Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > linux.kernel > #1322283 > unrolled thread

[PATCH] x86/head64: micro optimization: use memset instead of for loop

Started byAlexander Kuleshov <kuleshovmail@gmail.com>
First post2016-01-30 09:10 +0100
Last post2016-02-09 14:20 +0100
Articles 6 — 4 participants

Back to article view | Back to linux.kernel


Contents

  [PATCH] x86/head64: micro optimization: use memset instead of for loop Alexander Kuleshov <kuleshovmail@gmail.com> - 2016-01-30 09:10 +0100
    [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables() tip-bot for Alexander Kuleshov <tipbot@zytor.com> - 2016-01-30 09:30 +0100
      Re: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables() Yinghai Lu <yinghai@kernel.org> - 2016-02-05 01:30 +0100
        Re: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables() Ingo Molnar <mingo@kernel.org> - 2016-02-09 13:30 +0100
          [PATCH v2] x86/head64: micro optimization: use memset instead of for loop Alexander Kuleshov <kuleshovmail@gmail.com> - 2016-02-09 14:20 +0100
            Re: [PATCH v2] x86/head64: micro optimization: use memset instead of  for loop Ingo Molnar <mingo@kernel.org> - 2016-02-09 14:20 +0100

#1322283 — [PATCH] x86/head64: micro optimization: use memset instead of for loop

FromAlexander Kuleshov <kuleshovmail@gmail.com>
Date2016-01-30 09:10 +0100
Subject[PATCH] x86/head64: micro optimization: use memset instead of for loop
Message-ID<qWyLv-8gv-5@gated-at.bofh.it>
This patch simplifies clearing of page table directories, by
the using of the fast memset() from the arch/x86/lib/memset_64.S.
Besides this, we 25 bytes is 25 bytes less than original.

text       data     bss              dec            hex filename

9735144 4970776 15474688        30180608        1cc8500 vmlinux.old
9735119 4970776 15474688        30180583        1cc84e7 vmlinux

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
---
 arch/x86/kernel/head64.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index f129a9a..4260ac4 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -40,13 +40,8 @@ pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
 /* Wipe all early page tables except for the kernel symbol map */
 static void __init reset_early_page_tables(void)
 {
-	unsigned long i;
-
-	for (i = 0; i < PTRS_PER_PGD-1; i++)
-		early_level4_pgt[i].pgd = 0;
-
+	memset(early_level4_pgt, 0, sizeof(pgd_t) * (PTRS_PER_PGD-1));
 	next_early_pgt = 0;
-
 	write_cr3(__pa_nodebug(early_level4_pgt));
 }
 
@@ -54,7 +49,6 @@ static void __init reset_early_page_tables(void)
 int __init early_make_pgtable(unsigned long address)
 {
 	unsigned long physaddr = address - __PAGE_OFFSET;
-	unsigned long i;
 	pgdval_t pgd, *pgd_p;
 	pudval_t pud, *pud_p;
 	pmdval_t pmd, *pmd_p;
@@ -81,8 +75,7 @@ again:
 		}
 
 		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PUD; i++)
-			pud_p[i] = 0;
+		memset(pud_p, 0, sizeof(pud_p) * PTRS_PER_PUD);
 		*pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pud_p += pud_index(address);
@@ -97,8 +90,7 @@ again:
 		}
 
 		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PMD; i++)
-			pmd_p[i] = 0;
+		memset(pmd_p, 0, sizeof(pmd_p) * PTRS_PER_PMD);
 		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pmd = (physaddr & PMD_MASK) + early_pmd_flags;
-- 
2.7.0.25.gfc10eb5

[toc] | [next] | [standalone]


#1322285 — [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()

Fromtip-bot for Alexander Kuleshov <tipbot@zytor.com>
Date2016-01-30 09:30 +0100
Subject[tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()
Message-ID<qWz4S-8oS-3@gated-at.bofh.it>
In reply to#1322283
Commit-ID:  5e9ebbd87a99ecc6abb74325b0ac63c46891f6f3
Gitweb:     http://git.kernel.org/tip/5e9ebbd87a99ecc6abb74325b0ac63c46891f6f3
Author:     Alexander Kuleshov <kuleshovmail@gmail.com>
AuthorDate: Sat, 30 Jan 2016 14:01:12 +0600
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Sat, 30 Jan 2016 09:20:55 +0100

x86/boot: Micro-optimize reset_early_page_tables()

Save 25 bytes of code and make the bootup a tiny bit faster:

     text    data bss             dec             filename
  9735144 4970776 15474688        30180608        vmlinux.old
  9735119 4970776 15474688        30180583        vmlinux

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
Cc: Alexander Popov <alpopov@ptsecurity.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1454140872-16926-1-git-send-email-kuleshovmail@gmail.com
[ Fixed various small details. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/kernel/head64.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index f129a9a..35843ca 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -40,13 +40,8 @@ pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
 /* Wipe all early page tables except for the kernel symbol map */
 static void __init reset_early_page_tables(void)
 {
-	unsigned long i;
-
-	for (i = 0; i < PTRS_PER_PGD-1; i++)
-		early_level4_pgt[i].pgd = 0;
-
+	memset(early_level4_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
 	next_early_pgt = 0;
-
 	write_cr3(__pa_nodebug(early_level4_pgt));
 }
 
@@ -54,7 +49,6 @@ static void __init reset_early_page_tables(void)
 int __init early_make_pgtable(unsigned long address)
 {
 	unsigned long physaddr = address - __PAGE_OFFSET;
-	unsigned long i;
 	pgdval_t pgd, *pgd_p;
 	pudval_t pud, *pud_p;
 	pmdval_t pmd, *pmd_p;
@@ -81,8 +75,7 @@ again:
 		}
 
 		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PUD; i++)
-			pud_p[i] = 0;
+		memset(pud_p, 0, sizeof(pud_p) * PTRS_PER_PUD);
 		*pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pud_p += pud_index(address);
@@ -97,8 +90,7 @@ again:
 		}
 
 		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PMD; i++)
-			pmd_p[i] = 0;
+		memset(pmd_p, 0, sizeof(pmd_p) * PTRS_PER_PMD);
 		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pmd = (physaddr & PMD_MASK) + early_pmd_flags;

[toc] | [prev] | [next] | [standalone]


#1327375 — Re: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()

FromYinghai Lu <yinghai@kernel.org>
Date2016-02-05 01:30 +0100
SubjectRe: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()
Message-ID<qYCrE-10Z-11@gated-at.bofh.it>
In reply to#1322285
On Sat, Jan 30, 2016 at 12:26 AM, tip-bot for Alexander Kuleshov
<tipbot@zytor.com> wrote:
> ---
>  arch/x86/kernel/head64.c | 14 +++-----------
>  1 file changed, 3 insertions(+), 11 deletions(-)
>
> diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
> index f129a9a..35843ca 100644
> --- a/arch/x86/kernel/head64.c
> +++ b/arch/x86/kernel/head64.c
> @@ -40,13 +40,8 @@ pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);

.
>
> @@ -54,7 +49,6 @@ static void __init reset_early_page_tables(void)
>  int __init early_make_pgtable(unsigned long address)
>  {
>         unsigned long physaddr = address - __PAGE_OFFSET;
> -       unsigned long i;
>         pgdval_t pgd, *pgd_p;
>         pudval_t pud, *pud_p;
>         pmdval_t pmd, *pmd_p;
> @@ -81,8 +75,7 @@ again:
>                 }
>
>                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
> -               for (i = 0; i < PTRS_PER_PUD; i++)
> -                       pud_p[i] = 0;
> +               memset(pud_p, 0, sizeof(pud_p) * PTRS_PER_PUD);

should be
                  memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);

>                 *pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
>         }
>         pud_p += pud_index(address);
> @@ -97,8 +90,7 @@ again:
>                 }
>
>                 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
> -               for (i = 0; i < PTRS_PER_PMD; i++)
> -                       pmd_p[i] = 0;
> +               memset(pmd_p, 0, sizeof(pmd_p) * PTRS_PER_PMD);

should be
                    memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);

>                 *pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
>         }
>         pmd = (physaddr & PMD_MASK) + early_pmd_flags;

[toc] | [prev] | [next] | [standalone]


#1330182 — Re: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()

FromIngo Molnar <mingo@kernel.org>
Date2016-02-09 13:30 +0100
SubjectRe: [tip:x86/boot] x86/boot: Micro-optimize reset_early_page_tables()
Message-ID<r0fAC-3Wc-33@gated-at.bofh.it>
In reply to#1327375
* Yinghai Lu <yinghai@kernel.org> wrote:

> On Sat, Jan 30, 2016 at 12:26 AM, tip-bot for Alexander Kuleshov
> <tipbot@zytor.com> wrote:
> > ---
> >  arch/x86/kernel/head64.c | 14 +++-----------
> >  1 file changed, 3 insertions(+), 11 deletions(-)
> >
> > diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
> > index f129a9a..35843ca 100644
> > --- a/arch/x86/kernel/head64.c
> > +++ b/arch/x86/kernel/head64.c
> > @@ -40,13 +40,8 @@ pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
> 
> .
> >
> > @@ -54,7 +49,6 @@ static void __init reset_early_page_tables(void)
> >  int __init early_make_pgtable(unsigned long address)
> >  {
> >         unsigned long physaddr = address - __PAGE_OFFSET;
> > -       unsigned long i;
> >         pgdval_t pgd, *pgd_p;
> >         pudval_t pud, *pud_p;
> >         pmdval_t pmd, *pmd_p;
> > @@ -81,8 +75,7 @@ again:
> >                 }
> >
> >                 pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
> > -               for (i = 0; i < PTRS_PER_PUD; i++)
> > -                       pud_p[i] = 0;
> > +               memset(pud_p, 0, sizeof(pud_p) * PTRS_PER_PUD);
> 
> should be
>                   memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
> 
> >                 *pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
> >         }
> >         pud_p += pud_index(address);
> > @@ -97,8 +90,7 @@ again:
> >                 }
> >
> >                 pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
> > -               for (i = 0; i < PTRS_PER_PMD; i++)
> > -                       pmd_p[i] = 0;
> > +               memset(pmd_p, 0, sizeof(pmd_p) * PTRS_PER_PMD);
> 
> should be
>                     memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);

Indeed. Fortunately the two sizes are the same AFAICS, but this should 
nevertheless be fixed.

Thanks,

	Ingo

[toc] | [prev] | [next] | [standalone]


#1330220 — [PATCH v2] x86/head64: micro optimization: use memset instead of for loop

FromAlexander Kuleshov <kuleshovmail@gmail.com>
Date2016-02-09 14:20 +0100
Subject[PATCH v2] x86/head64: micro optimization: use memset instead of for loop
Message-ID<r0gn0-4x9-7@gated-at.bofh.it>
In reply to#1330182
This patch simplifies clearing of page table directories, by
the using of the fast memset() from the arch/x86/lib/memset_64.S.
Besides this, we 25 bytes is 25 bytes less than original.

text       data     bss              dec            hex filename

9735144 4970776 15474688        30180608        1cc8500 vmlinux.old
9735119 4970776 15474688        30180583        1cc84e7 vmlinux

Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
---
Changelog:

v2: We got sizeof pud_p/pmd_p in the v1, but actually we need
to take sizeof of pud_p/pmd_p pointer, although their sizes are
the same.

 arch/x86/kernel/head64.c | 14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/head64.c b/arch/x86/kernel/head64.c
index 2c0f340..1f4422d 100644
--- a/arch/x86/kernel/head64.c
+++ b/arch/x86/kernel/head64.c
@@ -40,13 +40,8 @@ pmdval_t early_pmd_flags = __PAGE_KERNEL_LARGE & ~(_PAGE_GLOBAL | _PAGE_NX);
 /* Wipe all early page tables except for the kernel symbol map */
 static void __init reset_early_page_tables(void)
 {
-	unsigned long i;
-
-	for (i = 0; i < PTRS_PER_PGD-1; i++)
-		early_level4_pgt[i].pgd = 0;
-
+	memset(early_level4_pgt, 0, sizeof(pgd_t)*(PTRS_PER_PGD-1));
 	next_early_pgt = 0;
-
 	write_cr3(__pa_nodebug(early_level4_pgt));
 }
 
@@ -54,7 +49,6 @@ static void __init reset_early_page_tables(void)
 int __init early_make_pgtable(unsigned long address)
 {
 	unsigned long physaddr = address - __PAGE_OFFSET;
-	unsigned long i;
 	pgdval_t pgd, *pgd_p;
 	pudval_t pud, *pud_p;
 	pmdval_t pmd, *pmd_p;
@@ -81,8 +75,7 @@ again:
 		}
 
 		pud_p = (pudval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PUD; i++)
-			pud_p[i] = 0;
+		memset(pud_p, 0, sizeof(*pud_p) * PTRS_PER_PUD);
 		*pgd_p = (pgdval_t)pud_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pud_p += pud_index(address);
@@ -97,8 +90,7 @@ again:
 		}
 
 		pmd_p = (pmdval_t *)early_dynamic_pgts[next_early_pgt++];
-		for (i = 0; i < PTRS_PER_PMD; i++)
-			pmd_p[i] = 0;
+		memset(pmd_p, 0, sizeof(*pmd_p) * PTRS_PER_PMD);
 		*pud_p = (pudval_t)pmd_p - __START_KERNEL_map + phys_base + _KERNPG_TABLE;
 	}
 	pmd = (physaddr & PMD_MASK) + early_pmd_flags;
-- 
2.7.0.25.gfc10eb5

[toc] | [prev] | [next] | [standalone]


#1330221 — Re: [PATCH v2] x86/head64: micro optimization: use memset instead of for loop

FromIngo Molnar <mingo@kernel.org>
Date2016-02-09 14:20 +0100
SubjectRe: [PATCH v2] x86/head64: micro optimization: use memset instead of for loop
Message-ID<r0gn0-4x9-5@gated-at.bofh.it>
In reply to#1330220
* Alexander Kuleshov <kuleshovmail@gmail.com> wrote:

> This patch simplifies clearing of page table directories, by
> the using of the fast memset() from the arch/x86/lib/memset_64.S.
> Besides this, we 25 bytes is 25 bytes less than original.
> 
> text       data     bss              dec            hex filename
> 
> 9735144 4970776 15474688        30180608        1cc8500 vmlinux.old
> 9735119 4970776 15474688        30180583        1cc84e7 vmlinux
> 
> Signed-off-by: Alexander Kuleshov <kuleshovmail@gmail.com>
> ---
> Changelog:
> 
> v2: We got sizeof pud_p/pmd_p in the v1, but actually we need
> to take sizeof of pud_p/pmd_p pointer, although their sizes are
> the same.

Please send a delta patch (with changelog) as the v1 commit is already deep within 
the tip:x86/boot branch.

Thanks,

	Ingo

[toc] | [prev] | [standalone]


Back to top | Article view | linux.kernel


csiph-web