tracing: Create a sparse bitmask for pid filtering

When the trace_pid_list was created, the default pid max was 32768.
Creating a bitmask that can hold one bit for all 32768 took up 4096 (one
page). Having a one page bitmask was not much of a problem, and that was
used for mapping pids. But today, systems are bigger and can run more
tasks, and now the default pid_max is usually set to 4194304. Which means
to handle that many pids requires 524288 bytes. Worse yet, the pid_max can
be set to 2^30 (1073741824 or 1G) which would take 134217728 (128M) of
memory to store this array.

Since the pid_list array is very sparsely populated, it is a huge waste of
memory to store all possible bits for each pid when most will not be set.

Instead, use a page table scheme to store the array, and allow this to
handle up to 30 bit pids.

The pid_mask will start out with 256 entries for the first 8 MSB bits.
This will cost 1K for 32 bit architectures and 2K for 64 bit. Each of
these will have a 256 array to store the next 8 bits of the pid (another
1 or 2K). These will hold an 2K byte bitmask (which will cover the LSB
14 bits or 16384 pids).

When the trace_pid_list is allocated, it will have the 1/2K upper bits
allocated, and then it will allocate a cache for the next upper chunks and
the lower chunks (default 6 of each). Then when a bit is "set", these
chunks will be pulled from the free list and added to the array. If the
free list gets down to a lever (default 2), it will trigger an irqwork
that will refill the cache back up.

On clearing a bit, if the clear causes the bitmask to be zero, that chunk
will then be placed back into the free cache for later use, keeping the
need to allocate more down to a minimum.

Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>

1 files changed, 368 insertions, 33 deletions

diff --git a/kernel/trace/pid_list.c b/kernel/trace/pid_list.c
index 4483ef70b562..cbf8031b2b99 100644
--- a/kernel/trace/pid_list.c
+++ b/kernel/trace/pid_list.c
@@ -2,10 +2,119 @@
 /*
  * Copyright (C) 2021 VMware Inc, Steven Rostedt <rostedt@goodmis.org>
  */
-#include <linux/vmalloc.h>
+#include <linux/spinlock.h>
+#include <linux/irq_work.h>
 #include <linux/slab.h>
 #include "trace.h"
 
+/* See pid_list.h for details */
+
+static inline union lower_chunk *get_lower_chunk(struct trace_pid_list *pid_list)
+{
+	union lower_chunk *chunk;
+
+	lockdep_assert_held(&pid_list->lock);
+
+	if (!pid_list->lower_list)
+		return NULL;
+
+	chunk = pid_list->lower_list;
+	pid_list->lower_list = chunk->next;
+	pid_list->free_lower_chunks--;
+	WARN_ON_ONCE(pid_list->free_lower_chunks < 0);
+	chunk->next = NULL;
+	/*
+	 * If a refill needs to happen, it can not happen here
+	 * as the scheduler run queue locks are held.
+	 */
+	if (pid_list->free_lower_chunks <= CHUNK_REALLOC)
+		irq_work_queue(&pid_list->refill_irqwork);
+
+	return chunk;
+}
+
+static inline union upper_chunk *get_upper_chunk(struct trace_pid_list *pid_list)
+{
+	union upper_chunk *chunk;
+
+	lockdep_assert_held(&pid_list->lock);
+
+	if (!pid_list->upper_list)
+		return NULL;
+
+	chunk = pid_list->upper_list;
+	pid_list->upper_list = chunk->next;
+	pid_list->free_upper_chunks--;
+	WARN_ON_ONCE(pid_list->free_upper_chunks < 0);
+	chunk->next = NULL;
+	/*
+	 * If a refill needs to happen, it can not happen here
+	 * as the scheduler run queue locks are held.
+	 */
+	if (pid_list->free_upper_chunks <= CHUNK_REALLOC)
+		irq_work_queue(&pid_list->refill_irqwork);
+
+	return chunk;
+}
+
+static inline void put_lower_chunk(struct trace_pid_list *pid_list,
+				   union lower_chunk *chunk)
+{
+	lockdep_assert_held(&pid_list->lock);
+
+	chunk->next = pid_list->lower_list;
+	pid_list->lower_list = chunk;
+	pid_list->free_lower_chunks++;
+}
+
+static inline void put_upper_chunk(struct trace_pid_list *pid_list,
+				   union upper_chunk *chunk)
+{
+	lockdep_assert_held(&pid_list->lock);
+
+	chunk->next = pid_list->upper_list;
+	pid_list->upper_list = chunk;
+	pid_list->free_upper_chunks++;
+}
+
+static inline bool upper_empty(union upper_chunk *chunk)
+{
+	/*
+	 * If chunk->data has no lower chunks, it will be the same
+	 * as a zeroed bitmask. Use find_first_bit() to test it
+	 * and if it doesn't find any bits set, then the array
+	 * is empty.
+	 */
+	int bit = find_first_bit((unsigned long *)chunk->data,
+				 sizeof(chunk->data) * 8);
+	return bit >= sizeof(chunk->data) * 8;
+}
+
+static inline int pid_split(unsigned int pid, unsigned int *upper1,
+			     unsigned int *upper2, unsigned int *lower)
+{
+	/* MAX_PID should cover all pids */
+	BUILD_BUG_ON(MAX_PID < PID_MAX_LIMIT);
+
+	/* In case a bad pid is passed in, then fail */
+	if (unlikely(pid >= MAX_PID))
+		return -1;
+
+	*upper1 = (pid >> UPPER1_SHIFT) & UPPER_MASK;
+	*upper2 = (pid >> UPPER2_SHIFT) & UPPER_MASK;
+	*lower = pid & LOWER_MASK;
+
+	return 0;
+}
+
+static inline unsigned int pid_join(unsigned int upper1,
+				    unsigned int upper2, unsigned int lower)
+{
+	return ((upper1 & UPPER_MASK) << UPPER1_SHIFT) |
+		((upper2 & UPPER_MASK) << UPPER2_SHIFT) |
+		(lower & LOWER_MASK);
+}
+
 /**
  * trace_pid_list_is_set - test if the pid is set in the list
  * @pid_list: The pid list to test
@@ -19,14 +128,30 @@
  */
 bool trace_pid_list_is_set(struct trace_pid_list *pid_list, unsigned int pid)
 {
-	/*
-	 * If pid_max changed after filtered_pids was created, we
-	 * by default ignore all pids greater than the previous pid_max.
-	 */
-	if (pid >= pid_list->pid_max)
+	union upper_chunk *upper_chunk;
+	union lower_chunk *lower_chunk;
+	unsigned long flags;
+	unsigned int upper1;
+	unsigned int upper2;
+	unsigned int lower;
+	bool ret = false;
+
+	if (!pid_list)
 		return false;
 
-	return test_bit(pid, pid_list->pids);
+	if (pid_split(pid, &upper1, &upper2, &lower) < 0)
+		return false;
+
+	raw_spin_lock_irqsave(&pid_list->lock, flags);
+	upper_chunk = pid_list->upper[upper1];
+	if (upper_chunk) {
+		lower_chunk = upper_chunk->data[upper2];
+		if (lower_chunk)
+			ret = test_bit(lower, lower_chunk->data);
+	}
+	raw_spin_unlock_irqrestore(&pid_list->lock, flags);
+
+	return ret;
 }
 
 /**
@@ -42,13 +167,44 @@ bool trace_pid_list_is_set(struct trace_pid_list *pid_list, unsigned int pid)
  */
 int trace_pid_list_set(struct trace_pid_list *pid_list, unsigned int pid)
 {
-	/* Sorry, but we don't support pid_max changing after setting */
-	if (pid >= pid_list->pid_max)
-		return -EINVAL;
+	union upper_chunk *upper_chunk;
+	union lower_chunk *lower_chunk;
+	unsigned long flags;
+	unsigned int upper1;
+	unsigned int upper2;
+	unsigned int lower;
+	int ret;
 
-	set_bit(pid, pid_list->pids);
+	if (!pid_list)
+		return -ENODEV;
 
-	return 0;
+	if (pid_split(pid, &upper1, &upper2, &lower) < 0)
+		return -EINVAL;
+
+	raw_spin_lock_irqsave(&pid_list->lock, flags);
+	upper_chunk = pid_list->upper[upper1];
+	if (!upper_chunk) {
+		upper_chunk = get_upper_chunk(pid_list);
+		if (!upper_chunk) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		pid_list->upper[upper1] = upper_chunk;
+	}
+	lower_chunk = upper_chunk->data[upper2];
+	if (!lower_chunk) {
+		lower_chunk = get_lower_chunk(pid_list);
+		if (!lower_chunk) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		upper_chunk->data[upper2] = lower_chunk;
+	}
+	set_bit(lower, lower_chunk->data);
+	ret = 0;
+ out:
+	raw_spin_unlock_irqrestore(&pid_list->lock, flags);
+	return ret;
 }
 
 /**
@@ -64,12 +220,41 @@ int trace_pid_list_set(struct trace_pid_list *pid_list, unsigned int pid)
  */
 int trace_pid_list_clear(struct trace_pid_list *pid_list, unsigned int pid)
 {
-	/* Sorry, but we don't support pid_max changing after setting */
-	if (pid >= pid_list->pid_max)
+	union upper_chunk *upper_chunk;
+	union lower_chunk *lower_chunk;
+	unsigned long flags;
+	unsigned int upper1;
+	unsigned int upper2;
+	unsigned int lower;
+
+	if (!pid_list)
+		return -ENODEV;
+
+	if (pid_split(pid, &upper1, &upper2, &lower) < 0)
 		return -EINVAL;
 
-	clear_bit(pid, pid_list->pids);
+	raw_spin_lock_irqsave(&pid_list->lock, flags);
+	upper_chunk = pid_list->upper[upper1];
+	if (!upper_chunk)
+		goto out;
 
+	lower_chunk = upper_chunk->data[upper2];
+	if (!lower_chunk)
+		goto out;
+
+	clear_bit(lower, lower_chunk->data);
+
+	/* if there's no more bits set, add it to the free list */
+	if (find_first_bit(lower_chunk->data, LOWER_MAX) >= LOWER_MAX) {
+		put_lower_chunk(pid_list, lower_chunk);
+		upper_chunk->data[upper2] = NULL;
+		if (upper_empty(upper_chunk)) {
+			put_upper_chunk(pid_list, upper_chunk);
+			pid_list->upper[upper1] = NULL;
+		}
+	}
+ out:
+	raw_spin_unlock_irqrestore(&pid_list->lock, flags);
 	return 0;
 }
 
@@ -88,13 +273,45 @@ int trace_pid_list_clear(struct trace_pid_list *pid_list, unsigned int pid)
 int trace_pid_list_next(struct trace_pid_list *pid_list, unsigned int pid,
 			unsigned int *next)
 {
-	pid = find_next_bit(pid_list->pids, pid_list->pid_max, pid);
+	union upper_chunk *upper_chunk;
+	union lower_chunk *lower_chunk;
+	unsigned long flags;
+	unsigned int upper1;
+	unsigned int upper2;
+	unsigned int lower;
+
+	if (!pid_list)
+		return -ENODEV;
+
+	if (pid_split(pid, &upper1, &upper2, &lower) < 0)
+		return -EINVAL;
 
-	if (pid < pid_list->pid_max) {
-		*next = pid;
-		return 0;
+	raw_spin_lock_irqsave(&pid_list->lock, flags);
+	for (; upper1 <= UPPER_MASK; upper1++, upper2 = 0) {
+		upper_chunk = pid_list->upper[upper1];
+
+		if (!upper_chunk)
+			continue;
+
+		for (; upper2 <= UPPER_MASK; upper2++, lower = 0) {
+			lower_chunk = upper_chunk->data[upper2];
+			if (!lower_chunk)
+				continue;
+
+			lower = find_next_bit(lower_chunk->data, LOWER_MAX,
+					    lower);
+			if (lower < LOWER_MAX)
+				goto found;
+		}
 	}
-	return -1;
+
+ found:
+	raw_spin_unlock_irqrestore(&pid_list->lock, flags);
+	if (upper1 > UPPER_MASK)
+		return -1;
+
+	*next = pid_join(upper1, upper2, lower);
+	return 0;
 }
 
 /**
@@ -109,15 +326,79 @@ int trace_pid_list_next(struct trace_pid_list *pid_list, unsigned int pid,
  */
 int trace_pid_list_first(struct trace_pid_list *pid_list, unsigned int *pid)
 {
-	unsigned int first;
+	return trace_pid_list_next(pid_list, 0, pid);
+}
+
+static void pid_list_refill_irq(struct irq_work *iwork)
+{
+	struct trace_pid_list *pid_list = container_of(iwork, struct trace_pid_list,
+						       refill_irqwork);
+	union upper_chunk *upper;
+	union lower_chunk *lower;
+	union upper_chunk **upper_next = &upper;
+	union lower_chunk **lower_next = &lower;
+	int upper_count;
+	int lower_count;
+	int ucnt = 0;
+	int lcnt = 0;
 
-	first = find_first_bit(pid_list->pids, pid_list->pid_max);
+ again:
+	raw_spin_lock(&pid_list->lock);
+	upper_count = CHUNK_ALLOC - pid_list->free_upper_chunks;
+	lower_count = CHUNK_ALLOC - pid_list->free_lower_chunks;
+	raw_spin_unlock(&pid_list->lock);
+
+	if (upper_count <= 0 && lower_count <= 0)
+		return;
 
-	if (first < pid_list->pid_max) {
-		*pid = first;
-		return 0;
+	while (upper_count-- > 0) {
+		union upper_chunk *chunk;
+
+		chunk = kzalloc(sizeof(*chunk), GFP_KERNEL);
+		if (!chunk)
+			break;
+		*upper_next = chunk;
+		upper_next = &chunk->next;
+		ucnt++;
 	}
-	return -1;
+
+	while (lower_count-- > 0) {
+		union lower_chunk *chunk;
+
+		chunk = kzalloc(sizeof(*chunk), GFP_KERNEL);
+		if (!chunk)
+			break;
+		*lower_next = chunk;
+		lower_next = &chunk->next;
+		lcnt++;
+	}
+
+	raw_spin_lock(&pid_list->lock);
+	if (upper) {
+		*upper_next = pid_list->upper_list;
+		pid_list->upper_list = upper;
+		pid_list->free_upper_chunks += ucnt;
+	}
+	if (lower) {
+		*lower_next = pid_list->lower_list;
+		pid_list->lower_list = lower;
+		pid_list->free_lower_chunks += lcnt;
+	}
+	raw_spin_unlock(&pid_list->lock);
+
+	/*
+	 * On success of allocating all the chunks, both counters
+	 * will be less than zero. If they are not, then an allocation
+	 * failed, and we should not try again.
+	 */
+	if (upper_count >= 0 || lower_count >= 0)
+		return;
+	/*
+	 * When the locks were released, free chunks could have
+	 * been used and allocation needs to be done again. Might as
+	 * well allocate it now.
+	 */
+	goto again;
 }
 
 /**
@@ -130,18 +411,41 @@ int trace_pid_list_first(struct trace_pid_list *pid_list, unsigned int *pid)
 struct trace_pid_list *trace_pid_list_alloc(void)
 {
 	struct trace_pid_list *pid_list;
+	int i;
+
+	/* According to linux/thread.h, pids can be no bigger that 30 bits */
+	WARN_ON_ONCE(pid_max > (1 << 30));
 
-	pid_list = kmalloc(sizeof(*pid_list), GFP_KERNEL);
+	pid_list = kzalloc(sizeof(*pid_list), GFP_KERNEL);
 	if (!pid_list)
 		return NULL;
 
-	pid_list->pid_max = READ_ONCE(pid_max);
+	init_irq_work(&pid_list->refill_irqwork, pid_list_refill_irq);
 
-	pid_list->pids = vzalloc((pid_list->pid_max + 7) >> 3);
-	if (!pid_list->pids) {
-		kfree(pid_list);
-		return NULL;
+	raw_spin_lock_init(&pid_list->lock);
+
+	for (i = 0; i < CHUNK_ALLOC; i++) {
+		union upper_chunk *chunk;
+
+		chunk = kzalloc(sizeof(*chunk), GFP_KERNEL);
+		if (!chunk)
+			break;
+		chunk->next = pid_list->upper_list;
+		pid_list->upper_list = chunk;
+		pid_list->free_upper_chunks++;
 	}
+
+	for (i = 0; i < CHUNK_ALLOC; i++) {
+		union lower_chunk *chunk;
+
+		chunk = kzalloc(sizeof(*chunk), GFP_KERNEL);
+		if (!chunk)
+			break;
+		chunk->next = pid_list->lower_list;
+		pid_list->lower_list = chunk;
+		pid_list->free_lower_chunks++;
+	}
+
 	return pid_list;
 }
 
@@ -152,9 +456,40 @@ struct trace_pid_list *trace_pid_list_alloc(void)
  */
 void trace_pid_list_free(struct trace_pid_list *pid_list)
 {
+	union upper_chunk *upper;
+	union lower_chunk *lower;
+	int i, j;
+
 	if (!pid_list)
 		return;
 
-	vfree(pid_list->pids);
+	irq_work_sync(&pid_list->refill_irqwork);
+
+	while (pid_list->lower_list) {
+		union lower_chunk *chunk;
+
+		chunk = pid_list->lower_list;
+		pid_list->lower_list = pid_list->lower_list->next;
+		kfree(chunk);
+	}
+
+	while (pid_list->upper_list) {
+		union upper_chunk *chunk;
+
+		chunk = pid_list->upper_list;
+		pid_list->upper_list = pid_list->upper_list->next;
+		kfree(chunk);
+	}
+
+	for (i = 0; i < UPPER1_SIZE; i++) {
+		upper = pid_list->upper[i];
+		if (upper) {
+			for (j = 0; j < UPPER2_SIZE; j++) {
+				lower = upper->data[j];
+				kfree(lower);
+			}
+			kfree(upper);
+		}
+	}
 	kfree(pid_list);
 }