net/xps: Fix calculation of initial number of xps queues
The existing code breaks on architectures where the L1 cache size
(L1_CACHE_BYTES) is smaller or equal the size of struct xps_map.
The new code ensures that we get at minimum one initial xps queue, or even more
as long as it fits into the next multiple of L1_CACHE_SIZE.
Signed-off-by: Helge Deller <deller@gmx.de>
Acked-by: Alexander Duyck <aduyck@mirantis.com>
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 2d15e38..2212c82 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -718,8 +718,8 @@
u16 queues[0];
};
#define XPS_MAP_SIZE(_num) (sizeof(struct xps_map) + ((_num) * sizeof(u16)))
-#define XPS_MIN_MAP_ALLOC ((L1_CACHE_BYTES - sizeof(struct xps_map)) \
- / sizeof(u16))
+#define XPS_MIN_MAP_ALLOC ((L1_CACHE_ALIGN(offsetof(struct xps_map, queues[1])) \
+ - sizeof(struct xps_map)) / sizeof(u16))
/*
* This structure holds all XPS maps for device. Maps are indexed by CPU.