2005-04-16 18:20:36 -04:00
|
|
|
/* $Id: fault.c,v 1.59 2002/02/09 19:49:31 davem Exp $
|
|
|
|
* arch/sparc64/mm/fault.c: Page fault handlers for the 64-bit Sparc.
|
|
|
|
*
|
|
|
|
* Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
|
|
|
|
* Copyright (C) 1997, 1999 Jakub Jelinek (jj@ultra.linux.cz)
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <asm/head.h>
|
|
|
|
|
|
|
|
#include <linux/string.h>
|
|
|
|
#include <linux/types.h>
|
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/ptrace.h>
|
|
|
|
#include <linux/mman.h>
|
|
|
|
#include <linux/signal.h>
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/interrupt.h>
|
2005-09-06 18:19:30 -04:00
|
|
|
#include <linux/kprobes.h>
|
2006-07-22 04:12:09 -04:00
|
|
|
#include <linux/kallsyms.h>
|
2007-05-08 03:27:03 -04:00
|
|
|
#include <linux/kdebug.h>
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
#include <asm/page.h>
|
|
|
|
#include <asm/pgtable.h>
|
|
|
|
#include <asm/openprom.h>
|
|
|
|
#include <asm/oplib.h>
|
|
|
|
#include <asm/uaccess.h>
|
|
|
|
#include <asm/asi.h>
|
|
|
|
#include <asm/lsu.h>
|
|
|
|
#include <asm/sections.h>
|
[SPARC64]: Fix and re-enable dynamic TSB sizing.
This is good for up to %50 performance improvement of some test cases.
The problem has been the race conditions, and hopefully I've plugged
them all up here.
1) There was a serious race in switch_mm() wrt. lazy TLB
switching to and from kernel threads.
We could erroneously skip a tsb_context_switch() and thus
use a stale TSB across a TSB grow event.
There is a big comment now in that function describing
exactly how it can happen.
2) All code paths that do something with the TSB need to be
guarded with the mm->context.lock spinlock. This makes
page table flushing paths properly synchronize with both
TSB growing and TLB context changes.
3) TSB growing events are moved to the end of successful fault
processing. Previously it was in update_mmu_cache() but
that is deadlock prone. At the end of do_sparc64_fault()
we hold no spinlocks that could deadlock the TSB grow
sequence. We also have dropped the address space semaphore.
While we're here, add prefetching to the copy_tsb() routine
and put it in assembler into the tsb.S file. This piece of
code is quite time critical.
There are some small negative side effects to this code which
can be improved upon. In particular we grab the mm->context.lock
even for the tsb insert done by update_mmu_cache() now and that's
a bit excessive. We can get rid of that locking, and the same
lock taking in flush_tsb_user(), by disabling PSTATE_IE around
the whole operation including the capturing of the tsb pointer
and tsb_nentries value. That would work because anyone growing
the TSB won't free up the old TSB until all cpus respond to the
TSB change cross call.
I'm not quite so confident in that optimization to put it in
right now, but eventually we might be able to and the description
is here for reference.
This code seems very solid now. It passes several parallel GCC
bootstrap builds, and our favorite "nut cruncher" stress test which is
a full "make -j8192" build of a "make allmodconfig" kernel. That puts
about 256 processes on each cpu's run queue, makes lots of process cpu
migrations occur, causes lots of page table and TLB flushing activity,
incurs many context version number changes, and it swaps the machine
real far out to disk even though there is 16GB of ram on this test
system. :-)
Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-16 05:02:32 -05:00
|
|
|
#include <asm/mmu_context.h>
|
2005-04-16 18:20:36 -04:00
|
|
|
|
2006-06-26 03:25:27 -04:00
|
|
|
#ifdef CONFIG_KPROBES
|
2007-05-08 21:25:14 -04:00
|
|
|
static inline int notify_page_fault(struct pt_regs *regs)
|
2006-06-26 03:25:27 -04:00
|
|
|
{
|
2007-05-08 21:25:14 -04:00
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
/* kprobe_running() needs smp_processor_id() */
|
|
|
|
if (!user_mode(regs)) {
|
|
|
|
preempt_disable();
|
|
|
|
if (kprobe_running() && kprobe_fault_handler(regs, 0))
|
|
|
|
ret = 1;
|
|
|
|
preempt_enable();
|
|
|
|
}
|
|
|
|
return ret;
|
2006-06-26 03:25:27 -04:00
|
|
|
}
|
|
|
|
#else
|
2007-05-08 21:25:14 -04:00
|
|
|
static inline int notify_page_fault(struct pt_regs *regs)
|
2006-06-26 03:25:27 -04:00
|
|
|
{
|
2007-05-08 21:25:14 -04:00
|
|
|
return 0;
|
2006-06-26 03:25:27 -04:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-04-16 18:20:36 -04:00
|
|
|
/*
|
|
|
|
* To debug kernel to catch accesses to certain virtual/physical addresses.
|
|
|
|
* Mode = 0 selects physical watchpoints, mode = 1 selects virtual watchpoints.
|
|
|
|
* flags = VM_READ watches memread accesses, flags = VM_WRITE watches memwrite accesses.
|
|
|
|
* Caller passes in a 64bit aligned addr, with mask set to the bytes that need to be
|
|
|
|
* watched. This is only useful on a single cpu machine for now. After the watchpoint
|
|
|
|
* is detected, the process causing it will be killed, thus preventing an infinite loop.
|
|
|
|
*/
|
|
|
|
void set_brkpt(unsigned long addr, unsigned char mask, int flags, int mode)
|
|
|
|
{
|
|
|
|
unsigned long lsubits;
|
|
|
|
|
|
|
|
__asm__ __volatile__("ldxa [%%g0] %1, %0"
|
|
|
|
: "=r" (lsubits)
|
|
|
|
: "i" (ASI_LSU_CONTROL));
|
|
|
|
lsubits &= ~(LSU_CONTROL_PM | LSU_CONTROL_VM |
|
|
|
|
LSU_CONTROL_PR | LSU_CONTROL_VR |
|
|
|
|
LSU_CONTROL_PW | LSU_CONTROL_VW);
|
|
|
|
|
|
|
|
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
|
|
|
|
"membar #Sync"
|
|
|
|
: /* no outputs */
|
|
|
|
: "r" (addr), "r" (mode ? VIRT_WATCHPOINT : PHYS_WATCHPOINT),
|
|
|
|
"i" (ASI_DMMU));
|
|
|
|
|
|
|
|
lsubits |= ((unsigned long)mask << (mode ? 25 : 33));
|
|
|
|
if (flags & VM_READ)
|
|
|
|
lsubits |= (mode ? LSU_CONTROL_VR : LSU_CONTROL_PR);
|
|
|
|
if (flags & VM_WRITE)
|
|
|
|
lsubits |= (mode ? LSU_CONTROL_VW : LSU_CONTROL_PW);
|
|
|
|
__asm__ __volatile__("stxa %0, [%%g0] %1\n\t"
|
|
|
|
"membar #Sync"
|
|
|
|
: /* no outputs */
|
|
|
|
: "r" (lsubits), "i" (ASI_LSU_CONTROL)
|
|
|
|
: "memory");
|
|
|
|
}
|
|
|
|
|
2005-09-06 18:19:30 -04:00
|
|
|
static void __kprobes unhandled_fault(unsigned long address,
|
|
|
|
struct task_struct *tsk,
|
|
|
|
struct pt_regs *regs)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
if ((unsigned long) address < PAGE_SIZE) {
|
|
|
|
printk(KERN_ALERT "Unable to handle kernel NULL "
|
|
|
|
"pointer dereference\n");
|
|
|
|
} else {
|
|
|
|
printk(KERN_ALERT "Unable to handle kernel paging request "
|
|
|
|
"at virtual address %016lx\n", (unsigned long)address);
|
|
|
|
}
|
|
|
|
printk(KERN_ALERT "tsk->{mm,active_mm}->context = %016lx\n",
|
|
|
|
(tsk->mm ?
|
|
|
|
CTX_HWBITS(tsk->mm->context) :
|
|
|
|
CTX_HWBITS(tsk->active_mm->context)));
|
|
|
|
printk(KERN_ALERT "tsk->{mm,active_mm}->pgd = %016lx\n",
|
|
|
|
(tsk->mm ? (unsigned long) tsk->mm->pgd :
|
|
|
|
(unsigned long) tsk->active_mm->pgd));
|
|
|
|
die_if_kernel("Oops", regs);
|
|
|
|
}
|
|
|
|
|
2006-02-13 21:07:45 -05:00
|
|
|
static void bad_kernel_pc(struct pt_regs *regs, unsigned long vaddr)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
unsigned long *ksp;
|
|
|
|
|
|
|
|
printk(KERN_CRIT "OOPS: Bogus kernel PC [%016lx] in fault handler\n",
|
|
|
|
regs->tpc);
|
2006-07-22 04:12:09 -04:00
|
|
|
printk(KERN_CRIT "OOPS: RPC [%016lx]\n", regs->u_regs[15]);
|
|
|
|
print_symbol("RPC: <%s>\n", regs->u_regs[15]);
|
2006-02-13 21:07:45 -05:00
|
|
|
printk(KERN_CRIT "OOPS: Fault was to vaddr[%lx]\n", vaddr);
|
2005-04-16 18:20:36 -04:00
|
|
|
__asm__("mov %%sp, %0" : "=r" (ksp));
|
|
|
|
show_stack(current, ksp);
|
|
|
|
unhandled_fault(regs->tpc, current, regs);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We now make sure that mmap_sem is held in all paths that call
|
|
|
|
* this. Additionally, to prevent kswapd from ripping ptes from
|
|
|
|
* under us, raise interrupts around the time that we look at the
|
|
|
|
* pte, kswapd will have to wait to get his smp ipi response from
|
2005-11-08 13:00:55 -05:00
|
|
|
* us. vmtruncate likewise. This saves us having to get pte lock.
|
2005-04-16 18:20:36 -04:00
|
|
|
*/
|
|
|
|
static unsigned int get_user_insn(unsigned long tpc)
|
|
|
|
{
|
|
|
|
pgd_t *pgdp = pgd_offset(current->mm, tpc);
|
|
|
|
pud_t *pudp;
|
|
|
|
pmd_t *pmdp;
|
|
|
|
pte_t *ptep, pte;
|
|
|
|
unsigned long pa;
|
|
|
|
u32 insn = 0;
|
|
|
|
unsigned long pstate;
|
|
|
|
|
|
|
|
if (pgd_none(*pgdp))
|
|
|
|
goto outret;
|
|
|
|
pudp = pud_offset(pgdp, tpc);
|
|
|
|
if (pud_none(*pudp))
|
|
|
|
goto outret;
|
|
|
|
pmdp = pmd_offset(pudp, tpc);
|
|
|
|
if (pmd_none(*pmdp))
|
|
|
|
goto outret;
|
|
|
|
|
|
|
|
/* This disables preemption for us as well. */
|
|
|
|
__asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
|
|
|
|
__asm__ __volatile__("wrpr %0, %1, %%pstate"
|
|
|
|
: : "r" (pstate), "i" (PSTATE_IE));
|
|
|
|
ptep = pte_offset_map(pmdp, tpc);
|
|
|
|
pte = *ptep;
|
|
|
|
if (!pte_present(pte))
|
|
|
|
goto out;
|
|
|
|
|
2006-02-12 00:57:54 -05:00
|
|
|
pa = (pte_pfn(pte) << PAGE_SHIFT);
|
2005-04-16 18:20:36 -04:00
|
|
|
pa += (tpc & ~PAGE_MASK);
|
|
|
|
|
|
|
|
/* Use phys bypass so we don't pollute dtlb/dcache. */
|
|
|
|
__asm__ __volatile__("lduwa [%1] %2, %0"
|
|
|
|
: "=r" (insn)
|
|
|
|
: "r" (pa), "i" (ASI_PHYS_USE_EC));
|
|
|
|
|
|
|
|
out:
|
|
|
|
pte_unmap(ptep);
|
|
|
|
__asm__ __volatile__("wrpr %0, 0x0, %%pstate" : : "r" (pstate));
|
|
|
|
outret:
|
|
|
|
return insn;
|
|
|
|
}
|
|
|
|
|
|
|
|
extern unsigned long compute_effective_address(struct pt_regs *, unsigned int, unsigned int);
|
|
|
|
|
|
|
|
static void do_fault_siginfo(int code, int sig, struct pt_regs *regs,
|
|
|
|
unsigned int insn, int fault_code)
|
|
|
|
{
|
|
|
|
siginfo_t info;
|
|
|
|
|
|
|
|
info.si_code = code;
|
|
|
|
info.si_signo = sig;
|
|
|
|
info.si_errno = 0;
|
|
|
|
if (fault_code & FAULT_CODE_ITLB)
|
|
|
|
info.si_addr = (void __user *) regs->tpc;
|
|
|
|
else
|
|
|
|
info.si_addr = (void __user *)
|
|
|
|
compute_effective_address(regs, insn, 0);
|
|
|
|
info.si_trapno = 0;
|
|
|
|
force_sig_info(sig, &info, current);
|
|
|
|
}
|
|
|
|
|
|
|
|
extern int handle_ldf_stq(u32, struct pt_regs *);
|
|
|
|
extern int handle_ld_nf(u32, struct pt_regs *);
|
|
|
|
|
|
|
|
static unsigned int get_fault_insn(struct pt_regs *regs, unsigned int insn)
|
|
|
|
{
|
|
|
|
if (!insn) {
|
|
|
|
if (!regs->tpc || (regs->tpc & 0x3))
|
|
|
|
return 0;
|
|
|
|
if (regs->tstate & TSTATE_PRIV) {
|
|
|
|
insn = *(unsigned int *) regs->tpc;
|
|
|
|
} else {
|
|
|
|
insn = get_user_insn(regs->tpc);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return insn;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void do_kernel_fault(struct pt_regs *regs, int si_code, int fault_code,
|
|
|
|
unsigned int insn, unsigned long address)
|
|
|
|
{
|
|
|
|
unsigned char asi = ASI_P;
|
|
|
|
|
|
|
|
if ((!insn) && (regs->tstate & TSTATE_PRIV))
|
|
|
|
goto cannot_handle;
|
|
|
|
|
|
|
|
/* If user insn could be read (thus insn is zero), that
|
|
|
|
* is fine. We will just gun down the process with a signal
|
|
|
|
* in that case.
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (!(fault_code & (FAULT_CODE_WRITE|FAULT_CODE_ITLB)) &&
|
|
|
|
(insn & 0xc0800000) == 0xc0800000) {
|
|
|
|
if (insn & 0x2000)
|
|
|
|
asi = (regs->tstate >> 24);
|
|
|
|
else
|
|
|
|
asi = (insn >> 5);
|
|
|
|
if ((asi & 0xf2) == 0x82) {
|
|
|
|
if (insn & 0x1000000) {
|
|
|
|
handle_ldf_stq(insn, regs);
|
|
|
|
} else {
|
|
|
|
/* This was a non-faulting load. Just clear the
|
|
|
|
* destination register(s) and continue with the next
|
|
|
|
* instruction. -jj
|
|
|
|
*/
|
|
|
|
handle_ld_nf(insn, regs);
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Is this in ex_table? */
|
|
|
|
if (regs->tstate & TSTATE_PRIV) {
|
2005-09-28 23:21:11 -04:00
|
|
|
const struct exception_table_entry *entry;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
if (asi == ASI_P && (insn & 0xc0800000) == 0xc0800000) {
|
|
|
|
if (insn & 0x2000)
|
|
|
|
asi = (regs->tstate >> 24);
|
|
|
|
else
|
|
|
|
asi = (insn >> 5);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Look in asi.h: All _S asis have LS bit set */
|
|
|
|
if ((asi & 0x1) &&
|
2005-09-28 23:21:11 -04:00
|
|
|
(entry = search_exception_tables(regs->tpc))) {
|
|
|
|
regs->tpc = entry->fixup;
|
2005-04-16 18:20:36 -04:00
|
|
|
regs->tnpc = regs->tpc + 4;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
/* The si_code was set to make clear whether
|
|
|
|
* this was a SEGV_MAPERR or SEGV_ACCERR fault.
|
|
|
|
*/
|
|
|
|
do_fault_siginfo(si_code, SIGSEGV, regs, insn, fault_code);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
cannot_handle:
|
|
|
|
unhandled_fault (address, current, regs);
|
|
|
|
}
|
|
|
|
|
2005-09-06 18:19:30 -04:00
|
|
|
asmlinkage void __kprobes do_sparc64_fault(struct pt_regs *regs)
|
2005-04-16 18:20:36 -04:00
|
|
|
{
|
|
|
|
struct mm_struct *mm = current->mm;
|
|
|
|
struct vm_area_struct *vma;
|
|
|
|
unsigned int insn = 0;
|
2007-07-19 04:47:05 -04:00
|
|
|
int si_code, fault_code, fault;
|
[SPARC64]: Fix and re-enable dynamic TSB sizing.
This is good for up to %50 performance improvement of some test cases.
The problem has been the race conditions, and hopefully I've plugged
them all up here.
1) There was a serious race in switch_mm() wrt. lazy TLB
switching to and from kernel threads.
We could erroneously skip a tsb_context_switch() and thus
use a stale TSB across a TSB grow event.
There is a big comment now in that function describing
exactly how it can happen.
2) All code paths that do something with the TSB need to be
guarded with the mm->context.lock spinlock. This makes
page table flushing paths properly synchronize with both
TSB growing and TLB context changes.
3) TSB growing events are moved to the end of successful fault
processing. Previously it was in update_mmu_cache() but
that is deadlock prone. At the end of do_sparc64_fault()
we hold no spinlocks that could deadlock the TSB grow
sequence. We also have dropped the address space semaphore.
While we're here, add prefetching to the copy_tsb() routine
and put it in assembler into the tsb.S file. This piece of
code is quite time critical.
There are some small negative side effects to this code which
can be improved upon. In particular we grab the mm->context.lock
even for the tsb insert done by update_mmu_cache() now and that's
a bit excessive. We can get rid of that locking, and the same
lock taking in flush_tsb_user(), by disabling PSTATE_IE around
the whole operation including the capturing of the tsb pointer
and tsb_nentries value. That would work because anyone growing
the TSB won't free up the old TSB until all cpus respond to the
TSB change cross call.
I'm not quite so confident in that optimization to put it in
right now, but eventually we might be able to and the description
is here for reference.
This code seems very solid now. It passes several parallel GCC
bootstrap builds, and our favorite "nut cruncher" stress test which is
a full "make -j8192" build of a "make allmodconfig" kernel. That puts
about 256 processes on each cpu's run queue, makes lots of process cpu
migrations occur, causes lots of page table and TLB flushing activity,
incurs many context version number changes, and it swaps the machine
real far out to disk even though there is 16GB of ram on this test
system. :-)
Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-16 05:02:32 -05:00
|
|
|
unsigned long address, mm_rss;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
fault_code = get_thread_fault_code();
|
|
|
|
|
2007-05-08 21:25:14 -04:00
|
|
|
if (notify_page_fault(regs))
|
2005-04-16 18:20:36 -04:00
|
|
|
return;
|
|
|
|
|
|
|
|
si_code = SEGV_MAPERR;
|
|
|
|
address = current_thread_info()->fault_address;
|
|
|
|
|
|
|
|
if ((fault_code & FAULT_CODE_ITLB) &&
|
|
|
|
(fault_code & FAULT_CODE_DTLB))
|
|
|
|
BUG();
|
|
|
|
|
|
|
|
if (regs->tstate & TSTATE_PRIV) {
|
|
|
|
unsigned long tpc = regs->tpc;
|
|
|
|
|
|
|
|
/* Sanity check the PC. */
|
|
|
|
if ((tpc >= KERNBASE && tpc < (unsigned long) _etext) ||
|
|
|
|
(tpc >= MODULES_VADDR && tpc < MODULES_END)) {
|
|
|
|
/* Valid, no problems... */
|
|
|
|
} else {
|
2006-02-13 21:07:45 -05:00
|
|
|
bad_kernel_pc(regs, address);
|
2005-04-16 18:20:36 -04:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we're in an interrupt or have no user
|
|
|
|
* context, we must not take the fault..
|
|
|
|
*/
|
|
|
|
if (in_atomic() || !mm)
|
|
|
|
goto intr_or_no_mm;
|
|
|
|
|
|
|
|
if (test_thread_flag(TIF_32BIT)) {
|
|
|
|
if (!(regs->tstate & TSTATE_PRIV))
|
|
|
|
regs->tpc &= 0xffffffff;
|
|
|
|
address &= 0xffffffff;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!down_read_trylock(&mm->mmap_sem)) {
|
|
|
|
if ((regs->tstate & TSTATE_PRIV) &&
|
|
|
|
!search_exception_tables(regs->tpc)) {
|
|
|
|
insn = get_fault_insn(regs, insn);
|
|
|
|
goto handle_kernel_fault;
|
|
|
|
}
|
|
|
|
down_read(&mm->mmap_sem);
|
|
|
|
}
|
|
|
|
|
|
|
|
vma = find_vma(mm, address);
|
|
|
|
if (!vma)
|
|
|
|
goto bad_area;
|
|
|
|
|
|
|
|
/* Pure DTLB misses do not tell us whether the fault causing
|
|
|
|
* load/store/atomic was a write or not, it only says that there
|
|
|
|
* was no match. So in such a case we (carefully) read the
|
|
|
|
* instruction to try and figure this out. It's an optimization
|
|
|
|
* so it's ok if we can't do this.
|
|
|
|
*
|
|
|
|
* Special hack, window spill/fill knows the exact fault type.
|
|
|
|
*/
|
|
|
|
if (((fault_code &
|
|
|
|
(FAULT_CODE_DTLB | FAULT_CODE_WRITE | FAULT_CODE_WINFIXUP)) == FAULT_CODE_DTLB) &&
|
|
|
|
(vma->vm_flags & VM_WRITE) != 0) {
|
|
|
|
insn = get_fault_insn(regs, 0);
|
|
|
|
if (!insn)
|
|
|
|
goto continue_fault;
|
2006-03-28 16:32:24 -05:00
|
|
|
/* All loads, stores and atomics have bits 30 and 31 both set
|
|
|
|
* in the instruction. Bit 21 is set in all stores, but we
|
|
|
|
* have to avoid prefetches which also have bit 21 set.
|
|
|
|
*/
|
2005-04-16 18:20:36 -04:00
|
|
|
if ((insn & 0xc0200000) == 0xc0200000 &&
|
2006-03-28 16:32:24 -05:00
|
|
|
(insn & 0x01780000) != 0x01680000) {
|
2005-04-16 18:20:36 -04:00
|
|
|
/* Don't bother updating thread struct value,
|
|
|
|
* because update_mmu_cache only cares which tlb
|
|
|
|
* the access came from.
|
|
|
|
*/
|
|
|
|
fault_code |= FAULT_CODE_WRITE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
continue_fault:
|
|
|
|
|
|
|
|
if (vma->vm_start <= address)
|
|
|
|
goto good_area;
|
|
|
|
if (!(vma->vm_flags & VM_GROWSDOWN))
|
|
|
|
goto bad_area;
|
|
|
|
if (!(fault_code & FAULT_CODE_WRITE)) {
|
|
|
|
/* Non-faulting loads shouldn't expand stack. */
|
|
|
|
insn = get_fault_insn(regs, insn);
|
|
|
|
if ((insn & 0xc0800000) == 0xc0800000) {
|
|
|
|
unsigned char asi;
|
|
|
|
|
|
|
|
if (insn & 0x2000)
|
|
|
|
asi = (regs->tstate >> 24);
|
|
|
|
else
|
|
|
|
asi = (insn >> 5);
|
|
|
|
if ((asi & 0xf2) == 0x82)
|
|
|
|
goto bad_area;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (expand_stack(vma, address))
|
|
|
|
goto bad_area;
|
|
|
|
/*
|
|
|
|
* Ok, we have a good vm_area for this memory access, so
|
|
|
|
* we can handle it..
|
|
|
|
*/
|
|
|
|
good_area:
|
|
|
|
si_code = SEGV_ACCERR;
|
|
|
|
|
|
|
|
/* If we took a ITLB miss on a non-executable page, catch
|
|
|
|
* that here.
|
|
|
|
*/
|
|
|
|
if ((fault_code & FAULT_CODE_ITLB) && !(vma->vm_flags & VM_EXEC)) {
|
|
|
|
BUG_ON(address != regs->tpc);
|
|
|
|
BUG_ON(regs->tstate & TSTATE_PRIV);
|
|
|
|
goto bad_area;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fault_code & FAULT_CODE_WRITE) {
|
|
|
|
if (!(vma->vm_flags & VM_WRITE))
|
|
|
|
goto bad_area;
|
|
|
|
|
|
|
|
/* Spitfire has an icache which does not snoop
|
|
|
|
* processor stores. Later processors do...
|
|
|
|
*/
|
|
|
|
if (tlb_type == spitfire &&
|
|
|
|
(vma->vm_flags & VM_EXEC) != 0 &&
|
|
|
|
vma->vm_file != NULL)
|
|
|
|
set_thread_fault_code(fault_code |
|
|
|
|
FAULT_CODE_BLKCOMMIT);
|
|
|
|
} else {
|
|
|
|
/* Allow reads even for write-only mappings */
|
|
|
|
if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
|
|
|
|
goto bad_area;
|
|
|
|
}
|
|
|
|
|
2007-07-19 04:47:05 -04:00
|
|
|
fault = handle_mm_fault(mm, vma, address, (fault_code & FAULT_CODE_WRITE));
|
|
|
|
if (unlikely(fault & VM_FAULT_ERROR)) {
|
|
|
|
if (fault & VM_FAULT_OOM)
|
|
|
|
goto out_of_memory;
|
|
|
|
else if (fault & VM_FAULT_SIGBUS)
|
|
|
|
goto do_sigbus;
|
2005-04-16 18:20:36 -04:00
|
|
|
BUG();
|
|
|
|
}
|
2007-07-19 04:47:05 -04:00
|
|
|
if (fault & VM_FAULT_MAJOR)
|
|
|
|
current->maj_flt++;
|
|
|
|
else
|
|
|
|
current->min_flt++;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
up_read(&mm->mmap_sem);
|
[SPARC64]: Fix and re-enable dynamic TSB sizing.
This is good for up to %50 performance improvement of some test cases.
The problem has been the race conditions, and hopefully I've plugged
them all up here.
1) There was a serious race in switch_mm() wrt. lazy TLB
switching to and from kernel threads.
We could erroneously skip a tsb_context_switch() and thus
use a stale TSB across a TSB grow event.
There is a big comment now in that function describing
exactly how it can happen.
2) All code paths that do something with the TSB need to be
guarded with the mm->context.lock spinlock. This makes
page table flushing paths properly synchronize with both
TSB growing and TLB context changes.
3) TSB growing events are moved to the end of successful fault
processing. Previously it was in update_mmu_cache() but
that is deadlock prone. At the end of do_sparc64_fault()
we hold no spinlocks that could deadlock the TSB grow
sequence. We also have dropped the address space semaphore.
While we're here, add prefetching to the copy_tsb() routine
and put it in assembler into the tsb.S file. This piece of
code is quite time critical.
There are some small negative side effects to this code which
can be improved upon. In particular we grab the mm->context.lock
even for the tsb insert done by update_mmu_cache() now and that's
a bit excessive. We can get rid of that locking, and the same
lock taking in flush_tsb_user(), by disabling PSTATE_IE around
the whole operation including the capturing of the tsb pointer
and tsb_nentries value. That would work because anyone growing
the TSB won't free up the old TSB until all cpus respond to the
TSB change cross call.
I'm not quite so confident in that optimization to put it in
right now, but eventually we might be able to and the description
is here for reference.
This code seems very solid now. It passes several parallel GCC
bootstrap builds, and our favorite "nut cruncher" stress test which is
a full "make -j8192" build of a "make allmodconfig" kernel. That puts
about 256 processes on each cpu's run queue, makes lots of process cpu
migrations occur, causes lots of page table and TLB flushing activity,
incurs many context version number changes, and it swaps the machine
real far out to disk even though there is 16GB of ram on this test
system. :-)
Signed-off-by: David S. Miller <davem@davemloft.net>
2006-03-16 05:02:32 -05:00
|
|
|
|
|
|
|
mm_rss = get_mm_rss(mm);
|
2006-03-22 03:49:59 -05:00
|
|
|
#ifdef CONFIG_HUGETLB_PAGE
|
|
|
|
mm_rss -= (mm->context.huge_pte_count * (HPAGE_SIZE / PAGE_SIZE));
|
|
|
|
#endif
|
2006-03-27 04:07:55 -05:00
|
|
|
if (unlikely(mm_rss >
|
2006-03-22 03:49:59 -05:00
|
|
|
mm->context.tsb_block[MM_TSB_BASE].tsb_rss_limit))
|
|
|
|
tsb_grow(mm, MM_TSB_BASE, mm_rss);
|
|
|
|
#ifdef CONFIG_HUGETLB_PAGE
|
|
|
|
mm_rss = mm->context.huge_pte_count;
|
2006-03-27 04:07:55 -05:00
|
|
|
if (unlikely(mm_rss >
|
2006-03-22 03:49:59 -05:00
|
|
|
mm->context.tsb_block[MM_TSB_HUGE].tsb_rss_limit))
|
|
|
|
tsb_grow(mm, MM_TSB_HUGE, mm_rss);
|
|
|
|
#endif
|
2005-09-29 00:06:47 -04:00
|
|
|
return;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Something tried to access memory that isn't in our memory map..
|
|
|
|
* Fix it, but check if it's kernel or user first..
|
|
|
|
*/
|
|
|
|
bad_area:
|
|
|
|
insn = get_fault_insn(regs, insn);
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
|
|
|
|
handle_kernel_fault:
|
|
|
|
do_kernel_fault(regs, si_code, fault_code, insn, address);
|
2005-09-29 00:06:47 -04:00
|
|
|
return;
|
2005-04-16 18:20:36 -04:00
|
|
|
|
|
|
|
/*
|
|
|
|
* We ran out of memory, or some other thing happened to us that made
|
|
|
|
* us unable to handle the page fault gracefully.
|
|
|
|
*/
|
|
|
|
out_of_memory:
|
|
|
|
insn = get_fault_insn(regs, insn);
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
printk("VM: killing process %s\n", current->comm);
|
|
|
|
if (!(regs->tstate & TSTATE_PRIV))
|
|
|
|
do_exit(SIGKILL);
|
|
|
|
goto handle_kernel_fault;
|
|
|
|
|
|
|
|
intr_or_no_mm:
|
|
|
|
insn = get_fault_insn(regs, 0);
|
|
|
|
goto handle_kernel_fault;
|
|
|
|
|
|
|
|
do_sigbus:
|
|
|
|
insn = get_fault_insn(regs, insn);
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Send a sigbus, regardless of whether we were in kernel
|
|
|
|
* or user mode.
|
|
|
|
*/
|
|
|
|
do_fault_siginfo(BUS_ADRERR, SIGBUS, regs, insn, fault_code);
|
|
|
|
|
|
|
|
/* Kernel mode? Handle exceptions or die */
|
|
|
|
if (regs->tstate & TSTATE_PRIV)
|
|
|
|
goto handle_kernel_fault;
|
|
|
|
}
|