Stability hack. Under Solaris when KM_SLEEP is set kmem_cache_alloc()

may not fail.  To get this behavior I'd added a retry to the shim layer
even though it is abusive to the VM, at least it should prevent the crash.
Additionally I added a proc counter so I can easily check how often this
is happening.  It should be fairly rare, but likely will get worse and
worse the longer the machine has been up.


git-svn-id: https://outreach.scidac.gov/svn/spl/trunk@104 7e1ea52c-4ff2-0310-8f11-9dd32ca42a1c

2 files changed, 37 insertions, 0 deletions

diff --git a/modules/spl/spl-kmem.c b/modules/spl/spl-kmem.c
index 75a421ac7..80c4488b1 100644
--- a/modules/spl/spl-kmem.c
+++ b/modules/spl/spl-kmem.c
@@ -16,6 +16,7 @@ unsigned long kmem_alloc_max = 0;
 atomic64_t vmem_alloc_used;
 unsigned long vmem_alloc_max = 0;
 int kmem_warning_flag = 1;
+atomic64_t kmem_cache_alloc_failed;
 
 spinlock_t kmem_lock;
 struct hlist_head kmem_table[KMEM_TABLE_SIZE];
@@ -268,6 +269,7 @@ kmem_cache_generic_shrinker(int nr_to_scan, unsigned int gfp_mask)
  */
 #undef kmem_cache_create
 #undef kmem_cache_destroy
+#undef kmem_cache_alloc
 
 kmem_cache_t *
 __kmem_cache_create(char *name, size_t size, size_t align,
@@ -360,6 +362,30 @@ __kmem_cache_destroy(kmem_cache_t *cache)
 }
 EXPORT_SYMBOL(__kmem_cache_destroy);
 
+/* Under Solaris if the KM_SLEEP flag is passed we absolutely must
+ * sleep until we are allocated the memory.  Under Linux you can still
+ * get a memory allocation failure, so I'm forced to keep requesting
+ * the memory even if the system is under substantial memory pressure
+ * of fragmentation prevents the allocation from succeeded.  This is
+ * not the correct fix, or even a good one.  But it will do for now.
+ */
+void *
+__kmem_cache_alloc(kmem_cache_t *cache, gfp_t flags)
+{
+	void *rc;
+	ENTRY;
+
+restart:
+	rc = kmem_cache_alloc(cache, flags);
+        if ((rc == NULL) && (flags & KM_SLEEP)) {
+		atomic64_inc(&kmem_cache_alloc_failed);
+		GOTO(restart, rc);
+	}
+
+	RETURN(rc);
+}
+EXPORT_SYMBOL(__kmem_cache_alloc);
+
 void
 __kmem_reap(void)
 {
@@ -395,6 +421,8 @@ kmem_init(void)
 
                 for (i = 0; i < VMEM_TABLE_SIZE; i++)
                         INIT_HLIST_HEAD(&vmem_table[i]);
+
+		atomic64_set(&kmem_cache_alloc_failed, 0);
         }
 #endif
 	RETURN(0);
diff --git a/modules/spl/spl-proc.c b/modules/spl/spl-proc.c
index 07e18102a..a8d3485a6 100644
--- a/modules/spl/spl-proc.c
+++ b/modules/spl/spl-proc.c
@@ -60,6 +60,7 @@ enum {
         CTL_KMEM_KMEMMAX,         /* Max alloc'd by kmem bytes */
         CTL_KMEM_VMEMUSED,        /* Currently alloc'd vmem bytes */
         CTL_KMEM_VMEMMAX,         /* Max alloc'd by vmem bytes */
+	CTL_KMEM_ALLOC_FAILED,    /* Cache allocation failed */
 #endif
 
 	CTL_MUTEX_STATS,          /* Global mutex statistics */
@@ -660,6 +661,14 @@ static struct ctl_table spl_kmem_table[] = {
                 .mode     = 0444,
                 .proc_handler = &proc_doulongvec_minmax,
         },
+        {
+                .ctl_name = CTL_KMEM_ALLOC_FAILED,
+                .procname = "kmem_alloc_failed",
+                .data     = &kmem_cache_alloc_failed,
+                .maxlen   = sizeof(atomic64_t),
+                .mode     = 0444,
+                .proc_handler = &proc_doatomic64,
+        },
 	{0},
 };
 #endif /* DEBUG_KMEM */