The forward and backward dependencies consume the same number of list_entries. Since backward dependencies are removed, the max number of list_entries can be halved safely: MAX_LOCKDEP_ENTRIES /= 2, which goes from 16384 to 8192 or 32768 to 16384.
Besides memory space reduction, the performance gain is also promising with this new IRQ usage checking algorithm. We briefly show the performance improvement on a simple workload: Linux kernel build (i.e., make clean; reboot; make vmlinux -j8). It is expected that for complex real-world workloads with heavier lock usage and bigger dependency graph, the benefits should be more. Results: ------ Before ------ direct dependencies: 6900 [max: 32768] max bfs queue depth: 272 find-mask forwards checks: 2875 find-mask backwards checks: 50229 ----- After ----- direct dependencies: 3444 [max: 16384] ( - 50 % ) max bfs queue depth: 223 ( - 18 % ) find-mask forwards checks: 370 ( - 87 % ) find-mask backwards checks: 0 ( - 100 % ) Signed-off-by: Yuyang Du <[email protected]> --- kernel/locking/lockdep_internals.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/kernel/locking/lockdep_internals.h b/kernel/locking/lockdep_internals.h index 30d021b..3b4b3dd 100644 --- a/kernel/locking/lockdep_internals.h +++ b/kernel/locking/lockdep_internals.h @@ -92,11 +92,11 @@ enum { * table (if it's not there yet), and we check it for lock order * conflicts and deadlocks. */ -#define MAX_LOCKDEP_ENTRIES 16384UL +#define MAX_LOCKDEP_ENTRIES 8194UL #define MAX_LOCKDEP_CHAINS_BITS 15 #define MAX_STACK_TRACE_ENTRIES 262144UL #else -#define MAX_LOCKDEP_ENTRIES 32768UL +#define MAX_LOCKDEP_ENTRIES 16384UL #define MAX_LOCKDEP_CHAINS_BITS 16 -- 1.8.3.1
