From: "Martin J. Bligh" <mbligh@google.com>
To: Chuck Ebbert <76306.1226@compuserve.com>
Cc: linux-kernel <linux-kernel@vger.kernel.org>,
Rohit Seth <rohitseth@google.com>, Andi Kleen <ak@suse.de>,
Andrew Morton <akpm@osdl.org>, Ingo Molnar <mingo@elte.hu>
Subject: Re: [RFC, patch] i386: vgetcpu() for NUMA, take 2
Date: Sat, 24 Jun 2006 08:37:05 -0700 [thread overview]
Message-ID: <449D5C21.5060600@google.com> (raw)
In-Reply-To: <200606240426_MC3-1-C354-E4BC@compuserve.com>
Chuck Ebbert wrote:
> This is attempt #2 at vgetcpu() NUMA support for i386. It uses a
> GDT entry to hold cpu and node number for fast userspace access.
>
> changes since #1:
> proper function prototype (same as x86_64)
> changed alignment of vsyscall functions to 16 bytes
> (sigreturn needs to stay fixed, others can move)
>
> to-do:
> CFI annotations
> test NUMA on real NUMA hardware (someone please test)
What was the point of returning wrong info to userspace really quickly?
;-) (ie you could have migrated CPUs, so it's totally unreliable)
Is this just for some statistical monitoring thing?
M.
> Test program:
>
> /* vgetcpu.c: test how fast vgetcpu runs
> * boot kernel with vgetcpu patch first, then:
> * gcc -O3 -o vgetcpu vgetcpu.c <srcpath>/arch/i386/kernel/vsyscall-int80.so
> * (don't forget the optimization (-O3))
> */
> #define _GNU_SOURCE
> #include <stdio.h>
> #include <stdlib.h>
>
> extern int __attribute__ ((regparm(2))) __vgetcpu(int *cpu, int *node);
>
> #define rdtscll(t) asm("rdtsc" : "=A" (t))
>
> int main(int argc, char * const argv[])
> {
> long long tsc1, tsc2;
> int i, cpu = 999, node = 999, iters = 99999;
>
> if (__vgetcpu(&cpu, &node) || node == 999 || cpu == 999) {
> printf("vgetcpu failed!\n");
> _exit(1);
> }
> printf("node: %d, cpu: %d\n", node, cpu);
>
> rdtscll(tsc1);
> for (i = 0; i < iters; i++)
> __vgetcpu(&cpu, &node);
> rdtscll(tsc2);
>
> printf("vgetcpu took %llu clocks per call\n", (tsc2 - tsc1) / iters);
>
> return 0;
> }
>
>
> Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com>
>
> arch/i386/kernel/cpu/common.c | 3 ++
> arch/i386/kernel/head.S | 11 +++++++-
> arch/i386/kernel/smpboot.c | 2 +
> arch/i386/kernel/vsyscall-getcpu.S | 42 ++++++++++++++++++++++++++++++++++
> arch/i386/kernel/vsyscall-int80.S | 2 +
> arch/i386/kernel/vsyscall-sigreturn.S | 3 --
> arch/i386/kernel/vsyscall-sysenter.S | 2 +
> arch/i386/kernel/vsyscall.lds.S | 1
> include/asm-i386/segment.h | 4 ++-
> 9 files changed, 65 insertions(+), 5 deletions(-)
>
> --- 2.6.17-32.orig/arch/i386/kernel/cpu/common.c
> +++ 2.6.17-32/arch/i386/kernel/cpu/common.c
> @@ -642,6 +642,9 @@ void __cpuinit cpu_init(void)
> ((((__u64)stk16_off) << 32) & 0xff00000000000000ULL) |
> (CPU_16BIT_STACK_SIZE - 1);
>
> + /* Set up GDT entry for per-cpu data */
> + gdt[GDT_ENTRY_VGETCPU].a |= cpu & 0xff;
> +
> cpu_gdt_descr->size = GDT_SIZE - 1;
> cpu_gdt_descr->address = (unsigned long)gdt;
>
> --- 2.6.17-32.orig/arch/i386/kernel/head.S
> +++ 2.6.17-32/arch/i386/kernel/head.S
> @@ -480,7 +480,7 @@ ENTRY(boot_gdt_table)
> .quad 0x00cf92000000ffff /* kernel 4GB data at 0x00000000 */
>
> /*
> - * The Global Descriptor Table contains 28 quadwords, per-CPU.
> + * The Global Descriptor Table contains 32 quadwords, per-CPU.
> */
> .align L1_CACHE_BYTES
> ENTRY(cpu_gdt_table)
> @@ -525,7 +525,14 @@ ENTRY(cpu_gdt_table)
> .quad 0x004092000000ffff /* 0xc8 APM DS data */
>
> .quad 0x0000920000000000 /* 0xd0 - ESPFIX 16-bit SS */
> - .quad 0x0000000000000000 /* 0xd8 - unused */
> +
> + /*
> + * Use GDT entries to store per-cpu data for user space (DPL 3.)
> + * 32-bit data segment, byte granularity, base 0, limit set at runtime.
> + * Userspace will use LSL to access this data, stored in the limit field.
> + */
> + .quad 0x0040f20000000000 /* 0xd8 - nodeid and logical CPU number */
> +
> .quad 0x0000000000000000 /* 0xe0 - unused */
> .quad 0x0000000000000000 /* 0xe8 - unused */
> .quad 0x0000000000000000 /* 0xf0 - unused */
> --- /dev/null
> +++ 2.6.17-32/arch/i386/kernel/vsyscall-getcpu.S
> @@ -0,0 +1,42 @@
> +/*
> + * fastcall int __vgetcpu(int *cpu, int *node)
> + *
> + * This file is #include'd by vsyscall-*.S to place vgetcpu after the
> + * sigreturn code.
> + *
> + * Puts logical CPU number in *cpu, node ID in *node;
> + * returns 0 for success and -EFAULT on error.
> + *
> + * CPU number and node ID are 8 bits each, with 4 total bits available
> + * for future growth of either field.
> + */
> +
> +#include <linux/errno.h>
> +#include <asm/segment.h>
> +
> + .text
> + .balign 16
> + .globl __vgetcpu
> + .type __vgetcpu,@function
> +__vgetcpu:
> +.LSTART_vgetcpu:
> + mov $((GDT_ENTRY_VGETCPU<<3)|3),%cx
> + lsl %ecx,%ecx
> + jnz 1f
> + push %ecx
> + and $0xff,%ecx /* 8-bit cpu number */
> + mov %ecx,(%eax)
> + pop %ecx
> + xor %eax,%eax
> + shr $8,%ecx /* assume top 4 bits are zero */
> + mov %ecx,(%edx)
> + ret
> +1:
> + push $-EFAULT /* saves 2 bytes of .text */
> + pop %eax
> + ret
> +.LEND_vgetcpu:
> + .size __vgetcpu,.-.LSTART_vgetcpu
> + .previous
> +
> +/* ZZZ: need CFI annotations here */
> --- 2.6.17-32.orig/arch/i386/kernel/vsyscall-int80.S
> +++ 2.6.17-32/arch/i386/kernel/vsyscall-int80.S
> @@ -51,3 +51,5 @@ __kernel_vsyscall:
> * Get the common code for the sigreturn entry points.
> */
> #include "vsyscall-sigreturn.S"
> +
> +#include "vsyscall-getcpu.S"
> --- 2.6.17-32.orig/arch/i386/kernel/vsyscall-sysenter.S
> +++ 2.6.17-32/arch/i386/kernel/vsyscall-sysenter.S
> @@ -120,3 +120,5 @@ SYSENTER_RETURN:
> * Get the common code for the sigreturn entry points.
> */
> #include "vsyscall-sigreturn.S"
> +
> +#include "vsyscall-getcpu.S"
> --- 2.6.17-32.orig/arch/i386/kernel/vsyscall.lds.S
> +++ 2.6.17-32/arch/i386/kernel/vsyscall.lds.S
> @@ -57,6 +57,7 @@ VERSION
> __kernel_vsyscall;
> __kernel_sigreturn;
> __kernel_rt_sigreturn;
> + __vgetcpu;
>
> local: *;
> };
> --- 2.6.17-32.orig/arch/i386/kernel/smpboot.c
> +++ 2.6.17-32/arch/i386/kernel/smpboot.c
> @@ -615,6 +615,7 @@ static inline void map_cpu_to_node(int c
> printk("Mapping cpu %d to node %d\n", cpu, node);
> cpu_set(cpu, node_2_cpu_mask[node]);
> cpu_2_node[cpu] = node;
> + get_cpu_gdt_table(cpu)[GDT_ENTRY_VGETCPU].a |= (node & 0xff) << 8;
> }
>
> /* undo a mapping between cpu and node. */
> @@ -626,6 +627,7 @@ static inline void unmap_cpu_to_node(int
> for (node = 0; node < MAX_NUMNODES; node ++)
> cpu_clear(cpu, node_2_cpu_mask[node]);
> cpu_2_node[cpu] = 0;
> + get_cpu_gdt_table(cpu)[GDT_ENTRY_VGETCPU].a &= ~(0xff << 8);
> }
> #else /* !CONFIG_NUMA */
>
> --- 2.6.17-32.orig/include/asm-i386/segment.h
> +++ 2.6.17-32/include/asm-i386/segment.h
> @@ -39,7 +39,7 @@
> * 25 - APM BIOS support
> *
> * 26 - ESPFIX small SS
> - * 27 - unused
> + * 27 - vgetcpu() data
> * 28 - unused
> * 29 - unused
> * 30 - unused
> @@ -74,6 +74,8 @@
> #define GDT_ENTRY_ESPFIX_SS (GDT_ENTRY_KERNEL_BASE + 14)
> #define __ESPFIX_SS (GDT_ENTRY_ESPFIX_SS * 8)
>
> +#define GDT_ENTRY_VGETCPU (GDT_ENTRY_KERNEL_BASE + 15)
> +
> #define GDT_ENTRY_DOUBLEFAULT_TSS 31
>
> /*
> --- 2.6.17-32.orig/arch/i386/kernel/vsyscall-sigreturn.S
> +++ 2.6.17-32/arch/i386/kernel/vsyscall-sigreturn.S
> @@ -26,7 +26,7 @@ __kernel_sigreturn:
> .LEND_sigreturn:
> .size __kernel_sigreturn,.-.LSTART_sigreturn
>
> - .balign 32
> + .balign 16
> .globl __kernel_rt_sigreturn
> .type __kernel_rt_sigreturn,@function
> __kernel_rt_sigreturn:
> @@ -35,7 +35,6 @@ __kernel_rt_sigreturn:
> int $0x80
> .LEND_rt_sigreturn:
> .size __kernel_rt_sigreturn,.-.LSTART_rt_sigreturn
> - .balign 32
> .previous
>
> .section .eh_frame,"a",@progbits
next prev parent reply other threads:[~2006-06-24 15:38 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2006-06-24 8:23 [RFC, patch] i386: vgetcpu() for NUMA, take 2 Chuck Ebbert
2006-06-24 15:37 ` Martin J. Bligh [this message]
2006-06-25 15:19 ` Andi Kleen
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=449D5C21.5060600@google.com \
--to=mbligh@google.com \
--cc=76306.1226@compuserve.com \
--cc=ak@suse.de \
--cc=akpm@osdl.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@elte.hu \
--cc=rohitseth@google.com \
/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