[RFC 0/2] ACPI, APEI, GHES, printk support for recoverable error via NMI

[RFC 0/2] ACPI, APEI, GHES, printk support for recoverable error via NMI

[Huang Ying]

This patchset is based on the lockless list patchset.

[RFC 1/2] lib, Make gen_pool memory allocator lockless
[RFC 2/2] ACPI, APEI, GHES, printk support for recoverable error via NMI

[RFC 1/2] lib, Make gen_pool memory allocator lockless

[Huang Ying]

This version of the gen_pool memory allocator supports lockless
operation.

This makes it safe to use in NMI handlers and other special
unblockable contexts that could otherwise deadlock on locks.  This is
implemented by using atomic operations and retries on any conflicts.
The disadvantage is that there may be livelocks in extreme cases.  For
better scalability, one gen_pool allocator can be used for each CPU.

The lockless operation only works if there is enough memory available.
If new memory is added to the pool a lock has to be still taken.  So
any user relying on locklessness has to ensure that sufficient memory
is preallocated.

The basic atomic operation of this allocator is cmpxchg on long.  On
architectures that don't have NMI-safe cmpxchg implementation, the
allocator can NOT be used in NMI handler.  So code uses the allocator
in NMI handler should depend on CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
---
 include/linux/bitmap.h   |    1 
 include/linux/genalloc.h |   46 +++++++-
 lib/bitmap.c             |    2 
 lib/genalloc.c           |  254 ++++++++++++++++++++++++++++++++++++++---------
 4 files changed, 248 insertions(+), 55 deletions(-)

--- a/include/linux/bitmap.h
+++ b/include/linux/bitmap.h
@@ -142,6 +142,7 @@ extern void bitmap_release_region(unsign
 extern int bitmap_allocate_region(unsigned long *bitmap, int pos, int order);
 extern void bitmap_copy_le(void *dst, const unsigned long *src, int nbits);
 
+#define BITMAP_FIRST_WORD_MASK(start) (~0UL << ((start) % BITS_PER_LONG))
 #define BITMAP_LAST_WORD_MASK(nbits)					\
 (									\
 	((nbits) % BITS_PER_LONG) ?					\
--- a/include/linux/genalloc.h
+++ b/include/linux/genalloc.h
@@ -1,8 +1,28 @@
+#ifndef GENALLOC_H
+#define GENALLOC_H
 /*
- * Basic general purpose allocator for managing special purpose memory
- * not managed by the regular kmalloc/kfree interface.
- * Uses for this includes on-device special memory, uncached memory
- * etc.
+ * Basic general purpose allocator for managing special purpose
+ * memory, for example, memory that is not managed by the regular
+ * kmalloc/kfree interface.  Uses for this includes on-device special
+ * memory, uncached memory etc.
+ *
+ * It is safe to use the allocator in NMI handlers and other special
+ * unblockable contexts that could otherwise deadlock on locks.  This
+ * is implemented by using atomic operations and retries on any
+ * conflicts.  The disadvantage is that there may be livelocks in
+ * extreme cases.  For better scalability, one allocator can be used
+ * for each CPU.
+ *
+ * The lockless operation only works if there is enough memory
+ * available.  If new memory is added to the pool a lock has to be
+ * still taken.  So any user relying on locklessness has to ensure
+ * that sufficient memory is preallocated.
+ *
+ * The basic atomic operation of this allocator is cmpxchg on long.
+ * On architectures that don't have NMI-safe cmpxchg implementation,
+ * the allocator can NOT be used in NMI handler.  So code uses the
+ * allocator in NMI handler should depend on
+ * CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG.
  *
  * This source code is licensed under the GNU General Public License,
  * Version 2.  See the file COPYING for more details.
@@ -13,7 +33,7 @@
  *  General purpose special memory pool descriptor.
  */
 struct gen_pool {
-	rwlock_t lock;
+	spinlock_t lock;
 	struct list_head chunks;	/* list of chunks in this pool */
 	int min_alloc_order;		/* minimum allocation order */
 };
@@ -22,15 +42,29 @@ struct gen_pool {
  *  General purpose special memory pool chunk descriptor.
  */
 struct gen_pool_chunk {
-	spinlock_t lock;
 	struct list_head next_chunk;	/* next chunk in pool */
+	atomic_t avail;
 	unsigned long start_addr;	/* starting address of memory chunk */
 	unsigned long end_addr;		/* ending address of memory chunk */
 	unsigned long bits[0];		/* bitmap for allocating memory chunk */
 };
 
+/**
+ * gen_pool_for_each_chunk - iterate over chunks of generic memory pool
+ * @chunk:	the struct gen_pool_chunk * to use as a loop cursor
+ * @pool:	the generic memory pool
+ *
+ * Not lockless, proper mutual exclusion is needed to use this macro
+ * with other gen_pool function simultaneously.
+ */
+#define gen_pool_for_each_chunk(chunk, pool)			\
+	list_for_each_entry_rcu(chunk, &(pool)->chunks, next_chunk)
+
 extern struct gen_pool *gen_pool_create(int, int);
 extern int gen_pool_add(struct gen_pool *, unsigned long, size_t, int);
 extern void gen_pool_destroy(struct gen_pool *);
 extern unsigned long gen_pool_alloc(struct gen_pool *, size_t);
 extern void gen_pool_free(struct gen_pool *, unsigned long, size_t);
+extern size_t gen_pool_avail(struct gen_pool *);
+extern size_t gen_pool_size(struct gen_pool *);
+#endif /* GENALLOC_H */
--- a/lib/bitmap.c
+++ b/lib/bitmap.c
@@ -271,8 +271,6 @@ int __bitmap_weight(const unsigned long
 }
 EXPORT_SYMBOL(__bitmap_weight);
 
-#define BITMAP_FIRST_WORD_MASK(start) (~0UL << ((start) % BITS_PER_LONG))
-
 void bitmap_set(unsigned long *map, int start, int nr)
 {
 	unsigned long *p = map + BIT_WORD(start);
--- a/lib/genalloc.c
+++ b/lib/genalloc.c
@@ -1,8 +1,33 @@
 /*
- * Basic general purpose allocator for managing special purpose memory
- * not managed by the regular kmalloc/kfree interface.
- * Uses for this includes on-device special memory, uncached memory
- * etc.
+ * Basic general purpose allocator for managing special purpose
+ * memory, for example, memory that is not managed by the regular
+ * kmalloc/kfree interface.  Uses for this includes on-device special
+ * memory, uncached memory etc.
+ *
+ * It is safe to use the allocator in NMI handlers and other special
+ * unblockable contexts that could otherwise deadlock on locks.  This
+ * is implemented by using atomic operations and retries on any
+ * conflicts.  The disadvantage is that there may be livelocks in
+ * extreme cases.  For better scalability, one allocator can be used
+ * for each CPU.
+ *
+ * The lockless operation only works if there is enough memory
+ * available.  If new memory is added to the pool a lock has to be
+ * still taken.  So any user relying on locklessness has to ensure
+ * that sufficient memory is preallocated.
+ *
+ * The basic atomic operation of this allocator is cmpxchg on long.
+ * On architectures that don't have NMI-safe cmpxchg implementation,
+ * the allocator can NOT be used in NMI handler.  So code uses the
+ * allocator in NMI handler should depend on
+ * CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG.
+ *
+ * rcu_read_lock and rcu_read_unlock is not used int gen_pool_alloc,
+ * gen_pool_free, gen_pool_avail and gen_pool_size etc, because chunks
+ * are only added into pool, not deleted from pool unless the pool
+ * itself is destroyed.  If chunk will be deleted from pool,
+ * rcu_read_lock and rcu_read_unlock should be uses in these
+ * functions.
  *
  * Copyright 2005 (C) Jes Sorensen <jes@trained-monkey.org>
  *
@@ -13,8 +38,107 @@
 #include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/bitmap.h>
+#include <linux/rculist.h>
+#include <linux/interrupt.h>
 #include <linux/genalloc.h>
 
+static int set_bits_ll(unsigned long *addr, unsigned long mask_to_set)
+{
+	unsigned long val, nval;
+
+	nval = *addr;
+	do {
+		val = nval;
+		if (val & mask_to_set)
+			return -EBUSY;
+	} while ((nval = cmpxchg(addr, val, val | mask_to_set)) != val);
+
+	return 0;
+}
+
+static int clear_bits_ll(unsigned long *addr, unsigned long mask_to_clear)
+{
+	unsigned long val, nval;
+
+	nval = *addr;
+	do {
+		val = nval;
+		if ((val & mask_to_clear) != mask_to_clear)
+			return -EBUSY;
+	} while ((nval = cmpxchg(addr, val, val & ~mask_to_clear)) != val);
+
+	return 0;
+}
+
+/*
+ * bitmap_set_ll - set the specified number of bits at the specified position
+ * @map: pointer to a bitmap
+ * @start: a bit position in @map
+ * @nr: number of bits to set
+ *
+ * Set @nr bits start from @start in @map lock-lessly. Several users
+ * can set/clear the same bitmap simultaneously without lock. If two
+ * users set the same bit, one user will return remain bits, otherwise
+ * return 0.
+ */
+static int bitmap_set_ll(unsigned long *map, int start, int nr)
+{
+	unsigned long *p = map + BIT_WORD(start);
+	const int size = start + nr;
+	int bits_to_set = BITS_PER_LONG - (start % BITS_PER_LONG);
+	unsigned long mask_to_set = BITMAP_FIRST_WORD_MASK(start);
+
+	while (nr - bits_to_set >= 0) {
+		if (set_bits_ll(p, mask_to_set))
+			return nr;
+		nr -= bits_to_set;
+		bits_to_set = BITS_PER_LONG;
+		mask_to_set = ~0UL;
+		p++;
+	}
+	if (nr) {
+		mask_to_set &= BITMAP_LAST_WORD_MASK(size);
+		if (set_bits_ll(p, mask_to_set))
+			return nr;
+	}
+
+	return 0;
+}
+
+/*
+ * bitmap_clear_ll - clear the specified number of bits at the specified position
+ * @map: pointer to a bitmap
+ * @start: a bit position in @map
+ * @nr: number of bits to set
+ *
+ * Clear @nr bits start from @start in @map lock-lessly. Several users
+ * can set/clear the same bitmap simultaneously without lock. If two
+ * users clear the same bit, one user will return remain bits,
+ * otherwise return 0.
+ */
+static int bitmap_clear_ll(unsigned long *map, int start, int nr)
+{
+	unsigned long *p = map + BIT_WORD(start);
+	const int size = start + nr;
+	int bits_to_clear = BITS_PER_LONG - (start % BITS_PER_LONG);
+	unsigned long mask_to_clear = BITMAP_FIRST_WORD_MASK(start);
+
+	while (nr - bits_to_clear >= 0) {
+		if (clear_bits_ll(p, mask_to_clear))
+			return nr;
+		nr -= bits_to_clear;
+		bits_to_clear = BITS_PER_LONG;
+		mask_to_clear = ~0UL;
+		p++;
+	}
+	if (nr) {
+		mask_to_clear &= BITMAP_LAST_WORD_MASK(size);
+		if (clear_bits_ll(p, mask_to_clear))
+			return nr;
+	}
+
+	return 0;
+}
 
 /**
  * gen_pool_create - create a new special memory pool
@@ -30,7 +154,7 @@ struct gen_pool *gen_pool_create(int min
 
 	pool = kmalloc_node(sizeof(struct gen_pool), GFP_KERNEL, nid);
 	if (pool != NULL) {
-		rwlock_init(&pool->lock);
+		spin_lock_init(&pool->lock);
 		INIT_LIST_HEAD(&pool->chunks);
 		pool->min_alloc_order = min_alloc_order;
 	}
@@ -58,15 +182,15 @@ int gen_pool_add(struct gen_pool *pool,
 
 	chunk = kmalloc_node(nbytes, GFP_KERNEL | __GFP_ZERO, nid);
 	if (unlikely(chunk == NULL))
-		return -1;
+		return -ENOMEM;
 
-	spin_lock_init(&chunk->lock);
 	chunk->start_addr = addr;
 	chunk->end_addr = addr + size;
+	atomic_set(&chunk->avail, size);
 
-	write_lock(&pool->lock);
-	list_add(&chunk->next_chunk, &pool->chunks);
-	write_unlock(&pool->lock);
+	spin_lock(&pool->lock);
+	list_add_rcu(&chunk->next_chunk, &pool->chunks);
+	spin_unlock(&pool->lock);
 
 	return 0;
 }
@@ -86,7 +210,6 @@ void gen_pool_destroy(struct gen_pool *p
 	int order = pool->min_alloc_order;
 	int bit, end_bit;
 
-
 	list_for_each_safe(_chunk, _next_chunk, &pool->chunks) {
 		chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk);
 		list_del(&chunk->next_chunk);
@@ -108,43 +231,47 @@ EXPORT_SYMBOL(gen_pool_destroy);
  * @size: number of bytes to allocate from the pool
  *
  * Allocate the requested number of bytes from the specified pool.
- * Uses a first-fit algorithm.
+ * Uses a first-fit algorithm. Can not be used in NMI handler on
+ * architectures without NMI-safe cmpxchg implementation.
  */
 unsigned long gen_pool_alloc(struct gen_pool *pool, size_t size)
 {
-	struct list_head *_chunk;
 	struct gen_pool_chunk *chunk;
-	unsigned long addr, flags;
+	unsigned long addr;
 	int order = pool->min_alloc_order;
-	int nbits, start_bit, end_bit;
+	int nbits, start_bit = 0, end_bit, remain;
+
+#ifndef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
+	BUG_ON(in_nmi());
+#endif
 
 	if (size == 0)
 		return 0;
 
 	nbits = (size + (1UL << order) - 1) >> order;
-
-	read_lock(&pool->lock);
-	list_for_each(_chunk, &pool->chunks) {
-		chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk);
+	list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk) {
+		if (size > atomic_read(&chunk->avail))
+			continue;
 
 		end_bit = (chunk->end_addr - chunk->start_addr) >> order;
-
-		spin_lock_irqsave(&chunk->lock, flags);
-		start_bit = bitmap_find_next_zero_area(chunk->bits, end_bit, 0,
-						nbits, 0);
-		if (start_bit >= end_bit) {
-			spin_unlock_irqrestore(&chunk->lock, flags);
+retry:
+		start_bit = bitmap_find_next_zero_area(chunk->bits, end_bit,
+						       start_bit, nbits, 0);
+		if (start_bit >= end_bit)
 			continue;
+		remain = bitmap_set_ll(chunk->bits, start_bit, nbits);
+		if (remain) {
+			remain = bitmap_clear_ll(chunk->bits, start_bit,
+						 nbits - remain);
+			BUG_ON(remain);
+			goto retry;
 		}
 
 		addr = chunk->start_addr + ((unsigned long)start_bit << order);
-
-		bitmap_set(chunk->bits, start_bit, nbits);
-		spin_unlock_irqrestore(&chunk->lock, flags);
-		read_unlock(&pool->lock);
+		size = nbits << order;
+		atomic_sub(size, &chunk->avail);
 		return addr;
 	}
-	read_unlock(&pool->lock);
 	return 0;
 }
 EXPORT_SYMBOL(gen_pool_alloc);
@@ -155,33 +282,66 @@ EXPORT_SYMBOL(gen_pool_alloc);
  * @addr: starting address of memory to free back to pool
  * @size: size in bytes of memory to free
  *
- * Free previously allocated special memory back to the specified pool.
+ * Free previously allocated special memory back to the specified
+ * pool.  Can not be used in NMI handler on architectures without
+ * NMI-safe cmpxchg implementation.
  */
 void gen_pool_free(struct gen_pool *pool, unsigned long addr, size_t size)
 {
-	struct list_head *_chunk;
 	struct gen_pool_chunk *chunk;
-	unsigned long flags;
 	int order = pool->min_alloc_order;
-	int bit, nbits;
-
-	nbits = (size + (1UL << order) - 1) >> order;
+	int start_bit, nbits, remain;
 
-	read_lock(&pool->lock);
-	list_for_each(_chunk, &pool->chunks) {
-		chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk);
+#ifndef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
+	BUG_ON(in_nmi());
+#endif
 
+	nbits = (size + (1UL << order) - 1) >> order;
+	list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk) {
 		if (addr >= chunk->start_addr && addr < chunk->end_addr) {
 			BUG_ON(addr + size > chunk->end_addr);
-			spin_lock_irqsave(&chunk->lock, flags);
-			bit = (addr - chunk->start_addr) >> order;
-			while (nbits--)
-				__clear_bit(bit++, chunk->bits);
-			spin_unlock_irqrestore(&chunk->lock, flags);
-			break;
+			start_bit = (addr - chunk->start_addr) >> order;
+			remain = bitmap_clear_ll(chunk->bits, start_bit, nbits);
+			BUG_ON(remain);
+			size = nbits << order;
+			atomic_add(size, &chunk->avail);
+			return;
 		}
 	}
-	BUG_ON(nbits > 0);
-	read_unlock(&pool->lock);
+	BUG();
 }
 EXPORT_SYMBOL(gen_pool_free);
+
+/**
+ * gen_pool_avail - get available free space of the pool
+ * @pool: pool to get available free space
+ *
+ * Return available free space of the specified pool.
+ */
+size_t gen_pool_avail(struct gen_pool *pool)
+{
+	struct gen_pool_chunk *chunk;
+	size_t avail = 0;
+
+	list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk)
+		avail += atomic_read(&chunk->avail);
+	return avail;
+}
+EXPORT_SYMBOL_GPL(gen_pool_avail);
+
+/**
+ * gen_pool_size - get size in bytes of memory managed by the pool
+ * @pool: pool to get size
+ *
+ * Return size in bytes of memory managed by the pool.
+ */
+size_t gen_pool_size(struct gen_pool *pool)
+{
+	struct gen_pool_chunk *chunk;
+	size_t size = 0;
+
+	list_for_each_entry_rcu(chunk, &pool->chunks, next_chunk)
+		size += chunk->end_addr - chunk->start_addr;
+	return size;
+}
+EXPORT_SYMBOL_GPL(gen_pool_size);

[RFC 2/2] ACPI, APEI, GHES, printk support for recoverable error via NMI

[Huang Ying]

Some APEI GHES recoverable errors are reported via NMI, but printk is
not safe in NMI context.

To solve the issue, a lock-less memory allocator is used to allocate
memory in NMI handler, save the error record into the allocated
memory, put the error record into a lock-less list.  On the other
hand, a irq_work is used to delay the operation from NMI context to
IRQ context.  The irq_work IRQ handler will remove nodes from
lock-less list, printk the error record and do some further processing
include recovery operation, then free the memory.

Signed-off-by: Huang Ying <ying.huang@intel.com>
---
 drivers/acpi/apei/Kconfig |    1 
 drivers/acpi/apei/ghes.c  |  186 ++++++++++++++++++++++++++++++++++++++++++----
 2 files changed, 174 insertions(+), 13 deletions(-)

--- a/drivers/acpi/apei/Kconfig
+++ b/drivers/acpi/apei/Kconfig
@@ -11,6 +11,7 @@ config ACPI_APEI_GHES
 	tristate "APEI Generic Hardware Error Source"
 	depends on ACPI_APEI && X86
 	select ACPI_HED
+	select GENERIC_ALLOCATOR
 	help
 	  Generic Hardware Error Source provides a way to report
 	  platform hardware errors (such as that from chipset). It
--- a/drivers/acpi/apei/ghes.c
+++ b/drivers/acpi/apei/ghes.c
@@ -42,6 +42,9 @@
 #include <linux/mutex.h>
 #include <linux/ratelimit.h>
 #include <linux/vmalloc.h>
+#include <linux/irq_work.h>
+#include <linux/llist.h>
+#include <linux/genalloc.h>
 #include <acpi/apei.h>
 #include <acpi/atomicio.h>
 #include <acpi/hed.h>
@@ -53,6 +56,15 @@
 #define GHES_PFX	"GHES: "
 
 #define GHES_ESTATUS_MAX_SIZE		65536
+#define GHES_ESOURCE_PREALLOC_MAX_SIZE	65536
+
+#define GHES_ESTATUS_POOL_MIN_ALLOC_ORDER	3
+
+#define GHES_ESTATUS_NODE_LEN(estatus_len)			\
+	(sizeof(struct ghes_estatus_node) + (estatus_len))
+#define GHES_ESTATUS_FROM_NODE(estatus_node)				\
+	((struct acpi_hest_generic_status *)				\
+	 ((struct ghes_estatus_node *)(estatus_node) + 1))
 
 /*
  * One struct ghes is created for each generic hardware error source.
@@ -77,6 +89,11 @@ struct ghes {
 	};
 };
 
+struct ghes_estatus_node {
+	struct llist_node llnode;
+	struct acpi_hest_generic *generic;
+};
+
 static int ghes_panic_timeout	__read_mostly = 30;
 
 /*
@@ -121,6 +138,19 @@ static struct vm_struct *ghes_ioremap_ar
 static DEFINE_RAW_SPINLOCK(ghes_ioremap_lock_nmi);
 static DEFINE_SPINLOCK(ghes_ioremap_lock_irq);
 
+/*
+ * printk is not safe in NMI context.  So in NMI handler, we allocate
+ * required memory from lock-less memory allocator
+ * (ghes_estatus_pool), save estatus into it, put them into lock-less
+ * list (ghes_estatus_llist), then delay printk into IRQ context via
+ * irq_work (ghes_proc_irq_work).  ghes_estatus_size_request record
+ * required pool size by all NMI error source.
+ */
+static struct gen_pool *ghes_estatus_pool;
+static unsigned long ghes_estatus_pool_size_request;
+static struct llist_head ghes_estatus_llist;
+static struct irq_work ghes_proc_irq_work;
+
 static int ghes_ioremap_init(void)
 {
 	ghes_ioremap_area = __get_vm_area(PAGE_SIZE * GHES_IOREMAP_PAGES,
@@ -180,6 +210,50 @@ static void ghes_iounmap_irq(void __iome
 	__flush_tlb_one(vaddr);
 }
 
+static int ghes_estatus_pool_init(void)
+{
+	ghes_estatus_pool = gen_pool_create(GHES_ESTATUS_POOL_MIN_ALLOC_ORDER, -1);
+	if (!ghes_estatus_pool)
+		return -ENOMEM;
+	return 0;
+}
+
+static void ghes_estatus_pool_exit(void)
+{
+	struct gen_pool_chunk *chunk;
+
+	gen_pool_for_each_chunk(chunk, ghes_estatus_pool)
+		free_page(chunk->start_addr);
+	gen_pool_destroy(ghes_estatus_pool);
+}
+
+static int ghes_estatus_pool_expand(unsigned long len)
+{
+	unsigned long i, pages, size, addr;
+	int ret;
+
+	ghes_estatus_pool_size_request += PAGE_ALIGN(len);
+	size = gen_pool_size(ghes_estatus_pool);
+	if (size >= ghes_estatus_pool_size_request)
+		return 0;
+	pages = (ghes_estatus_pool_size_request - size) / PAGE_SIZE;
+	for (i = 0; i < pages; i++) {
+		addr = __get_free_page(GFP_KERNEL);
+		if (!addr)
+			return -ENOMEM;
+		ret = gen_pool_add(ghes_estatus_pool, addr, PAGE_SIZE, -1);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static void ghes_estatus_pool_shrink(unsigned long len)
+{
+	ghes_estatus_pool_size_request -= PAGE_ALIGN(len);
+}
+
 static struct ghes *ghes_new(struct acpi_hest_generic *generic)
 {
 	struct ghes *ghes;
@@ -341,13 +415,13 @@ static void ghes_clear_estatus(struct gh
 	ghes->flags &= ~GHES_TO_CLEAR;
 }
 
-static void ghes_do_proc(struct ghes *ghes)
+static void ghes_do_proc(const struct acpi_hest_generic_status *estatus)
 {
 	int sev, processed = 0;
 	struct acpi_hest_generic_data *gdata;
 
-	sev = ghes_severity(ghes->estatus->error_severity);
-	apei_estatus_for_each_section(ghes->estatus, gdata) {
+	sev = ghes_severity(estatus->error_severity);
+	apei_estatus_for_each_section(estatus, gdata) {
 #ifdef CONFIG_X86_MCE
 		if (!uuid_le_cmp(*(uuid_le *)gdata->section_type,
 				 CPER_SEC_PLATFORM_MEM)) {
@@ -360,13 +434,15 @@ static void ghes_do_proc(struct ghes *gh
 	}
 }
 
-static void ghes_print_estatus(const char *pfx, struct ghes *ghes)
+static void ghes_print_estatus(const char *pfx,
+			       const struct acpi_hest_generic *generic,
+			       const struct acpi_hest_generic_status *estatus)
 {
 	/* Not more than 2 messages every 5 seconds */
 	static DEFINE_RATELIMIT_STATE(ratelimit, 5*HZ, 2);
 
 	if (pfx == NULL) {
-		if (ghes_severity(ghes->estatus->error_severity) <=
+		if (ghes_severity(estatus->error_severity) <=
 		    GHES_SEV_CORRECTED)
 			pfx = KERN_WARNING HW_ERR;
 		else
@@ -375,8 +451,8 @@ static void ghes_print_estatus(const cha
 	if (__ratelimit(&ratelimit)) {
 		printk(
 	"%s""Hardware error from APEI Generic Hardware Error Source: %d\n",
-	pfx, ghes->generic->header.source_id);
-		apei_estatus_print(pfx, ghes->estatus);
+	pfx, generic->header.source_id);
+		apei_estatus_print(pfx, estatus);
 	}
 }
 
@@ -387,8 +463,8 @@ static int ghes_proc(struct ghes *ghes)
 	rc = ghes_read_estatus(ghes, 0);
 	if (rc)
 		goto out;
-	ghes_print_estatus(NULL, ghes);
-	ghes_do_proc(ghes);
+	ghes_print_estatus(NULL, ghes->generic, ghes->estatus);
+	ghes_do_proc(ghes->estatus);
 
 out:
 	ghes_clear_estatus(ghes);
@@ -447,6 +523,40 @@ static int ghes_notify_sci(struct notifi
 	return ret;
 }
 
+static void ghes_proc_in_irq(struct irq_work *irq_work)
+{
+	struct llist_node *llnode, *next, *tail = NULL;
+	struct ghes_estatus_node *estatus_node;
+	struct acpi_hest_generic_status *estatus;
+	u32 len, node_len;
+
+	/*
+	 * Because the time order of estatus in list is reversed,
+	 * revert it back to proper order.
+	 */
+	llnode = llist_del_all(&ghes_estatus_llist);
+	while (llnode) {
+		next = llnode->next;
+		llnode->next = tail;
+		tail = llnode;
+		llnode = next;
+	}
+	llnode = tail;
+	while (llnode) {
+		next = llnode->next;
+		estatus_node = llist_entry(llnode, struct ghes_estatus_node,
+					   llnode);
+		estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
+		len = apei_estatus_len(estatus);
+		node_len = GHES_ESTATUS_NODE_LEN(len);
+		ghes_do_proc(estatus);
+		ghes_print_estatus(NULL, estatus_node->generic, estatus);
+		gen_pool_free(ghes_estatus_pool, (unsigned long)estatus_node,
+			      node_len);
+		llnode = next;
+	}
+}
+
 static int ghes_notify_nmi(struct notifier_block *this,
 				  unsigned long cmd, void *data)
 {
@@ -476,7 +586,8 @@ static int ghes_notify_nmi(struct notifi
 
 	if (sev_global >= GHES_SEV_PANIC) {
 		oops_begin();
-		ghes_print_estatus(KERN_EMERG HW_ERR, ghes_global);
+		ghes_print_estatus(KERN_EMERG HW_ERR, ghes_global->generic,
+				   ghes_global->estatus);
 		/* reboot to log the error! */
 		if (panic_timeout == 0)
 			panic_timeout = ghes_panic_timeout;
@@ -484,12 +595,31 @@ static int ghes_notify_nmi(struct notifi
 	}
 
 	list_for_each_entry_rcu(ghes, &ghes_nmi, list) {
+#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
+		u32 len, node_len;
+		struct ghes_estatus_node *estatus_node;
+		struct acpi_hest_generic_status *estatus;
+#endif
 		if (!(ghes->flags & GHES_TO_CLEAR))
 			continue;
-		/* Do not print estatus because printk is not NMI safe */
-		ghes_do_proc(ghes);
+#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
+		/* Save estatus for further processing in IRQ context */
+		len = apei_estatus_len(ghes->estatus);
+		node_len = GHES_ESTATUS_NODE_LEN(len);
+		estatus_node = (void *)gen_pool_alloc(ghes_estatus_pool,
+						      node_len);
+		if (estatus_node) {
+			estatus_node->generic = ghes->generic;
+			estatus = GHES_ESTATUS_FROM_NODE(estatus_node);
+			memcpy(estatus, ghes->estatus, len);
+			llist_add(&estatus_node->llnode, &ghes_estatus_llist);
+		}
+#endif
 		ghes_clear_estatus(ghes);
 	}
+#ifdef CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG
+	irq_work_queue(&ghes_proc_irq_work);
+#endif
 
 out:
 	raw_spin_unlock(&ghes_nmi_lock);
@@ -504,10 +634,26 @@ static struct notifier_block ghes_notifi
 	.notifier_call = ghes_notify_nmi,
 };
 
+static unsigned long ghes_esource_prealloc_size(
+	const struct acpi_hest_generic *generic)
+{
+	unsigned long block_length, prealloc_records, prealloc_size;
+
+	block_length = min_t(unsigned long, generic->error_block_length,
+			     GHES_ESTATUS_MAX_SIZE);
+	prealloc_records = max_t(unsigned long,
+				 generic->records_to_preallocate, 1);
+	prealloc_size = min_t(unsigned long, block_length * prealloc_records,
+			      GHES_ESOURCE_PREALLOC_MAX_SIZE);
+
+	return prealloc_size;
+}
+
 static int __devinit ghes_probe(struct platform_device *ghes_dev)
 {
 	struct acpi_hest_generic *generic;
 	struct ghes *ghes = NULL;
+	unsigned long len;
 	int rc = -EINVAL;
 
 	generic = *(struct acpi_hest_generic **)ghes_dev->dev.platform_data;
@@ -573,6 +719,8 @@ static int __devinit ghes_probe(struct p
 		mutex_unlock(&ghes_list_mutex);
 		break;
 	case ACPI_HEST_NOTIFY_NMI:
+		len = ghes_esource_prealloc_size(generic);
+		ghes_estatus_pool_expand(len);
 		mutex_lock(&ghes_list_mutex);
 		if (list_empty(&ghes_nmi))
 			register_die_notifier(&ghes_notifier_nmi);
@@ -597,6 +745,7 @@ static int __devexit ghes_remove(struct
 {
 	struct ghes *ghes;
 	struct acpi_hest_generic *generic;
+	unsigned long len;
 
 	ghes = platform_get_drvdata(ghes_dev);
 	generic = ghes->generic;
@@ -627,6 +776,8 @@ static int __devexit ghes_remove(struct
 		 * freed after NMI handler finishes.
 		 */
 		synchronize_rcu();
+		len = ghes_esource_prealloc_size(generic);
+		ghes_estatus_pool_shrink(len);
 		break;
 	default:
 		BUG();
@@ -662,15 +813,23 @@ static int __init ghes_init(void)
 		return -EINVAL;
 	}
 
+	init_irq_work(&ghes_proc_irq_work, ghes_proc_in_irq);
+
 	rc = ghes_ioremap_init();
 	if (rc)
 		goto err;
 
-	rc = platform_driver_register(&ghes_platform_driver);
+	rc = ghes_estatus_pool_init();
 	if (rc)
 		goto err_ioremap_exit;
 
+	rc = platform_driver_register(&ghes_platform_driver);
+	if (rc)
+		goto err_pool_exit;
+
 	return 0;
+err_pool_exit:
+	ghes_estatus_pool_exit();
 err_ioremap_exit:
 	ghes_ioremap_exit();
 err:
@@ -680,6 +839,7 @@ err:
 static void __exit ghes_exit(void)
 {
 	platform_driver_unregister(&ghes_platform_driver);
+	ghes_estatus_pool_exit();
 	ghes_ioremap_exit();
 }
 

RE: [RFC 1/2] lib, Make gen_pool memory allocator lockless

[Luck, Tony]

+static int set_bits_ll(unsigned long *addr, unsigned long mask_to_set)
+{
+	unsigned long val, nval;
+
+	nval = *addr;
+	do {
+		val = nval;
+		if (val & mask_to_set)
+			return -EBUSY;
+	} while ((nval = cmpxchg(addr, val, val | mask_to_set)) != val);
+
+	return 0;
+}

Spin loops like this should have a call to cpu_relax(), shouldn't they?

-Tony

RE: [RFC 1/2] lib, Make gen_pool memory allocator lockless

[Huang Ying]

On Thu, 2011-02-10 at 03:03 +0800, Luck, Tony wrote:
> +static int set_bits_ll(unsigned long *addr, unsigned long mask_to_set)
> +{
> +	unsigned long val, nval;
> +
> +	nval = *addr;
> +	do {
> +		val = nval;
> +		if (val & mask_to_set)
> +			return -EBUSY;
> +	} while ((nval = cmpxchg(addr, val, val | mask_to_set)) != val);
> +
> +	return 0;
> +}
> 
> Spin loops like this should have a call to cpu_relax(), shouldn't they?

Thanks!  Sounds reasonable to me.  Will change it.

Best Regards,
Huang Ying