2005-10-10 08:50:37 -04:00
|
|
|
/*
|
|
|
|
*
|
|
|
|
* Common boot and setup code.
|
|
|
|
*
|
|
|
|
* Copyright (C) 2001 PPC64 Team, IBM Corp
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU General Public License
|
|
|
|
* as published by the Free Software Foundation; either version
|
|
|
|
* 2 of the License, or (at your option) any later version.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#undef DEBUG
|
|
|
|
|
|
|
|
#include <linux/config.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/string.h>
|
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/kernel.h>
|
|
|
|
#include <linux/reboot.h>
|
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/initrd.h>
|
|
|
|
#include <linux/ide.h>
|
|
|
|
#include <linux/seq_file.h>
|
|
|
|
#include <linux/ioport.h>
|
|
|
|
#include <linux/console.h>
|
|
|
|
#include <linux/utsname.h>
|
|
|
|
#include <linux/tty.h>
|
|
|
|
#include <linux/root_dev.h>
|
|
|
|
#include <linux/notifier.h>
|
|
|
|
#include <linux/cpu.h>
|
|
|
|
#include <linux/unistd.h>
|
|
|
|
#include <linux/serial.h>
|
|
|
|
#include <linux/serial_8250.h>
|
[PATCH] powerpc/64: per cpu data optimisations
The current ppc64 per cpu data implementation is quite slow. eg:
lhz 11,18(13) /* smp_processor_id() */
ld 9,.LC63-.LCTOC1(30) /* per_cpu__variable_name */
ld 8,.LC61-.LCTOC1(30) /* __per_cpu_offset */
sldi 11,11,3 /* form index into __per_cpu_offset */
mr 10,9
ldx 9,11,8 /* __per_cpu_offset[smp_processor_id()] */
ldx 0,10,9 /* load per cpu data */
5 loads for something that is supposed to be fast, pretty awful. One
reason for the large number of loads is that we have to synthesize 2
64bit constants (per_cpu__variable_name and __per_cpu_offset).
By putting __per_cpu_offset into the paca we can avoid the 2 loads
associated with it:
ld 11,56(13) /* paca->data_offset */
ld 9,.LC59-.LCTOC1(30) /* per_cpu__variable_name */
ldx 0,9,11 /* load per cpu data
Longer term we can should be able to do even better than 3 loads.
If per_cpu__variable_name wasnt a 64bit constant and paca->data_offset
was in a register we could cut it down to one load. A suggestion from
Rusty is to use gcc's __thread extension here. In order to do this we
would need to free up r13 (the __thread register and where the paca
currently is). So far Ive had a few unsuccessful attempts at doing that :)
The patch also allocates per cpu memory node local on NUMA machines.
This patch from Rusty has been sitting in my queue _forever_ but stalled
when I hit the compiler bug. Sorry about that.
Finally I also only allocate per cpu data for possible cpus, which comes
straight out of the x86-64 port. On a pseries kernel (with NR_CPUS == 128)
and 4 possible cpus we see some nice gains:
total used free shared buffers cached
Mem: 4012228 212860 3799368 0 0 162424
total used free shared buffers cached
Mem: 4016200 212984 3803216 0 0 162424
A saving of 3.75MB. Quite nice for smaller machines. Note: we now have
to be careful of per cpu users that touch data for !possible cpus.
At this stage it might be worth making the NUMA and possible cpu
optimisations generic, but per cpu init is done so early we have to be
careful that all architectures have their possible map setup correctly.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-01-10 21:16:44 -05:00
|
|
|
#include <linux/bootmem.h>
|
2005-10-10 08:50:37 -04:00
|
|
|
#include <asm/io.h>
|
2005-12-04 02:39:37 -05:00
|
|
|
#include <asm/kdump.h>
|
2005-10-10 08:50:37 -04:00
|
|
|
#include <asm/prom.h>
|
|
|
|
#include <asm/processor.h>
|
|
|
|
#include <asm/pgtable.h>
|
|
|
|
#include <asm/smp.h>
|
|
|
|
#include <asm/elf.h>
|
|
|
|
#include <asm/machdep.h>
|
|
|
|
#include <asm/paca.h>
|
|
|
|
#include <asm/time.h>
|
|
|
|
#include <asm/cputable.h>
|
|
|
|
#include <asm/sections.h>
|
|
|
|
#include <asm/btext.h>
|
|
|
|
#include <asm/nvram.h>
|
|
|
|
#include <asm/setup.h>
|
|
|
|
#include <asm/system.h>
|
|
|
|
#include <asm/rtas.h>
|
|
|
|
#include <asm/iommu.h>
|
|
|
|
#include <asm/serial.h>
|
|
|
|
#include <asm/cache.h>
|
|
|
|
#include <asm/page.h>
|
|
|
|
#include <asm/mmu.h>
|
|
|
|
#include <asm/lmb.h>
|
2005-11-01 21:51:41 -05:00
|
|
|
#include <asm/iseries/it_lp_naca.h>
|
2005-10-10 08:50:37 -04:00
|
|
|
#include <asm/firmware.h>
|
2005-10-28 08:53:37 -04:00
|
|
|
#include <asm/xmon.h>
|
2005-11-06 17:49:43 -05:00
|
|
|
#include <asm/udbg.h>
|
2005-11-11 08:06:06 -05:00
|
|
|
#include <asm/kexec.h>
|
2005-10-10 08:50:37 -04:00
|
|
|
|
2005-11-08 19:01:06 -05:00
|
|
|
#include "setup.h"
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
#ifdef DEBUG
|
|
|
|
#define DBG(fmt...) udbg_printf(fmt)
|
|
|
|
#else
|
|
|
|
#define DBG(fmt...)
|
|
|
|
#endif
|
|
|
|
|
|
|
|
int have_of = 1;
|
|
|
|
int boot_cpuid = 0;
|
|
|
|
dev_t boot_dev;
|
|
|
|
u64 ppc64_pft_size;
|
|
|
|
|
2005-12-08 20:40:17 -05:00
|
|
|
/* Pick defaults since we might want to patch instructions
|
|
|
|
* before we've read this from the device tree.
|
|
|
|
*/
|
|
|
|
struct ppc64_caches ppc64_caches = {
|
|
|
|
.dline_size = 0x80,
|
|
|
|
.log_dline_size = 7,
|
|
|
|
.iline_size = 0x80,
|
|
|
|
.log_iline_size = 7
|
|
|
|
};
|
2005-10-10 08:50:37 -04:00
|
|
|
EXPORT_SYMBOL_GPL(ppc64_caches);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* These are used in binfmt_elf.c to put aux entries on the stack
|
|
|
|
* for each elf executable being started.
|
|
|
|
*/
|
|
|
|
int dcache_bsize;
|
|
|
|
int icache_bsize;
|
|
|
|
int ucache_bsize;
|
|
|
|
|
|
|
|
#ifdef CONFIG_MAGIC_SYSRQ
|
|
|
|
unsigned long SYSRQ_KEY;
|
|
|
|
#endif /* CONFIG_MAGIC_SYSRQ */
|
|
|
|
|
|
|
|
|
|
|
|
static int ppc64_panic_event(struct notifier_block *, unsigned long, void *);
|
|
|
|
static struct notifier_block ppc64_panic_block = {
|
|
|
|
.notifier_call = ppc64_panic_event,
|
|
|
|
.priority = INT_MIN /* may not return; must be done last */
|
|
|
|
};
|
|
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
|
|
|
|
static int smt_enabled_cmdline;
|
|
|
|
|
|
|
|
/* Look for ibm,smt-enabled OF option */
|
|
|
|
static void check_smt_enabled(void)
|
|
|
|
{
|
|
|
|
struct device_node *dn;
|
|
|
|
char *smt_option;
|
|
|
|
|
|
|
|
/* Allow the command line to overrule the OF option */
|
|
|
|
if (smt_enabled_cmdline)
|
|
|
|
return;
|
|
|
|
|
|
|
|
dn = of_find_node_by_path("/options");
|
|
|
|
|
|
|
|
if (dn) {
|
|
|
|
smt_option = (char *)get_property(dn, "ibm,smt-enabled", NULL);
|
|
|
|
|
|
|
|
if (smt_option) {
|
|
|
|
if (!strcmp(smt_option, "on"))
|
|
|
|
smt_enabled_at_boot = 1;
|
|
|
|
else if (!strcmp(smt_option, "off"))
|
|
|
|
smt_enabled_at_boot = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Look for smt-enabled= cmdline option */
|
|
|
|
static int __init early_smt_enabled(char *p)
|
|
|
|
{
|
|
|
|
smt_enabled_cmdline = 1;
|
|
|
|
|
|
|
|
if (!p)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!strcmp(p, "on") || !strcmp(p, "1"))
|
|
|
|
smt_enabled_at_boot = 1;
|
|
|
|
else if (!strcmp(p, "off") || !strcmp(p, "0"))
|
|
|
|
smt_enabled_at_boot = 0;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
early_param("smt-enabled", early_smt_enabled);
|
|
|
|
|
2005-11-04 18:33:55 -05:00
|
|
|
#else
|
|
|
|
#define check_smt_enabled()
|
2005-10-10 08:50:37 -04:00
|
|
|
#endif /* CONFIG_SMP */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Early initialization entry point. This is called by head.S
|
|
|
|
* with MMU translation disabled. We rely on the "feature" of
|
|
|
|
* the CPU that ignores the top 2 bits of the address in real
|
|
|
|
* mode so we can access kernel globals normally provided we
|
|
|
|
* only toy with things in the RMO region. From here, we do
|
|
|
|
* some early parsing of the device-tree to setup out LMB
|
|
|
|
* data structures, and allocate & initialize the hash table
|
|
|
|
* and segment tables so we can start running with translation
|
|
|
|
* enabled.
|
|
|
|
*
|
|
|
|
* It is this function which will call the probe() callback of
|
|
|
|
* the various platform types and copy the matching one to the
|
|
|
|
* global ppc_md structure. Your platform can eventually do
|
|
|
|
* some very early initializations from the probe() routine, but
|
|
|
|
* this is not recommended, be very careful as, for example, the
|
|
|
|
* device-tree is not accessible via normal means at this point.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void __init early_setup(unsigned long dt_ptr)
|
|
|
|
{
|
2006-01-10 19:54:09 -05:00
|
|
|
/* Enable early debugging if any specified (see udbg.h) */
|
|
|
|
udbg_early_init();
|
2005-10-10 08:50:37 -04:00
|
|
|
|
2006-03-28 07:15:54 -05:00
|
|
|
DBG(" -> early_setup(), dt_ptr: 0x%lx\n", dt_ptr);
|
2005-10-10 08:50:37 -04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Do early initializations using the flattened device
|
|
|
|
* tree, like retreiving the physical memory map or
|
|
|
|
* calculating/retreiving the hash table size
|
|
|
|
*/
|
|
|
|
early_init_devtree(__va(dt_ptr));
|
|
|
|
|
2006-03-25 01:25:17 -05:00
|
|
|
/* Now we know the logical id of our boot cpu, setup the paca. */
|
|
|
|
setup_boot_paca();
|
|
|
|
|
|
|
|
/* Fix up paca fields required for the boot cpu */
|
|
|
|
get_paca()->cpu_start = 1;
|
|
|
|
get_paca()->stab_real = __pa((u64)&initial_stab);
|
|
|
|
get_paca()->stab_addr = (u64)&initial_stab;
|
|
|
|
|
2006-03-28 07:15:54 -05:00
|
|
|
/* Probe the machine type */
|
|
|
|
probe_machine();
|
2005-10-10 08:50:37 -04:00
|
|
|
|
2005-12-04 02:39:37 -05:00
|
|
|
#ifdef CONFIG_CRASH_DUMP
|
|
|
|
kdump_setup();
|
|
|
|
#endif
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
DBG("Found, Initializing memory management...\n");
|
|
|
|
|
|
|
|
/*
|
2005-11-06 19:06:55 -05:00
|
|
|
* Initialize the MMU Hash table and create the linear mapping
|
|
|
|
* of memory. Has to be done before stab/slb initialization as
|
|
|
|
* this is currently where the page size encoding is obtained
|
2005-10-10 08:50:37 -04:00
|
|
|
*/
|
2005-11-06 19:06:55 -05:00
|
|
|
htab_initialize();
|
2005-10-10 08:50:37 -04:00
|
|
|
|
|
|
|
/*
|
2005-11-06 19:06:55 -05:00
|
|
|
* Initialize stab / SLB management except on iSeries
|
2005-10-10 08:50:37 -04:00
|
|
|
*/
|
2006-04-01 11:45:04 -05:00
|
|
|
if (cpu_has_feature(CPU_FTR_SLB))
|
|
|
|
slb_initialize();
|
|
|
|
else if (!firmware_has_feature(FW_FEATURE_ISERIES))
|
|
|
|
stab_initialize(get_paca()->stab_real);
|
2005-10-10 08:50:37 -04:00
|
|
|
|
|
|
|
DBG(" <- early_setup()\n");
|
|
|
|
}
|
|
|
|
|
2005-11-09 21:37:51 -05:00
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
void early_setup_secondary(void)
|
|
|
|
{
|
|
|
|
struct paca_struct *lpaca = get_paca();
|
|
|
|
|
|
|
|
/* Mark enabled in PACA */
|
|
|
|
lpaca->proc_enabled = 0;
|
|
|
|
|
|
|
|
/* Initialize hash table for that CPU */
|
|
|
|
htab_initialize_secondary();
|
|
|
|
|
|
|
|
/* Initialize STAB/SLB. We use a virtual address as it works
|
|
|
|
* in real mode on pSeries and we want a virutal address on
|
|
|
|
* iSeries anyway
|
|
|
|
*/
|
|
|
|
if (cpu_has_feature(CPU_FTR_SLB))
|
|
|
|
slb_initialize();
|
|
|
|
else
|
|
|
|
stab_initialize(lpaca->stab_addr);
|
|
|
|
}
|
|
|
|
|
|
|
|
#endif /* CONFIG_SMP */
|
2005-10-10 08:50:37 -04:00
|
|
|
|
2005-11-03 20:09:42 -05:00
|
|
|
#if defined(CONFIG_SMP) || defined(CONFIG_KEXEC)
|
|
|
|
void smp_release_cpus(void)
|
|
|
|
{
|
|
|
|
extern unsigned long __secondary_hold_spinloop;
|
2005-12-05 16:49:00 -05:00
|
|
|
unsigned long *ptr;
|
2005-11-03 20:09:42 -05:00
|
|
|
|
|
|
|
DBG(" -> smp_release_cpus()\n");
|
|
|
|
|
|
|
|
/* All secondary cpus are spinning on a common spinloop, release them
|
|
|
|
* all now so they can start to spin on their individual paca
|
|
|
|
* spinloops. For non SMP kernels, the secondary cpus never get out
|
|
|
|
* of the common spinloop.
|
|
|
|
* This is useless but harmless on iSeries, secondaries are already
|
|
|
|
* waiting on their paca spinloops. */
|
|
|
|
|
2005-12-05 16:49:00 -05:00
|
|
|
ptr = (unsigned long *)((unsigned long)&__secondary_hold_spinloop
|
|
|
|
- PHYSICAL_START);
|
|
|
|
*ptr = 1;
|
2005-11-03 20:09:42 -05:00
|
|
|
mb();
|
|
|
|
|
|
|
|
DBG(" <- smp_release_cpus()\n");
|
|
|
|
}
|
|
|
|
#endif /* CONFIG_SMP || CONFIG_KEXEC */
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
/*
|
2005-11-09 21:37:51 -05:00
|
|
|
* Initialize some remaining members of the ppc64_caches and systemcfg
|
|
|
|
* structures
|
2005-10-10 08:50:37 -04:00
|
|
|
* (at least until we get rid of them completely). This is mostly some
|
|
|
|
* cache informations about the CPU that will be used by cache flush
|
|
|
|
* routines and/or provided to userland
|
|
|
|
*/
|
|
|
|
static void __init initialize_cache_info(void)
|
|
|
|
{
|
|
|
|
struct device_node *np;
|
|
|
|
unsigned long num_cpus = 0;
|
|
|
|
|
|
|
|
DBG(" -> initialize_cache_info()\n");
|
|
|
|
|
|
|
|
for (np = NULL; (np = of_find_node_by_type(np, "cpu"));) {
|
|
|
|
num_cpus += 1;
|
|
|
|
|
|
|
|
/* We're assuming *all* of the CPUs have the same
|
|
|
|
* d-cache and i-cache sizes... -Peter
|
|
|
|
*/
|
|
|
|
|
|
|
|
if ( num_cpus == 1 ) {
|
|
|
|
u32 *sizep, *lsizep;
|
|
|
|
u32 size, lsize;
|
|
|
|
const char *dc, *ic;
|
|
|
|
|
|
|
|
/* Then read cache informations */
|
2006-03-28 07:15:54 -05:00
|
|
|
if (machine_is(powermac)) {
|
2005-10-10 08:50:37 -04:00
|
|
|
dc = "d-cache-block-size";
|
|
|
|
ic = "i-cache-block-size";
|
|
|
|
} else {
|
|
|
|
dc = "d-cache-line-size";
|
|
|
|
ic = "i-cache-line-size";
|
|
|
|
}
|
|
|
|
|
|
|
|
size = 0;
|
|
|
|
lsize = cur_cpu_spec->dcache_bsize;
|
|
|
|
sizep = (u32 *)get_property(np, "d-cache-size", NULL);
|
|
|
|
if (sizep != NULL)
|
|
|
|
size = *sizep;
|
|
|
|
lsizep = (u32 *) get_property(np, dc, NULL);
|
|
|
|
if (lsizep != NULL)
|
|
|
|
lsize = *lsizep;
|
|
|
|
if (sizep == 0 || lsizep == 0)
|
|
|
|
DBG("Argh, can't find dcache properties ! "
|
|
|
|
"sizep: %p, lsizep: %p\n", sizep, lsizep);
|
|
|
|
|
2005-11-11 05:15:21 -05:00
|
|
|
ppc64_caches.dsize = size;
|
|
|
|
ppc64_caches.dline_size = lsize;
|
2005-10-10 08:50:37 -04:00
|
|
|
ppc64_caches.log_dline_size = __ilog2(lsize);
|
|
|
|
ppc64_caches.dlines_per_page = PAGE_SIZE / lsize;
|
|
|
|
|
|
|
|
size = 0;
|
|
|
|
lsize = cur_cpu_spec->icache_bsize;
|
|
|
|
sizep = (u32 *)get_property(np, "i-cache-size", NULL);
|
|
|
|
if (sizep != NULL)
|
|
|
|
size = *sizep;
|
|
|
|
lsizep = (u32 *)get_property(np, ic, NULL);
|
|
|
|
if (lsizep != NULL)
|
|
|
|
lsize = *lsizep;
|
|
|
|
if (sizep == 0 || lsizep == 0)
|
|
|
|
DBG("Argh, can't find icache properties ! "
|
|
|
|
"sizep: %p, lsizep: %p\n", sizep, lsizep);
|
|
|
|
|
2005-11-11 05:15:21 -05:00
|
|
|
ppc64_caches.isize = size;
|
|
|
|
ppc64_caches.iline_size = lsize;
|
2005-10-10 08:50:37 -04:00
|
|
|
ppc64_caches.log_iline_size = __ilog2(lsize);
|
|
|
|
ppc64_caches.ilines_per_page = PAGE_SIZE / lsize;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
DBG(" <- initialize_cache_info()\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Do some initial setup of the system. The parameters are those which
|
|
|
|
* were passed in from the bootloader.
|
|
|
|
*/
|
|
|
|
void __init setup_system(void)
|
|
|
|
{
|
|
|
|
DBG(" -> setup_system()\n");
|
|
|
|
|
2006-02-03 03:05:47 -05:00
|
|
|
#ifdef CONFIG_KEXEC
|
|
|
|
kdump_move_device_tree();
|
|
|
|
#endif
|
2005-10-10 08:50:37 -04:00
|
|
|
/*
|
|
|
|
* Unflatten the device-tree passed by prom_init or kexec
|
|
|
|
*/
|
|
|
|
unflatten_device_tree();
|
|
|
|
|
2005-11-11 08:06:06 -05:00
|
|
|
#ifdef CONFIG_KEXEC
|
|
|
|
kexec_setup(); /* requires unflattened device tree. */
|
|
|
|
#endif
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
/*
|
|
|
|
* Fill the ppc64_caches & systemcfg structures with informations
|
2006-01-09 18:10:13 -05:00
|
|
|
* retrieved from the device-tree. Need to be called before
|
2005-10-10 08:50:37 -04:00
|
|
|
* finish_device_tree() since the later requires some of the
|
|
|
|
* informations filled up here to properly parse the interrupt
|
|
|
|
* tree.
|
|
|
|
* It also sets up the cache line sizes which allows to call
|
|
|
|
* routines like flush_icache_range (used by the hash init
|
|
|
|
* later on).
|
|
|
|
*/
|
|
|
|
initialize_cache_info();
|
|
|
|
|
|
|
|
#ifdef CONFIG_PPC_RTAS
|
|
|
|
/*
|
|
|
|
* Initialize RTAS if available
|
|
|
|
*/
|
|
|
|
rtas_initialize();
|
|
|
|
#endif /* CONFIG_PPC_RTAS */
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check if we have an initrd provided via the device-tree
|
|
|
|
*/
|
|
|
|
check_for_initrd();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Do some platform specific early initializations, that includes
|
|
|
|
* setting up the hash table pointers. It also sets up some interrupt-mapping
|
|
|
|
* related options that will be used by finish_device_tree()
|
|
|
|
*/
|
|
|
|
ppc_md.init_early();
|
|
|
|
|
2005-11-23 01:56:06 -05:00
|
|
|
/*
|
|
|
|
* We can discover serial ports now since the above did setup the
|
|
|
|
* hash table management for us, thus ioremap works. We do that early
|
|
|
|
* so that further code can be debugged
|
|
|
|
*/
|
|
|
|
find_legacy_serial_ports();
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
/*
|
|
|
|
* "Finish" the device-tree, that is do the actual parsing of
|
|
|
|
* some of the properties like the interrupt map
|
|
|
|
*/
|
|
|
|
finish_device_tree();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Initialize xmon
|
|
|
|
*/
|
|
|
|
#ifdef CONFIG_XMON_DEFAULT
|
|
|
|
xmon_init(1);
|
|
|
|
#endif
|
|
|
|
/*
|
|
|
|
* Register early console
|
|
|
|
*/
|
|
|
|
register_early_udbg_console();
|
|
|
|
|
|
|
|
/* Save unparsed command line copy for /proc/cmdline */
|
|
|
|
strlcpy(saved_command_line, cmd_line, COMMAND_LINE_SIZE);
|
|
|
|
|
|
|
|
parse_early_param();
|
|
|
|
|
2005-11-04 18:33:55 -05:00
|
|
|
check_smt_enabled();
|
|
|
|
smp_setup_cpu_maps();
|
2005-10-10 08:50:37 -04:00
|
|
|
|
2006-02-15 22:13:50 -05:00
|
|
|
#ifdef CONFIG_SMP
|
2005-10-10 08:50:37 -04:00
|
|
|
/* Release secondary cpus out of their spinloops at 0x60 now that
|
|
|
|
* we can map physical -> logical CPU ids
|
|
|
|
*/
|
|
|
|
smp_release_cpus();
|
2006-02-15 22:13:50 -05:00
|
|
|
#endif
|
2005-10-10 08:50:37 -04:00
|
|
|
|
|
|
|
printk("Starting Linux PPC64 %s\n", system_utsname.version);
|
|
|
|
|
|
|
|
printk("-----------------------------------------------------\n");
|
|
|
|
printk("ppc64_pft_size = 0x%lx\n", ppc64_pft_size);
|
2005-11-11 05:15:21 -05:00
|
|
|
printk("ppc64_interrupt_controller = 0x%ld\n",
|
|
|
|
ppc64_interrupt_controller);
|
|
|
|
printk("physicalMemorySize = 0x%lx\n", lmb_phys_mem_size());
|
2005-10-10 08:50:37 -04:00
|
|
|
printk("ppc64_caches.dcache_line_size = 0x%x\n",
|
2005-11-11 05:15:21 -05:00
|
|
|
ppc64_caches.dline_size);
|
2005-10-10 08:50:37 -04:00
|
|
|
printk("ppc64_caches.icache_line_size = 0x%x\n",
|
2005-11-11 05:15:21 -05:00
|
|
|
ppc64_caches.iline_size);
|
2005-10-10 08:50:37 -04:00
|
|
|
printk("htab_address = 0x%p\n", htab_address);
|
|
|
|
printk("htab_hash_mask = 0x%lx\n", htab_hash_mask);
|
2005-12-04 02:39:23 -05:00
|
|
|
#if PHYSICAL_START > 0
|
|
|
|
printk("physical_start = 0x%x\n", PHYSICAL_START);
|
|
|
|
#endif
|
2005-10-10 08:50:37 -04:00
|
|
|
printk("-----------------------------------------------------\n");
|
|
|
|
|
|
|
|
DBG(" <- setup_system()\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
static int ppc64_panic_event(struct notifier_block *this,
|
|
|
|
unsigned long event, void *ptr)
|
|
|
|
{
|
|
|
|
ppc_md.panic((char *)ptr); /* May not return */
|
|
|
|
return NOTIFY_DONE;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifdef CONFIG_IRQSTACKS
|
|
|
|
static void __init irqstack_early_init(void)
|
|
|
|
{
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* interrupt stacks must be under 256MB, we cannot afford to take
|
|
|
|
* SLB misses on them.
|
|
|
|
*/
|
2006-03-28 17:50:51 -05:00
|
|
|
for_each_possible_cpu(i) {
|
2005-11-06 19:06:55 -05:00
|
|
|
softirq_ctx[i] = (struct thread_info *)
|
|
|
|
__va(lmb_alloc_base(THREAD_SIZE,
|
|
|
|
THREAD_SIZE, 0x10000000));
|
|
|
|
hardirq_ctx[i] = (struct thread_info *)
|
|
|
|
__va(lmb_alloc_base(THREAD_SIZE,
|
|
|
|
THREAD_SIZE, 0x10000000));
|
2005-10-10 08:50:37 -04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
#define irqstack_early_init()
|
|
|
|
#endif
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Stack space used when we detect a bad kernel stack pointer, and
|
|
|
|
* early in SMP boots before relocation is enabled.
|
|
|
|
*/
|
|
|
|
static void __init emergency_stack_init(void)
|
|
|
|
{
|
|
|
|
unsigned long limit;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Emergency stacks must be under 256MB, we cannot afford to take
|
|
|
|
* SLB misses on them. The ABI also requires them to be 128-byte
|
|
|
|
* aligned.
|
|
|
|
*
|
|
|
|
* Since we use these as temporary stacks during secondary CPU
|
|
|
|
* bringup, we need to get at them in real mode. This means they
|
|
|
|
* must also be within the RMO region.
|
|
|
|
*/
|
|
|
|
limit = min(0x10000000UL, lmb.rmo_size);
|
|
|
|
|
2006-03-28 17:50:51 -05:00
|
|
|
for_each_possible_cpu(i)
|
2005-11-06 19:06:55 -05:00
|
|
|
paca[i].emergency_sp =
|
|
|
|
__va(lmb_alloc_base(HW_PAGE_SIZE, 128, limit)) + HW_PAGE_SIZE;
|
2005-10-10 08:50:37 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Called into from start_kernel, after lock_kernel has been called.
|
|
|
|
* Initializes bootmem, which is unsed to manage page allocation until
|
|
|
|
* mem_init is called.
|
|
|
|
*/
|
|
|
|
void __init setup_arch(char **cmdline_p)
|
|
|
|
{
|
|
|
|
extern void do_init_bootmem(void);
|
|
|
|
|
|
|
|
ppc64_boot_msg(0x12, "Setup Arch");
|
|
|
|
|
|
|
|
*cmdline_p = cmd_line;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Set cache line size based on type of cpu as a default.
|
|
|
|
* Systems with OF can look in the properties on the cpu node(s)
|
|
|
|
* for a possibly more accurate value.
|
|
|
|
*/
|
|
|
|
dcache_bsize = ppc64_caches.dline_size;
|
|
|
|
icache_bsize = ppc64_caches.iline_size;
|
|
|
|
|
|
|
|
/* reboot on panic */
|
|
|
|
panic_timeout = 180;
|
|
|
|
|
|
|
|
if (ppc_md.panic)
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 04:16:30 -05:00
|
|
|
atomic_notifier_chain_register(&panic_notifier_list,
|
|
|
|
&ppc64_panic_block);
|
2005-10-10 08:50:37 -04:00
|
|
|
|
|
|
|
init_mm.start_code = PAGE_OFFSET;
|
|
|
|
init_mm.end_code = (unsigned long) _etext;
|
|
|
|
init_mm.end_data = (unsigned long) _edata;
|
|
|
|
init_mm.brk = klimit;
|
|
|
|
|
|
|
|
irqstack_early_init();
|
|
|
|
emergency_stack_init();
|
|
|
|
|
|
|
|
stabs_alloc();
|
|
|
|
|
|
|
|
/* set up the bootmem stuff with available memory */
|
|
|
|
do_init_bootmem();
|
|
|
|
sparse_init();
|
|
|
|
|
2005-10-20 07:00:20 -04:00
|
|
|
#ifdef CONFIG_DUMMY_CONSOLE
|
|
|
|
conswitchp = &dummy_con;
|
|
|
|
#endif
|
|
|
|
|
2005-10-10 08:50:37 -04:00
|
|
|
ppc_md.setup_arch();
|
|
|
|
|
|
|
|
paging_init();
|
|
|
|
ppc64_boot_msg(0x15, "Setup Done");
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* ToDo: do something useful if ppc_md is not yet setup. */
|
|
|
|
#define PPC64_LINUX_FUNCTION 0x0f000000
|
|
|
|
#define PPC64_IPL_MESSAGE 0xc0000000
|
|
|
|
#define PPC64_TERM_MESSAGE 0xb0000000
|
|
|
|
|
|
|
|
static void ppc64_do_msg(unsigned int src, const char *msg)
|
|
|
|
{
|
|
|
|
if (ppc_md.progress) {
|
|
|
|
char buf[128];
|
|
|
|
|
|
|
|
sprintf(buf, "%08X\n", src);
|
|
|
|
ppc_md.progress(buf, 0);
|
|
|
|
snprintf(buf, 128, "%s", msg);
|
|
|
|
ppc_md.progress(buf, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Print a boot progress message. */
|
|
|
|
void ppc64_boot_msg(unsigned int src, const char *msg)
|
|
|
|
{
|
|
|
|
ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_IPL_MESSAGE|src, msg);
|
|
|
|
printk("[boot]%04x %s\n", src, msg);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Print a termination message (print only -- does not stop the kernel) */
|
|
|
|
void ppc64_terminate_msg(unsigned int src, const char *msg)
|
|
|
|
{
|
|
|
|
ppc64_do_msg(PPC64_LINUX_FUNCTION|PPC64_TERM_MESSAGE|src, msg);
|
|
|
|
printk("[terminate]%04x %s\n", src, msg);
|
|
|
|
}
|
|
|
|
|
|
|
|
void cpu_die(void)
|
|
|
|
{
|
|
|
|
if (ppc_md.cpu_die)
|
|
|
|
ppc_md.cpu_die();
|
|
|
|
}
|
[PATCH] powerpc/64: per cpu data optimisations
The current ppc64 per cpu data implementation is quite slow. eg:
lhz 11,18(13) /* smp_processor_id() */
ld 9,.LC63-.LCTOC1(30) /* per_cpu__variable_name */
ld 8,.LC61-.LCTOC1(30) /* __per_cpu_offset */
sldi 11,11,3 /* form index into __per_cpu_offset */
mr 10,9
ldx 9,11,8 /* __per_cpu_offset[smp_processor_id()] */
ldx 0,10,9 /* load per cpu data */
5 loads for something that is supposed to be fast, pretty awful. One
reason for the large number of loads is that we have to synthesize 2
64bit constants (per_cpu__variable_name and __per_cpu_offset).
By putting __per_cpu_offset into the paca we can avoid the 2 loads
associated with it:
ld 11,56(13) /* paca->data_offset */
ld 9,.LC59-.LCTOC1(30) /* per_cpu__variable_name */
ldx 0,9,11 /* load per cpu data
Longer term we can should be able to do even better than 3 loads.
If per_cpu__variable_name wasnt a 64bit constant and paca->data_offset
was in a register we could cut it down to one load. A suggestion from
Rusty is to use gcc's __thread extension here. In order to do this we
would need to free up r13 (the __thread register and where the paca
currently is). So far Ive had a few unsuccessful attempts at doing that :)
The patch also allocates per cpu memory node local on NUMA machines.
This patch from Rusty has been sitting in my queue _forever_ but stalled
when I hit the compiler bug. Sorry about that.
Finally I also only allocate per cpu data for possible cpus, which comes
straight out of the x86-64 port. On a pseries kernel (with NR_CPUS == 128)
and 4 possible cpus we see some nice gains:
total used free shared buffers cached
Mem: 4012228 212860 3799368 0 0 162424
total used free shared buffers cached
Mem: 4016200 212984 3803216 0 0 162424
A saving of 3.75MB. Quite nice for smaller machines. Note: we now have
to be careful of per cpu users that touch data for !possible cpus.
At this stage it might be worth making the NUMA and possible cpu
optimisations generic, but per cpu init is done so early we have to be
careful that all architectures have their possible map setup correctly.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-01-10 21:16:44 -05:00
|
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
|
|
void __init setup_per_cpu_areas(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
unsigned long size;
|
|
|
|
char *ptr;
|
|
|
|
|
|
|
|
/* Copy section for each CPU (we discard the original) */
|
|
|
|
size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
|
|
|
|
#ifdef CONFIG_MODULES
|
|
|
|
if (size < PERCPU_ENOUGH_ROOM)
|
|
|
|
size = PERCPU_ENOUGH_ROOM;
|
|
|
|
#endif
|
|
|
|
|
2006-03-28 17:50:51 -05:00
|
|
|
for_each_possible_cpu(i) {
|
[PATCH] powerpc/64: per cpu data optimisations
The current ppc64 per cpu data implementation is quite slow. eg:
lhz 11,18(13) /* smp_processor_id() */
ld 9,.LC63-.LCTOC1(30) /* per_cpu__variable_name */
ld 8,.LC61-.LCTOC1(30) /* __per_cpu_offset */
sldi 11,11,3 /* form index into __per_cpu_offset */
mr 10,9
ldx 9,11,8 /* __per_cpu_offset[smp_processor_id()] */
ldx 0,10,9 /* load per cpu data */
5 loads for something that is supposed to be fast, pretty awful. One
reason for the large number of loads is that we have to synthesize 2
64bit constants (per_cpu__variable_name and __per_cpu_offset).
By putting __per_cpu_offset into the paca we can avoid the 2 loads
associated with it:
ld 11,56(13) /* paca->data_offset */
ld 9,.LC59-.LCTOC1(30) /* per_cpu__variable_name */
ldx 0,9,11 /* load per cpu data
Longer term we can should be able to do even better than 3 loads.
If per_cpu__variable_name wasnt a 64bit constant and paca->data_offset
was in a register we could cut it down to one load. A suggestion from
Rusty is to use gcc's __thread extension here. In order to do this we
would need to free up r13 (the __thread register and where the paca
currently is). So far Ive had a few unsuccessful attempts at doing that :)
The patch also allocates per cpu memory node local on NUMA machines.
This patch from Rusty has been sitting in my queue _forever_ but stalled
when I hit the compiler bug. Sorry about that.
Finally I also only allocate per cpu data for possible cpus, which comes
straight out of the x86-64 port. On a pseries kernel (with NR_CPUS == 128)
and 4 possible cpus we see some nice gains:
total used free shared buffers cached
Mem: 4012228 212860 3799368 0 0 162424
total used free shared buffers cached
Mem: 4016200 212984 3803216 0 0 162424
A saving of 3.75MB. Quite nice for smaller machines. Note: we now have
to be careful of per cpu users that touch data for !possible cpus.
At this stage it might be worth making the NUMA and possible cpu
optimisations generic, but per cpu init is done so early we have to be
careful that all architectures have their possible map setup correctly.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-01-10 21:16:44 -05:00
|
|
|
ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
|
|
|
|
if (!ptr)
|
|
|
|
panic("Cannot allocate cpu data for CPU %d\n", i);
|
|
|
|
|
|
|
|
paca[i].data_offset = ptr - __per_cpu_start;
|
|
|
|
memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|