num_entries = *orc_ip_size / sizeof(int);
+ struct orc_sort_param *param = (struct orc_sort_param *)arg;
+ unsigned int num_entries = param->orc_num_entries;
+ unsigned int orc_ip_size = num_entries * sizeof(int);
unsigned int orc_size = num_entries * sizeof(struct orc_entry);
+ unsigned int lookup_num_blocks = param->lookup_table_size / sizeof(int);
+ unsigned int *orc_lookup = param->orc_lookup_table;
+ unsigned long lookup_start_ip = param->start_ip;
+ unsigned long lookup_stop_ip = param->start_ip + param->text_size;
+ struct orc_entry *orc;
- idxs = (int *)malloc(*orc_ip_size);
+ idxs = (int *)malloc(orc_ip_size);
if (!idxs) {
snprintf(g_err, ERRSTR_MAXSZ, "malloc idxs: %s",
strerror(errno));
pthread_exit(g_err);
}
- tmp_orc_ip_table = (int *)malloc(*orc_ip_size);
+ tmp_orc_ip_table = (int *)malloc(orc_ip_size);
if (!tmp_orc_ip_table) {
snprintf(g_err, ERRSTR_MAXSZ, "malloc tmp_orc_ip_table: %s",
strerror(errno));
@@ -173,6 +217,31 @@ static void *sort_orctable(void *arg)
g_orc_table[i] = tmp_orc_table[idxs[i]];
}
+#define LOOKUP_BLOCK_ORDER 8
+#define LOOKUP_BLOCK_SIZE (1 << LOOKUP_BLOCK_ORDER)
+
+ for (i = 0; i < lookup_num_blocks-1; i++) {
+ orc = __orc_find(g_orc_ip_table, g_orc_table,
+ num_entries,
+ lookup_start_ip + (LOOKUP_BLOCK_SIZE * i));
+ if (!orc) {
+ snprintf(g_err, ERRSTR_MAXSZ,
+ "Corrupt .orc_unwind table\n");
+ pthread_exit(g_err);
+ }
+
+ orc_lookup[i] = orc - g_orc_table;
+ }
+
+ /* Initialize the ending block: */
+ orc = __orc_find(g_orc_ip_table, g_orc_table, num_entries,
+ lookup_stop_ip);
+ if (!orc) {
+ snprintf(g_err, ERRSTR_MAXSZ, "Corrupt .orc_unwind table\n");
+ pthread_exit(g_err);
+ }
+ orc_lookup[lookup_num_blocks-1] = orc - g_orc_table;
+
free(idxs);
free(tmp_orc_ip_table);
free(tmp_orc_table);
@@ -221,6 +290,8 @@ static int do_sort(Elf_Ehdr *ehdr,
unsigned int orc_ip_size = 0;
unsigned int orc_size = 0;
unsigned int orc_num_entries = 0;
+ unsigned long orc_ip_addr = 0;
+ struct orc_sort_param param;
#endif
shstrndx = r2(&ehdr->e_shstrndx);
@@ -259,17 +330,27 @@ static int do_sort(Elf_Ehdr *ehdr,
orc_ip_size = s->sh_size;
g_orc_ip_table = (int *)((void *)ehdr +
s->sh_offset);
+ orc_ip_addr = s->sh_addr;
}
if (!strcmp(secstrings + idx, ".orc_unwind")) {
orc_size = s->sh_size;
g_orc_table = (struct orc_entry *)((void *)ehdr +
s->sh_offset);
}
+ if (!strcmp(secstrings + idx, ".orc_lookup")) {
+ param.lookup_table_size = s->sh_size;
+ param.orc_lookup_table = (unsigned int *)
+ ((void *)ehdr + s->sh_offset);
+ }
+ if (!strcmp(secstrings + idx, ".text")) {
+ param.text_size = s->sh_size;
+ param.start_ip = s->sh_addr;
+ }
#endif
} /* for loop */
#if defined(SORTTABLE_64) && defined(UNWINDER_ORC_ENABLED)
- if (!g_orc_ip_table || !g_orc_table) {
+ if (!g_orc_ip_table || !g_orc_table || !param.orc_lookup_table) {
fprintf(stderr,
"incomplete ORC unwind tables in file: %s\n", fname);
goto out;
@@ -285,9 +366,13 @@ static int do_sort(Elf_Ehdr *ehdr,
goto out;
}
+ /* Make orc_ip return virtual address at execution. */
+ orc_ip_table_offset = orc_ip_addr - (unsigned long)g_orc_ip_table;
+
/* create thread to sort ORC unwind tables concurrently */
+ param.orc_num_entries = orc_num_entries;
if (pthread_create(&orc_sort_thread, NULL,
- sort_orctable, &orc_ip_size)) {
+ sort_orctable, ¶m)) {
fprintf(stderr,
"pthread_create orc_sort_thread failed '%s': %s\n",
strerror(errno), fname);
--
2.14.4.44.g2045bb6
The orc fast lookup table is built by scripts/sorttable tool. All that
is left is setting lookup_num_blocks.
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
arch/x86/kernel/unwind_orc.c | 41 ++---------------------------------------
1 file changed, 2 insertions(+), 39 deletions(-)
diff --git a/arch/x86/kernel/unwind_orc.c b/arch/x86/kernel/unwind_orc.c
index 7f969b2d240f..e4cf124c7a51 100644
--- a/arch/x86/kernel/unwind_orc.c
+++ b/arch/x86/kernel/unwind_orc.c
@@ -264,48 +264,11 @@ void unwind_module_init(struct module *mod, void *_orc_ip, size_t orc_ip_size,
void __init unwind_init(void)
{
- size_t orc_ip_size = (void *)__stop_orc_unwind_ip - (void *)__start_orc_unwind_ip;
- size_t orc_size = (void *)__stop_orc_unwind - (void *)__start_orc_unwind;
- size_t num_entries = orc_ip_size / sizeof(int);
- struct orc_entry *orc;
- int i;
-
- if (!num_entries || orc_ip_size % sizeof(int) != 0 ||
- orc_size % sizeof(struct orc_entry) != 0 ||
- num_entries != orc_size / sizeof(struct orc_entry)) {
- orc_warn("WARNING: Bad or missing .orc_unwind table. Disabling unwinder.\n");
- return;
- }
-
/*
- * Note, the orc_unwind and orc_unwind_ip tables were already
- * sorted at build time via the 'sorttable' tool.
- * It's ready for binary search straight away, no need to sort it.
+ * The fast lookup table is built via sorttable tool. Initialize
+ * lookup_num_blocks only.
*/
-
- /* Initialize the fast lookup table: */
lookup_num_blocks = orc_lookup_end - orc_lookup;
- for (i = 0; i < lookup_num_blocks-1; i++) {
- orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind,
- num_entries,
- LOOKUP_START_IP + (LOOKUP_BLOCK_SIZE * i));
- if (!orc) {
- orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
- return;
- }
-
- orc_lookup[i] = orc - __start_orc_unwind;
- }
-
- /* Initialize the ending block: */
- orc = __orc_find(__start_orc_unwind_ip, __start_orc_unwind, num_entries,
- LOOKUP_STOP_IP);
- if (!orc) {
- orc_warn("WARNING: Corrupt .orc_unwind table. Disabling unwinder.\n");
- return;
- }
- orc_lookup[lookup_num_blocks-1] = orc - __start_orc_unwind;
-
orc_init = true;
}
--
2.14.4.44.g2045bb6
In order to edit orc_lookup table via sorttable, type of section
orc_lookup needs to be SHT_PROGBITS instead of SHT_NOBITS.
Linker script doesn't seem to allow manual specification of the section
type, so just write a byte into the section instead.
Signed-off-by: Josh Poimboeuf <[email protected]>
Signed-off-by: Huaixin Chang <[email protected]>
Signed-off-by: Shile Zhang <[email protected]>
---
include/asm-generic/vmlinux.lds.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index db600ef218d7..49f4f5bc6165 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -826,6 +826,8 @@
. += (((SIZEOF(.text) + LOOKUP_BLOCK_SIZE - 1) / \
LOOKUP_BLOCK_SIZE) + 1) * 4; \
orc_lookup_end = .; \
+ /* HACK: force SHT_PROGBITS so sorttable can edit: */ \
+ BYTE(1); \
}
#else
#define ORC_UNWIND_TABLE
--
2.14.4.44.g2045bb6
Hi Josh, will you please have a look at this patchset?
There might be another way to set SHT_PROGBITS of section .orc_lookup by writing section headers when orc_unwind and orc_unwind_ip tables are writen. It might be as follows:
diff --git a/tools/objtool/orc_gen.c b/tools/objtool/orc_gen.c
index 3f98dcfbc177..860d4dcec8e6 100644
--- a/tools/objtool/orc_gen.c
+++ b/tools/objtool/orc_gen.c
@@ -183,6 +183,10 @@ int create_orc_sections(struct objtool_file *file)
u_sec = elf_create_section(file->elf, ".orc_unwind",
sizeof(struct orc_entry), idx);
+ /* make flags of section orc_lookup right */
+ if (!elf_create_section(file->elf, ".orc_lookup", sizeof(int), 0))
+ return -1;
+
What do you think about this way of setting SHT_PROGBITS?
> On Jun 3, 2020, at 10:39 PM, Huaixin Chang <[email protected]> wrote:
>
> Move building of fast lookup table from boot to sorttable tool. This saves us
> 6380us boot time on Intel(R) Xeon(R) CPU E5-2682 v4 @ 2.50GHz with cores. It
> adds a little more than 7ms to boot time when testing on the same CPU.
>
> Changelog v3:
> 1. Modify annotation of unwind_init().
>
> Changelog v2:
> 1. Type of section orc_lookup needs to be SHT_PROGBITS.
> 2. unwind_init() cannot be removed totally as setting lookup_num_blocks is needed.
>
> Huaixin Chang (3):
> scripts/sorttable: Change section type of orc_lookup to SHT_PROGBITS
> scripts/sorttable: Build orc fast lookup table via sorttable tool
> x86/unwind/orc: Simplify unwind_init() for x86 boot
>
> arch/x86/kernel/unwind_orc.c | 41 +---------------
> include/asm-generic/vmlinux.lds.h | 2 +
> scripts/sorttable.h | 99 ++++++++++++++++++++++++++++++++++++---
> 3 files changed, 96 insertions(+), 46 deletions(-)
>
> --
> 2.14.4.44.g2045bb6