Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mpe/linux

Pull powerpc fixes from Michael Ellerman:
 "Some more powerpc fixes if you please"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mpe/linux:
  powerpc: use device_online/offline() instead of cpu_up/down()
  powerpc/powernv: Properly fix LPC debugfs endianness
  powerpc: do_notify_resume can be called with bad thread_info flags argument
  powerpc/fadump: Fix endianess issues in firmware assisted dump handling
  powerpc: Fix section mismatch warning

arch/powerpc/include/asm/fadump.h

@@ -72,37 +72,37 @@
 /* Utility macros */
 #define SKIP_TO_NEXT_CPU(reg_entry)					\
 ({									\
-	while (reg_entry->reg_id != REG_ID("CPUEND"))	\
+	while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND"))	\
 		reg_entry++;						\
 	reg_entry++;							\
 })
 
 /* Kernel Dump section info */
 struct fadump_section {
-	u32	request_flag;
-	u16	source_data_type;
-	u16	error_flags;
-	u64	source_address;
-	u64	source_len;
-	u64	bytes_dumped;
-	u64	destination_address;
+	__be32	request_flag;
+	__be16	source_data_type;
+	__be16	error_flags;
+	__be64	source_address;
+	__be64	source_len;
+	__be64	bytes_dumped;
+	__be64	destination_address;
 };
 
 /* ibm,configure-kernel-dump header. */
 struct fadump_section_header {
-	u32	dump_format_version;
-	u16	dump_num_sections;
-	u16	dump_status_flag;
-	u32	offset_first_dump_section;
+	__be32	dump_format_version;
+	__be16	dump_num_sections;
+	__be16	dump_status_flag;
+	__be32	offset_first_dump_section;
 
 	/* Fields for disk dump option. */
-	u32	dd_block_size;
-	u64	dd_block_offset;
-	u64	dd_num_blocks;
-	u32	dd_offset_disk_path;
+	__be32	dd_block_size;
+	__be64	dd_block_offset;
+	__be64	dd_num_blocks;
+	__be32	dd_offset_disk_path;
 
 	/* Maximum time allowed to prevent an automatic dump-reboot. */
-	u32	max_time_auto;
+	__be32	max_time_auto;
 };
 
 /*
@@ -174,15 +174,15 @@ static inline u64 str_to_u64(const char *str)
 
 /* Register save area header. */
 struct fadump_reg_save_area_header {
-	u64		magic_number;
-	u32		version;
-	u32		num_cpu_offset;
+	__be64		magic_number;
+	__be32		version;
+	__be32		num_cpu_offset;
 };
 
 /* Register entry. */
 struct fadump_reg_entry {
-	u64		reg_id;
-	u64		reg_value;
+	__be64		reg_id;
+	__be64		reg_value;
 };
 
 /* fadump crash info structure */

arch/powerpc/kernel/entry_64.S

@@ -659,7 +659,13 @@ _GLOBAL(ret_from_except_lite)
 3:
 #endif
 	bl	save_nvgprs
+	/*
+	 * Use a non volatile GPR to save and restore our thread_info flags
+	 * across the call to restore_interrupts.
+	 */
+	mr	r30,r4
 	bl	restore_interrupts
+	mr	r4,r30
 	addi	r3,r1,STACK_FRAME_OVERHEAD
 	bl	do_notify_resume
 	b	ret_from_except

arch/powerpc/mm/init_32.c

@@ -103,7 +103,7 @@ unsigned long __max_low_memory = MAX_LOW_MEM;
 /*
  * Check for command-line options that affect what MMU_init will do.
  */
-void MMU_setup(void)
+void __init MMU_setup(void)
 {
 	/* Check for nobats option (used in mapin_ram). */
 	if (strstr(boot_command_line, "nobats")) {

arch/powerpc/platforms/powernv/opal-lpc.c

@@ -216,14 +216,54 @@ static ssize_t lpc_debug_read(struct file *filp, char __user *ubuf,
 				   &data, len);
 		if (rc)
 			return -ENXIO;
+
+		/*
+		 * Now there is some trickery with the data returned by OPAL
+		 * as it's the desired data right justified in a 32-bit BE
+		 * word.
+		 *
+		 * This is a very bad interface and I'm to blame for it :-(
+		 *
+		 * So we can't just apply a 32-bit swap to what comes from OPAL,
+		 * because user space expects the *bytes* to be in their proper
+		 * respective positions (ie, LPC position).
+		 *
+		 * So what we really want to do here is to shift data right
+		 * appropriately on a LE kernel.
+		 *
+		 * IE. If the LPC transaction has bytes B0, B1, B2 and B3 in that
+		 * order, we have in memory written to by OPAL at the "data"
+		 * pointer:
+		 *
+		 *               Bytes:      OPAL "data"   LE "data"
+		 *   32-bit:   B0 B1 B2 B3   B0B1B2B3      B3B2B1B0
+		 *   16-bit:   B0 B1         0000B0B1      B1B00000
+		 *    8-bit:   B0            000000B0      B0000000
+		 *
+		 * So a BE kernel will have the leftmost of the above in the MSB
+		 * and rightmost in the LSB and can just then "cast" the u32 "data"
+		 * down to the appropriate quantity and write it.
+		 *
+		 * However, an LE kernel can't. It doesn't need to swap because a
+		 * load from data followed by a store to user are going to preserve
+		 * the byte ordering which is the wire byte order which is what the
+		 * user wants, but in order to "crop" to the right size, we need to
+		 * shift right first.
+		 */
 		switch(len) {
 		case 4:
 			rc = __put_user((u32)data, (u32 __user *)ubuf);
 			break;
 		case 2:
+#ifdef __LITTLE_ENDIAN__
+			data >>= 16;
+#endif
 			rc = __put_user((u16)data, (u16 __user *)ubuf);
 			break;
 		default:
+#ifdef __LITTLE_ENDIAN__
+			data >>= 24;
+#endif
 			rc = __put_user((u8)data, (u8 __user *)ubuf);
 			break;
 		}
@@ -263,12 +303,31 @@ static ssize_t lpc_debug_write(struct file *filp, const char __user *ubuf,
 			else if (todo > 1 && (pos & 1) == 0)
 				len = 2;
 		}
+
+		/*
+		 * Similarly to the read case, we have some trickery here but
+		 * it's different to handle. We need to pass the value to OPAL in
+		 * a register whose layout depends on the access size. We want
+		 * to reproduce the memory layout of the user, however we aren't
+		 * doing a load from user and a store to another memory location
+		 * which would achieve that. Here we pass the value to OPAL via
+		 * a register which is expected to contain the "BE" interpretation
+		 * of the byte sequence. IE: for a 32-bit access, byte 0 should be
+		 * in the MSB. So here we *do* need to byteswap on LE.
+		 *
+		 *           User bytes:    LE "data"  OPAL "data"
+		 *  32-bit:  B0 B1 B2 B3    B3B2B1B0   B0B1B2B3
+		 *  16-bit:  B0 B1          0000B1B0   0000B0B1
+		 *   8-bit:  B0             000000B0   000000B0
+		 */
 		switch(len) {
 		case 4:
 			rc = __get_user(data, (u32 __user *)ubuf);
+			data = cpu_to_be32(data);
 			break;
 		case 2:
 			rc = __get_user(data, (u16 __user *)ubuf);
+			data = cpu_to_be16(data);
 			break;
 		default:
 			rc = __get_user(data, (u8 __user *)ubuf);

arch/powerpc/platforms/pseries/dlpar.c

@@ -382,7 +382,7 @@ static int dlpar_online_cpu(struct device_node *dn)
 			BUG_ON(get_cpu_current_state(cpu)
 					!= CPU_STATE_OFFLINE);
 			cpu_maps_update_done();
-			rc = cpu_up(cpu);
+			rc = device_online(get_cpu_device(cpu));
 			if (rc)
 				goto out;
 			cpu_maps_update_begin();
@@ -467,7 +467,7 @@ static int dlpar_offline_cpu(struct device_node *dn)
 			if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
 				set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
 				cpu_maps_update_done();
-				rc = cpu_down(cpu);
+				rc = device_offline(get_cpu_device(cpu));
 				if (rc)
 					goto out;
 				cpu_maps_update_begin();

arch/powerpc/platforms/pseries/lpar.c

@@ -43,6 +43,7 @@
 #include <asm/trace.h>
 #include <asm/firmware.h>
 #include <asm/plpar_wrappers.h>
+#include <asm/fadump.h>
 
 #include "pseries.h"
 
@@ -247,8 +248,17 @@ static void pSeries_lpar_hptab_clear(void)
 	}
 
 #ifdef __LITTLE_ENDIAN__
-	/* Reset exceptions to big endian */
-	if (firmware_has_feature(FW_FEATURE_SET_MODE)) {
+	/*
+	 * Reset exceptions to big endian.
+	 *
+	 * FIXME this is a hack for kexec, we need to reset the exception
+	 * endian before starting the new kernel and this is a convenient place
+	 * to do it.
+	 *
+	 * This is also called on boot when a fadump happens. In that case we
+	 * must not change the exception endian mode.
+	 */
+	if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active()) {
 		long rc;
 
 		rc = pseries_big_endian_exceptions();