Merge tag 'for-linus' of https://github.com/openrisc/linux

Pull OpenRISC updates from Stafford Horne:

 - A few sparse warning fixups and other cleanups I noticed when working
   on a recent TLB bug found on a new OpenRISC core bring up.

 - A few fixup's from me and Jason A Donenfeld to help shutdown OpenRISC
   platforms when running CI tests

* tag 'for-linus' of https://github.com/openrisc/linux:
  openrisc: Allow power off handler overriding
  openrisc: Remove unused IMMU tlb workardound
  openrisc/fault: Fix symbol scope warnings
  openrisc/delay: Add include to fix symbol not declared warning
  openrisc/time: Fix symbol scope warnings
  openrisc/traps: Declare unhandled_exception for asmlinkage
  openrisc/traps: Remove die_if_kernel function
  openrisc/traps: Declare file scope symbols as static
  openrisc: Update litex defconfig to support glibc userland
  openrisc: Pretty print show_registers memory dumps
  openrisc: Add syscall details to emergency syscall debugging
  openrisc: Add support for liteuart emergency printing
  openrisc: Cleanup emergency print handling
  openrisc: Add gcc machine instruction flag configuration
  openrisc: define nop command for simulator reboot
  openrisc: remove bogus nops and shutdowns
  openrisc: fix typos in comments

11 files changed, 272 insertions, 315 deletions

diff --git a/arch/openrisc/Kconfig b/arch/openrisc/Kconfig
index 99f0e4a4cbbd..e814df4c483c 100644
--- a/arch/openrisc/Kconfig
+++ b/arch/openrisc/Kconfig
@@ -113,6 +113,59 @@ config OPENRISC_HAVE_INST_DIV
 	default y
 	help
 	  Select this if your implementation has a hardware divide instruction
+
+config OPENRISC_HAVE_INST_CMOV
+	bool "Have instruction l.cmov for conditional move"
+	default n
+	help
+	  This config enables gcc to generate l.cmov instructions when compiling
+	  the kernel which in general will improve performance and reduce the
+	  binary size.
+
+	  Select this if your implementation has support for the Class II
+	  l.cmov conistional move instruction.
+
+	  Say N if you are unsure.
+
+config OPENRISC_HAVE_INST_ROR
+	bool "Have instruction l.ror for rotate right"
+	default n
+	help
+	  This config enables gcc to generate l.ror instructions when compiling
+	  the kernel which in general will improve performance and reduce the
+	  binary size.
+
+	  Select this if your implementation has support for the Class II
+	  l.ror rotate right instruction.
+
+	  Say N if you are unsure.
+
+config OPENRISC_HAVE_INST_RORI
+	bool "Have instruction l.rori for rotate right with immediate"
+	default n
+	help
+	  This config enables gcc to generate l.rori instructions when compiling
+	  the kernel which in general will improve performance and reduce the
+	  binary size.
+
+	  Select this if your implementation has support for the Class II
+	  l.rori rotate right with immediate instruction.
+
+	  Say N if you are unsure.
+
+config OPENRISC_HAVE_INST_SEXT
+	bool "Have instructions l.ext* for sign extension"
+	default n
+	help
+	  This config enables gcc to generate l.ext* instructions when compiling
+	  the kernel which in general will improve performance and reduce the
+	  binary size.
+
+	  Select this if your implementation has support for the Class II
+	  l.exths, l.extbs, l.exthz and l.extbz size extend instructions.
+
+	  Say N if you are unsure.
+
 endmenu
 
 config NR_CPUS
diff --git a/arch/openrisc/Makefile b/arch/openrisc/Makefile
index 760b734fb822..b446510173cd 100644
--- a/arch/openrisc/Makefile
+++ b/arch/openrisc/Makefile
@@ -21,6 +21,7 @@ OBJCOPYFLAGS    := -O binary -R .note -R .comment -S
 LIBGCC 		:= $(shell $(CC) $(KBUILD_CFLAGS) -print-libgcc-file-name)
 
 KBUILD_CFLAGS	+= -pipe -ffixed-r10 -D__linux__
+KBUILD_CFLAGS	+= -msfimm -mshftimm
 
 all: vmlinux.bin
 
@@ -38,6 +39,22 @@ else
 	KBUILD_CFLAGS += $(call cc-option,-msoft-div)
 endif
 
+ifeq ($(CONFIG_OPENRISC_HAVE_INST_CMOV),y)
+	KBUILD_CFLAGS += $(call cc-option,-mcmov)
+endif
+
+ifeq ($(CONFIG_OPENRISC_HAVE_INST_ROR),y)
+	KBUILD_CFLAGS += $(call cc-option,-mror)
+endif
+
+ifeq ($(CONFIG_OPENRISC_HAVE_INST_RORI),y)
+	KBUILD_CFLAGS += $(call cc-option,-mrori)
+endif
+
+ifeq ($(CONFIG_OPENRISC_HAVE_INST_SEXT),y)
+	KBUILD_CFLAGS += $(call cc-option,-msext)
+endif
+
 head-y 		:= arch/openrisc/kernel/head.o
 
 libs-y		+= $(LIBGCC)
diff --git a/arch/openrisc/configs/or1klitex_defconfig b/arch/openrisc/configs/or1klitex_defconfig
index d695879a4d26..d3fb964b4f85 100644
--- a/arch/openrisc/configs/or1klitex_defconfig
+++ b/arch/openrisc/configs/or1klitex_defconfig
@@ -1,22 +1,54 @@
+CONFIG_SYSVIPC=y
+CONFIG_POSIX_MQUEUE=y
+CONFIG_CGROUPS=y
+CONFIG_NAMESPACES=y
+CONFIG_USER_NS=y
 CONFIG_BLK_DEV_INITRD=y
 CONFIG_CC_OPTIMIZE_FOR_SIZE=y
+CONFIG_SGETMASK_SYSCALL=y
 CONFIG_EMBEDDED=y
 CONFIG_OPENRISC_BUILTIN_DTB="or1klitex"
 CONFIG_HZ_100=y
+CONFIG_OPENRISC_HAVE_SHADOW_GPRS=y
 CONFIG_NET=y
 CONFIG_PACKET=y
+CONFIG_PACKET_DIAG=y
 CONFIG_UNIX=y
+CONFIG_UNIX_DIAG=y
 CONFIG_INET=y
+CONFIG_IP_MULTICAST=y
+CONFIG_IP_ADVANCED_ROUTER=y
+CONFIG_INET_UDP_DIAG=y
+CONFIG_INET_RAW_DIAG=y
+# CONFIG_WIRELESS is not set
+# CONFIG_ETHTOOL_NETLINK is not set
 CONFIG_DEVTMPFS=y
 CONFIG_DEVTMPFS_MOUNT=y
 CONFIG_OF_OVERLAY=y
 CONFIG_NETDEVICES=y
 CONFIG_LITEX_LITEETH=y
+# CONFIG_WLAN is not set
 CONFIG_SERIAL_LITEUART=y
 CONFIG_SERIAL_LITEUART_CONSOLE=y
 CONFIG_TTY_PRINTK=y
+# CONFIG_GPIO_CDEV is not set
+CONFIG_MMC=y
+CONFIG_MMC_LITEX=y
+# CONFIG_VHOST_MENU is not set
+# CONFIG_IOMMU_SUPPORT is not set
 CONFIG_LITEX_SOC_CONTROLLER=y
+CONFIG_EXT2_FS=y
+CONFIG_EXT3_FS=y
+CONFIG_MSDOS_FS=y
+CONFIG_VFAT_FS=y
+CONFIG_EXFAT_FS=y
 CONFIG_TMPFS=y
+CONFIG_NFS_FS=y
+CONFIG_NFS_V3_ACL=y
+CONFIG_NFS_V4=y
+CONFIG_NLS_CODEPAGE_437=y
+CONFIG_NLS_ISO8859_1=y
+CONFIG_LSM="lockdown,yama,loadpin,safesetid,integrity,bpf"
 CONFIG_PRINTK_TIME=y
 CONFIG_PANIC_ON_OOPS=y
 CONFIG_SOFTLOCKUP_DETECTOR=y
diff --git a/arch/openrisc/kernel/entry.S b/arch/openrisc/kernel/entry.S
index 3ca1b1f490b9..54a87bba35ca 100644
--- a/arch/openrisc/kernel/entry.S
+++ b/arch/openrisc/kernel/entry.S
@@ -601,7 +601,7 @@ UNHANDLED_EXCEPTION(_vector_0xb00,0xb00)
  */
 
 _string_syscall_return:
-	.string "syscall return %ld \n\r\0"
+	.string "syscall r9:0x%08x -> syscall(%ld) return %ld\0"
 	.align 4
 
 ENTRY(_sys_call_handler)
@@ -679,15 +679,25 @@ _syscall_return:
 _syscall_debug:
 	l.movhi r3,hi(_string_syscall_return)
 	l.ori   r3,r3,lo(_string_syscall_return)
-	l.ori   r27,r0,1
+	l.ori   r27,r0,2
 	l.sw    -4(r1),r27
 	l.sw    -8(r1),r11
-	l.addi  r1,r1,-8
+	l.lwz	r29,PT_ORIG_GPR11(r1)
+	l.sw    -12(r1),r29
+	l.lwz	r29,PT_GPR9(r1)
+	l.sw    -16(r1),r29
 	l.movhi r27,hi(_printk)
 	l.ori   r27,r27,lo(_printk)
 	l.jalr  r27
-	 l.nop
-	l.addi  r1,r1,8
+	 l.addi  r1,r1,-16
+	l.addi  r1,r1,16
+#endif
+#if 0
+_syscall_show_regs:
+	l.movhi r27,hi(show_registers)
+	l.ori   r27,r27,lo(show_registers)
+	l.jalr  r27
+	 l.or	r3,r1,r1
 #endif
 
 _syscall_check_trace_leave:
diff --git a/arch/openrisc/kernel/head.S b/arch/openrisc/kernel/head.S
index 2fa6cefa62ca..e11699f3d6bd 100644
--- a/arch/openrisc/kernel/head.S
+++ b/arch/openrisc/kernel/head.S
@@ -297,19 +297,23 @@
 	/* temporary store r3, r9 into r1, r10 */		;\
 	l.addi	r1,r3,0x0					;\
 	l.addi	r10,r9,0x0					;\
-	/* the string referenced by r3 must be low enough */	;\
+	LOAD_SYMBOL_2_GPR(r9,_string_unhandled_exception)	;\
+	tophys	(r3,r9)						;\
 	l.jal	_emergency_print				;\
-	l.ori	r3,r0,lo(_string_unhandled_exception)		;\
+	 l.nop							;\
 	l.mfspr	r3,r0,SPR_NPC					;\
 	l.jal	_emergency_print_nr				;\
-	l.andi	r3,r3,0x1f00					;\
-	/* the string referenced by r3 must be low enough */	;\
+	 l.andi	r3,r3,0x1f00					;\
+	LOAD_SYMBOL_2_GPR(r9,_string_epc_prefix)		;\
+	tophys	(r3,r9)						;\
 	l.jal	_emergency_print				;\
-	l.ori	r3,r0,lo(_string_epc_prefix)			;\
+	 l.nop							;\
 	l.jal	_emergency_print_nr				;\
-	l.mfspr	r3,r0,SPR_EPCR_BASE				;\
+	 l.mfspr r3,r0,SPR_EPCR_BASE				;\
+	LOAD_SYMBOL_2_GPR(r9,_string_nl)			;\
+	tophys	(r3,r9)						;\
 	l.jal	_emergency_print				;\
-	l.ori	r3,r0,lo(_string_nl)				;\
+	 l.nop							;\
 	/* end of printing */					;\
 	l.addi	r3,r1,0x0					;\
 	l.addi	r9,r10,0x0					;\
@@ -1330,274 +1334,110 @@ i_pte_not_present:
 
 /* =================================================[ debugging aids ]=== */
 
-	.align 64
-_immu_trampoline:
-	.space 64
-_immu_trampoline_top:
-
-#define TRAMP_SLOT_0		(0x0)
-#define TRAMP_SLOT_1		(0x4)
-#define TRAMP_SLOT_2		(0x8)
-#define TRAMP_SLOT_3		(0xc)
-#define TRAMP_SLOT_4		(0x10)
-#define TRAMP_SLOT_5		(0x14)
-#define TRAMP_FRAME_SIZE	(0x18)
-
-ENTRY(_immu_trampoline_workaround)
-	// r2 EEA
-	// r6 is physical EEA
-	tophys(r6,r2)
-
-	LOAD_SYMBOL_2_GPR(r5,_immu_trampoline)
-	tophys	(r3,r5)			// r3 is trampoline (physical)
-
-	LOAD_SYMBOL_2_GPR(r4,0x15000000)
-	l.sw	TRAMP_SLOT_0(r3),r4
-	l.sw	TRAMP_SLOT_1(r3),r4
-	l.sw	TRAMP_SLOT_4(r3),r4
-	l.sw	TRAMP_SLOT_5(r3),r4
-
-					// EPC = EEA - 0x4
-	l.lwz	r4,0x0(r6)		// load op @ EEA + 0x0 (fc address)
-	l.sw	TRAMP_SLOT_3(r3),r4	// store it to _immu_trampoline_data
-	l.lwz	r4,-0x4(r6)		// load op @ EEA - 0x4 (f8 address)
-	l.sw	TRAMP_SLOT_2(r3),r4	// store it to _immu_trampoline_data
-
-	l.srli  r5,r4,26                // check opcode for write access
-	l.sfeqi r5,0                    // l.j
-	l.bf    0f
-	l.sfeqi r5,0x11                 // l.jr
-	l.bf    1f
-	l.sfeqi r5,1                    // l.jal
-	l.bf    2f
-	l.sfeqi r5,0x12                 // l.jalr
-	l.bf    3f
-	l.sfeqi r5,3                    // l.bnf
-	l.bf    4f
-	l.sfeqi r5,4                    // l.bf
-	l.bf    5f
-99:
-	l.nop
-	l.j	99b			// should never happen
-	l.nop	1
-
-	// r2 is EEA
-	// r3 is trampoline address (physical)
-	// r4 is instruction
-	// r6 is physical(EEA)
-	//
-	// r5
-
-2:	// l.jal
-
-	/* 19 20 aa aa	l.movhi r9,0xaaaa
-	 * a9 29 bb bb  l.ori	r9,0xbbbb
-	 *
-	 * where 0xaaaabbbb is EEA + 0x4 shifted right 2
-	 */
-
-	l.addi	r6,r2,0x4		// this is 0xaaaabbbb
-
-					// l.movhi r9,0xaaaa
-	l.ori	r5,r0,0x1920		// 0x1920 == l.movhi r9
-	l.sh	(TRAMP_SLOT_0+0x0)(r3),r5
-	l.srli	r5,r6,16
-	l.sh	(TRAMP_SLOT_0+0x2)(r3),r5
-
-					// l.ori   r9,0xbbbb
-	l.ori	r5,r0,0xa929		// 0xa929 == l.ori r9
-	l.sh	(TRAMP_SLOT_1+0x0)(r3),r5
-	l.andi	r5,r6,0xffff
-	l.sh	(TRAMP_SLOT_1+0x2)(r3),r5
-
-	/* falthrough, need to set up new jump offset */
-
-
-0:	// l.j
-	l.slli	r6,r4,6			// original offset shifted left 6 - 2
-//	l.srli	r6,r6,6			// original offset shifted right 2
-
-	l.slli	r4,r2,4			// old jump position: EEA shifted left 4
-//	l.srli	r4,r4,6			// old jump position: shifted right 2
-
-	l.addi	r5,r3,0xc		// new jump position (physical)
-	l.slli	r5,r5,4			// new jump position: shifted left 4
-
-	// calculate new jump offset
-	// new_off = old_off + (old_jump - new_jump)
-
-	l.sub	r5,r4,r5		// old_jump - new_jump
-	l.add	r5,r6,r5		// orig_off + (old_jump - new_jump)
-	l.srli	r5,r5,6			// new offset shifted right 2
-
-	// r5 is new jump offset
-					// l.j has opcode 0x0...
-	l.sw	TRAMP_SLOT_2(r3),r5	// write it back
-
-	l.j	trampoline_out
-	l.nop
-
-/* ----------------------------- */
-
-3:	// l.jalr
-
-	/* 19 20 aa aa	l.movhi r9,0xaaaa
-	 * a9 29 bb bb  l.ori	r9,0xbbbb
-	 *
-	 * where 0xaaaabbbb is EEA + 0x4 shifted right 2
-	 */
-
-	l.addi	r6,r2,0x4		// this is 0xaaaabbbb
-
-					// l.movhi r9,0xaaaa
-	l.ori	r5,r0,0x1920		// 0x1920 == l.movhi r9
-	l.sh	(TRAMP_SLOT_0+0x0)(r3),r5
-	l.srli	r5,r6,16
-	l.sh	(TRAMP_SLOT_0+0x2)(r3),r5
-
-					// l.ori   r9,0xbbbb
-	l.ori	r5,r0,0xa929		// 0xa929 == l.ori r9
-	l.sh	(TRAMP_SLOT_1+0x0)(r3),r5
-	l.andi	r5,r6,0xffff
-	l.sh	(TRAMP_SLOT_1+0x2)(r3),r5
-
-	l.lhz	r5,(TRAMP_SLOT_2+0x0)(r3)	// load hi part of jump instruction
-	l.andi	r5,r5,0x3ff		// clear out opcode part
-	l.ori	r5,r5,0x4400		// opcode changed from l.jalr -> l.jr
-	l.sh	(TRAMP_SLOT_2+0x0)(r3),r5 // write it back
-
-	/* falthrough */
-
-1:	// l.jr
-	l.j	trampoline_out
-	l.nop
-
-/* ----------------------------- */
-
-4:	// l.bnf
-5:	// l.bf
-	l.slli	r6,r4,6			// original offset shifted left 6 - 2
-//	l.srli	r6,r6,6			// original offset shifted right 2
-
-	l.slli	r4,r2,4			// old jump position: EEA shifted left 4
-//	l.srli	r4,r4,6			// old jump position: shifted right 2
-
-	l.addi	r5,r3,0xc		// new jump position (physical)
-	l.slli	r5,r5,4			// new jump position: shifted left 4
-
-	// calculate new jump offset
-	// new_off = old_off + (old_jump - new_jump)
-
-	l.add	r6,r6,r4		// (orig_off + old_jump)
-	l.sub	r6,r6,r5		// (orig_off + old_jump) - new_jump
-	l.srli	r6,r6,6			// new offset shifted right 2
-
-	// r6 is new jump offset
-	l.lwz	r4,(TRAMP_SLOT_2+0x0)(r3)	// load jump instruction
-	l.srli	r4,r4,16
-	l.andi	r4,r4,0xfc00		// get opcode part
-	l.slli	r4,r4,16
-	l.or	r6,r4,r6		// l.b(n)f new offset
-	l.sw	TRAMP_SLOT_2(r3),r6	// write it back
-
-	/* we need to add l.j to EEA + 0x8 */
-	tophys	(r4,r2)			// may not be needed (due to shifts down_
-	l.addi	r4,r4,(0x8 - 0x8)	// jump target = r2 + 0x8 (compensate for 0x8)
-					// jump position = r5 + 0x8 (0x8 compensated)
-	l.sub	r4,r4,r5		// jump offset = target - new_position + 0x8
-
-	l.slli	r4,r4,4			// the amount of info in imediate of jump
-	l.srli	r4,r4,6			// jump instruction with offset
-	l.sw	TRAMP_SLOT_4(r3),r4	// write it to 4th slot
-
-	/* fallthrough */
-
-trampoline_out:
-	// set up new EPC to point to our trampoline code
-	LOAD_SYMBOL_2_GPR(r5,_immu_trampoline)
-	l.mtspr	r0,r5,SPR_EPCR_BASE
-
-	// immu_trampoline is (4x) CACHE_LINE aligned
-	// and only 6 instructions long,
-	// so we need to invalidate only 2 lines
-
-	/* Establish cache block size
-	   If BS=0, 16;
-	   If BS=1, 32;
-	   r14 contain block size
-	*/
-	l.mfspr r21,r0,SPR_ICCFGR
-	l.andi	r21,r21,SPR_ICCFGR_CBS
-	l.srli	r21,r21,7
-	l.ori	r23,r0,16
-	l.sll	r14,r23,r21
-
-	l.mtspr	r0,r5,SPR_ICBIR
-	l.add	r5,r5,r14
-	l.mtspr	r0,r5,SPR_ICBIR
-
-	l.jr	r9
-	l.nop
-
-
 /*
- * DSCR: prints a string referenced by r3.
+ * DESC: Prints ASCII character stored in r7
  *
- * PRMS: r3     	- address of the first character of null
- *			terminated string to be printed
+ * PRMS: r7	- a 32-bit value with an ASCII character in the first byte
+ *		position.
  *
- * PREQ: UART at UART_BASE_ADD has to be initialized
+ * PREQ: The UART at UART_BASE_ADD has to be initialized
  *
- * POST: caller should be aware that r3, r9 are changed
+ * POST: internally used but restores:
+ * 	 r4	- to store UART_BASE_ADD
+ *	 r5	- for loading OFF_TXFULL / THRE,TEMT
+ *	 r6	- for storing bitmask (SERIAL_8250)
  */
-ENTRY(_emergency_print)
+ENTRY(_emergency_putc)
 	EMERGENCY_PRINT_STORE_GPR4
 	EMERGENCY_PRINT_STORE_GPR5
 	EMERGENCY_PRINT_STORE_GPR6
-	EMERGENCY_PRINT_STORE_GPR7
-2:
-	l.lbz	r7,0(r3)
-	l.sfeq	r7,r0
-	l.bf	9f
-	l.nop
 
-// putc:
 	l.movhi r4,hi(UART_BASE_ADD)
+	l.ori	r4,r4,lo(UART_BASE_ADD)
+
+#if defined(CONFIG_SERIAL_LITEUART)
+	/* Check OFF_TXFULL status */
+1:      l.lwz	r5,4(r4)
+	l.andi	r5,r5,0xff
+	l.sfnei	r5,0
+	l.bf	1b
+	 l.nop
 
+	/* Write character */
+	l.andi	r7,r7,0xff
+	l.sw	0(r4),r7
+#elif defined(CONFIG_SERIAL_8250)
+	/* Check UART LSR THRE (hold) bit */
 	l.addi  r6,r0,0x20
 1:      l.lbz   r5,5(r4)
 	l.andi  r5,r5,0x20
 	l.sfeq  r5,r6
 	l.bnf   1b
-	l.nop
+	 l.nop
 
+	/* Write character */
 	l.sb    0(r4),r7
 
+	/* Check UART LSR THRE|TEMT (hold, empty) bits */
 	l.addi  r6,r0,0x60
 1:      l.lbz   r5,5(r4)
 	l.andi  r5,r5,0x60
 	l.sfeq  r5,r6
 	l.bnf   1b
-	l.nop
+	 l.nop
+#endif
+	EMERGENCY_PRINT_LOAD_GPR6
+	EMERGENCY_PRINT_LOAD_GPR5
+	EMERGENCY_PRINT_LOAD_GPR4
+	l.jr	r9
+	 l.nop
+
+/*
+ * DSCR: prints a string referenced by r3.
+ *
+ * PRMS: r3     	- address of the first character of null
+ *			terminated string to be printed
+ *
+ * PREQ: UART at UART_BASE_ADD has to be initialized
+ *
+ * POST: caller should be aware that r3, r9 are changed
+ */
+ENTRY(_emergency_print)
+	EMERGENCY_PRINT_STORE_GPR7
+	EMERGENCY_PRINT_STORE_GPR9
+
+	/* Load character to r7, check for null terminator */
+2:	l.lbz	r7,0(r3)
+	l.sfeqi	r7,0x0
+	l.bf	9f
+	 l.nop
+
+	l.jal	_emergency_putc
+	 l.nop
 
 	/* next character */
 	l.j	2b
-	l.addi	r3,r3,0x1
+	 l.addi	r3,r3,0x1
 
 9:
+	EMERGENCY_PRINT_LOAD_GPR9
 	EMERGENCY_PRINT_LOAD_GPR7
-	EMERGENCY_PRINT_LOAD_GPR6
-	EMERGENCY_PRINT_LOAD_GPR5
-	EMERGENCY_PRINT_LOAD_GPR4
 	l.jr	r9
-	l.nop
+	 l.nop
 
+/*
+ * DSCR: prints a number in r3 in hex.
+ *
+ * PRMS: r3     	- a 32-bit unsigned integer
+ *
+ * PREQ: UART at UART_BASE_ADD has to be initialized
+ *
+ * POST: caller should be aware that r3, r9 are changed
+ */
 ENTRY(_emergency_print_nr)
-	EMERGENCY_PRINT_STORE_GPR4
-	EMERGENCY_PRINT_STORE_GPR5
-	EMERGENCY_PRINT_STORE_GPR6
 	EMERGENCY_PRINT_STORE_GPR7
 	EMERGENCY_PRINT_STORE_GPR8
+	EMERGENCY_PRINT_STORE_GPR9
 
 	l.addi	r8,r0,32		// shift register
 
@@ -1609,58 +1449,39 @@ ENTRY(_emergency_print_nr)
 	/* don't skip the last zero if number == 0x0 */
 	l.sfeqi	r8,0x4
 	l.bf	2f
-	l.nop
+	 l.nop
 
 	l.sfeq	r7,r0
 	l.bf	1b
-	l.nop
+	 l.nop
 
 2:
 	l.srl	r7,r3,r8
 
 	l.andi	r7,r7,0xf
 	l.sflts	r8,r0
-	l.bf	9f
+	 l.bf	9f
 
+	/* Numbers greater than 9 translate to a-f */
 	l.sfgtui r7,0x9
 	l.bnf	8f
-	l.nop
+	 l.nop
 	l.addi	r7,r7,0x27
 
-8:
-	l.addi	r7,r7,0x30
-// putc:
-	l.movhi r4,hi(UART_BASE_ADD)
-
-	l.addi  r6,r0,0x20
-1:      l.lbz   r5,5(r4)
-	l.andi  r5,r5,0x20
-	l.sfeq  r5,r6
-	l.bnf   1b
-	l.nop
-
-	l.sb    0(r4),r7
-
-	l.addi  r6,r0,0x60
-1:      l.lbz   r5,5(r4)
-	l.andi  r5,r5,0x60
-	l.sfeq  r5,r6
-	l.bnf   1b
-	l.nop
+	/* Convert to ascii and output character */
+8:	l.jal	_emergency_putc
+	 l.addi	r7,r7,0x30
 
 	/* next character */
 	l.j	2b
 	l.addi	r8,r8,-0x4
 
 9:
+	EMERGENCY_PRINT_LOAD_GPR9
 	EMERGENCY_PRINT_LOAD_GPR8
 	EMERGENCY_PRINT_LOAD_GPR7
-	EMERGENCY_PRINT_LOAD_GPR6
-	EMERGENCY_PRINT_LOAD_GPR5
-	EMERGENCY_PRINT_LOAD_GPR4
 	l.jr	r9
-	l.nop
-
+	 l.nop
 
 /*
  * This should be used for debugging only.
@@ -1685,7 +1506,9 @@ ENTRY(_emergency_print_nr)
 
 ENTRY(_early_uart_init)
 	l.movhi	r3,hi(UART_BASE_ADD)
+	l.ori	r3,r3,lo(UART_BASE_ADD)
 
+#if defined(CONFIG_SERIAL_8250)
 	l.addi	r4,r0,0x7
 	l.sb	0x2(r3),r4
 
@@ -1703,9 +1526,10 @@ ENTRY(_early_uart_init)
 	l.addi  r4,r0,((UART_DIVISOR) & 0x000000ff)
 	l.sb	UART_DLL(r3),r4
 	l.sb	0x3(r3),r5
+#endif
 
 	l.jr	r9
-	l.nop
+	 l.nop
 
 	.align	0x1000
 	.global _secondary_evbar
@@ -1720,13 +1544,13 @@ _secondary_evbar:
 
 	.section .rodata
 _string_unhandled_exception:
-	.string "\n\rRunarunaround: Unhandled exception 0x\0"
+	.string "\r\nRunarunaround: Unhandled exception 0x\0"
 
 _string_epc_prefix:
 	.string ": EPC=0x\0"
 
 _string_nl:
-	.string "\n\r\0"
+	.string "\r\n\0"
 
 
 /* ========================================[ page aligned structures ]=== */
diff --git a/arch/openrisc/kernel/process.c b/arch/openrisc/kernel/process.c
index 3c0c91bcdcba..1d4c0921aafa 100644
--- a/arch/openrisc/kernel/process.c
+++ b/arch/openrisc/kernel/process.c
@@ -52,6 +52,8 @@ void machine_restart(char *cmd)
 {
 	do_kernel_restart(cmd);
 
+	__asm__("l.nop 13");
+
 	/* Give a grace period for failure to restart of 1s */
 	mdelay(1000);
 
@@ -61,6 +63,16 @@ void machine_restart(char *cmd)
 }
 
 /*
+ * This is used if pm_power_off has not been set by a power management
+ * driver, in this case we can assume we are on a simulator.  On
+ * OpenRISC simulators l.nop 1 will trigger the simulator exit.
+ */
+static void default_power_off(void)
+{
+	__asm__("l.nop 1");
+}
+
+/*
  * Similar to machine_power_off, but don't shut off power.  Add code
  * here to freeze the system for e.g. post-mortem debug purpose when
  * possible.  This halt has nothing to do with the idle halt.
@@ -75,7 +87,10 @@ void machine_halt(void)
 void machine_power_off(void)
 {
 	printk(KERN_INFO "*** MACHINE POWER OFF ***\n");
-	__asm__("l.nop 1");
+	if (pm_power_off != NULL)
+		pm_power_off();
+	else
+		default_power_off();
 }
 
 /*
@@ -89,7 +104,7 @@ void arch_cpu_idle(void)
 		mtspr(SPR_PMR, mfspr(SPR_PMR) | SPR_PMR_DME);
 }
 
-void (*pm_power_off) (void) = machine_power_off;
+void (*pm_power_off)(void) = NULL;
 EXPORT_SYMBOL(pm_power_off);
 
 /*
diff --git a/arch/openrisc/kernel/time.c b/arch/openrisc/kernel/time.c
index 6d18989d63d0..8e26c1af5441 100644
--- a/arch/openrisc/kernel/time.c
+++ b/arch/openrisc/kernel/time.c
@@ -23,6 +23,7 @@
 #include <linux/of_clk.h>
 
 #include <asm/cpuinfo.h>
+#include <asm/time.h>
 
 /* Test the timer ticks to count, used in sync routine */
 inline void openrisc_timer_set(unsigned long count)
@@ -61,7 +62,7 @@ static int openrisc_timer_set_next_event(unsigned long delta,
  * timers) we cannot enable the PERIODIC feature.  The tick timer can run using
  * one-shot events, so no problem.
  */
-DEFINE_PER_CPU(struct clock_event_device, clockevent_openrisc_timer);
+static DEFINE_PER_CPU(struct clock_event_device, clockevent_openrisc_timer);
 
 void openrisc_clockevent_init(void)
 {
diff --git a/arch/openrisc/kernel/traps.c b/arch/openrisc/kernel/traps.c
index 0446a3c34372..fd9a0f2b66c4 100644
--- a/arch/openrisc/kernel/traps.c
+++ b/arch/openrisc/kernel/traps.c
@@ -34,11 +34,11 @@
 #include <asm/unwinder.h>
 #include <asm/sections.h>
 
-int kstack_depth_to_print = 0x180;
+static int kstack_depth_to_print = 0x180;
 int lwa_flag;
-unsigned long __user *lwa_addr;
+static unsigned long __user *lwa_addr;
 
-void print_trace(void *data, unsigned long addr, int reliable)
+static void print_trace(void *data, unsigned long addr, int reliable)
 {
 	const char *loglvl = data;
 
@@ -46,6 +46,14 @@ void print_trace(void *data, unsigned long addr, int reliable)
 	       (void *) addr);
 }
 
+static void print_data(unsigned long base_addr, unsigned long word, int i)
+{
+	if (i == 0)
+		printk("(%08lx:)\t%08lx", base_addr + (i * 4), word);
+	else
+		printk(" %08lx:\t%08lx", base_addr + (i * 4), word);
+}
+
 /* displays a short stack trace */
 void show_stack(struct task_struct *task, unsigned long *esp, const char *loglvl)
 {
@@ -99,22 +107,36 @@ void show_registers(struct pt_regs *regs)
 		printk("\nStack: ");
 		show_stack(NULL, (unsigned long *)esp, KERN_EMERG);
 
+		if (esp < PAGE_OFFSET)
+			goto bad_stack;
+
+		printk("\n");
+		for (i = -8; i < 24; i += 1) {
+			unsigned long word;
+
+			if (__get_user(word, &((unsigned long *)esp)[i])) {
+bad_stack:
+				printk(" Bad Stack value.");
+				break;
+			}
+
+			print_data(esp, word, i);
+		}
+
 		printk("\nCode: ");
 		if (regs->pc < PAGE_OFFSET)
 			goto bad;
 
-		for (i = -24; i < 24; i++) {
-			unsigned char c;
-			if (__get_user(c, &((unsigned char *)regs->pc)[i])) {
+		for (i = -6; i < 6; i += 1) {
+			unsigned long word;
+
+			if (__get_user(word, &((unsigned long *)regs->pc)[i])) {
 bad:
 				printk(" Bad PC value.");
 				break;
 			}
 
-			if (i == 0)
-				printk("(%02x) ", c);
-			else
-				printk("%02x ", c);
+			print_data(regs->pc, word, i);
 		}
 	}
 	printk("\n");
@@ -185,13 +207,11 @@ void nommu_dump_state(struct pt_regs *regs,
 	printk("\nCode: ");
 
 	for (i = -24; i < 24; i++) {
-		unsigned char c;
-		c = ((unsigned char *)(__pa(regs->pc)))[i];
+		unsigned long word;
+
+		word = ((unsigned long *)(__pa(regs->pc)))[i];
 
-		if (i == 0)
-			printk("(%02x) ", c);
-		else
-			printk("%02x ", c);
+		print_data(regs->pc, word, i);
 	}
 	printk("\n");
 }
@@ -215,16 +235,7 @@ void __noreturn die(const char *str, struct pt_regs *regs, long err)
 	make_task_dead(SIGSEGV);
 }
 
-/* This is normally the 'Oops' routine */
-void die_if_kernel(const char *str, struct pt_regs *regs, long err)
-{
-	if (user_mode(regs))
-		return;
-
-	die(str, regs, err);
-}
-
-void unhandled_exception(struct pt_regs *regs, int ea, int vector)
+asmlinkage void unhandled_exception(struct pt_regs *regs, int ea, int vector)
 {
 	printk("Unable to handle exception at EA =0x%x, vector 0x%x",
 	       ea, vector);
diff --git a/arch/openrisc/lib/delay.c b/arch/openrisc/lib/delay.c
index 036ae57180ef..5e89e4131304 100644
--- a/arch/openrisc/lib/delay.c
+++ b/arch/openrisc/lib/delay.c
@@ -15,6 +15,7 @@
 #include <linux/kernel.h>
 #include <linux/export.h>
 #include <linux/init.h>
+#include <linux/timex.h>
 #include <asm/param.h>
 #include <asm/delay.h>
 #include <asm/timex.h>
diff --git a/arch/openrisc/mm/fault.c b/arch/openrisc/mm/fault.c
index 80bb66ad42f6..53b760af3bb7 100644
--- a/arch/openrisc/mm/fault.c
+++ b/arch/openrisc/mm/fault.c
@@ -18,15 +18,13 @@
 #include <linux/perf_event.h>
 
 #include <linux/uaccess.h>
+#include <asm/mmu_context.h>
 #include <asm/siginfo.h>
 #include <asm/signal.h>
 
 #define NUM_TLB_ENTRIES 64
 #define TLB_OFFSET(add) (((add) >> PAGE_SHIFT) & (NUM_TLB_ENTRIES-1))
 
-unsigned long pte_misses;	/* updated by do_page_fault() */
-unsigned long pte_errors;	/* updated by do_page_fault() */
-
 /* __PHX__ :: - check the vmalloc_fault in do_page_fault()
  *            - also look into include/asm/mmu_context.h
  */
@@ -223,8 +221,6 @@ no_context:
 	{
 		const struct exception_table_entry *entry;
 
-		__asm__ __volatile__("l.nop 42");
-
 		if ((entry = search_exception_tables(regs->pc)) != NULL) {
 			/* Adjust the instruction pointer in the stackframe */
 			regs->pc = entry->fixup;
@@ -252,9 +248,6 @@ no_context:
 	 */
 
 out_of_memory:
-	__asm__ __volatile__("l.nop 42");
-	__asm__ __volatile__("l.nop 1");
-
 	mmap_read_unlock(mm);
 	if (!user_mode(regs))
 		goto no_context;
diff --git a/arch/openrisc/mm/tlb.c b/arch/openrisc/mm/tlb.c
index 2b6feabf6381..e2f2a3c3bb22 100644
--- a/arch/openrisc/mm/tlb.c
+++ b/arch/openrisc/mm/tlb.c
@@ -128,7 +128,7 @@ void local_flush_tlb_mm(struct mm_struct *mm)
 
 	/* Was seeing bugs with the mm struct passed to us. Scrapped most of
 	   this function. */
-	/* Several architctures do this */
+	/* Several architectures do this */
 	local_flush_tlb_all();
 }