linux_dsm_epyc7002/arch/mips/mm/tlb-r4k.c
David Daney 6dd9344cfc MIPS: Implement Read Inhibit/eXecute Inhibit
The SmartMIPS ASE specifies how Read Inhibit (RI) and eXecute Inhibit
(XI) bits in the page tables work.  The upper two bits of EntryLo{0,1}
are RI and XI when the feature is enabled in the PageGrain register.
SmartMIPS only covers 32-bit systems.  Cavium Octeon+ extends this to
64-bit systems by continuing to place the RI and XI bits in the top of
EntryLo even when EntryLo is 64-bits wide.

Because we need to carry the RI and XI bits in the PTE, the layout of
the PTE is changed.  There is a two instruction overhead in the TLB
refill hot path to get the EntryLo bits into the proper position.
Also the TLB load exception has to probe the TLB to check if RI or XI
caused the exception.

Also of note is that the layout of the PTE bits is done at compile and
runtime rather than statically.  In the 32-bit case this allows for
the same number of PFN bits as before the patch as the _PAGE_HUGE is
not supported in 32-bit kernels (we have _PAGE_NO_EXEC and
_PAGE_NO_READ instead of _PAGE_READ and _PAGE_HUGE).

The patch is tested on Cavium Octeon+, but should also work on 32-bit
systems with the Smart-MIPS ASE.

Signed-off-by: David Daney <ddaney@caviumnetworks.com>
To: linux-mips@linux-mips.org
Patchwork: http://patchwork.linux-mips.org/patch/952/
Patchwork: http://patchwork.linux-mips.org/patch/956/
Patchwork: http://patchwork.linux-mips.org/patch/962/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
2010-02-27 12:53:26 +01:00

472 lines
11 KiB
C

/*
* 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) 1996 David S. Miller (dm@engr.sgi.com)
* Copyright (C) 1997, 1998, 1999, 2000 Ralf Baechle ralf@gnu.org
* Carsten Langgaard, carstenl@mips.com
* Copyright (C) 2002 MIPS Technologies, Inc. All rights reserved.
*/
#include <linux/init.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/mm.h>
#include <linux/hugetlb.h>
#include <asm/cpu.h>
#include <asm/bootinfo.h>
#include <asm/mmu_context.h>
#include <asm/pgtable.h>
#include <asm/system.h>
extern void build_tlb_refill_handler(void);
/*
* Make sure all entries differ. If they're not different
* MIPS32 will take revenge ...
*/
#define UNIQUE_ENTRYHI(idx) (CKSEG0 + ((idx) << (PAGE_SHIFT + 1)))
/* Atomicity and interruptability */
#ifdef CONFIG_MIPS_MT_SMTC
#include <asm/smtc.h>
#include <asm/mipsmtregs.h>
#define ENTER_CRITICAL(flags) \
{ \
unsigned int mvpflags; \
local_irq_save(flags);\
mvpflags = dvpe()
#define EXIT_CRITICAL(flags) \
evpe(mvpflags); \
local_irq_restore(flags); \
}
#else
#define ENTER_CRITICAL(flags) local_irq_save(flags)
#define EXIT_CRITICAL(flags) local_irq_restore(flags)
#endif /* CONFIG_MIPS_MT_SMTC */
#if defined(CONFIG_CPU_LOONGSON2)
/*
* LOONGSON2 has a 4 entry itlb which is a subset of dtlb,
* unfortrunately, itlb is not totally transparent to software.
*/
#define FLUSH_ITLB write_c0_diag(4);
#define FLUSH_ITLB_VM(vma) { if ((vma)->vm_flags & VM_EXEC) write_c0_diag(4); }
#else
#define FLUSH_ITLB
#define FLUSH_ITLB_VM(vma)
#endif
void local_flush_tlb_all(void)
{
unsigned long flags;
unsigned long old_ctx;
int entry;
ENTER_CRITICAL(flags);
/* Save old context and create impossible VPN2 value */
old_ctx = read_c0_entryhi();
write_c0_entrylo0(0);
write_c0_entrylo1(0);
entry = read_c0_wired();
/* Blast 'em all away. */
while (entry < current_cpu_data.tlbsize) {
/* Make sure all entries differ. */
write_c0_entryhi(UNIQUE_ENTRYHI(entry));
write_c0_index(entry);
mtc0_tlbw_hazard();
tlb_write_indexed();
entry++;
}
tlbw_use_hazard();
write_c0_entryhi(old_ctx);
FLUSH_ITLB;
EXIT_CRITICAL(flags);
}
/* All entries common to a mm share an asid. To effectively flush
these entries, we just bump the asid. */
void local_flush_tlb_mm(struct mm_struct *mm)
{
int cpu;
preempt_disable();
cpu = smp_processor_id();
if (cpu_context(cpu, mm) != 0) {
drop_mmu_context(mm, cpu);
}
preempt_enable();
}
void local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
unsigned long end)
{
struct mm_struct *mm = vma->vm_mm;
int cpu = smp_processor_id();
if (cpu_context(cpu, mm) != 0) {
unsigned long size, flags;
ENTER_CRITICAL(flags);
size = (end - start + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
size = (size + 1) >> 1;
if (size <= current_cpu_data.tlbsize/2) {
int oldpid = read_c0_entryhi();
int newpid = cpu_asid(cpu, mm);
start &= (PAGE_MASK << 1);
end += ((PAGE_SIZE << 1) - 1);
end &= (PAGE_MASK << 1);
while (start < end) {
int idx;
write_c0_entryhi(start | newpid);
start += (PAGE_SIZE << 1);
mtc0_tlbw_hazard();
tlb_probe();
tlb_probe_hazard();
idx = read_c0_index();
write_c0_entrylo0(0);
write_c0_entrylo1(0);
if (idx < 0)
continue;
/* Make sure all entries differ. */
write_c0_entryhi(UNIQUE_ENTRYHI(idx));
mtc0_tlbw_hazard();
tlb_write_indexed();
}
tlbw_use_hazard();
write_c0_entryhi(oldpid);
} else {
drop_mmu_context(mm, cpu);
}
FLUSH_ITLB;
EXIT_CRITICAL(flags);
}
}
void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
unsigned long size, flags;
ENTER_CRITICAL(flags);
size = (end - start + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
size = (size + 1) >> 1;
if (size <= current_cpu_data.tlbsize / 2) {
int pid = read_c0_entryhi();
start &= (PAGE_MASK << 1);
end += ((PAGE_SIZE << 1) - 1);
end &= (PAGE_MASK << 1);
while (start < end) {
int idx;
write_c0_entryhi(start);
start += (PAGE_SIZE << 1);
mtc0_tlbw_hazard();
tlb_probe();
tlb_probe_hazard();
idx = read_c0_index();
write_c0_entrylo0(0);
write_c0_entrylo1(0);
if (idx < 0)
continue;
/* Make sure all entries differ. */
write_c0_entryhi(UNIQUE_ENTRYHI(idx));
mtc0_tlbw_hazard();
tlb_write_indexed();
}
tlbw_use_hazard();
write_c0_entryhi(pid);
} else {
local_flush_tlb_all();
}
FLUSH_ITLB;
EXIT_CRITICAL(flags);
}
void local_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
{
int cpu = smp_processor_id();
if (cpu_context(cpu, vma->vm_mm) != 0) {
unsigned long flags;
int oldpid, newpid, idx;
newpid = cpu_asid(cpu, vma->vm_mm);
page &= (PAGE_MASK << 1);
ENTER_CRITICAL(flags);
oldpid = read_c0_entryhi();
write_c0_entryhi(page | newpid);
mtc0_tlbw_hazard();
tlb_probe();
tlb_probe_hazard();
idx = read_c0_index();
write_c0_entrylo0(0);
write_c0_entrylo1(0);
if (idx < 0)
goto finish;
/* Make sure all entries differ. */
write_c0_entryhi(UNIQUE_ENTRYHI(idx));
mtc0_tlbw_hazard();
tlb_write_indexed();
tlbw_use_hazard();
finish:
write_c0_entryhi(oldpid);
FLUSH_ITLB_VM(vma);
EXIT_CRITICAL(flags);
}
}
/*
* This one is only used for pages with the global bit set so we don't care
* much about the ASID.
*/
void local_flush_tlb_one(unsigned long page)
{
unsigned long flags;
int oldpid, idx;
ENTER_CRITICAL(flags);
oldpid = read_c0_entryhi();
page &= (PAGE_MASK << 1);
write_c0_entryhi(page);
mtc0_tlbw_hazard();
tlb_probe();
tlb_probe_hazard();
idx = read_c0_index();
write_c0_entrylo0(0);
write_c0_entrylo1(0);
if (idx >= 0) {
/* Make sure all entries differ. */
write_c0_entryhi(UNIQUE_ENTRYHI(idx));
mtc0_tlbw_hazard();
tlb_write_indexed();
tlbw_use_hazard();
}
write_c0_entryhi(oldpid);
FLUSH_ITLB;
EXIT_CRITICAL(flags);
}
/*
* We will need multiple versions of update_mmu_cache(), one that just
* updates the TLB with the new pte(s), and another which also checks
* for the R4k "end of page" hardware bug and does the needy.
*/
void __update_tlb(struct vm_area_struct * vma, unsigned long address, pte_t pte)
{
unsigned long flags;
pgd_t *pgdp;
pud_t *pudp;
pmd_t *pmdp;
pte_t *ptep;
int idx, pid;
/*
* Handle debugger faulting in for debugee.
*/
if (current->active_mm != vma->vm_mm)
return;
ENTER_CRITICAL(flags);
pid = read_c0_entryhi() & ASID_MASK;
address &= (PAGE_MASK << 1);
write_c0_entryhi(address | pid);
pgdp = pgd_offset(vma->vm_mm, address);
mtc0_tlbw_hazard();
tlb_probe();
tlb_probe_hazard();
pudp = pud_offset(pgdp, address);
pmdp = pmd_offset(pudp, address);
idx = read_c0_index();
#ifdef CONFIG_HUGETLB_PAGE
/* this could be a huge page */
if (pmd_huge(*pmdp)) {
unsigned long lo;
write_c0_pagemask(PM_HUGE_MASK);
ptep = (pte_t *)pmdp;
lo = pte_to_entrylo(pte_val(*ptep));
write_c0_entrylo0(lo);
write_c0_entrylo1(lo + (HPAGE_SIZE >> 7));
mtc0_tlbw_hazard();
if (idx < 0)
tlb_write_random();
else
tlb_write_indexed();
write_c0_pagemask(PM_DEFAULT_MASK);
} else
#endif
{
ptep = pte_offset_map(pmdp, address);
#if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32)
write_c0_entrylo0(ptep->pte_high);
ptep++;
write_c0_entrylo1(ptep->pte_high);
#else
write_c0_entrylo0(pte_to_entrylo(pte_val(*ptep++)));
write_c0_entrylo1(pte_to_entrylo(pte_val(*ptep)));
#endif
mtc0_tlbw_hazard();
if (idx < 0)
tlb_write_random();
else
tlb_write_indexed();
}
tlbw_use_hazard();
FLUSH_ITLB_VM(vma);
EXIT_CRITICAL(flags);
}
void __init add_wired_entry(unsigned long entrylo0, unsigned long entrylo1,
unsigned long entryhi, unsigned long pagemask)
{
unsigned long flags;
unsigned long wired;
unsigned long old_pagemask;
unsigned long old_ctx;
ENTER_CRITICAL(flags);
/* Save old context and create impossible VPN2 value */
old_ctx = read_c0_entryhi();
old_pagemask = read_c0_pagemask();
wired = read_c0_wired();
write_c0_wired(wired + 1);
write_c0_index(wired);
tlbw_use_hazard(); /* What is the hazard here? */
write_c0_pagemask(pagemask);
write_c0_entryhi(entryhi);
write_c0_entrylo0(entrylo0);
write_c0_entrylo1(entrylo1);
mtc0_tlbw_hazard();
tlb_write_indexed();
tlbw_use_hazard();
write_c0_entryhi(old_ctx);
tlbw_use_hazard(); /* What is the hazard here? */
write_c0_pagemask(old_pagemask);
local_flush_tlb_all();
EXIT_CRITICAL(flags);
}
/*
* Used for loading TLB entries before trap_init() has started, when we
* don't actually want to add a wired entry which remains throughout the
* lifetime of the system
*/
static int temp_tlb_entry __cpuinitdata;
__init int add_temporary_entry(unsigned long entrylo0, unsigned long entrylo1,
unsigned long entryhi, unsigned long pagemask)
{
int ret = 0;
unsigned long flags;
unsigned long wired;
unsigned long old_pagemask;
unsigned long old_ctx;
ENTER_CRITICAL(flags);
/* Save old context and create impossible VPN2 value */
old_ctx = read_c0_entryhi();
old_pagemask = read_c0_pagemask();
wired = read_c0_wired();
if (--temp_tlb_entry < wired) {
printk(KERN_WARNING
"No TLB space left for add_temporary_entry\n");
ret = -ENOSPC;
goto out;
}
write_c0_index(temp_tlb_entry);
write_c0_pagemask(pagemask);
write_c0_entryhi(entryhi);
write_c0_entrylo0(entrylo0);
write_c0_entrylo1(entrylo1);
mtc0_tlbw_hazard();
tlb_write_indexed();
tlbw_use_hazard();
write_c0_entryhi(old_ctx);
write_c0_pagemask(old_pagemask);
out:
EXIT_CRITICAL(flags);
return ret;
}
static int __cpuinitdata ntlb;
static int __init set_ntlb(char *str)
{
get_option(&str, &ntlb);
return 1;
}
__setup("ntlb=", set_ntlb);
void __cpuinit tlb_init(void)
{
/*
* You should never change this register:
* - On R4600 1.7 the tlbp never hits for pages smaller than
* the value in the c0_pagemask register.
* - The entire mm handling assumes the c0_pagemask register to
* be set to fixed-size pages.
*/
write_c0_pagemask(PM_DEFAULT_MASK);
write_c0_wired(0);
if (current_cpu_type() == CPU_R10000 ||
current_cpu_type() == CPU_R12000 ||
current_cpu_type() == CPU_R14000)
write_c0_framemask(0);
if (kernel_uses_smartmips_rixi) {
/*
* Enable the no read, no exec bits, and enable large virtual
* address.
*/
u32 pg = PG_RIE | PG_XIE;
#ifdef CONFIG_64BIT
pg |= PG_ELPA;
#endif
write_c0_pagegrain(pg);
}
temp_tlb_entry = current_cpu_data.tlbsize - 1;
/* From this point on the ARC firmware is dead. */
local_flush_tlb_all();
/* Did I tell you that ARC SUCKS? */
if (ntlb) {
if (ntlb > 1 && ntlb <= current_cpu_data.tlbsize) {
int wired = current_cpu_data.tlbsize - ntlb;
write_c0_wired(wired);
write_c0_index(wired-1);
printk("Restricting TLB to %d entries\n", ntlb);
} else
printk("Ignoring invalid argument ntlb=%d\n", ntlb);
}
build_tlb_refill_handler();
}