| Message ID | 1434405656-10465-4-git-send-email-jacob.e.keller@intel.com |
|---|---|
| State | Superseded |
| Delegated to: | Jeff Kirsher |
| Headers | show |
On Mon, 2015-06-15 at 15:00 -0700, Jacob Keller wrote: > This patch fixes a corner case issue with the PF/VF mailbox code. > Currently, fm10k_mbx_reset_work clears various state about the > mailbox. However, it did not clear the Tx FIFO head/tail pointers. > Thus, > when the PF finally re-connects (say due to VF reset), it in > -advertently > might send some previous data in the Tx FIFO again. This was > discovered > by attempting to perform a static VF MAC address assignment on the PF > device. The VF would receive the message and then initiate a reset. > During this process it disconnects and reconnects to the mailbox. If > the > PF doesn't respond fast enough, it may not see the disconnect > message, > and only sees a new connect message. In this flow, it does not > properly > clear the Tx FIFO pointers. Thus, it will re-transmit the last > message > in the Tx FIFO. However, it does not increment the mailbox Tx > messages > because it has already reset mbx->pulled and thus does not think it > is > actually sending any data. The fix is to simply reset the Tx FIFO > pointers whenever we call fm10k_mbx_reset_work. > > As a consequence, we no longer need both fm10k_mbx_fifo_drop_all and > calls > to fm10k_mbx_update_max_size with a size of 0. This patch will need a revision 2. There is some other issues this patch creates which we need to properly resolve. Clearing the Tx FIFO causes us to potentially drop untransmitted work and thus fail to send repsonses with no error indication returning to the base driver code. I am working on a corrected fix now. Regards, Jake
diff --git a/drivers/net/ethernet/intel/fm10k/fm10k_mbx.c b/drivers/net/ethernet/intel/fm10k/fm10k_mbx.c index 1a4b52637de9..f9016bf8b5b1 100644 --- a/drivers/net/ethernet/intel/fm10k/fm10k_mbx.c +++ b/drivers/net/ethernet/intel/fm10k/fm10k_mbx.c @@ -126,18 +126,6 @@ static u16 fm10k_fifo_head_drop(struct fm10k_mbx_fifo *fifo) } /** - * fm10k_fifo_drop_all - Drop all messages in FIFO - * @fifo: pointer to FIFO - * - * This function resets the head pointer to drop all messages in the FIFO, - * and ensure the FIFO is empty. - **/ -static void fm10k_fifo_drop_all(struct fm10k_mbx_fifo *fifo) -{ - fifo->head = fifo->tail; -} - -/** * fm10k_mbx_index_len - Convert a head/tail index into a length value * @mbx: pointer to mailbox * @head: head index @@ -1054,6 +1042,8 @@ static void fm10k_mbx_reset_work(struct fm10k_mbx_info *mbx) mbx->pulled = 0; mbx->tail_len = 0; mbx->head_len = 0; + mbx->tx.tail = 0; + mbx->tx.head = 0; mbx->rx.tail = 0; mbx->rx.head = 0; } @@ -1384,7 +1374,6 @@ static void fm10k_mbx_disconnect(struct fm10k_hw *hw, * drop all left over messages in the FIFO. */ fm10k_mbx_connect_reset(mbx); - fm10k_fifo_drop_all(&mbx->tx); fm10k_write_reg(hw, mbx->mbmem_reg, 0); } @@ -1725,7 +1714,6 @@ static void fm10k_sm_mbx_disconnect(struct fm10k_hw *hw, mbx->state = FM10K_STATE_CLOSED; mbx->remote = 0; fm10k_mbx_reset_work(mbx); - fm10k_mbx_update_max_size(mbx, 0); fm10k_write_reg(hw, mbx->mbmem_reg, 0); }
This patch fixes a corner case issue with the PF/VF mailbox code. Currently, fm10k_mbx_reset_work clears various state about the mailbox. However, it did not clear the Tx FIFO head/tail pointers. Thus, when the PF finally re-connects (say due to VF reset), it in-advertently might send some previous data in the Tx FIFO again. This was discovered by attempting to perform a static VF MAC address assignment on the PF device. The VF would receive the message and then initiate a reset. During this process it disconnects and reconnects to the mailbox. If the PF doesn't respond fast enough, it may not see the disconnect message, and only sees a new connect message. In this flow, it does not properly clear the Tx FIFO pointers. Thus, it will re-transmit the last message in the Tx FIFO. However, it does not increment the mailbox Tx messages because it has already reset mbx->pulled and thus does not think it is actually sending any data. The fix is to simply reset the Tx FIFO pointers whenever we call fm10k_mbx_reset_work. As a consequence, we no longer need both fm10k_mbx_fifo_drop_all and calls to fm10k_mbx_update_max_size with a size of 0. Signed-off-by: Jacob Keller <jacob.e.keller@intel.com> --- drivers/net/ethernet/intel/fm10k/fm10k_mbx.c | 16 ++-------------- 1 file changed, 2 insertions(+), 14 deletions(-)