AcoSail/kernel_optimize_test
8
0
Fork 0
forked from luck/tmp_suning_uos_patched
Code Pull Requests Activity
bf72aeba2f
kernel_optimize_test/arch/powerpc/mm/hash_native_64.c
Benjamin Herrenschmidt c5cf0e30bf [PATCH] powerpc: Fix buglet with MMU hash management 
Our MMU hash management code would not set the "C" bit (changed bit) in
the hardware PTE when updating a RO PTE into a RW PTE. That would cause
the hardware to possibly to a write back to the hash table to set it on
the first store access, which in addition to being a performance issue,
might also hit a bug when running with native hash management (non-HV)
as our code is specifically optimized for the case where no write back
happens.

Thus there is a very small therocial window were a hash PTE can become
corrupted if that HPTE has just been upgraded to read write, a store
access happens on it, and that races with another processor evicting
that same slot. Since eviction (caused by an almost full hash) is
extremely rare, the bug is very unlikely to happen fortunately.

This fixes by allowing the updating of the protection bits in the native
hash handling to also set (but not clear) the "C" bit, and, in order to
also improve performances in the general case, by always setting that
bit on newly inserted hash PTE so that writeback really never happens.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-06-09 21:20:59 +10:00

535 lines
12 KiB
C
Raw Blame History

/*
* native hashtable management.
*
* SMP scalability work:
* Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
*
* 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_LOW
#include <linux/spinlock.h>
#include <linux/bitops.h>
#include <linux/threads.h>
#include <linux/smp.h>
#include <asm/abs_addr.h>
#include <asm/machdep.h>
#include <asm/mmu.h>
#include <asm/mmu_context.h>
#include <asm/pgtable.h>
#include <asm/tlbflush.h>
#include <asm/tlb.h>
#include <asm/cputable.h>
#include <asm/udbg.h>
#ifdef DEBUG_LOW
#define DBG_LOW(fmt...) udbg_printf(fmt)
#else
#define DBG_LOW(fmt...)
#endif
#define HPTE_LOCK_BIT 3
static DEFINE_SPINLOCK(native_tlbie_lock);
static inline void __tlbie(unsigned long va, unsigned int psize)
{
unsigned int penc;
/* clear top 16 bits, non SLS segment */
va &= ~(0xffffULL << 48);
switch (psize) {
case MMU_PAGE_4K:
va &= ~0xffful;
asm volatile("tlbie %0,0" : : "r" (va) : "memory");
break;
default:
penc = mmu_psize_defs[psize].penc;
va &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
va |= (0x7f >> (8 - penc)) << 12;
asm volatile("tlbie %0,1" : : "r" (va) : "memory");
break;
}
}
static inline void __tlbiel(unsigned long va, unsigned int psize)
{
unsigned int penc;
/* clear top 16 bits, non SLS segment */
va &= ~(0xffffULL << 48);
switch (psize) {
case MMU_PAGE_4K:
va &= ~0xffful;
asm volatile(".long 0x7c000224 | (%0 << 11) | (0 << 21)"
: : "r"(va) : "memory");
break;
default:
penc = mmu_psize_defs[psize].penc;
va &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
va |= (0x7f >> (8 - penc)) << 12;
asm volatile(".long 0x7c000224 | (%0 << 11) | (1 << 21)"
: : "r"(va) : "memory");
break;
}
}
static inline void tlbie(unsigned long va, int psize, int local)
{
unsigned int use_local = local && cpu_has_feature(CPU_FTR_TLBIEL);
int lock_tlbie = !cpu_has_feature(CPU_FTR_LOCKLESS_TLBIE);
if (use_local)
use_local = mmu_psize_defs[psize].tlbiel;
if (lock_tlbie && !use_local)
spin_lock(&native_tlbie_lock);
asm volatile("ptesync": : :"memory");
if (use_local) {
__tlbiel(va, psize);
asm volatile("ptesync": : :"memory");
} else {
__tlbie(va, psize);
asm volatile("eieio; tlbsync; ptesync": : :"memory");
}
if (lock_tlbie && !use_local)
spin_unlock(&native_tlbie_lock);
}
static inline void native_lock_hpte(hpte_t *hptep)
{
unsigned long *word = &hptep->v;
while (1) {
if (!test_and_set_bit(HPTE_LOCK_BIT, word))
break;
while(test_bit(HPTE_LOCK_BIT, word))
cpu_relax();
}
}
static inline void native_unlock_hpte(hpte_t *hptep)
{
unsigned long *word = &hptep->v;
asm volatile("lwsync":::"memory");
clear_bit(HPTE_LOCK_BIT, word);
}
long native_hpte_insert(unsigned long hpte_group, unsigned long va,
unsigned long pa, unsigned long rflags,
unsigned long vflags, int psize)
{
hpte_t *hptep = htab_address + hpte_group;
unsigned long hpte_v, hpte_r;
int i;
if (!(vflags & HPTE_V_BOLTED)) {
DBG_LOW(" insert(group=%lx, va=%016lx, pa=%016lx,"
" rflags=%lx, vflags=%lx, psize=%d)\n",
hpte_group, va, pa, rflags, vflags, psize);
}
for (i = 0; i < HPTES_PER_GROUP; i++) {
if (! (hptep->v & HPTE_V_VALID)) {
/* retry with lock held */
native_lock_hpte(hptep);
if (! (hptep->v & HPTE_V_VALID))
break;
native_unlock_hpte(hptep);
}
hptep++;
}
if (i == HPTES_PER_GROUP)
return -1;
hpte_v = hpte_encode_v(va, psize) | vflags | HPTE_V_VALID;
hpte_r = hpte_encode_r(pa, psize) | rflags;
if (!(vflags & HPTE_V_BOLTED)) {
DBG_LOW(" i=%x hpte_v=%016lx, hpte_r=%016lx\n",
i, hpte_v, hpte_r);
}
hptep->r = hpte_r;
/* Guarantee the second dword is visible before the valid bit */
__asm__ __volatile__ ("eieio" : : : "memory");
/*
* Now set the first dword including the valid bit
* NOTE: this also unlocks the hpte
*/
hptep->v = hpte_v;
__asm__ __volatile__ ("ptesync" : : : "memory");
return i | (!!(vflags & HPTE_V_SECONDARY) << 3);
}
static long native_hpte_remove(unsigned long hpte_group)
{
hpte_t *hptep;
int i;
int slot_offset;
unsigned long hpte_v;
DBG_LOW(" remove(group=%lx)\n", hpte_group);
/* pick a random entry to start at */
slot_offset = mftb() & 0x7;
for (i = 0; i < HPTES_PER_GROUP; i++) {
hptep = htab_address + hpte_group + slot_offset;
hpte_v = hptep->v;
if ((hpte_v & HPTE_V_VALID) && !(hpte_v & HPTE_V_BOLTED)) {
/* retry with lock held */
native_lock_hpte(hptep);
hpte_v = hptep->v;
if ((hpte_v & HPTE_V_VALID)
&& !(hpte_v & HPTE_V_BOLTED))
break;
native_unlock_hpte(hptep);
}
slot_offset++;
slot_offset &= 0x7;
}
if (i == HPTES_PER_GROUP)
return -1;
/* Invalidate the hpte. NOTE: this also unlocks it */
hptep->v = 0;
return i;
}
static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
unsigned long va, int psize, int local)
{
hpte_t *hptep = htab_address + slot;
unsigned long hpte_v, want_v;
int ret = 0;
want_v = hpte_encode_v(va, psize);
DBG_LOW(" update(va=%016lx, avpnv=%016lx, hash=%016lx, newpp=%x)",
va, want_v & HPTE_V_AVPN, slot, newpp);
native_lock_hpte(hptep);
hpte_v = hptep->v;
/* Even if we miss, we need to invalidate the TLB */
if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
DBG_LOW(" -> miss\n");
native_unlock_hpte(hptep);
ret = -1;
} else {
DBG_LOW(" -> hit\n");
/* Update the HPTE */
hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
(newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_C));
native_unlock_hpte(hptep);
}
/* Ensure it is out of the tlb too. */
tlbie(va, psize, local);
return ret;
}
static long native_hpte_find(unsigned long va, int psize)
{
hpte_t *hptep;
unsigned long hash;
unsigned long i, j;
long slot;
unsigned long want_v, hpte_v;
hash = hpt_hash(va, mmu_psize_defs[psize].shift);
want_v = hpte_encode_v(va, psize);
for (j = 0; j < 2; j++) {
slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
for (i = 0; i < HPTES_PER_GROUP; i++) {
hptep = htab_address + slot;
hpte_v = hptep->v;
if (HPTE_V_COMPARE(hpte_v, want_v)
&& (hpte_v & HPTE_V_VALID)
&& ( !!(hpte_v & HPTE_V_SECONDARY) == j)) {
/* HPTE matches */
if (j)
slot = -slot;
return slot;
}
++slot;
}
hash = ~hash;
}
return -1;
}
/*
* Update the page protection bits. Intended to be used to create
* guard pages for kernel data structures on pages which are bolted
* in the HPT. Assumes pages being operated on will not be stolen.
*
* No need to lock here because we should be the only user.
*/
static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
int psize)
{
unsigned long vsid, va;
long slot;
hpte_t *hptep;
vsid = get_kernel_vsid(ea);
va = (vsid << 28) | (ea & 0x0fffffff);
slot = native_hpte_find(va, psize);
if (slot == -1)
panic("could not find page to bolt\n");
hptep = htab_address + slot;
/* Update the HPTE */
hptep->r = (hptep->r & ~(HPTE_R_PP | HPTE_R_N)) |
(newpp & (HPTE_R_PP | HPTE_R_N));
/* Ensure it is out of the tlb too. */
tlbie(va, psize, 0);
}
static void native_hpte_invalidate(unsigned long slot, unsigned long va,
int psize, int local)
{
hpte_t *hptep = htab_address + slot;
unsigned long hpte_v;
unsigned long want_v;
unsigned long flags;
local_irq_save(flags);
DBG_LOW(" invalidate(va=%016lx, hash: %x)\n", va, slot);
want_v = hpte_encode_v(va, psize);
native_lock_hpte(hptep);
hpte_v = hptep->v;
/* Even if we miss, we need to invalidate the TLB */
if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
native_unlock_hpte(hptep);
else
/* Invalidate the hpte. NOTE: this also unlocks it */
hptep->v = 0;
/* Invalidate the TLB */
tlbie(va, psize, local);
local_irq_restore(flags);
}
/*
* XXX This need fixing based on page size. It's only used by
* native_hpte_clear() for now which needs fixing too so they
* make a good pair...
*/
static unsigned long slot2va(unsigned long hpte_v, unsigned long slot)
{
unsigned long avpn = HPTE_V_AVPN_VAL(hpte_v);
unsigned long va;
va = avpn << 23;
if (! (hpte_v & HPTE_V_LARGE)) {
unsigned long vpi, pteg;
pteg = slot / HPTES_PER_GROUP;
if (hpte_v & HPTE_V_SECONDARY)
pteg = ~pteg;
vpi = ((va >> 28) ^ pteg) & htab_hash_mask;
va |= vpi << PAGE_SHIFT;
}
return va;
}
/*
* clear all mappings on kexec. All cpus are in real mode (or they will
* be when they isi), and we are the only one left. We rely on our kernel
* mapping being 0xC0's and the hardware ignoring those two real bits.
*
* TODO: add batching support when enabled. remember, no dynamic memory here,
* athough there is the control page available...
*
* XXX FIXME: 4k only for now !
*/
static void native_hpte_clear(void)
{
unsigned long slot, slots, flags;
hpte_t *hptep = htab_address;
unsigned long hpte_v;
unsigned long pteg_count;
pteg_count = htab_hash_mask + 1;
local_irq_save(flags);
/* we take the tlbie lock and hold it. Some hardware will
* deadlock if we try to tlbie from two processors at once.
*/
spin_lock(&native_tlbie_lock);
slots = pteg_count * HPTES_PER_GROUP;
for (slot = 0; slot < slots; slot++, hptep++) {
/*
* we could lock the pte here, but we are the only cpu
* running, right? and for crash dump, we probably
* don't want to wait for a maybe bad cpu.
*/
hpte_v = hptep->v;
/*
* Call __tlbie() here rather than tlbie() since we
* already hold the native_tlbie_lock.
*/
if (hpte_v & HPTE_V_VALID) {
hptep->v = 0;
__tlbie(slot2va(hpte_v, slot), MMU_PAGE_4K);
}
}
asm volatile("eieio; tlbsync; ptesync":::"memory");
spin_unlock(&native_tlbie_lock);
local_irq_restore(flags);
}
/*
* Batched hash table flush, we batch the tlbie's to avoid taking/releasing
* the lock all the time
*/
static void native_flush_hash_range(unsigned long number, int local)
{
unsigned long va, hash, index, hidx, shift, slot;
hpte_t *hptep;
unsigned long hpte_v;
unsigned long want_v;
unsigned long flags;
real_pte_t pte;
struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
unsigned long psize = batch->psize;
int i;
local_irq_save(flags);
for (i = 0; i < number; i++) {
va = batch->vaddr[i];
pte = batch->pte[i];
pte_iterate_hashed_subpages(pte, psize, va, index, shift) {
hash = hpt_hash(va, shift);
hidx = __rpte_to_hidx(pte, index);
if (hidx & _PTEIDX_SECONDARY)
hash = ~hash;
slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
slot += hidx & _PTEIDX_GROUP_IX;
hptep = htab_address + slot;
want_v = hpte_encode_v(va, psize);
native_lock_hpte(hptep);
hpte_v = hptep->v;
if (!HPTE_V_COMPARE(hpte_v, want_v) ||
!(hpte_v & HPTE_V_VALID))
native_unlock_hpte(hptep);
else
hptep->v = 0;
} pte_iterate_hashed_end();
}
if (cpu_has_feature(CPU_FTR_TLBIEL) &&
mmu_psize_defs[psize].tlbiel && local) {
asm volatile("ptesync":::"memory");
for (i = 0; i < number; i++) {
va = batch->vaddr[i];
pte = batch->pte[i];
pte_iterate_hashed_subpages(pte, psize, va, index,
shift) {
__tlbiel(va, psize);
} pte_iterate_hashed_end();
}
asm volatile("ptesync":::"memory");
} else {
int lock_tlbie = !cpu_has_feature(CPU_FTR_LOCKLESS_TLBIE);
if (lock_tlbie)
spin_lock(&native_tlbie_lock);
asm volatile("ptesync":::"memory");
for (i = 0; i < number; i++) {
va = batch->vaddr[i];
pte = batch->pte[i];
pte_iterate_hashed_subpages(pte, psize, va, index,
shift) {
__tlbie(va, psize);
} pte_iterate_hashed_end();
}
asm volatile("eieio; tlbsync; ptesync":::"memory");
if (lock_tlbie)
spin_unlock(&native_tlbie_lock);
}
local_irq_restore(flags);
}
#ifdef CONFIG_PPC_PSERIES
/* Disable TLB batching on nighthawk */
static inline int tlb_batching_enabled(void)
{
struct device_node *root = of_find_node_by_path("/");
int enabled = 1;
if (root) {
const char *model = get_property(root, "model", NULL);
if (model && !strcmp(model, "IBM,9076-N81"))
enabled = 0;
of_node_put(root);
}
return enabled;
}
#else
static inline int tlb_batching_enabled(void)
{
return 1;
}
#endif
void hpte_init_native(void)
{
ppc_md.hpte_invalidate = native_hpte_invalidate;
ppc_md.hpte_updatepp = native_hpte_updatepp;
ppc_md.hpte_updateboltedpp = native_hpte_updateboltedpp;
ppc_md.hpte_insert = native_hpte_insert;
ppc_md.hpte_remove = native_hpte_remove;
ppc_md.hpte_clear_all = native_hpte_clear;
if (tlb_batching_enabled())
ppc_md.flush_hash_range = native_flush_hash_range;
htab_finish_init();
}
View Git Blame Copy Permalink