On Wed, Dec 29, 2021 at 12:17:30PM +0100, Pavel Machek wrote:
hi!
From: Jiasheng Jiang jiasheng@iscas.ac.cn
[ Upstream commit bdf1b5c3884f6a0dc91b0dbdb8c3b7d205f449e0 ]
Because of the possible failure of the kcalloc, it should be better to set rx_queue->page_ptr_mask to 0 when it happens in order to maintain the consistency.
This is confusing/wrong, or at least not a complete fix.
+++ b/drivers/net/ethernet/sfc/rx_common.c @@ -150,7 +150,10 @@ static void efx_init_rx_recycle_ring(struct efx_rx_queue *rx_queue) efx->rx_bufs_per_page); rx_queue->page_ring = kcalloc(page_ring_size, sizeof(*rx_queue->page_ring), GFP_KERNEL);
- rx_queue->page_ptr_mask = page_ring_size - 1;
- if (!rx_queue->page_ring)
rx_queue->page_ptr_mask = 0;- else
rx_queue->page_ptr_mask = page_ring_size - 1;}
So we have !rx_queue->page_ring. But in efx_reuse_page, we do
index = rx_queue->page_remove & rx_queue->page_ptr_mask;page = rx_queue->page_ring[index];
So index is now zero, but we'll derefernce null pointer anyway.
Good point. I've posted a patch for this to netdev. https://lore.kernel.org/netdev/164103678041.26263.8354809911405746465.stgit@...
Thanks, Martin