core: Move free-list locking to a separate per-region lock

Currently, we have a single lock for the entire mem_region system; this
protects both the global region list, and the allocations from each
region.

This means we can't allocate memory while traversing the global region
list, as any malloc/realloc/free will try to acquire the mem_region lock
again.

This change separates the locking into different functions. We keep the
mem_region_lock to protect the regions list, and introduce a per-region
lock to protect allocations from the regions' free_lists.

Then we remove the open-coded invocations of mem_alloc, where we'd
avoided malloc() due to the above locking issue.

Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Signed-off-by: Stewart Smith <stewart@linux.vnet.ibm.com>

1 files changed, 3 insertions, 0 deletions

diff --git a/include/mem_region.h b/include/mem_region.h
index d6dd2965..e5cf0f66 100644
--- a/include/mem_region.h
+++ b/include/mem_region.h
@@ -19,6 +19,8 @@
 #include <ccan/list/list.h>
 #include <stdint.h>
 
+#include <lock.h>
+
 enum mem_region_type {
 	/* ranges allocatable by mem_alloc: this will be most of memory */
 	REGION_SKIBOOT_HEAP,
@@ -41,6 +43,7 @@ struct mem_region {
 	struct dt_node *mem_node;
 	enum mem_region_type type;
 	struct list_head free_list;
+	struct lock free_list_lock;
 };
 
 extern struct lock mem_region_lock;