sn2_smp.c

linux 内核源代码

/*
 * SN2 Platform specific SMP Support
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (C) 2000-2006 Silicon Graphics, Inc. All rights reserved.
 */

#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/spinlock.h>
#include <linux/threads.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/irq.h>
#include <linux/mmzone.h>
#include <linux/module.h>
#include <linux/bitops.h>
#include <linux/nodemask.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>

#include <asm/processor.h>
#include <asm/irq.h>
#include <asm/sal.h>
#include <asm/system.h>
#include <asm/delay.h>
#include <asm/io.h>
#include <asm/smp.h>
#include <asm/tlb.h>
#include <asm/numa.h>
#include <asm/hw_irq.h>
#include <asm/current.h>
#include <asm/sn/sn_cpuid.h>
#include <asm/sn/sn_sal.h>
#include <asm/sn/addrs.h>
#include <asm/sn/shub_mmr.h>
#include <asm/sn/nodepda.h>
#include <asm/sn/rw_mmr.h>
#include <asm/sn/sn_feature_sets.h>

DEFINE_PER_CPU(struct ptc_stats, ptcstats);
DECLARE_PER_CPU(struct ptc_stats, ptcstats);

static  __cacheline_aligned DEFINE_SPINLOCK(sn2_global_ptc_lock);

/* 0 = old algorithm (no IPI flushes), 1 = ipi deadlock flush, 2 = ipi instead of SHUB ptc, >2 = always ipi */
static int sn2_flush_opt = 0;

extern unsigned long
sn2_ptc_deadlock_recovery_core(volatile unsigned long *, unsigned long,
			       volatile unsigned long *, unsigned long,
			       volatile unsigned long *, unsigned long);
void
sn2_ptc_deadlock_recovery(short *, short, short, int,
			  volatile unsigned long *, unsigned long,
			  volatile unsigned long *, unsigned long);

/*
 * Note: some is the following is captured here to make degugging easier
 * (the macros make more sense if you see the debug patch - not posted)
 */
#define sn2_ptctest	0
#define local_node_uses_ptc_ga(sh1)	((sh1) ? 1 : 0)
#define max_active_pio(sh1)		((sh1) ? 32 : 7)
#define reset_max_active_on_deadlock()	1
#define PTC_LOCK(sh1)			((sh1) ? &sn2_global_ptc_lock : &sn_nodepda->ptc_lock)

struct ptc_stats {
	unsigned long ptc_l;
	unsigned long change_rid;
	unsigned long shub_ptc_flushes;
	unsigned long nodes_flushed;
	unsigned long deadlocks;
	unsigned long deadlocks2;
	unsigned long lock_itc_clocks;
	unsigned long shub_itc_clocks;
	unsigned long shub_itc_clocks_max;
	unsigned long shub_ptc_flushes_not_my_mm;
	unsigned long shub_ipi_flushes;
	unsigned long shub_ipi_flushes_itc_clocks;
};

#define sn2_ptctest	0

static inline unsigned long wait_piowc(void)
{
	volatile unsigned long *piows;
	unsigned long zeroval, ws;

	piows = pda->pio_write_status_addr;
	zeroval = pda->pio_write_status_val;
	do {
		cpu_relax();
	} while (((ws = *piows) & SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK) != zeroval);
	return (ws & SH_PIO_WRITE_STATUS_WRITE_DEADLOCK_MASK) != 0;
}

/**
 * sn_migrate - SN-specific task migration actions
 * @task: Task being migrated to new CPU
 *
 * SN2 PIO writes from separate CPUs are not guaranteed to arrive in order.
 * Context switching user threads which have memory-mapped MMIO may cause
 * PIOs to issue from separate CPUs, thus the PIO writes must be drained
 * from the previous CPU's Shub before execution resumes on the new CPU.
 */
void sn_migrate(struct task_struct *task)
{
	pda_t *last_pda = pdacpu(task_thread_info(task)->last_cpu);
	volatile unsigned long *adr = last_pda->pio_write_status_addr;
	unsigned long val = last_pda->pio_write_status_val;

	/* Drain PIO writes from old CPU's Shub */
	while (unlikely((*adr & SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK)
			!= val))
		cpu_relax();
}

void sn_tlb_migrate_finish(struct mm_struct *mm)
{
	/* flush_tlb_mm is inefficient if more than 1 users of mm */
	if (mm == current->mm && mm && atomic_read(&mm->mm_users) == 1)
		flush_tlb_mm(mm);
}

static void
sn2_ipi_flush_all_tlb(struct mm_struct *mm)
{
	unsigned long itc;

	itc = ia64_get_itc();
	smp_flush_tlb_cpumask(mm->cpu_vm_mask);
	itc = ia64_get_itc() - itc;
	__get_cpu_var(ptcstats).shub_ipi_flushes_itc_clocks += itc;
	__get_cpu_var(ptcstats).shub_ipi_flushes++;
}

/**
 * sn2_global_tlb_purge - globally purge translation cache of virtual address range
 * @mm: mm_struct containing virtual address range
 * @start: start of virtual address range
 * @end: end of virtual address range
 * @nbits: specifies number of bytes to purge per instruction (num = 1<<(nbits & 0xfc))
 *
 * Purges the translation caches of all processors of the given virtual address
 * range.
 *
 * Note:
 * 	- cpu_vm_mask is a bit mask that indicates which cpus have loaded the context.
 * 	- cpu_vm_mask is converted into a nodemask of the nodes containing the
 * 	  cpus in cpu_vm_mask.
 *	- if only one bit is set in cpu_vm_mask & it is the current cpu & the
 *	  process is purging its own virtual address range, then only the
 *	  local TLB needs to be flushed. This flushing can be done using
 *	  ptc.l. This is the common case & avoids the global spinlock.
 *	- if multiple cpus have loaded the context, then flushing has to be
 *	  done with ptc.g/MMRs under protection of the global ptc_lock.
 */

void
sn2_global_tlb_purge(struct mm_struct *mm, unsigned long start,
		     unsigned long end, unsigned long nbits)
{
	int i, ibegin, shub1, cnode, mynasid, cpu, lcpu = 0, nasid;
	int mymm = (mm == current->active_mm && mm == current->mm);
	int use_cpu_ptcga;
	volatile unsigned long *ptc0, *ptc1;
	unsigned long itc, itc2, flags, data0 = 0, data1 = 0, rr_value, old_rr = 0;
	short nasids[MAX_NUMNODES], nix;
	nodemask_t nodes_flushed;
	int active, max_active, deadlock, flush_opt = sn2_flush_opt;

	if (flush_opt > 2) {
		sn2_ipi_flush_all_tlb(mm);
		return;
	}

	nodes_clear(nodes_flushed);
	i = 0;

	for_each_cpu_mask(cpu, mm->cpu_vm_mask) {
		cnode = cpu_to_node(cpu);
		node_set(cnode, nodes_flushed);
		lcpu = cpu;
		i++;
	}

	if (i == 0)
		return;

	preempt_disable();

	if (likely(i == 1 && lcpu == smp_processor_id() && mymm)) {
		do {
			ia64_ptcl(start, nbits << 2);
			start += (1UL << nbits);
		} while (start < end);
		ia64_srlz_i();
		__get_cpu_var(ptcstats).ptc_l++;
		preempt_enable();
		return;
	}

	if (atomic_read(&mm->mm_users) == 1 && mymm) {
		flush_tlb_mm(mm);
		__get_cpu_var(ptcstats).change_rid++;
		preempt_enable();
		return;
	}

	if (flush_opt == 2) {
		sn2_ipi_flush_all_tlb(mm);
		preempt_enable();
		return;
	}

	itc = ia64_get_itc();
	nix = 0;
	for_each_node_mask(cnode, nodes_flushed)
		nasids[nix++] = cnodeid_to_nasid(cnode);

	rr_value = (mm->context << 3) | REGION_NUMBER(start);

	shub1 = is_shub1();
	if (shub1) {
		data0 = (1UL << SH1_PTC_0_A_SHFT) |
		    	(nbits << SH1_PTC_0_PS_SHFT) |
			(rr_value << SH1_PTC_0_RID_SHFT) |
		    	(1UL << SH1_PTC_0_START_SHFT);
		ptc0 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH1_PTC_0);
		ptc1 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH1_PTC_1);
	} else {
		data0 = (1UL << SH2_PTC_A_SHFT) |
			(nbits << SH2_PTC_PS_SHFT) |
		    	(1UL << SH2_PTC_START_SHFT);
		ptc0 = (long *)GLOBAL_MMR_PHYS_ADDR(0, SH2_PTC + 
			(rr_value << SH2_PTC_RID_SHFT));
		ptc1 = NULL;
	}
	

	mynasid = get_nasid();
	use_cpu_ptcga = local_node_uses_ptc_ga(shub1);
	max_active = max_active_pio(shub1);

	itc = ia64_get_itc();
	spin_lock_irqsave(PTC_LOCK(shub1), flags);
	itc2 = ia64_get_itc();

	__get_cpu_var(ptcstats).lock_itc_clocks += itc2 - itc;
	__get_cpu_var(ptcstats).shub_ptc_flushes++;
	__get_cpu_var(ptcstats).nodes_flushed += nix;
	if (!mymm)
		 __get_cpu_var(ptcstats).shub_ptc_flushes_not_my_mm++;

	if (use_cpu_ptcga && !mymm) {
		old_rr = ia64_get_rr(start);
		ia64_set_rr(start, (old_rr & 0xff) | (rr_value << 8));
		ia64_srlz_d();
	}

	wait_piowc();
	do {
		if (shub1)
			data1 = start | (1UL << SH1_PTC_1_START_SHFT);
		else
			data0 = (data0 & ~SH2_PTC_ADDR_MASK) | (start & SH2_PTC_ADDR_MASK);
		deadlock = 0;
		active = 0;
		for (ibegin = 0, i = 0; i < nix; i++) {
			nasid = nasids[i];
			if (use_cpu_ptcga && unlikely(nasid == mynasid)) {
				ia64_ptcga(start, nbits << 2);
				ia64_srlz_i();
			} else {
				ptc0 = CHANGE_NASID(nasid, ptc0);
				if (ptc1)
					ptc1 = CHANGE_NASID(nasid, ptc1);
				pio_atomic_phys_write_mmrs(ptc0, data0, ptc1, data1);
				active++;
			}
			if (active >= max_active || i == (nix - 1)) {