[ofa-general] IB_WC_WR_FLUSH_ERR on receive requests

[Isaac Huang]

Hello,

Recently we observed a handful of IB_WC_WR_FLUSH_ERR errors on receive
requests. I grepped through OFED 1.2.5.1 code and found that this
error was mostly set in low level hw drivers. It also seemed that some
ULPs didn't treat IB_WC_WR_FLUSH_ERR as a critical error, e.g. in
ulp/ipoib/ipoib_cm.c:ipoib_ib_handle_rx_wc():
         if (unlikely(wc->status != IB_WC_SUCCESS)) {
                 if (wc->status != IB_WC_WR_FLUSH_ERR)
                         ipoib_warn(priv, "failed recv event "
                                    "(status=%d, wrid=%d vend_err %x)\n",
                                    wc->status, wr_id, wc->vendor_err);

In what situations does this IB_WC_WR_FLUSH_ERR usually happen? Does it
indicate any issue in the HCA hardware?

Thanks,
Isaac