|
|
@@ -17,7 +17,7 @@
|
|
|
#include "trace.h"
|
|
|
|
|
|
static RADIX_TREE(pids, GFP_ATOMIC);
|
|
|
-static struct mutex pids_lock;
|
|
|
+static spinlock_t pids_lock;
|
|
|
static struct last_io_info last_io;
|
|
|
|
|
|
static inline void __print_last_io(void)
|
|
|
@@ -61,23 +61,29 @@ void f2fs_trace_pid(struct page *page)
|
|
|
|
|
|
set_page_private(page, (unsigned long)pid);
|
|
|
|
|
|
+retry:
|
|
|
if (radix_tree_preload(GFP_NOFS))
|
|
|
return;
|
|
|
|
|
|
- mutex_lock(&pids_lock);
|
|
|
+ spin_lock(&pids_lock);
|
|
|
p = radix_tree_lookup(&pids, pid);
|
|
|
if (p == current)
|
|
|
goto out;
|
|
|
if (p)
|
|
|
radix_tree_delete(&pids, pid);
|
|
|
|
|
|
- f2fs_radix_tree_insert(&pids, pid, current);
|
|
|
+ if (radix_tree_insert(&pids, pid, current)) {
|
|
|
+ spin_unlock(&pids_lock);
|
|
|
+ radix_tree_preload_end();
|
|
|
+ cond_resched();
|
|
|
+ goto retry;
|
|
|
+ }
|
|
|
|
|
|
trace_printk("%3x:%3x %4x %-16s\n",
|
|
|
MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
|
|
|
pid, current->comm);
|
|
|
out:
|
|
|
- mutex_unlock(&pids_lock);
|
|
|
+ spin_unlock(&pids_lock);
|
|
|
radix_tree_preload_end();
|
|
|
}
|
|
|
|
|
|
@@ -122,7 +128,7 @@ void f2fs_trace_ios(struct f2fs_io_info *fio, int flush)
|
|
|
|
|
|
void f2fs_build_trace_ios(void)
|
|
|
{
|
|
|
- mutex_init(&pids_lock);
|
|
|
+ spin_lock_init(&pids_lock);
|
|
|
}
|
|
|
|
|
|
#define PIDVEC_SIZE 128
|
|
|
@@ -150,7 +156,7 @@ void f2fs_destroy_trace_ios(void)
|
|
|
pid_t next_pid = 0;
|
|
|
unsigned int found;
|
|
|
|
|
|
- mutex_lock(&pids_lock);
|
|
|
+ spin_lock(&pids_lock);
|
|
|
while ((found = gang_lookup_pids(pid, next_pid, PIDVEC_SIZE))) {
|
|
|
unsigned idx;
|
|
|
|
|
|
@@ -158,5 +164,5 @@ void f2fs_destroy_trace_ios(void)
|
|
|
for (idx = 0; idx < found; idx++)
|
|
|
radix_tree_delete(&pids, pid[idx]);
|
|
|
}
|
|
|
- mutex_unlock(&pids_lock);
|
|
|
+ spin_unlock(&pids_lock);
|
|
|
}
|