forge/linux-pinenote
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
linux-pinenote/arch/sh/kernel/traps_64.c
Paul Mundt 0375a73c6f sh64: Attempt to make reserved insn trap handler resemble C. 
This has been long overdue. No functional changes.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>
2012-06-14 14:46:36 +09:00

816 lines
21 KiB
C
Raw Blame History

/*
* arch/sh/kernel/traps_64.c
*
* Copyright (C) 2000, 2001 Paolo Alberelli
* Copyright (C) 2003, 2004 Paul Mundt
* Copyright (C) 2003, 2004 Richard Curnow
*
* 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.
*/
#include <linux/sched.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/ptrace.h>
#include <linux/timer.h>
#include <linux/mm.h>
#include <linux/smp.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <linux/spinlock.h>
#include <linux/kallsyms.h>
#include <linux/interrupt.h>
#include <linux/sysctl.h>
#include <linux/module.h>
#include <linux/perf_event.h>
#include <asm/uaccess.h>
#include <asm/io.h>
#include <asm/alignment.h>
#include <asm/processor.h>
#include <asm/pgtable.h>
#include <asm/fpu.h>
static int read_opcode(reg_size_t pc, insn_size_t *result_opcode, int from_user_mode)
{
int get_user_error;
unsigned long aligned_pc;
insn_size_t opcode;
if ((pc & 3) == 1) {
/* SHmedia */
aligned_pc = pc & ~3;
if (from_user_mode) {
if (!access_ok(VERIFY_READ, aligned_pc, sizeof(insn_size_t))) {
get_user_error = -EFAULT;
} else {
get_user_error = __get_user(opcode, (insn_size_t *)aligned_pc);
*result_opcode = opcode;
}
return get_user_error;
} else {
/* If the fault was in the kernel, we can either read
* this directly, or if not, we fault.
*/
*result_opcode = *(insn_size_t *)aligned_pc;
return 0;
}
} else if ((pc & 1) == 0) {
/* SHcompact */
/* TODO : provide handling for this. We don't really support
user-mode SHcompact yet, and for a kernel fault, this would
have to come from a module built for SHcompact. */
return -EFAULT;
} else {
/* misaligned */
return -EFAULT;
}
}
static int address_is_sign_extended(__u64 a)
{
__u64 b;
#if (NEFF == 32)
b = (__u64)(__s64)(__s32)(a & 0xffffffffUL);
return (b == a) ? 1 : 0;
#else
#error "Sign extend check only works for NEFF==32"
#endif
}
/* return -1 for fault, 0 for OK */
static int generate_and_check_address(struct pt_regs *regs,
insn_size_t opcode,
int displacement_not_indexed,
int width_shift,
__u64 *address)
{
__u64 base_address, addr;
int basereg;
switch (1 << width_shift) {
case 1: inc_unaligned_byte_access(); break;
case 2: inc_unaligned_word_access(); break;
case 4: inc_unaligned_dword_access(); break;
case 8: inc_unaligned_multi_access(); break;
}
basereg = (opcode >> 20) & 0x3f;
base_address = regs->regs[basereg];
if (displacement_not_indexed) {
__s64 displacement;
displacement = (opcode >> 10) & 0x3ff;
displacement = ((displacement << 54) >> 54); /* sign extend */
addr = (__u64)((__s64)base_address + (displacement << width_shift));
} else {
__u64 offset;
int offsetreg;
offsetreg = (opcode >> 10) & 0x3f;
offset = regs->regs[offsetreg];
addr = base_address + offset;
}
/* Check sign extended */
if (!address_is_sign_extended(addr))
return -1;
/* Check accessible. For misaligned access in the kernel, assume the
address is always accessible (and if not, just fault when the
load/store gets done.) */
if (user_mode(regs)) {
inc_unaligned_user_access();
if (addr >= TASK_SIZE)
return -1;
} else
inc_unaligned_kernel_access();
*address = addr;
perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, addr);
unaligned_fixups_notify(current, opcode, regs);
return 0;
}
static void misaligned_kernel_word_load(__u64 address, int do_sign_extend, __u64 *result)
{
unsigned short x;
unsigned char *p, *q;
p = (unsigned char *) (int) address;
q = (unsigned char *) &x;
q[0] = p[0];
q[1] = p[1];
if (do_sign_extend) {
*result = (__u64)(__s64) *(short *) &x;
} else {
*result = (__u64) x;
}
}
static void misaligned_kernel_word_store(__u64 address, __u64 value)
{
unsigned short x;
unsigned char *p, *q;
p = (unsigned char *) (int) address;
q = (unsigned char *) &x;
x = (__u16) value;
p[0] = q[0];
p[1] = q[1];
}
static int misaligned_load(struct pt_regs *regs,
insn_size_t opcode,
int displacement_not_indexed,
int width_shift,
int do_sign_extend)
{
/* Return -1 for a fault, 0 for OK */
int error;
int destreg;
__u64 address;
error = generate_and_check_address(regs, opcode,
displacement_not_indexed, width_shift, &address);
if (error < 0)
return error;
destreg = (opcode >> 4) & 0x3f;
if (user_mode(regs)) {
__u64 buffer;
if (!access_ok(VERIFY_READ, (unsigned long) address, 1UL<<width_shift)) {
return -1;
}
if (__copy_user(&buffer, (const void *)(int)address, (1 << width_shift)) > 0) {
return -1; /* fault */
}
switch (width_shift) {
case 1:
if (do_sign_extend) {
regs->regs[destreg] = (__u64)(__s64) *(__s16 *) &buffer;
} else {
regs->regs[destreg] = (__u64) *(__u16 *) &buffer;
}
break;
case 2:
regs->regs[destreg] = (__u64)(__s64) *(__s32 *) &buffer;
break;
case 3:
regs->regs[destreg] = buffer;
break;
default:
printk("Unexpected width_shift %d in misaligned_load, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
} else {
/* kernel mode - we can take short cuts since if we fault, it's a genuine bug */
__u64 lo, hi;
switch (width_shift) {
case 1:
misaligned_kernel_word_load(address, do_sign_extend, &regs->regs[destreg]);
break;
case 2:
asm ("ldlo.l %1, 0, %0" : "=r" (lo) : "r" (address));
asm ("ldhi.l %1, 3, %0" : "=r" (hi) : "r" (address));
regs->regs[destreg] = lo | hi;
break;
case 3:
asm ("ldlo.q %1, 0, %0" : "=r" (lo) : "r" (address));
asm ("ldhi.q %1, 7, %0" : "=r" (hi) : "r" (address));
regs->regs[destreg] = lo | hi;
break;
default:
printk("Unexpected width_shift %d in misaligned_load, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
}
return 0;
}
static int misaligned_store(struct pt_regs *regs,
insn_size_t opcode,
int displacement_not_indexed,
int width_shift)
{
/* Return -1 for a fault, 0 for OK */
int error;
int srcreg;
__u64 address;
error = generate_and_check_address(regs, opcode,
displacement_not_indexed, width_shift, &address);
if (error < 0)
return error;
srcreg = (opcode >> 4) & 0x3f;
if (user_mode(regs)) {
__u64 buffer;
if (!access_ok(VERIFY_WRITE, (unsigned long) address, 1UL<<width_shift)) {
return -1;
}
switch (width_shift) {
case 1:
*(__u16 *) &buffer = (__u16) regs->regs[srcreg];
break;
case 2:
*(__u32 *) &buffer = (__u32) regs->regs[srcreg];
break;
case 3:
buffer = regs->regs[srcreg];
break;
default:
printk("Unexpected width_shift %d in misaligned_store, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
if (__copy_user((void *)(int)address, &buffer, (1 << width_shift)) > 0) {
return -1; /* fault */
}
} else {
/* kernel mode - we can take short cuts since if we fault, it's a genuine bug */
__u64 val = regs->regs[srcreg];
switch (width_shift) {
case 1:
misaligned_kernel_word_store(address, val);
break;
case 2:
asm ("stlo.l %1, 0, %0" : : "r" (val), "r" (address));
asm ("sthi.l %1, 3, %0" : : "r" (val), "r" (address));
break;
case 3:
asm ("stlo.q %1, 0, %0" : : "r" (val), "r" (address));
asm ("sthi.q %1, 7, %0" : : "r" (val), "r" (address));
break;
default:
printk("Unexpected width_shift %d in misaligned_store, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
}
return 0;
}
/* Never need to fix up misaligned FPU accesses within the kernel since that's a real
error. */
static int misaligned_fpu_load(struct pt_regs *regs,
insn_size_t opcode,
int displacement_not_indexed,
int width_shift,
int do_paired_load)
{
/* Return -1 for a fault, 0 for OK */
int error;
int destreg;
__u64 address;
error = generate_and_check_address(regs, opcode,
displacement_not_indexed, width_shift, &address);
if (error < 0)
return error;
destreg = (opcode >> 4) & 0x3f;
if (user_mode(regs)) {
__u64 buffer;
__u32 buflo, bufhi;
if (!access_ok(VERIFY_READ, (unsigned long) address, 1UL<<width_shift)) {
return -1;
}
if (__copy_user(&buffer, (const void *)(int)address, (1 << width_shift)) > 0) {
return -1; /* fault */
}
/* 'current' may be the current owner of the FPU state, so
context switch the registers into memory so they can be
indexed by register number. */
if (last_task_used_math == current) {
enable_fpu();
save_fpu(current);
disable_fpu();
last_task_used_math = NULL;
regs->sr |= SR_FD;
}
buflo = *(__u32*) &buffer;
bufhi = *(1 + (__u32*) &buffer);
switch (width_shift) {
case 2:
current->thread.xstate->hardfpu.fp_regs[destreg] = buflo;
break;
case 3:
if (do_paired_load) {
current->thread.xstate->hardfpu.fp_regs[destreg] = buflo;
current->thread.xstate->hardfpu.fp_regs[destreg+1] = bufhi;
} else {
#if defined(CONFIG_CPU_LITTLE_ENDIAN)
current->thread.xstate->hardfpu.fp_regs[destreg] = bufhi;
current->thread.xstate->hardfpu.fp_regs[destreg+1] = buflo;
#else
current->thread.xstate->hardfpu.fp_regs[destreg] = buflo;
current->thread.xstate->hardfpu.fp_regs[destreg+1] = bufhi;
#endif
}
break;
default:
printk("Unexpected width_shift %d in misaligned_fpu_load, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
return 0;
} else {
die ("Misaligned FPU load inside kernel", regs, 0);
return -1;
}
}
static int misaligned_fpu_store(struct pt_regs *regs,
insn_size_t opcode,
int displacement_not_indexed,
int width_shift,
int do_paired_load)
{
/* Return -1 for a fault, 0 for OK */
int error;
int srcreg;
__u64 address;
error = generate_and_check_address(regs, opcode,
displacement_not_indexed, width_shift, &address);
if (error < 0)
return error;
srcreg = (opcode >> 4) & 0x3f;
if (user_mode(regs)) {
__u64 buffer;
/* Initialise these to NaNs. */
__u32 buflo=0xffffffffUL, bufhi=0xffffffffUL;
if (!access_ok(VERIFY_WRITE, (unsigned long) address, 1UL<<width_shift)) {
return -1;
}
/* 'current' may be the current owner of the FPU state, so
context switch the registers into memory so they can be
indexed by register number. */
if (last_task_used_math == current) {
enable_fpu();
save_fpu(current);
disable_fpu();
last_task_used_math = NULL;
regs->sr |= SR_FD;
}
switch (width_shift) {
case 2:
buflo = current->thread.xstate->hardfpu.fp_regs[srcreg];
break;
case 3:
if (do_paired_load) {
buflo = current->thread.xstate->hardfpu.fp_regs[srcreg];
bufhi = current->thread.xstate->hardfpu.fp_regs[srcreg+1];
} else {
#if defined(CONFIG_CPU_LITTLE_ENDIAN)
bufhi = current->thread.xstate->hardfpu.fp_regs[srcreg];
buflo = current->thread.xstate->hardfpu.fp_regs[srcreg+1];
#else
buflo = current->thread.xstate->hardfpu.fp_regs[srcreg];
bufhi = current->thread.xstate->hardfpu.fp_regs[srcreg+1];
#endif
}
break;
default:
printk("Unexpected width_shift %d in misaligned_fpu_store, PC=%08lx\n",
width_shift, (unsigned long) regs->pc);
break;
}
*(__u32*) &buffer = buflo;
*(1 + (__u32*) &buffer) = bufhi;
if (__copy_user((void *)(int)address, &buffer, (1 << width_shift)) > 0) {
return -1; /* fault */
}
return 0;
} else {
die ("Misaligned FPU load inside kernel", regs, 0);
return -1;
}
}
static int misaligned_fixup(struct pt_regs *regs)
{
insn_size_t opcode;
int error;
int major, minor;
unsigned int user_action;
user_action = unaligned_user_action();
if (!(user_action & UM_FIXUP))
return -1;
error = read_opcode(regs->pc, &opcode, user_mode(regs));
if (error < 0) {
return error;
}
major = (opcode >> 26) & 0x3f;
minor = (opcode >> 16) & 0xf;
switch (major) {
case (0x84>>2): /* LD.W */
error = misaligned_load(regs, opcode, 1, 1, 1);
break;
case (0xb0>>2): /* LD.UW */
error = misaligned_load(regs, opcode, 1, 1, 0);
break;
case (0x88>>2): /* LD.L */
error = misaligned_load(regs, opcode, 1, 2, 1);
break;
case (0x8c>>2): /* LD.Q */
error = misaligned_load(regs, opcode, 1, 3, 0);
break;
case (0xa4>>2): /* ST.W */
error = misaligned_store(regs, opcode, 1, 1);
break;
case (0xa8>>2): /* ST.L */
error = misaligned_store(regs, opcode, 1, 2);
break;
case (0xac>>2): /* ST.Q */
error = misaligned_store(regs, opcode, 1, 3);
break;
case (0x40>>2): /* indexed loads */
switch (minor) {
case 0x1: /* LDX.W */
error = misaligned_load(regs, opcode, 0, 1, 1);
break;
case 0x5: /* LDX.UW */
error = misaligned_load(regs, opcode, 0, 1, 0);
break;
case 0x2: /* LDX.L */
error = misaligned_load(regs, opcode, 0, 2, 1);
break;
case 0x3: /* LDX.Q */
error = misaligned_load(regs, opcode, 0, 3, 0);
break;
default:
error = -1;
break;
}
break;
case (0x60>>2): /* indexed stores */
switch (minor) {
case 0x1: /* STX.W */
error = misaligned_store(regs, opcode, 0, 1);
break;
case 0x2: /* STX.L */
error = misaligned_store(regs, opcode, 0, 2);
break;
case 0x3: /* STX.Q */
error = misaligned_store(regs, opcode, 0, 3);
break;
default:
error = -1;
break;
}
break;
case (0x94>>2): /* FLD.S */
error = misaligned_fpu_load(regs, opcode, 1, 2, 0);
break;
case (0x98>>2): /* FLD.P */
error = misaligned_fpu_load(regs, opcode, 1, 3, 1);
break;
case (0x9c>>2): /* FLD.D */
error = misaligned_fpu_load(regs, opcode, 1, 3, 0);
break;
case (0x1c>>2): /* floating indexed loads */
switch (minor) {
case 0x8: /* FLDX.S */
error = misaligned_fpu_load(regs, opcode, 0, 2, 0);
break;
case 0xd: /* FLDX.P */
error = misaligned_fpu_load(regs, opcode, 0, 3, 1);
break;
case 0x9: /* FLDX.D */
error = misaligned_fpu_load(regs, opcode, 0, 3, 0);
break;
default:
error = -1;
break;
}
break;
case (0xb4>>2): /* FLD.S */
error = misaligned_fpu_store(regs, opcode, 1, 2, 0);
break;
case (0xb8>>2): /* FLD.P */
error = misaligned_fpu_store(regs, opcode, 1, 3, 1);
break;
case (0xbc>>2): /* FLD.D */
error = misaligned_fpu_store(regs, opcode, 1, 3, 0);
break;
case (0x3c>>2): /* floating indexed stores */
switch (minor) {
case 0x8: /* FSTX.S */
error = misaligned_fpu_store(regs, opcode, 0, 2, 0);
break;
case 0xd: /* FSTX.P */
error = misaligned_fpu_store(regs, opcode, 0, 3, 1);
break;
case 0x9: /* FSTX.D */
error = misaligned_fpu_store(regs, opcode, 0, 3, 0);
break;
default:
error = -1;
break;
}
break;
default:
/* Fault */
error = -1;
break;
}
if (error < 0) {
return error;
} else {
regs->pc += 4; /* Skip the instruction that's just been emulated */
return 0;
}
}
static void do_unhandled_exception(int signr, char *str, unsigned long error,
struct pt_regs *regs)
{
if (user_mode(regs))
force_sig(signr, current);
die_if_no_fixup(str, regs, error);
}
#define DO_ERROR(signr, str, name) \
asmlinkage void do_##name(unsigned long error_code, struct pt_regs *regs) \
{ \
do_unhandled_exception(signr, str, error_code, regs); \
}
DO_ERROR(SIGILL, "illegal slot instruction", illegal_slot_inst)
DO_ERROR(SIGSEGV, "address error (exec)", address_error_exec)
#if defined(CONFIG_SH64_ID2815_WORKAROUND)
#define OPCODE_INVALID 0
#define OPCODE_USER_VALID 1
#define OPCODE_PRIV_VALID 2
/* getcon/putcon - requires checking which control register is referenced. */
#define OPCODE_CTRL_REG 3
/* Table of valid opcodes for SHmedia mode.
Form a 10-bit value by concatenating the major/minor opcodes i.e.
opcode[31:26,20:16]. The 6 MSBs of this value index into the following
array. The 4 LSBs select the bit-pair in the entry (bits 1:0 correspond to
LSBs==4'b0000 etc). */
static unsigned long shmedia_opcode_table[64] = {
0x55554044,0x54445055,0x15141514,0x14541414,0x00000000,0x10001000,0x01110055,0x04050015,
0x00000444,0xc0000000,0x44545515,0x40405555,0x55550015,0x10005555,0x55555505,0x04050000,
0x00000555,0x00000404,0x00040445,0x15151414,0x00000000,0x00000000,0x00000000,0x00000000,
0x00000055,0x40404444,0x00000404,0xc0009495,0x00000000,0x00000000,0x00000000,0x00000000,
0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,
0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,
0x80005050,0x04005055,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,0x55555555,
0x81055554,0x00000404,0x55555555,0x55555555,0x00000000,0x00000000,0x00000000,0x00000000
};
/* Workaround SH5-101 cut2 silicon defect #2815 :
in some situations, inter-mode branches from SHcompact -> SHmedia
which should take ITLBMISS or EXECPROT exceptions at the target
falsely take RESINST at the target instead. */
void do_reserved_inst(unsigned long error_code, struct pt_regs *regs)
{
insn_size_t opcode = 0x6ff4fff0; /* guaranteed reserved opcode */
unsigned long pc, aligned_pc;
unsigned long index, shift;
unsigned long major, minor, combined;
unsigned long reserved_field;
int opcode_state;
int get_user_error;
int signr = SIGILL;
char *exception_name = "reserved_instruction";
pc = regs->pc;
/* SHcompact is not handled */
if (unlikely((pc & 3) == 0))
goto out;
/* SHmedia : check for defect. This requires executable vmas
to be readable too. */
aligned_pc = pc & ~3;
if (!access_ok(VERIFY_READ, aligned_pc, sizeof(insn_size_t)))
get_user_error = -EFAULT;
else
get_user_error = __get_user(opcode, (insn_size_t *)aligned_pc);
if (get_user_error < 0) {
/*
* Error trying to read opcode. This typically means a
* real fault, not a RESINST any more. So change the
* codes.
*/
exception_name = "address error (exec)";
signr = SIGSEGV;
goto out;
}
/* These bits are currently reserved as zero in all valid opcodes */
reserved_field = opcode & 0xf;
if (unlikely(reserved_field))
goto out; /* invalid opcode */
major = (opcode >> 26) & 0x3f;
minor = (opcode >> 16) & 0xf;
combined = (major << 4) | minor;
index = major;
shift = minor << 1;
opcode_state = (shmedia_opcode_table[index] >> shift) & 0x3;
switch (opcode_state) {
case OPCODE_INVALID:
/* Trap. */
break;
case OPCODE_USER_VALID:
/*
* Restart the instruction: the branch to the instruction
* will now be from an RTE not from SHcompact so the
* silicon defect won't be triggered.
*/
return;
case OPCODE_PRIV_VALID:
if (!user_mode(regs)) {
/*
* Should only ever get here if a module has
* SHcompact code inside it. If so, the same fix
* up is needed.
*/
return; /* same reason */
}
/*
* Otherwise, user mode trying to execute a privileged
* instruction - fall through to trap.
*/
break;
case OPCODE_CTRL_REG:
/* If in privileged mode, return as above. */
if (!user_mode(regs))
return;
/* In user mode ... */
if (combined == 0x9f) { /* GETCON */
unsigned long regno = (opcode >> 20) & 0x3f;
if (regno >= 62)
return;
/* reserved/privileged control register => trap */
} else if (combined == 0x1bf) { /* PUTCON */
unsigned long regno = (opcode >> 4) & 0x3f;
if (regno >= 62)
return;
/* reserved/privileged control register => trap */
}
break;
default:
/* Fall through to trap. */
break;
}
out:
do_unhandled_exception(signr, exception_name, error_code, regs);
}
#else /* CONFIG_SH64_ID2815_WORKAROUND */
/* If the workaround isn't needed, this is just a straightforward reserved
instruction */
DO_ERROR(SIGILL, "reserved instruction", reserved_inst)
#endif /* CONFIG_SH64_ID2815_WORKAROUND */
/* Called with interrupts disabled */
asmlinkage void do_exception_error(unsigned long ex, struct pt_regs *regs)
{
die_if_kernel("exception", regs, ex);
}
asmlinkage int do_unknown_trapa(unsigned long scId, struct pt_regs *regs)
{
/* Syscall debug */
printk("System call ID error: [0x1#args:8 #syscall:16 0x%lx]\n", scId);
die_if_kernel("unknown trapa", regs, scId);
return -ENOSYS;
}
/* Implement misaligned load/store handling for kernel (and optionally for user
mode too). Limitation : only SHmedia mode code is handled - there is no
handling at all for misaligned accesses occurring in SHcompact code yet. */
asmlinkage void do_address_error_load(unsigned long error_code, struct pt_regs *regs)
{
if (misaligned_fixup(regs) < 0)
do_unhandled_exception(SIGSEGV, "address error(load)",
error_code, regs);
}
asmlinkage void do_address_error_store(unsigned long error_code, struct pt_regs *regs)
{
if (misaligned_fixup(regs) < 0)
do_unhandled_exception(SIGSEGV, "address error(store)",
error_code, regs);
}
asmlinkage void do_debug_interrupt(unsigned long code, struct pt_regs *regs)
{
u64 peek_real_address_q(u64 addr);
u64 poke_real_address_q(u64 addr, u64 val);
unsigned long long DM_EXP_CAUSE_PHY = 0x0c100010;
unsigned long long exp_cause;
/* It's not worth ioremapping the debug module registers for the amount
of access we make to them - just go direct to their physical
addresses. */
exp_cause = peek_real_address_q(DM_EXP_CAUSE_PHY);
if (exp_cause & ~4)
printk("DM.EXP_CAUSE had unexpected bits set (=%08lx)\n",
(unsigned long)(exp_cause & 0xffffffff));
show_state();
/* Clear all DEBUGINT causes */
poke_real_address_q(DM_EXP_CAUSE_PHY, 0x0);
}
void __cpuinit per_cpu_trap_init(void)
{
/* Nothing to do for now, VBR initialization later. */
}
Reference in a new issue View git blame Copy permalink