Release 4.14 arch/powerpc/mm/mmu_context_nohash.c
/*
* This file contains the routines for handling the MMU on those
* PowerPC implementations where the MMU is not using the hash
* table, such as 8xx, 4xx, BookE's etc...
*
* Copyright 2008 Ben Herrenschmidt <benh@kernel.crashing.org>
* IBM Corp.
*
* Derived from previous arch/powerpc/mm/mmu_context.c
* and arch/powerpc/include/asm/mmu_context.h
*
* 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.
*
* TODO:
*
* - The global context lock will not scale very well
* - The maps should be dynamically allocated to allow for processors
* that support more PID bits at runtime
* - Implement flush_tlb_mm() by making the context stale and picking
* a new one
* - More aggressively clear stale map bits and maybe find some way to
* also clear mm->cpu_vm_mask bits when processes are migrated
*/
//#define DEBUG_MAP_CONSISTENCY
//#define DEBUG_CLAMP_LAST_CONTEXT 31
//#define DEBUG_HARDER
/* We don't use DEBUG because it tends to be compiled in always nowadays
* and this would generate way too much output
*/
#ifdef DEBUG_HARDER
#define pr_hard(args...) printk(KERN_DEBUG args)
#define pr_hardcont(args...) printk(KERN_CONT args)
#else
#define pr_hard(args...) do { } while(0)
#define pr_hardcont(args...) do { } while(0)
#endif
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/bootmem.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/slab.h>
#include <asm/mmu_context.h>
#include <asm/tlbflush.h>
#include "mmu_decl.h"
static unsigned int first_context, last_context;
static unsigned int next_context, nr_free_contexts;
static unsigned long *context_map;
static unsigned long *stale_map[NR_CPUS];
static struct mm_struct **context_mm;
static DEFINE_RAW_SPINLOCK(context_lock);
static bool no_selective_tlbil;
#define CTX_MAP_SIZE \
(sizeof(unsigned long) * (last_context / BITS_PER_LONG + 1))
/* Steal a context from a task that has one at the moment.
*
* This is used when we are running out of available PID numbers
* on the processors.
*
* This isn't an LRU system, it just frees up each context in
* turn (sort-of pseudo-random replacement :). This would be the
* place to implement an LRU scheme if anyone was motivated to do it.
* -- paulus
*
* For context stealing, we use a slightly different approach for
* SMP and UP. Basically, the UP one is simpler and doesn't use
* the stale map as we can just flush the local CPU
* -- benh
*/
#ifdef CONFIG_SMP
static unsigned int steal_context_smp(unsigned int id)
{
struct mm_struct *mm;
unsigned int cpu, max, i;
max = last_context - first_context;
/* Attempt to free next_context first and then loop until we manage */
while (max--) {
/* Pick up the victim mm */
mm = context_mm[id];
/* We have a candidate victim, check if it's active, on SMP
* we cannot steal active contexts
*/
if (mm->context.active) {
id++;
if (id > last_context)
id = first_context;
continue;
}
pr_hardcont(" | steal %d from 0x%p", id, mm);
/* Mark this mm has having no context anymore */
mm->context.id = MMU_NO_CONTEXT;
/* Mark it stale on all CPUs that used this mm. For threaded
* implementations, we set it on all threads on each core
* represented in the mask. A future implementation will use
* a core map instead but this will do for now.
*/
for_each_cpu(cpu, mm_cpumask(mm)) {
for (i = cpu_first_thread_sibling(cpu);
i <= cpu_last_thread_sibling(cpu); i++) {
if (stale_map[i])
__set_bit(id, stale_map[i]);
}
cpu = i - 1;
}
return id;
}
/* This will happen if you have more CPUs than available contexts,
* all we can do here is wait a bit and try again
*/
raw_spin_unlock(&context_lock);
cpu_relax();
raw_spin_lock(&context_lock);
/* This will cause the caller to try again */
return MMU_NO_CONTEXT;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 150 | 90.91% | 5 | 55.56% |
| Scott Wood | 9 | 5.45% | 1 | 11.11% |
| Rusty Russell | 2 | 1.21% | 1 | 11.11% |
| Vaidyanathan Srinivasan | 2 | 1.21% | 1 | 11.11% |
| Thomas Gleixner | 2 | 1.21% | 1 | 11.11% |
| Total | 165 | 100.00% | 9 | 100.00% |
#endif /* CONFIG_SMP */
static unsigned int steal_all_contexts(void)
{
struct mm_struct *mm;
int cpu = smp_processor_id();
unsigned int id;
for (id = first_context; id <= last_context; id++) {
/* Pick up the victim mm */
mm = context_mm[id];
pr_hardcont(" | steal %d from 0x%p", id, mm);
/* Mark this mm as having no context anymore */
mm->context.id = MMU_NO_CONTEXT;
if (id != first_context) {
context_mm[id] = NULL;
__clear_bit(id, context_map);
#ifdef DEBUG_MAP_CONSISTENCY
mm->context.active = 0;
#endif
}
__clear_bit(id, stale_map[cpu]);
}
/* Flush the TLB for all contexts (not to be used on SMP) */
_tlbil_all();
nr_free_contexts = last_context - first_context;
return first_context;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Christophe Leroy | 123 | 100.00% | 1 | 100.00% |
| Total | 123 | 100.00% | 1 | 100.00% |
/* Note that this will also be called on SMP if all other CPUs are
* offlined, which means that it may be called for cpu != 0. For
* this to work, we somewhat assume that CPUs that are onlined
* come up with a fully clean TLB (or are cleaned when offlined)
*/
static unsigned int steal_context_up(unsigned int id)
{
struct mm_struct *mm;
int cpu = smp_processor_id();
/* Pick up the victim mm */
mm = context_mm[id];
pr_hardcont(" | steal %d from 0x%p", id, mm);
/* Flush the TLB for that context */
local_flush_tlb_mm(mm);
/* Mark this mm has having no context anymore */
mm->context.id = MMU_NO_CONTEXT;
/* XXX This clear should ultimately be part of local_flush_tlb_mm */
__clear_bit(id, stale_map[cpu]);
return id;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 62 | 91.18% | 3 | 75.00% |
| Hideo Saito | 6 | 8.82% | 1 | 25.00% |
| Total | 68 | 100.00% | 4 | 100.00% |
#ifdef DEBUG_MAP_CONSISTENCY
static void context_check_map(void)
{
unsigned int id, nrf, nact;
nrf = nact = 0;
for (id = first_context; id <= last_context; id++) {
int used = test_bit(id, context_map);
if (!used)
nrf++;
if (used != (context_mm[id] != NULL))
pr_err("MMU: Context %d is %s and MM is %p !\n",
id, used ? "used" : "free", context_mm[id]);
if (context_mm[id] != NULL)
nact += context_mm[id]->context.active;
}
if (nrf != nr_free_contexts) {
pr_err("MMU: Free context count out of sync ! (%d vs %d)\n",
nr_free_contexts, nrf);
nr_free_contexts = nrf;
}
if (nact > num_online_cpus())
pr_err("MMU: More active contexts than CPUs ! (%d vs %d)\n",
nact, num_online_cpus());
if (first_context > 0 && !test_bit(0, context_map))
pr_err("MMU: Context 0 has been freed !!!\n");
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 163 | 100.00% | 2 | 100.00% |
| Total | 163 | 100.00% | 2 | 100.00% |
#else
static void context_check_map(void) { }
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 7 | 100.00% | 1 | 100.00% |
| Total | 7 | 100.00% | 1 | 100.00% |
#endif
void switch_mmu_context(struct mm_struct *prev, struct mm_struct *next,
struct task_struct *tsk)
{
unsigned int i, id, cpu = smp_processor_id();
unsigned long *map;
/* No lockless fast path .. yet */
raw_spin_lock(&context_lock);
pr_hard("[%d] activating context for mm @%p, active=%d, id=%d",
cpu, next, next->context.active, next->context.id);
#ifdef CONFIG_SMP
/* Mark us active and the previous one not anymore */
next->context.active++;
if (prev) {
pr_hardcont(" (old=0x%p a=%d)", prev, prev->context.active);
WARN_ON(prev->context.active < 1);
prev->context.active--;
}
again:
#endif /* CONFIG_SMP */
/* If we already have a valid assigned context, skip all that */
id = next->context.id;
if (likely(id != MMU_NO_CONTEXT)) {
#ifdef DEBUG_MAP_CONSISTENCY
if (context_mm[id] != next)
pr_err("MMU: mm 0x%p has id %d but context_mm[%d] says 0x%p\n",
next, id, id, context_mm[id]);
#endif
goto ctxt_ok;
}
/* We really don't have a context, let's try to acquire one */
id = next_context;
if (id > last_context)
id = first_context;
map = context_map;
/* No more free contexts, let's try to steal one */
if (nr_free_contexts == 0) {
#ifdef CONFIG_SMP
if (num_online_cpus() > 1) {
id = steal_context_smp(id);
if (id == MMU_NO_CONTEXT)
goto again;
goto stolen;
}
#endif /* CONFIG_SMP */
if (no_selective_tlbil)
id = steal_all_contexts();
else
id = steal_context_up(id);
goto stolen;
}
nr_free_contexts--;
/* We know there's at least one free context, try to find it */
while (__test_and_set_bit(id, map)) {
id = find_next_zero_bit(map, last_context+1, id);
if (id > last_context)
id = first_context;
}
stolen:
next_context = id + 1;
context_mm[id] = next;
next->context.id = id;
pr_hardcont(" | new id=%d,nrf=%d", id, nr_free_contexts);
context_check_map();
ctxt_ok:
/* If that context got marked stale on this CPU, then flush the
* local TLB for it and unmark it before we use it
*/
if (test_bit(id, stale_map[cpu])) {
pr_hardcont(" | stale flush %d [%d..%d]",
id, cpu_first_thread_sibling(cpu),
cpu_last_thread_sibling(cpu));
local_flush_tlb_mm(next);
/* XXX This clear should ultimately be part of local_flush_tlb_mm */
for (i = cpu_first_thread_sibling(cpu);
i <= cpu_last_thread_sibling(cpu); i++) {
if (stale_map[i])
__clear_bit(id, stale_map[i]);
}
}
/* Flick the MMU and release lock */
pr_hardcont(" -> %d\n", id);
set_context(id, next->pgd);
raw_spin_unlock(&context_lock);
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 387 | 90.85% | 4 | 36.36% |
| Kumar Gala | 11 | 2.58% | 2 | 18.18% |
| Christophe Leroy | 10 | 2.35% | 1 | 9.09% |
| Scott Wood | 7 | 1.64% | 1 | 9.09% |
| Aneesh Kumar K.V | 5 | 1.17% | 1 | 9.09% |
| Vaidyanathan Srinivasan | 4 | 0.94% | 1 | 9.09% |
| Thomas Gleixner | 2 | 0.47% | 1 | 9.09% |
| Total | 426 | 100.00% | 11 | 100.00% |
/*
* Set up the context for a new address space.
*/
int init_new_context(struct task_struct *t, struct mm_struct *mm)
{
pr_hard("initing context for mm @%p\n", mm);
mm->context.id = MMU_NO_CONTEXT;
mm->context.active = 0;
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 41 | 100.00% | 3 | 100.00% |
| Total | 41 | 100.00% | 3 | 100.00% |
/*
* We're finished using the context for an address space.
*/
void destroy_context(struct mm_struct *mm)
{
unsigned long flags;
unsigned int id;
if (mm->context.id == MMU_NO_CONTEXT)
return;
WARN_ON(mm->context.active != 0);
raw_spin_lock_irqsave(&context_lock, flags);
id = mm->context.id;
if (id != MMU_NO_CONTEXT) {
__clear_bit(id, context_map);
mm->context.id = MMU_NO_CONTEXT;
#ifdef DEBUG_MAP_CONSISTENCY
mm->context.active = 0;
#endif
context_mm[id] = NULL;
nr_free_contexts++;
}
raw_spin_unlock_irqrestore(&context_lock, flags);
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 108 | 98.18% | 4 | 80.00% |
| Thomas Gleixner | 2 | 1.82% | 1 | 20.00% |
| Total | 110 | 100.00% | 5 | 100.00% |
#ifdef CONFIG_SMP
static int mmu_ctx_cpu_prepare(unsigned int cpu)
{
/* We don't touch CPU 0 map, it's allocated at aboot and kept
* around forever
*/
if (cpu == boot_cpuid)
return 0;
pr_devel("MMU: Allocating stale context map for CPU %d\n", cpu);
stale_map[cpu] = kzalloc(CTX_MAP_SIZE, GFP_KERNEL);
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 32 | 76.19% | 1 | 25.00% |
| Sebastian Andrzej Siewior | 8 | 19.05% | 1 | 25.00% |
| Matthew McClintock | 1 | 2.38% | 1 | 25.00% |
| Michael Ellerman | 1 | 2.38% | 1 | 25.00% |
| Total | 42 | 100.00% | 4 | 100.00% |
static int mmu_ctx_cpu_dead(unsigned int cpu)
{
#ifdef CONFIG_HOTPLUG_CPU
if (cpu == boot_cpuid)
return 0;
pr_devel("MMU: Freeing stale context map for CPU %d\n", cpu);
kfree(stale_map[cpu]);
stale_map[cpu] = NULL;
/* We also clear the cpu_vm_mask bits of CPUs going away */
clear_tasks_mm_cpumask(cpu);
#endif
return 0;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 34 | 61.82% | 2 | 40.00% |
| Sebastian Andrzej Siewior | 19 | 34.55% | 1 | 20.00% |
| Anton Vorontsov | 1 | 1.82% | 1 | 20.00% |
| Michael Ellerman | 1 | 1.82% | 1 | 20.00% |
| Total | 55 | 100.00% | 5 | 100.00% |
#endif /* CONFIG_SMP */
/*
* Initialize the context management stuff.
*/
void __init mmu_context_init(void)
{
/* Mark init_mm as being active on all possible CPUs since
* we'll get called with prev == init_mm the first time
* we schedule on a given CPU
*/
init_mm.context.active = NR_CPUS;
/*
* The MPC8xx has only 16 contexts. We rotate through them on each
* task switch. A better way would be to keep track of tasks that
* own contexts, and implement an LRU usage. That way very active
* tasks don't always have to pay the TLB reload overhead. The
* kernel pages are mapped shared, so the kernel can run on behalf
* of any task that makes a kernel entry. Shared does not mean they
* are not protected, just that the ASID comparison is not performed.
* -- Dan
*
* The IBM4xx has 256 contexts, so we can just rotate through these
* as a way of "switching" contexts. If the TID of the TLB is zero,
* the PID/TID comparison is disabled, so we can use a TID of zero
* to represent all kernel pages as shared among all contexts.
* -- Dan
*
* The IBM 47x core supports 16-bit PIDs, thus 65535 contexts. We
* should normally never have to steal though the facility is
* present if needed.
* -- BenH
*/
if (mmu_has_feature(MMU_FTR_TYPE_8xx)) {
first_context = 0;
last_context = 15;
no_selective_tlbil = true;
} else if (mmu_has_feature(MMU_FTR_TYPE_47x)) {
first_context = 1;
last_context = 65535;
no_selective_tlbil = false;
} else {
first_context = 1;
last_context = 255;
no_selective_tlbil = false;
}
#ifdef DEBUG_CLAMP_LAST_CONTEXT
last_context = DEBUG_CLAMP_LAST_CONTEXT;
#endif
/*
* Allocate the maps used by context management
*/
context_map = memblock_virt_alloc(CTX_MAP_SIZE, 0);
context_mm = memblock_virt_alloc(sizeof(void *) * (last_context + 1), 0);
#ifndef CONFIG_SMP
stale_map[0] = memblock_virt_alloc(CTX_MAP_SIZE, 0);
#else
stale_map[boot_cpuid] = memblock_virt_alloc(CTX_MAP_SIZE, 0);
cpuhp_setup_state_nocalls(CPUHP_POWERPC_MMU_CTX_PREPARE,
"powerpc/mmu/ctx:prepare",
mmu_ctx_cpu_prepare, mmu_ctx_cpu_dead);
#endif
printk(KERN_INFO
"MMU: Allocated %zu bytes of context maps for %d contexts\n",
2 * CTX_MAP_SIZE + (sizeof(void *) * (last_context + 1)),
last_context - first_context + 1);
/*
* Some processors have too few contexts to reserve one for
* init_mm, and require using context 0 for a normal task.
* Other processors reserve the use of context zero for the kernel.
* This code assumes first_context < 32.
*/
context_map[0] = (1 << first_context) - 1;
next_context = first_context;
nr_free_contexts = last_context - first_context + 1;
}
Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 147 | 69.34% | 4 | 44.44% |
| Dave Kleikamp | 19 | 8.96% | 1 | 11.11% |
| Matthew McClintock | 14 | 6.60% | 1 | 11.11% |
| Michael Ellerman | 12 | 5.66% | 1 | 11.11% |
| Christophe Leroy | 12 | 5.66% | 1 | 11.11% |
| Sebastian Andrzej Siewior | 8 | 3.77% | 1 | 11.11% |
| Total | 212 | 100.00% | 9 | 100.00% |
Overall Contributors
| Person | Tokens | Prop | Commits | CommitProp |
| Benjamin Herrenschmidt | 1272 | 81.33% | 8 | 33.33% |
| Christophe Leroy | 152 | 9.72% | 1 | 4.17% |
| Sebastian Andrzej Siewior | 35 | 2.24% | 1 | 4.17% |
| Dave Kleikamp | 19 | 1.21% | 1 | 4.17% |
| Scott Wood | 16 | 1.02% | 1 | 4.17% |
| Matthew McClintock | 15 | 0.96% | 1 | 4.17% |
| Michael Ellerman | 14 | 0.90% | 2 | 8.33% |
| Kumar Gala | 11 | 0.70% | 2 | 8.33% |
| Thomas Gleixner | 7 | 0.45% | 1 | 4.17% |
| Hideo Saito | 6 | 0.38% | 1 | 4.17% |
| Vaidyanathan Srinivasan | 6 | 0.38% | 1 | 4.17% |
| Aneesh Kumar K.V | 5 | 0.32% | 1 | 4.17% |
| Tejun Heo | 3 | 0.19% | 1 | 4.17% |
| Rusty Russell | 2 | 0.13% | 1 | 4.17% |
| Anton Vorontsov | 1 | 0.06% | 1 | 4.17% |
| Total | 1564 | 100.00% | 24 | 100.00% |
Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.