Merge tag 'md-3.5' of git://neil.brown.name/md

Pull md updates from NeilBrown: "It's been a busy cycle for md - lots of fun stuff here.. if you like this kind of thing :-) Main features: - RAID10 arrays can be reshaped - adding and removing devices and changing chunks (not 'far' array though) - allow RAID5 arrays to be reshaped with a backup file (not tested yet, but the priciple works fine for RAID10). - arrays can be reshaped while a bitmap is present - you no longer need to remove it first - SSSE3 support for RAID6 syndrome calculations and of course a number of minor fixes etc." * tag 'md-3.5' of git://neil.brown.name/md: (56 commits) md/bitmap: record the space available for the bitmap in the superblock. md/raid10: Remove extras after reshape to smaller number of devices. md/raid5: improve removal of extra devices after reshape. md: check the return of mddev_find() MD RAID1: Further conditionalize 'fullsync' DM RAID: Use md_error() in place of simply setting Faulty bit DM RAID: Record and handle missing devices DM RAID: Set recovery flags on resume md/raid5: Allow reshape while a bitmap is present. md/raid10: resize bitmap when required during reshape. md: allow array to be resized while bitmap is present. md/bitmap: make sure reshape request are reflected in superblock. md/bitmap: add bitmap_resize function to allow bitmap resizing. md/bitmap: use DIV_ROUND_UP instead of open-code md/bitmap: create a 'struct bitmap_counts' substructure of 'struct bitmap' md/bitmap: make bitmap bitops atomic. md/bitmap: make _page_attr bitops atomic. md/bitmap: merge bitmap_file_unmap and bitmap_file_put. md/bitmap: remove async freeing of bitmap file. md/bitmap: convert some spin_lock_irqsave to spin_lock_irq ...
2026-05-01 15:00:59 -07:00 · 2012-05-23 17:08:40 -07:00
parent 2c13bc0f8f 1dff2b87a3
commit c80ddb5263
25 changed files with 3128 additions and 872 deletions
@@ -115,9 +115,10 @@ cfi-sections := $(call as-instr,.cfi_sections .debug_frame,-DCONFIG_AS_CFI_SECTI

 # does binutils support specific instructions?
 asinstr := $(call as-instr,fxsaveq (%rax),-DCONFIG_AS_FXSAVEQ=1)
+avx_instr := $(call as-instr,vxorps %ymm0$(comma)%ymm1$(comma)%ymm2,-DCONFIG_AS_AVX=1)

-KBUILD_AFLAGS += $(cfi) $(cfi-sigframe) $(cfi-sections) $(asinstr)
-KBUILD_CFLAGS += $(cfi) $(cfi-sigframe) $(cfi-sections) $(asinstr)
+KBUILD_AFLAGS += $(cfi) $(cfi-sigframe) $(cfi-sections) $(asinstr) $(avx_instr)
+KBUILD_CFLAGS += $(cfi) $(cfi-sigframe) $(cfi-sections) $(asinstr) $(avx_instr)

 LDFLAGS := -m elf_$(UTS_MACHINE)

@@ -861,6 +861,9 @@ static struct xor_block_template xor_block_pIII_sse = {
 	.do_5 = xor_sse_5,
 };

+/* Also try the AVX routines */
+#include "xor_avx.h"
+
 /* Also try the generic routines.  */
 #include <asm-generic/xor.h>

@@ -871,6 +874,7 @@ do {							\
 	xor_speed(&xor_block_8regs_p);			\
 	xor_speed(&xor_block_32regs);			\
 	xor_speed(&xor_block_32regs_p);			\
+	AVX_XOR_SPEED;					\
 	if (cpu_has_xmm)				\
 		xor_speed(&xor_block_pIII_sse);		\
 	if (cpu_has_mmx) {				\
@@ -883,6 +887,6 @@ do {							\
   We may also be able to load into the L1 only depending on how the cpu
   deals with a load to a line that is being prefetched.  */
 #define XOR_SELECT_TEMPLATE(FASTEST)			\
-	(cpu_has_xmm ? &xor_block_pIII_sse : FASTEST)
+	AVX_SELECT(cpu_has_xmm ? &xor_block_pIII_sse : FASTEST)

 #endif /* _ASM_X86_XOR_32_H */
@@ -347,15 +347,21 @@ static struct xor_block_template xor_block_sse = {
 	.do_5 = xor_sse_5,
 };

+
+/* Also try the AVX routines */
+#include "xor_avx.h"
+
 #undef XOR_TRY_TEMPLATES
 #define XOR_TRY_TEMPLATES			\
 do {						\
+	AVX_XOR_SPEED;				\
 	xor_speed(&xor_block_sse);		\
 } while (0)

 /* We force the use of the SSE xor block because it can write around L2.
   We may also be able to load into the L1 only depending on how the cpu
   deals with a load to a line that is being prefetched.  */
-#define XOR_SELECT_TEMPLATE(FASTEST) (&xor_block_sse)
+#define XOR_SELECT_TEMPLATE(FASTEST) \
+	AVX_SELECT(&xor_block_sse)

 #endif /* _ASM_X86_XOR_64_H */
@@ -0,0 +1,214 @@
+#ifndef _ASM_X86_XOR_AVX_H
+#define _ASM_X86_XOR_AVX_H
+
+/*
+ * Optimized RAID-5 checksumming functions for AVX
+ *
+ * Copyright (C) 2012 Intel Corporation
+ * Author: Jim Kukunas <james.t.kukunas@linux.intel.com>
+ *
+ * Based on Ingo Molnar and Zach Brown's respective MMX and SSE routines
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+
+#ifdef CONFIG_AS_AVX
+
+#include <linux/compiler.h>
+#include <asm/i387.h>
+
+#define ALIGN32 __aligned(32)
+
+#define YMM_SAVED_REGS 4
+
+#define YMMS_SAVE \
+do { \
+	preempt_disable(); \
+	cr0 = read_cr0(); \
+	clts(); \
+	asm volatile("vmovaps %%ymm0, %0" : "=m" (ymm_save[0]) : : "memory"); \
+	asm volatile("vmovaps %%ymm1, %0" : "=m" (ymm_save[32]) : : "memory"); \
+	asm volatile("vmovaps %%ymm2, %0" : "=m" (ymm_save[64]) : : "memory"); \
+	asm volatile("vmovaps %%ymm3, %0" : "=m" (ymm_save[96]) : : "memory"); \
+} while (0);
+
+#define YMMS_RESTORE \
+do { \
+	asm volatile("sfence" : : : "memory"); \
+	asm volatile("vmovaps %0, %%ymm3" : : "m" (ymm_save[96])); \
+	asm volatile("vmovaps %0, %%ymm2" : : "m" (ymm_save[64])); \
+	asm volatile("vmovaps %0, %%ymm1" : : "m" (ymm_save[32])); \
+	asm volatile("vmovaps %0, %%ymm0" : : "m" (ymm_save[0])); \
+	write_cr0(cr0); \
+	preempt_enable(); \
+} while (0);
+
+#define BLOCK4(i) \
+		BLOCK(32 * i, 0) \
+		BLOCK(32 * (i + 1), 1) \
+		BLOCK(32 * (i + 2), 2) \
+		BLOCK(32 * (i + 3), 3)
+
+#define BLOCK16() \
+		BLOCK4(0) \
+		BLOCK4(4) \
+		BLOCK4(8) \
+		BLOCK4(12)
+
+static void xor_avx_2(unsigned long bytes, unsigned long *p0, unsigned long *p1)
+{
+	unsigned long cr0, lines = bytes >> 9;
+	char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+
+	YMMS_SAVE
+
+	while (lines--) {
+#undef BLOCK
+#define BLOCK(i, reg) \
+do { \
+	asm volatile("vmovdqa %0, %%ymm" #reg : : "m" (p1[i / sizeof(*p1)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm"  #reg : : \
+		"m" (p0[i / sizeof(*p0)])); \
+	asm volatile("vmovdqa %%ymm" #reg ", %0" : \
+		"=m" (p0[i / sizeof(*p0)])); \
+} while (0);
+
+		BLOCK16()
+
+		p0 = (unsigned long *)((uintptr_t)p0 + 512);
+		p1 = (unsigned long *)((uintptr_t)p1 + 512);
+	}
+
+	YMMS_RESTORE
+}
+
+static void xor_avx_3(unsigned long bytes, unsigned long *p0, unsigned long *p1,
+	unsigned long *p2)
+{
+	unsigned long cr0, lines = bytes >> 9;
+	char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+
+	YMMS_SAVE
+
+	while (lines--) {
+#undef BLOCK
+#define BLOCK(i, reg) \
+do { \
+	asm volatile("vmovdqa %0, %%ymm" #reg : : "m" (p2[i / sizeof(*p2)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p1[i / sizeof(*p1)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p0[i / sizeof(*p0)])); \
+	asm volatile("vmovdqa %%ymm" #reg ", %0" : \
+		"=m" (p0[i / sizeof(*p0)])); \
+} while (0);
+
+		BLOCK16()
+
+		p0 = (unsigned long *)((uintptr_t)p0 + 512);
+		p1 = (unsigned long *)((uintptr_t)p1 + 512);
+		p2 = (unsigned long *)((uintptr_t)p2 + 512);
+	}
+
+	YMMS_RESTORE
+}
+
+static void xor_avx_4(unsigned long bytes, unsigned long *p0, unsigned long *p1,
+	unsigned long *p2, unsigned long *p3)
+{
+	unsigned long cr0, lines = bytes >> 9;
+	char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+
+	YMMS_SAVE
+
+	while (lines--) {
+#undef BLOCK
+#define BLOCK(i, reg) \
+do { \
+	asm volatile("vmovdqa %0, %%ymm" #reg : : "m" (p3[i / sizeof(*p3)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p2[i / sizeof(*p2)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p1[i / sizeof(*p1)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p0[i / sizeof(*p0)])); \
+	asm volatile("vmovdqa %%ymm" #reg ", %0" : \
+		"=m" (p0[i / sizeof(*p0)])); \
+} while (0);
+
+		BLOCK16();
+
+		p0 = (unsigned long *)((uintptr_t)p0 + 512);
+		p1 = (unsigned long *)((uintptr_t)p1 + 512);
+		p2 = (unsigned long *)((uintptr_t)p2 + 512);
+		p3 = (unsigned long *)((uintptr_t)p3 + 512);
+	}
+
+	YMMS_RESTORE
+}
+
+static void xor_avx_5(unsigned long bytes, unsigned long *p0, unsigned long *p1,
+	unsigned long *p2, unsigned long *p3, unsigned long *p4)
+{
+	unsigned long cr0, lines = bytes >> 9;
+	char ymm_save[32 * YMM_SAVED_REGS] ALIGN32;
+
+	YMMS_SAVE
+
+	while (lines--) {
+#undef BLOCK
+#define BLOCK(i, reg) \
+do { \
+	asm volatile("vmovdqa %0, %%ymm" #reg : : "m" (p4[i / sizeof(*p4)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p3[i / sizeof(*p3)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p2[i / sizeof(*p2)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p1[i / sizeof(*p1)])); \
+	asm volatile("vxorps %0, %%ymm" #reg ", %%ymm" #reg : : \
+		"m" (p0[i / sizeof(*p0)])); \
+	asm volatile("vmovdqa %%ymm" #reg ", %0" : \
+		"=m" (p0[i / sizeof(*p0)])); \
+} while (0);
+
+		BLOCK16()
+
+		p0 = (unsigned long *)((uintptr_t)p0 + 512);
+		p1 = (unsigned long *)((uintptr_t)p1 + 512);
+		p2 = (unsigned long *)((uintptr_t)p2 + 512);
+		p3 = (unsigned long *)((uintptr_t)p3 + 512);
+		p4 = (unsigned long *)((uintptr_t)p4 + 512);
+	}
+
+	YMMS_RESTORE
+}
+
+static struct xor_block_template xor_block_avx = {
+	.name = "avx",
+	.do_2 = xor_avx_2,
+	.do_3 = xor_avx_3,
+	.do_4 = xor_avx_4,
+	.do_5 = xor_avx_5,
+};
+
+#define AVX_XOR_SPEED \
+do { \
+	if (cpu_has_avx) \
+		xor_speed(&xor_block_avx); \
+} while (0)
+
+#define AVX_SELECT(FASTEST) \
+	(cpu_has_avx ? &xor_block_avx : FASTEST)
+
+#else
+
+#define AVX_XOR_SPEED {}
+
+#define AVX_SELECT(FASTEST) (FASTEST)
+
+#endif
+#endif
@@ -21,6 +21,7 @@
 #include <linux/gfp.h>
 #include <linux/raid/xor.h>
 #include <linux/jiffies.h>
+#include <linux/preempt.h>
 #include <asm/xor.h>

 /* The xor routines to use.  */
@@ -63,12 +64,14 @@ static void
 do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
 {
 	int speed;
-	unsigned long now;
+	unsigned long now, j;
 	int i, count, max;

 	tmpl->next = template_list;
 	template_list = tmpl;

+	preempt_disable();
+
 	/*
 	 * Count the number of XORs done during a whole jiffy, and use
 	 * this to calculate the speed of checksumming.  We use a 2-page
@@ -76,9 +79,11 @@ do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
 	 */
 	max = 0;
 	for (i = 0; i < 5; i++) {
-		now = jiffies;
+		j = jiffies;
 		count = 0;
-		while (jiffies == now) {
+		while ((now = jiffies) == j)
+			cpu_relax();
+		while (time_before(jiffies, now + 1)) {
 			mb(); /* prevent loop optimzation */
 			tmpl->do_2(BENCH_SIZE, b1, b2);
 			mb();
@@ -89,6 +94,8 @@ do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2)
 			max = count;
 	}

+	preempt_enable();
+
 	speed = max * (HZ * BENCH_SIZE / 1024);
 	tmpl->speed = speed;

@@ -111,9 +111,9 @@ typedef __u16 bitmap_counter_t;

 /* use these for bitmap->flags and bitmap->sb->state bit-fields */
 enum bitmap_state {
-	BITMAP_STALE  = 0x002,  /* the bitmap file is out of date or had -EIO */
-	BITMAP_WRITE_ERROR = 0x004, /* A write error has occurred */
-	BITMAP_HOSTENDIAN = 0x8000,
+	BITMAP_STALE	   = 1,  /* the bitmap file is out of date or had -EIO */
+	BITMAP_WRITE_ERROR = 2, /* A write error has occurred */
+	BITMAP_HOSTENDIAN  =15,
 };

 /* the superblock at the front of the bitmap file -- little endian */
@@ -128,8 +128,10 @@ typedef struct bitmap_super_s {
 	__le32 chunksize;    /* 52  the bitmap chunk size in bytes */
 	__le32 daemon_sleep; /* 56  seconds between disk flushes */
 	__le32 write_behind; /* 60  number of outstanding write-behind writes */
+	__le32 sectors_reserved; /* 64 number of 512-byte sectors that are
+				  * reserved for the bitmap. */

-	__u8  pad[256 - 64]; /* set to zero */
+	__u8  pad[256 - 68]; /* set to zero */
 } bitmap_super_t;

 /* notes:
@@ -159,36 +161,49 @@ struct bitmap_page {
 	 * pointer and use it as two counters itself
 	 */
 	unsigned int hijacked:1;
+	/*
+	 * If any counter in this page is '1' or '2' - and so could be
+	 * cleared then that page is marked as 'pending'
+	 */
+	unsigned int pending:1;
 	/*
 	 * count of dirty bits on the page
 	 */
-	unsigned int  count:31;
+	unsigned int  count:30;
 };

 /* the main bitmap structure - one per mddev */
 struct bitmap {
-	struct bitmap_page *bp;
-	unsigned long pages; /* total number of pages in the bitmap */
-	unsigned long missing_pages; /* number of pages not yet allocated */
+
+	struct bitmap_counts {
+		spinlock_t lock;
+		struct bitmap_page *bp;
+		unsigned long pages;		/* total number of pages
+						 * in the bitmap */
+		unsigned long missing_pages;	/* number of pages
+						 * not yet allocated */
+		unsigned long chunkshift;	/* chunksize = 2^chunkshift
+						 * (for bitops) */
+		unsigned long chunks;		/* Total number of data
+						 * chunks for the array */
+	} counts;

 	struct mddev *mddev; /* the md device that the bitmap is for */

-	/* bitmap chunksize -- how much data does each bit represent? */
-	unsigned long chunkshift; /* chunksize = 2^(chunkshift+9) (for bitops) */
-	unsigned long chunks; /* total number of data chunks for the array */
-
 	__u64	events_cleared;
 	int need_sync;

-	/* bitmap spinlock */
-	spinlock_t lock;
-
-	struct file *file; /* backing disk file */
-	struct page *sb_page; /* cached copy of the bitmap file superblock */
-	struct page **filemap; /* list of cache pages for the file */
-	unsigned long *filemap_attr; /* attributes associated w/ filemap pages */
-	unsigned long file_pages; /* number of pages in the file */
-	int last_page_size; /* bytes in the last page */
+	struct bitmap_storage {
+		struct file *file;		/* backing disk file */
+		struct page *sb_page;		/* cached copy of the bitmap
+						 * file superblock */
+		struct page **filemap;		/* list of cache pages for
+						 * the file */
+		unsigned long *filemap_attr;	/* attributes associated
+						 * w/ filemap pages */
+		unsigned long file_pages;	/* number of pages in the file*/
+		unsigned long bytes;		/* total bytes in the bitmap */
+	} storage;

 	unsigned long flags;

@@ -242,6 +257,9 @@ void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector);

 void bitmap_unplug(struct bitmap *bitmap);
 void bitmap_daemon_work(struct mddev *mddev);
+
+int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
+		  int chunksize, int init);
 #endif

 #endif
@@ -155,10 +155,7 @@ static void context_free(struct raid_set *rs)
 	for (i = 0; i < rs->md.raid_disks; i++) {
 		if (rs->dev[i].meta_dev)
 			dm_put_device(rs->ti, rs->dev[i].meta_dev);
-		if (rs->dev[i].rdev.sb_page)
-			put_page(rs->dev[i].rdev.sb_page);
-		rs->dev[i].rdev.sb_page = NULL;
-		rs->dev[i].rdev.sb_loaded = 0;
+		md_rdev_clear(&rs->dev[i].rdev);
 		if (rs->dev[i].data_dev)
 			dm_put_device(rs->ti, rs->dev[i].data_dev);
 	}
@@ -606,7 +603,7 @@ static int read_disk_sb(struct md_rdev *rdev, int size)
 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, 1)) {
 		DMERR("Failed to read superblock of device at position %d",
 		      rdev->raid_disk);
-		set_bit(Faulty, &rdev->flags);
+		md_error(rdev->mddev, rdev);
 		return -EINVAL;
 	}

@@ -617,16 +614,18 @@ static int read_disk_sb(struct md_rdev *rdev, int size)

 static void super_sync(struct mddev *mddev, struct md_rdev *rdev)
 {
-	struct md_rdev *r;
+	int i;
 	uint64_t failed_devices;
 	struct dm_raid_superblock *sb;
+	struct raid_set *rs = container_of(mddev, struct raid_set, md);

 	sb = page_address(rdev->sb_page);
 	failed_devices = le64_to_cpu(sb->failed_devices);

-	rdev_for_each(r, mddev)
-		if ((r->raid_disk >= 0) && test_bit(Faulty, &r->flags))
-			failed_devices |= (1ULL << r->raid_disk);
+	for (i = 0; i < mddev->raid_disks; i++)
+		if (!rs->dev[i].data_dev ||
+		    test_bit(Faulty, &(rs->dev[i].rdev.flags)))
+			failed_devices |= (1ULL << i);

 	memset(sb, 0, sizeof(*sb));

@@ -1252,12 +1251,13 @@ static void raid_resume(struct dm_target *ti)
 {
 	struct raid_set *rs = ti->private;

+	set_bit(MD_CHANGE_DEVS, &rs->md.flags);
 	if (!rs->bitmap_loaded) {
 		bitmap_load(&rs->md);
 		rs->bitmap_loaded = 1;
-	} else
-		md_wakeup_thread(rs->md.thread);
+	}

+	clear_bit(MD_RECOVERY_FROZEN, &rs->md.recovery);
 	mddev_resume(&rs->md);
 }

@@ -55,6 +55,7 @@ struct md_rdev {
 	int		sb_loaded;
 	__u64		sb_events;
 	sector_t	data_offset;	/* start of data in array */
+	sector_t	new_data_offset;/* only relevant while reshaping */
 	sector_t 	sb_start;	/* offset of the super block (in 512byte sectors) */
 	int		sb_size;	/* bytes in the superblock */
 	int		preferred_minor;	/* autorun support */
@@ -193,8 +194,9 @@ static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
 	return 0;
 }
 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
-			      int acknowledged);
-extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors);
+			      int is_new);
+extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
+				int is_new);
 extern void md_ack_all_badblocks(struct badblocks *bb);

 struct mddev {
@@ -262,6 +264,7 @@ struct mddev {
 	sector_t			reshape_position;
 	int				delta_disks, new_level, new_layout;
 	int				new_chunk_sectors;
+	int				reshape_backwards;

 	atomic_t			plug_cnt;	/* If device is expecting
 							 * more bios soon.
@@ -390,10 +393,13 @@ struct mddev {
 						 * For external metadata, offset
 						 * from start of device. 
 						 */
+		unsigned long		space; /* space available at this offset */
 		loff_t			default_offset; /* this is the offset to use when
 							 * hot-adding a bitmap.  It should
 							 * eventually be settable by sysfs.
 							 */
+		unsigned long		default_space; /* space available at
+							* default offset */
 		struct mutex		mutex;
 		unsigned long		chunksize;
 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
@@ -591,6 +597,7 @@ extern void md_write_start(struct mddev *mddev, struct bio *bi);
 extern void md_write_end(struct mddev *mddev);
 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
+extern void md_finish_reshape(struct mddev *mddev);

 extern int mddev_congested(struct mddev *mddev, int bits);
 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
@@ -615,6 +622,7 @@ extern int md_run(struct mddev *mddev);
 extern void md_stop(struct mddev *mddev);
 extern void md_stop_writes(struct mddev *mddev);
 extern int md_rdev_init(struct md_rdev *rdev);
+extern void md_rdev_clear(struct md_rdev *rdev);

 extern void mddev_suspend(struct mddev *mddev);
 extern void mddev_resume(struct mddev *mddev);
@@ -1859,7 +1859,9 @@ static void fix_read_error(struct r1conf *conf, int read_disk,

 			rdev = conf->mirrors[d].rdev;
 			if (rdev &&
-			    test_bit(In_sync, &rdev->flags) &&
+			    (test_bit(In_sync, &rdev->flags) ||
+			     (!test_bit(Faulty, &rdev->flags) &&
+			      rdev->recovery_offset >= sect + s)) &&
 			    is_badblock(rdev, sect, s,
 					&first_bad, &bad_sectors) == 0 &&
 			    sync_page_io(rdev, sect, s<<9,
@@ -2024,7 +2026,7 @@ static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio
 			continue;
 		if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 		    test_bit(R1BIO_MadeGood, &r1_bio->state)) {
-			rdev_clear_badblocks(rdev, r1_bio->sector, s);
+			rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
 		}
 		if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 		    test_bit(R1BIO_WriteError, &r1_bio->state)) {
@@ -2044,7 +2046,7 @@ static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
 			struct md_rdev *rdev = conf->mirrors[m].rdev;
 			rdev_clear_badblocks(rdev,
 					     r1_bio->sector,
-					     r1_bio->sectors);
+					     r1_bio->sectors, 0);
 			rdev_dec_pending(rdev, conf->mddev);
 		} else if (r1_bio->bios[m] != NULL) {
 			/* This drive got a write error.  We need to
@@ -2598,7 +2600,8 @@ static struct r1conf *setup_conf(struct mddev *mddev)
 		if (!disk->rdev ||
 		    !test_bit(In_sync, &disk->rdev->flags)) {
 			disk->head_position = 0;
-			if (disk->rdev)
+			if (disk->rdev &&
+			    (disk->rdev->saved_raid_disk < 0))
 				conf->fullsync = 1;
 		} else if (conf->last_used < 0)
 			/*
@@ -2750,9 +2753,16 @@ static int raid1_resize(struct mddev *mddev, sector_t sectors)
 	 * any io in the removed space completes, but it hardly seems
 	 * worth it.
 	 */
-	md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0));
-	if (mddev->array_sectors > raid1_size(mddev, sectors, 0))
+	sector_t newsize = raid1_size(mddev, sectors, 0);
+	if (mddev->external_size &&
+	    mddev->array_sectors > newsize)
 		return -EINVAL;
+	if (mddev->bitmap) {
+		int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
+		if (ret)
+			return ret;
+	}
+	md_set_array_sectors(mddev, newsize);
 	set_capacity(mddev->gendisk, mddev->array_sectors);
 	revalidate_disk(mddev->gendisk);
 	if (sectors > mddev->dev_sectors &&
@@ -14,32 +14,38 @@ struct mirror_info {
 struct r10conf {
 	struct mddev		*mddev;
 	struct mirror_info	*mirrors;
-	int			raid_disks;
+	struct mirror_info	*mirrors_new, *mirrors_old;
 	spinlock_t		device_lock;

 	/* geometry */
-	int			near_copies;  /* number of copies laid out
+	struct geom {
+		int		raid_disks;
+		int		near_copies;  /* number of copies laid out
 					       * raid0 style */
-	int 			far_copies;   /* number of copies laid out
+		int		far_copies;   /* number of copies laid out
 					       * at large strides across drives
 					       */
-	int			far_offset;   /* far_copies are offset by 1
+		int		far_offset;   /* far_copies are offset by 1
 					       * stripe instead of many
 					       */
-	int			copies;	      /* near_copies * far_copies.
-					       * must be <= raid_disks
-					       */
-	sector_t		stride;	      /* distance between far copies.
+		sector_t	stride;	      /* distance between far copies.
 					       * This is size / far_copies unless
 					       * far_offset, in which case it is
 					       * 1 stripe.
 					       */
+		int		chunk_shift; /* shift from chunks to sectors */
+		sector_t	chunk_mask;
+	} prev, geo;
+	int			copies;	      /* near_copies * far_copies.
+					       * must be <= raid_disks
+					       */

 	sector_t		dev_sectors;  /* temp copy of
 					       * mddev->dev_sectors */
-
-	int			chunk_shift; /* shift from chunks to sectors */
-	sector_t		chunk_mask;
+	sector_t		reshape_progress;
+	sector_t		reshape_safe;
+	unsigned long		reshape_checkpoint;
+	sector_t		offset_diff;

 	struct list_head	retry_list;
 	/* queue pending writes and submit them on unplug */
@@ -136,6 +142,7 @@ enum r10bio_state {
 	R10BIO_Uptodate,
 	R10BIO_IsSync,
 	R10BIO_IsRecover,
+	R10BIO_IsReshape,
 	R10BIO_Degraded,
 /* Set ReadError on bios that experience a read error
 * so that raid10d knows what to do with them.
@@ -146,5 +153,10 @@ enum r10bio_state {
 */
 	R10BIO_MadeGood,
 	R10BIO_WriteError,
+/* During a reshape we might be performing IO on the
+ * 'previous' part of the array, in which case this
+ * flag is set
+ */
+	R10BIO_Previous,
 };
 #endif
@@ -285,6 +285,7 @@ enum r5dev_flags {
 			 */
 	R5_Wantdrain,	/* dev->towrite needs to be drained */
 	R5_WantFUA,	/* Write should be FUA */
+	R5_SyncIO,	/* The IO is sync */
 	R5_WriteError,	/* got a write error - need to record it */
 	R5_MadeGood,	/* A bad block has been fixed by writing to it */
 	R5_ReadRepl,	/* Will/did read from replacement rather than orig */
@@ -385,6 +386,12 @@ struct r5conf {
 	short			generation; /* increments with every reshape */
 	unsigned long		reshape_checkpoint; /* Time we last updated
 						     * metadata */
+	long long		min_offset_diff; /* minimum difference between
+						  * data_offset and
+						  * new_data_offset across all
+						  * devices.  May be negative,
+						  * but is closest to zero.
+						  */

 	struct list_head	handle_list; /* stripes needing handling */
 	struct list_head	hold_list; /* preread ready stripes */
@@ -233,7 +233,10 @@ struct mdp_superblock_1 {
 	__le32	delta_disks;	/* change in number of raid_disks		*/
 	__le32	new_layout;	/* new layout					*/
 	__le32	new_chunk;	/* new chunk size (512byte sectors)		*/
-	__u8	pad1[128-124];	/* set to 0 when written */
+	__le32  new_offset;	/* signed number to add to data_offset in new
+				 * layout.  0 == no-change.  This can be
+				 * different on each device in the array.
+				 */

 	/* constant this-device information - 64 bytes */
 	__le64	data_offset;	/* sector start of data, often 0 */
@@ -281,10 +284,18 @@ struct mdp_superblock_1 {
 					    * active device with same 'role'.
 					    * 'recovery_offset' is also set.
 					    */
+#define	MD_FEATURE_RESHAPE_BACKWARDS	32 /* Reshape doesn't change number
+					    * of devices, but is going
+					    * backwards anyway.
+					    */
+#define	MD_FEATURE_NEW_OFFSET		64 /* new_offset must be honoured */
 #define	MD_FEATURE_ALL			(MD_FEATURE_BITMAP_OFFSET	\
 					|MD_FEATURE_RECOVERY_OFFSET	\
 					|MD_FEATURE_RESHAPE_ACTIVE	\
 					|MD_FEATURE_BAD_BLOCKS		\
-					|MD_FEATURE_REPLACEMENT)
+					|MD_FEATURE_REPLACEMENT		\
+					|MD_FEATURE_RESHAPE_BACKWARDS	\
+					|MD_FEATURE_NEW_OFFSET		\
+					)

 #endif 
@@ -99,8 +99,20 @@ extern const struct raid6_calls raid6_altivec2;
 extern const struct raid6_calls raid6_altivec4;
 extern const struct raid6_calls raid6_altivec8;

+struct raid6_recov_calls {
+	void (*data2)(int, size_t, int, int, void **);
+	void (*datap)(int, size_t, int, void **);
+	int  (*valid)(void);
+	const char *name;
+	int priority;
+};
+
+extern const struct raid6_recov_calls raid6_recov_intx1;
+extern const struct raid6_recov_calls raid6_recov_ssse3;
+
 /* Algorithm list */
 extern const struct raid6_calls * const raid6_algos[];
+extern const struct raid6_recov_calls *const raid6_recov_algos[];
 int raid6_select_algo(void);

 /* Return values from chk_syndrome */
@@ -111,14 +123,16 @@ int raid6_select_algo(void);

 /* Galois field tables */
 extern const u8 raid6_gfmul[256][256] __attribute__((aligned(256)));
+extern const u8 raid6_vgfmul[256][32] __attribute__((aligned(256)));
 extern const u8 raid6_gfexp[256]      __attribute__((aligned(256)));
 extern const u8 raid6_gfinv[256]      __attribute__((aligned(256)));
 extern const u8 raid6_gfexi[256]      __attribute__((aligned(256)));

 /* Recovery routines */
-void raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
+extern void (*raid6_2data_recov)(int disks, size_t bytes, int faila, int failb,
 		       void **ptrs);
-void raid6_datap_recov(int disks, size_t bytes, int faila, void **ptrs);
+extern void (*raid6_datap_recov)(int disks, size_t bytes, int faila,
+			void **ptrs);
 void raid6_dual_recov(int disks, size_t bytes, int faila, int failb,
 		      void **ptrs);

@@ -1,6 +1,6 @@
 obj-$(CONFIG_RAID6_PQ)	+= raid6_pq.o

-raid6_pq-y	+= algos.o recov.o tables.o int1.o int2.o int4.o \
+raid6_pq-y	+= algos.o recov.o recov_ssse3.o tables.o int1.o int2.o int4.o \
 		   int8.o int16.o int32.o altivec1.o altivec2.o altivec4.o \
 		   altivec8.o mmx.o sse1.o sse2.o
 hostprogs-y	+= mktables
@@ -17,11 +17,11 @@
 */

 #include <linux/raid/pq.h>
-#include <linux/module.h>
 #ifndef __KERNEL__
 #include <sys/mman.h>
 #include <stdio.h>
 #else
+#include <linux/module.h>
 #include <linux/gfp.h>
 #if !RAID6_USE_EMPTY_ZERO_PAGE
 /* In .bss so it's zeroed */
@@ -34,10 +34,6 @@ struct raid6_calls raid6_call;
 EXPORT_SYMBOL_GPL(raid6_call);

 const struct raid6_calls * const raid6_algos[] = {
-	&raid6_intx1,
-	&raid6_intx2,
-	&raid6_intx4,
-	&raid6_intx8,
 #if defined(__ia64__)
 	&raid6_intx16,
 	&raid6_intx32,
@@ -61,6 +57,24 @@ const struct raid6_calls * const raid6_algos[] = {
 	&raid6_altivec4,
 	&raid6_altivec8,
 #endif
+	&raid6_intx1,
+	&raid6_intx2,
+	&raid6_intx4,
+	&raid6_intx8,
+	NULL
+};
+
+void (*raid6_2data_recov)(int, size_t, int, int, void **);
+EXPORT_SYMBOL_GPL(raid6_2data_recov);
+
+void (*raid6_datap_recov)(int, size_t, int, void **);
+EXPORT_SYMBOL_GPL(raid6_datap_recov);
+
+const struct raid6_recov_calls *const raid6_recov_algos[] = {
+#if (defined(__i386__) || defined(__x86_64__)) && !defined(__arch_um__)
+	&raid6_recov_ssse3,
+#endif
+	&raid6_recov_intx1,
 	NULL
 };

@@ -72,59 +86,55 @@ const struct raid6_calls * const raid6_algos[] = {
 #define time_before(x, y) ((x) < (y))
 #endif

-/* Try to pick the best algorithm */
-/* This code uses the gfmul table as convenient data set to abuse */
-
-int __init raid6_select_algo(void)
+static inline const struct raid6_recov_calls *raid6_choose_recov(void)
 {
-	const struct raid6_calls * const * algo;
-	const struct raid6_calls * best;
-	char *syndromes;
-	void *dptrs[(65536/PAGE_SIZE)+2];
-	int i, disks;
-	unsigned long perf, bestperf;
-	int bestprefer;
-	unsigned long j0, j1;
+	const struct raid6_recov_calls *const *algo;
+	const struct raid6_recov_calls *best;

-	disks = (65536/PAGE_SIZE)+2;
-	for ( i = 0 ; i < disks-2 ; i++ ) {
-		dptrs[i] = ((char *)raid6_gfmul) + PAGE_SIZE*i;
-	}
+	for (best = NULL, algo = raid6_recov_algos; *algo; algo++)
+		if (!best || (*algo)->priority > best->priority)
+			if (!(*algo)->valid || (*algo)->valid())
+				best = *algo;

-	/* Normal code - use a 2-page allocation to avoid D$ conflict */
-	syndromes = (void *) __get_free_pages(GFP_KERNEL, 1);
+	if (best) {
+		raid6_2data_recov = best->data2;
+		raid6_datap_recov = best->datap;

-	if ( !syndromes ) {
-		printk("raid6: Yikes!  No memory available.\n");
-		return -ENOMEM;
-	}
+		printk("raid6: using %s recovery algorithm\n", best->name);
+	} else
+		printk("raid6: Yikes! No recovery algorithm found!\n");

-	dptrs[disks-2] = syndromes;
-	dptrs[disks-1] = syndromes + PAGE_SIZE;
+	return best;
+}

-	bestperf = 0;  bestprefer = 0;  best = NULL;
+static inline const struct raid6_calls *raid6_choose_gen(
+	void *(*const dptrs)[(65536/PAGE_SIZE)+2], const int disks)
+{
+	unsigned long perf, bestperf, j0, j1;
+	const struct raid6_calls *const *algo;
+	const struct raid6_calls *best;
+
+	for (bestperf = 0, best = NULL, algo = raid6_algos; *algo; algo++) {
+		if (!best || (*algo)->prefer >= best->prefer) {
+			if ((*algo)->valid && !(*algo)->valid())
+				continue;

-	for ( algo = raid6_algos ; *algo ; algo++ ) {
-		if ( !(*algo)->valid || (*algo)->valid() ) {
 			perf = 0;

 			preempt_disable();
 			j0 = jiffies;
-			while ( (j1 = jiffies) == j0 )
+			while ((j1 = jiffies) == j0)
 				cpu_relax();
 			while (time_before(jiffies,
 					    j1 + (1<<RAID6_TIME_JIFFIES_LG2))) {
-				(*algo)->gen_syndrome(disks, PAGE_SIZE, dptrs);
+				(*algo)->gen_syndrome(disks, PAGE_SIZE, *dptrs);
 				perf++;
 			}
 			preempt_enable();

-			if ( (*algo)->prefer > bestprefer ||
-			     ((*algo)->prefer == bestprefer &&
-			      perf > bestperf) ) {
-				best = *algo;
-				bestprefer = best->prefer;
+			if (perf > bestperf) {
 				bestperf = perf;
+				best = *algo;
 			}
 			printk("raid6: %-8s %5ld MB/s\n", (*algo)->name,
 			       (perf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
@@ -139,9 +149,46 @@ int __init raid6_select_algo(void)
 	} else
 		printk("raid6: Yikes!  No algorithm found!\n");

+	return best;
+}
+
+
+/* Try to pick the best algorithm */
+/* This code uses the gfmul table as convenient data set to abuse */
+
+int __init raid6_select_algo(void)
+{
+	const int disks = (65536/PAGE_SIZE)+2;
+
+	const struct raid6_calls *gen_best;
+	const struct raid6_recov_calls *rec_best;
+	char *syndromes;
+	void *dptrs[(65536/PAGE_SIZE)+2];
+	int i;
+
+	for (i = 0; i < disks-2; i++)
+		dptrs[i] = ((char *)raid6_gfmul) + PAGE_SIZE*i;
+
+	/* Normal code - use a 2-page allocation to avoid D$ conflict */
+	syndromes = (void *) __get_free_pages(GFP_KERNEL, 1);
+
+	if (!syndromes) {
+		printk("raid6: Yikes!  No memory available.\n");
+		return -ENOMEM;
+	}
+
+	dptrs[disks-2] = syndromes;
+	dptrs[disks-1] = syndromes + PAGE_SIZE;
+
+	/* select raid gen_syndrome function */
+	gen_best = raid6_choose_gen(&dptrs, disks);
+
+	/* select raid recover functions */
+	rec_best = raid6_choose_recov();
+
 	free_pages((unsigned long)syndromes, 1);

-	return best ? 0 : -EINVAL;
+	return gen_best && rec_best ? 0 : -EINVAL;
 }

 static void raid6_exit(void)
@@ -81,6 +81,31 @@ int main(int argc, char *argv[])
 	printf("EXPORT_SYMBOL(raid6_gfmul);\n");
 	printf("#endif\n");

+	/* Compute vector multiplication table */
+	printf("\nconst u8  __attribute__((aligned(256)))\n"
+		"raid6_vgfmul[256][32] =\n"
+		"{\n");
+	for (i = 0; i < 256; i++) {
+		printf("\t{\n");
+		for (j = 0; j < 16; j += 8) {
+			printf("\t\t");
+			for (k = 0; k < 8; k++)
+				printf("0x%02x,%c", gfmul(i, j + k),
+				       (k == 7) ? '\n' : ' ');
+		}
+		for (j = 0; j < 16; j += 8) {
+			printf("\t\t");
+			for (k = 0; k < 8; k++)
+				printf("0x%02x,%c", gfmul(i, (j + k) << 4),
+				       (k == 7) ? '\n' : ' ');
+		}
+		printf("\t},\n");
+	}
+	printf("};\n");
+	printf("#ifdef __KERNEL__\n");
+	printf("EXPORT_SYMBOL(raid6_vgfmul);\n");
+	printf("#endif\n");
+
 	/* Compute power-of-2 table (exponent) */
 	v = 1;
 	printf("\nconst u8 __attribute__((aligned(256)))\n"
--- a/Show More
+++ b/Show More