The low-level driver modified the page-list addresses for FRWR post send to big-endian, and set a "present" bit. This caused problems later when the ULP attempted to unmap the pages in the page-list (using the list addresses which were assumed to be still in CPU-endian order).
The cause of the crash was found by Vu Pham of Mellanox. The fix is along the lines suggested by Steve Wise in comment #21 in Bugzilla 1571. This patch fixes Bugzilla 1571. Signed-off-by: Jack Morgenstein <ja...@dev.mellanox.co.il> --- Roland, please take this for kernel 2.6.30. diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h index 9974e88..a8c0bc4 100644 --- a/drivers/infiniband/hw/mlx4/mlx4_ib.h +++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h @@ -86,6 +86,7 @@ struct mlx4_ib_mr { struct mlx4_ib_fast_reg_page_list { struct ib_fast_reg_page_list ibfrpl; + u64 *mapped_page_list; dma_addr_t map; }; diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c index 8e4d26d..fddf583 100644 --- a/drivers/infiniband/hw/mlx4/mr.c +++ b/drivers/infiniband/hw/mlx4/mr.c @@ -231,16 +231,22 @@ struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device if (!mfrpl) return ERR_PTR(-ENOMEM); - mfrpl->ibfrpl.page_list = dma_alloc_coherent(&dev->dev->pdev->dev, + mfrpl->ibfrpl.page_list = kmalloc(size, GFP_KERNEL); + if (!mfrpl->ibfrpl.page_list) + goto err_free; + + mfrpl->mapped_page_list = dma_alloc_coherent(&dev->dev->pdev->dev, size, &mfrpl->map, GFP_KERNEL); if (!mfrpl->ibfrpl.page_list) - goto err_free; + goto err_free_mfrpl; WARN_ON(mfrpl->map & 0x3f); return &mfrpl->ibfrpl; +err_free_mfrpl: + kfree(mfrpl->ibfrpl.page_list); err_free: kfree(mfrpl); return ERR_PTR(-ENOMEM); @@ -252,8 +258,9 @@ void mlx4_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list) struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list); int size = page_list->max_page_list_len * sizeof (u64); - dma_free_coherent(&dev->dev->pdev->dev, size, page_list->page_list, + dma_free_coherent(&dev->dev->pdev->dev, size, mfrpl->mapped_page_list, mfrpl->map); + kfree(mfrpl->ibfrpl.page_list); kfree(mfrpl); } diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index f385a24..20724ae 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -1365,7 +1365,7 @@ static void set_fmr_seg(struct mlx4_wqe_fmr_seg *fseg, struct ib_send_wr *wr) int i; for (i = 0; i < wr->wr.fast_reg.page_list_len; ++i) - wr->wr.fast_reg.page_list->page_list[i] = + mfrpl->mapped_page_list[i] = cpu_to_be64(wr->wr.fast_reg.page_list->page_list[i] | MLX4_MTT_FLAG_PRESENT); _______________________________________________ general mailing list general@lists.openfabrics.org http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general