kernel_optimize_test/arch/mips/kernel/unaligned.c
Deng-Cheng Zhu 7f788d2d53 MIPS: add support for software performance events
Software events are required as part of the measurable stuff by the
Linux performance counter subsystem. Here is the list of events added by
this patch:
PERF_COUNT_SW_PAGE_FAULTS
PERF_COUNT_SW_PAGE_FAULTS_MIN
PERF_COUNT_SW_PAGE_FAULTS_MAJ
PERF_COUNT_SW_ALIGNMENT_FAULTS
PERF_COUNT_SW_EMULATION_FAULTS

Signed-off-by: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
To: linux-mips@linux-mips.org
Cc: a.p.zijlstra@chello.nl
Cc: paulus@samba.org
Cc: mingo@elte.hu
Cc: acme@redhat.com
Cc: jamie.iles@picochip.com
Acked-by: David Daney <ddaney@caviumnetworks.com>
Reviewed-by: Matt Fleming <matt@console-pimps.org>
Patchwork: https://patchwork.linux-mips.org/patch/1686/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
2010-10-29 19:08:48 +01:00

577 lines
14 KiB
C

/*
* Handle unaligned accesses by emulation.
*
* 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, 1998, 1999, 2002 by Ralf Baechle
* Copyright (C) 1999 Silicon Graphics, Inc.
*
* This file contains exception handler for address error exception with the
* special capability to execute faulting instructions in software. The
* handler does not try to handle the case when the program counter points
* to an address not aligned to a word boundary.
*
* Putting data to unaligned addresses is a bad practice even on Intel where
* only the performance is affected. Much worse is that such code is non-
* portable. Due to several programs that die on MIPS due to alignment
* problems I decided to implement this handler anyway though I originally
* didn't intend to do this at all for user code.
*
* For now I enable fixing of address errors by default to make life easier.
* I however intend to disable this somewhen in the future when the alignment
* problems with user programs have been fixed. For programmers this is the
* right way to go.
*
* Fixing address errors is a per process option. The option is inherited
* across fork(2) and execve(2) calls. If you really want to use the
* option in your user programs - I discourage the use of the software
* emulation strongly - use the following code in your userland stuff:
*
* #include <sys/sysmips.h>
*
* ...
* sysmips(MIPS_FIXADE, x);
* ...
*
* The argument x is 0 for disabling software emulation, enabled otherwise.
*
* Below a little program to play around with this feature.
*
* #include <stdio.h>
* #include <sys/sysmips.h>
*
* struct foo {
* unsigned char bar[8];
* };
*
* main(int argc, char *argv[])
* {
* struct foo x = {0, 1, 2, 3, 4, 5, 6, 7};
* unsigned int *p = (unsigned int *) (x.bar + 3);
* int i;
*
* if (argc > 1)
* sysmips(MIPS_FIXADE, atoi(argv[1]));
*
* printf("*p = %08lx\n", *p);
*
* *p = 0xdeadface;
*
* for(i = 0; i <= 7; i++)
* printf("%02x ", x.bar[i]);
* printf("\n");
* }
*
* Coprocessor loads are not supported; I think this case is unimportant
* in the practice.
*
* TODO: Handle ndc (attempted store to doubleword in uncached memory)
* exception for the R6000.
* A store crossing a page boundary might be executed only partially.
* Undo the partial store in this case.
*/
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/signal.h>
#include <linux/smp.h>
#include <linux/sched.h>
#include <linux/debugfs.h>
#include <linux/perf_event.h>
#include <asm/asm.h>
#include <asm/branch.h>
#include <asm/byteorder.h>
#include <asm/cop2.h>
#include <asm/inst.h>
#include <asm/uaccess.h>
#include <asm/system.h>
#define STR(x) __STR(x)
#define __STR(x) #x
enum {
UNALIGNED_ACTION_QUIET,
UNALIGNED_ACTION_SIGNAL,
UNALIGNED_ACTION_SHOW,
};
#ifdef CONFIG_DEBUG_FS
static u32 unaligned_instructions;
static u32 unaligned_action;
#else
#define unaligned_action UNALIGNED_ACTION_QUIET
#endif
extern void show_registers(struct pt_regs *regs);
static void emulate_load_store_insn(struct pt_regs *regs,
void __user *addr, unsigned int __user *pc)
{
union mips_instruction insn;
unsigned long value;
unsigned int res;
perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS,
1, 0, regs, 0);
/*
* This load never faults.
*/
__get_user(insn.word, pc);
switch (insn.i_format.opcode) {
/*
* These are instructions that a compiler doesn't generate. We
* can assume therefore that the code is MIPS-aware and
* really buggy. Emulating these instructions would break the
* semantics anyway.
*/
case ll_op:
case lld_op:
case sc_op:
case scd_op:
/*
* For these instructions the only way to create an address
* error is an attempted access to kernel/supervisor address
* space.
*/
case ldl_op:
case ldr_op:
case lwl_op:
case lwr_op:
case sdl_op:
case sdr_op:
case swl_op:
case swr_op:
case lb_op:
case lbu_op:
case sb_op:
goto sigbus;
/*
* The remaining opcodes are the ones that are really of interest.
*/
case lh_op:
if (!access_ok(VERIFY_READ, addr, 2))
goto sigbus;
__asm__ __volatile__ (".set\tnoat\n"
#ifdef __BIG_ENDIAN
"1:\tlb\t%0, 0(%2)\n"
"2:\tlbu\t$1, 1(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tlb\t%0, 1(%2)\n"
"2:\tlbu\t$1, 0(%2)\n\t"
#endif
"sll\t%0, 0x8\n\t"
"or\t%0, $1\n\t"
"li\t%1, 0\n"
"3:\t.set\tat\n\t"
".section\t.fixup,\"ax\"\n\t"
"4:\tli\t%1, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=&r" (value), "=r" (res)
: "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
regs->regs[insn.i_format.rt] = value;
break;
case lw_op:
if (!access_ok(VERIFY_READ, addr, 4))
goto sigbus;
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
"1:\tlwl\t%0, (%2)\n"
"2:\tlwr\t%0, 3(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tlwl\t%0, 3(%2)\n"
"2:\tlwr\t%0, (%2)\n\t"
#endif
"li\t%1, 0\n"
"3:\t.section\t.fixup,\"ax\"\n\t"
"4:\tli\t%1, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=&r" (value), "=r" (res)
: "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
regs->regs[insn.i_format.rt] = value;
break;
case lhu_op:
if (!access_ok(VERIFY_READ, addr, 2))
goto sigbus;
__asm__ __volatile__ (
".set\tnoat\n"
#ifdef __BIG_ENDIAN
"1:\tlbu\t%0, 0(%2)\n"
"2:\tlbu\t$1, 1(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tlbu\t%0, 1(%2)\n"
"2:\tlbu\t$1, 0(%2)\n\t"
#endif
"sll\t%0, 0x8\n\t"
"or\t%0, $1\n\t"
"li\t%1, 0\n"
"3:\t.set\tat\n\t"
".section\t.fixup,\"ax\"\n\t"
"4:\tli\t%1, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=&r" (value), "=r" (res)
: "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
regs->regs[insn.i_format.rt] = value;
break;
case lwu_op:
#ifdef CONFIG_64BIT
/*
* A 32-bit kernel might be running on a 64-bit processor. But
* if we're on a 32-bit processor and an i-cache incoherency
* or race makes us see a 64-bit instruction here the sdl/sdr
* would blow up, so for now we don't handle unaligned 64-bit
* instructions on 32-bit kernels.
*/
if (!access_ok(VERIFY_READ, addr, 4))
goto sigbus;
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
"1:\tlwl\t%0, (%2)\n"
"2:\tlwr\t%0, 3(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tlwl\t%0, 3(%2)\n"
"2:\tlwr\t%0, (%2)\n\t"
#endif
"dsll\t%0, %0, 32\n\t"
"dsrl\t%0, %0, 32\n\t"
"li\t%1, 0\n"
"3:\t.section\t.fixup,\"ax\"\n\t"
"4:\tli\t%1, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=&r" (value), "=r" (res)
: "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
regs->regs[insn.i_format.rt] = value;
break;
#endif /* CONFIG_64BIT */
/* Cannot handle 64-bit instructions in 32-bit kernel */
goto sigill;
case ld_op:
#ifdef CONFIG_64BIT
/*
* A 32-bit kernel might be running on a 64-bit processor. But
* if we're on a 32-bit processor and an i-cache incoherency
* or race makes us see a 64-bit instruction here the sdl/sdr
* would blow up, so for now we don't handle unaligned 64-bit
* instructions on 32-bit kernels.
*/
if (!access_ok(VERIFY_READ, addr, 8))
goto sigbus;
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
"1:\tldl\t%0, (%2)\n"
"2:\tldr\t%0, 7(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tldl\t%0, 7(%2)\n"
"2:\tldr\t%0, (%2)\n\t"
#endif
"li\t%1, 0\n"
"3:\t.section\t.fixup,\"ax\"\n\t"
"4:\tli\t%1, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=&r" (value), "=r" (res)
: "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
regs->regs[insn.i_format.rt] = value;
break;
#endif /* CONFIG_64BIT */
/* Cannot handle 64-bit instructions in 32-bit kernel */
goto sigill;
case sh_op:
if (!access_ok(VERIFY_WRITE, addr, 2))
goto sigbus;
value = regs->regs[insn.i_format.rt];
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
".set\tnoat\n"
"1:\tsb\t%1, 1(%2)\n\t"
"srl\t$1, %1, 0x8\n"
"2:\tsb\t$1, 0(%2)\n\t"
".set\tat\n\t"
#endif
#ifdef __LITTLE_ENDIAN
".set\tnoat\n"
"1:\tsb\t%1, 0(%2)\n\t"
"srl\t$1,%1, 0x8\n"
"2:\tsb\t$1, 1(%2)\n\t"
".set\tat\n\t"
#endif
"li\t%0, 0\n"
"3:\n\t"
".section\t.fixup,\"ax\"\n\t"
"4:\tli\t%0, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=r" (res)
: "r" (value), "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
break;
case sw_op:
if (!access_ok(VERIFY_WRITE, addr, 4))
goto sigbus;
value = regs->regs[insn.i_format.rt];
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
"1:\tswl\t%1,(%2)\n"
"2:\tswr\t%1, 3(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tswl\t%1, 3(%2)\n"
"2:\tswr\t%1, (%2)\n\t"
#endif
"li\t%0, 0\n"
"3:\n\t"
".section\t.fixup,\"ax\"\n\t"
"4:\tli\t%0, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=r" (res)
: "r" (value), "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
break;
case sd_op:
#ifdef CONFIG_64BIT
/*
* A 32-bit kernel might be running on a 64-bit processor. But
* if we're on a 32-bit processor and an i-cache incoherency
* or race makes us see a 64-bit instruction here the sdl/sdr
* would blow up, so for now we don't handle unaligned 64-bit
* instructions on 32-bit kernels.
*/
if (!access_ok(VERIFY_WRITE, addr, 8))
goto sigbus;
value = regs->regs[insn.i_format.rt];
__asm__ __volatile__ (
#ifdef __BIG_ENDIAN
"1:\tsdl\t%1,(%2)\n"
"2:\tsdr\t%1, 7(%2)\n\t"
#endif
#ifdef __LITTLE_ENDIAN
"1:\tsdl\t%1, 7(%2)\n"
"2:\tsdr\t%1, (%2)\n\t"
#endif
"li\t%0, 0\n"
"3:\n\t"
".section\t.fixup,\"ax\"\n\t"
"4:\tli\t%0, %3\n\t"
"j\t3b\n\t"
".previous\n\t"
".section\t__ex_table,\"a\"\n\t"
STR(PTR)"\t1b, 4b\n\t"
STR(PTR)"\t2b, 4b\n\t"
".previous"
: "=r" (res)
: "r" (value), "r" (addr), "i" (-EFAULT));
if (res)
goto fault;
compute_return_epc(regs);
break;
#endif /* CONFIG_64BIT */
/* Cannot handle 64-bit instructions in 32-bit kernel */
goto sigill;
case lwc1_op:
case ldc1_op:
case swc1_op:
case sdc1_op:
/*
* I herewith declare: this does not happen. So send SIGBUS.
*/
goto sigbus;
/*
* COP2 is available to implementor for application specific use.
* It's up to applications to register a notifier chain and do
* whatever they have to do, including possible sending of signals.
*/
case lwc2_op:
cu2_notifier_call_chain(CU2_LWC2_OP, regs);
break;
case ldc2_op:
cu2_notifier_call_chain(CU2_LDC2_OP, regs);
break;
case swc2_op:
cu2_notifier_call_chain(CU2_SWC2_OP, regs);
break;
case sdc2_op:
cu2_notifier_call_chain(CU2_SDC2_OP, regs);
break;
default:
/*
* Pheeee... We encountered an yet unknown instruction or
* cache coherence problem. Die sucker, die ...
*/
goto sigill;
}
#ifdef CONFIG_DEBUG_FS
unaligned_instructions++;
#endif
return;
fault:
/* Did we have an exception handler installed? */
if (fixup_exception(regs))
return;
die_if_kernel("Unhandled kernel unaligned access", regs);
force_sig(SIGSEGV, current);
return;
sigbus:
die_if_kernel("Unhandled kernel unaligned access", regs);
force_sig(SIGBUS, current);
return;
sigill:
die_if_kernel("Unhandled kernel unaligned access or invalid instruction", regs);
force_sig(SIGILL, current);
}
asmlinkage void do_ade(struct pt_regs *regs)
{
unsigned int __user *pc;
mm_segment_t seg;
perf_sw_event(PERF_COUNT_SW_ALIGNMENT_FAULTS,
1, 0, regs, regs->cp0_badvaddr);
/*
* Did we catch a fault trying to load an instruction?
* Or are we running in MIPS16 mode?
*/
if ((regs->cp0_badvaddr == regs->cp0_epc) || (regs->cp0_epc & 0x1))
goto sigbus;
pc = (unsigned int __user *) exception_epc(regs);
if (user_mode(regs) && !test_thread_flag(TIF_FIXADE))
goto sigbus;
if (unaligned_action == UNALIGNED_ACTION_SIGNAL)
goto sigbus;
else if (unaligned_action == UNALIGNED_ACTION_SHOW)
show_registers(regs);
/*
* Do branch emulation only if we didn't forward the exception.
* This is all so but ugly ...
*/
seg = get_fs();
if (!user_mode(regs))
set_fs(KERNEL_DS);
emulate_load_store_insn(regs, (void __user *)regs->cp0_badvaddr, pc);
set_fs(seg);
return;
sigbus:
die_if_kernel("Kernel unaligned instruction access", regs);
force_sig(SIGBUS, current);
/*
* XXX On return from the signal handler we should advance the epc
*/
}
#ifdef CONFIG_DEBUG_FS
extern struct dentry *mips_debugfs_dir;
static int __init debugfs_unaligned(void)
{
struct dentry *d;
if (!mips_debugfs_dir)
return -ENODEV;
d = debugfs_create_u32("unaligned_instructions", S_IRUGO,
mips_debugfs_dir, &unaligned_instructions);
if (!d)
return -ENOMEM;
d = debugfs_create_u32("unaligned_action", S_IRUGO | S_IWUSR,
mips_debugfs_dir, &unaligned_action);
if (!d)
return -ENOMEM;
return 0;
}
__initcall(debugfs_unaligned);
#endif