From c1b43dca137f2154845122417fba86d4bae67182 Mon Sep 17 00:00:00 2001 From: Eli Cohen Date: Tue, 22 Mar 2011 22:38:41 +0000 Subject: [PATCH] mlx4: Add blue flame support for kernel consumers Using blue flame can improve latency by allowing the HW to more efficiently access the WQE. This patch presents two functions that are used to allocate or release HW resources for using blue flame; the caller need to supply a struct mlx4_bf object when allocating resources. Consumers that make use of this API should post doorbells to the UAR object pointed by the initialized struct mlx4_bf; Signed-off-by: Eli Cohen Signed-off-by: David S. Miller --- drivers/net/mlx4/main.c | 31 +++++++++++++++ drivers/net/mlx4/mlx4.h | 3 ++ drivers/net/mlx4/pd.c | 94 +++++++++++++++++++++++++++++++++++++++++++++ include/linux/mlx4/device.h | 13 +++++++ include/linux/mlx4/qp.h | 1 + 5 files changed, 142 insertions(+) diff --git a/drivers/net/mlx4/main.c b/drivers/net/mlx4/main.c index c8e2761..05c5671 100644 --- a/drivers/net/mlx4/main.c +++ b/drivers/net/mlx4/main.c @@ -39,6 +39,7 @@ #include #include #include +#include #include #include @@ -721,8 +722,31 @@ static void mlx4_free_icms(struct mlx4_dev *dev) mlx4_free_icm(dev, priv->fw.aux_icm, 0); } +static int map_bf_area(struct mlx4_dev *dev) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + resource_size_t bf_start; + resource_size_t bf_len; + int err = 0; + + bf_start = pci_resource_start(dev->pdev, 2) + (dev->caps.num_uars << PAGE_SHIFT); + bf_len = pci_resource_len(dev->pdev, 2) - (dev->caps.num_uars << PAGE_SHIFT); + priv->bf_mapping = io_mapping_create_wc(bf_start, bf_len); + if (!priv->bf_mapping) + err = -ENOMEM; + + return err; +} + +static void unmap_bf_area(struct mlx4_dev *dev) +{ + if (mlx4_priv(dev)->bf_mapping) + io_mapping_free(mlx4_priv(dev)->bf_mapping); +} + static void mlx4_close_hca(struct mlx4_dev *dev) { + unmap_bf_area(dev); mlx4_CLOSE_HCA(dev, 0); mlx4_free_icms(dev); mlx4_UNMAP_FA(dev); @@ -775,6 +799,9 @@ static int mlx4_init_hca(struct mlx4_dev *dev) goto err_stop_fw; } + if (map_bf_area(dev)) + mlx4_dbg(dev, "Failed to map blue flame area\n"); + init_hca.log_uar_sz = ilog2(dev->caps.num_uars); err = mlx4_init_icm(dev, &dev_cap, &init_hca, icm_size); @@ -805,6 +832,7 @@ err_free_icm: mlx4_free_icms(dev); err_stop_fw: + unmap_bf_area(dev); mlx4_UNMAP_FA(dev); mlx4_free_icm(dev, priv->fw.fw_icm, 0); @@ -1196,6 +1224,9 @@ static int __mlx4_init_one(struct pci_dev *pdev, const struct pci_device_id *id) pci_read_config_byte(pdev, PCI_REVISION_ID, &dev->rev_id); + INIT_LIST_HEAD(&priv->bf_list); + mutex_init(&priv->bf_mutex); + /* * Now reset the HCA before we touch the PCI capabilities or * attempt a firmware command, since a boot ROM may have left diff --git a/drivers/net/mlx4/mlx4.h b/drivers/net/mlx4/mlx4.h index d8bb441..bc9a216 100644 --- a/drivers/net/mlx4/mlx4.h +++ b/drivers/net/mlx4/mlx4.h @@ -353,6 +353,9 @@ struct mlx4_priv { struct mutex port_mutex; struct mlx4_msix_ctl msix_ctl; struct mlx4_steer *steer; + struct list_head bf_list; + struct mutex bf_mutex; + struct io_mapping *bf_mapping; }; static inline struct mlx4_priv *mlx4_priv(struct mlx4_dev *dev) diff --git a/drivers/net/mlx4/pd.c b/drivers/net/mlx4/pd.c index c4988d6..5210a0f 100644 --- a/drivers/net/mlx4/pd.c +++ b/drivers/net/mlx4/pd.c @@ -32,6 +32,7 @@ */ #include +#include #include @@ -77,6 +78,7 @@ int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar) return -ENOMEM; uar->pfn = (pci_resource_start(dev->pdev, 2) >> PAGE_SHIFT) + uar->index; + uar->map = NULL; return 0; } @@ -88,6 +90,98 @@ void mlx4_uar_free(struct mlx4_dev *dev, struct mlx4_uar *uar) } EXPORT_SYMBOL_GPL(mlx4_uar_free); +int mlx4_bf_alloc(struct mlx4_dev *dev, struct mlx4_bf *bf) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + struct mlx4_uar *uar; + int err = 0; + int idx; + + if (!priv->bf_mapping) + return -ENOMEM; + + mutex_lock(&priv->bf_mutex); + if (!list_empty(&priv->bf_list)) + uar = list_entry(priv->bf_list.next, struct mlx4_uar, bf_list); + else { + uar = kmalloc(sizeof *uar, GFP_KERNEL); + if (!uar) { + err = -ENOMEM; + goto out; + } + err = mlx4_uar_alloc(dev, uar); + if (err) + goto free_kmalloc; + + uar->map = ioremap(uar->pfn << PAGE_SHIFT, PAGE_SIZE); + if (!uar->map) { + err = -ENOMEM; + goto free_uar; + } + + uar->bf_map = io_mapping_map_wc(priv->bf_mapping, uar->index << PAGE_SHIFT); + if (!uar->bf_map) { + err = -ENOMEM; + goto unamp_uar; + } + uar->free_bf_bmap = 0; + list_add(&uar->bf_list, &priv->bf_list); + } + + bf->uar = uar; + idx = ffz(uar->free_bf_bmap); + uar->free_bf_bmap |= 1 << idx; + bf->uar = uar; + bf->offset = 0; + bf->buf_size = dev->caps.bf_reg_size / 2; + bf->reg = uar->bf_map + idx * dev->caps.bf_reg_size; + if (uar->free_bf_bmap == (1 << dev->caps.bf_regs_per_page) - 1) + list_del_init(&uar->bf_list); + + goto out; + +unamp_uar: + bf->uar = NULL; + iounmap(uar->map); + +free_uar: + mlx4_uar_free(dev, uar); + +free_kmalloc: + kfree(uar); + +out: + mutex_unlock(&priv->bf_mutex); + return err; +} +EXPORT_SYMBOL_GPL(mlx4_bf_alloc); + +void mlx4_bf_free(struct mlx4_dev *dev, struct mlx4_bf *bf) +{ + struct mlx4_priv *priv = mlx4_priv(dev); + int idx; + + if (!bf->uar || !bf->uar->bf_map) + return; + + mutex_lock(&priv->bf_mutex); + idx = (bf->reg - bf->uar->bf_map) / dev->caps.bf_reg_size; + bf->uar->free_bf_bmap &= ~(1 << idx); + if (!bf->uar->free_bf_bmap) { + if (!list_empty(&bf->uar->bf_list)) + list_del(&bf->uar->bf_list); + + io_mapping_unmap(bf->uar->bf_map); + iounmap(bf->uar->map); + mlx4_uar_free(dev, bf->uar); + kfree(bf->uar); + } else if (list_empty(&bf->uar->bf_list)) + list_add(&bf->uar->bf_list, &priv->bf_list); + + mutex_unlock(&priv->bf_mutex); +} +EXPORT_SYMBOL_GPL(mlx4_bf_free); + int mlx4_init_uar_table(struct mlx4_dev *dev) { if (dev->caps.num_uars <= 128) { diff --git a/include/linux/mlx4/device.h b/include/linux/mlx4/device.h index 56fa5e1..8985768 100644 --- a/include/linux/mlx4/device.h +++ b/include/linux/mlx4/device.h @@ -351,6 +351,17 @@ struct mlx4_fmr { struct mlx4_uar { unsigned long pfn; int index; + struct list_head bf_list; + unsigned free_bf_bmap; + void __iomem *map; + void __iomem *bf_map; +}; + +struct mlx4_bf { + unsigned long offset; + int buf_size; + struct mlx4_uar *uar; + void __iomem *reg; }; struct mlx4_cq { @@ -478,6 +489,8 @@ void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn); int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar); void mlx4_uar_free(struct mlx4_dev *dev, struct mlx4_uar *uar); +int mlx4_bf_alloc(struct mlx4_dev *dev, struct mlx4_bf *bf); +void mlx4_bf_free(struct mlx4_dev *dev, struct mlx4_bf *bf); int mlx4_mtt_init(struct mlx4_dev *dev, int npages, int page_shift, struct mlx4_mtt *mtt); diff --git a/include/linux/mlx4/qp.h b/include/linux/mlx4/qp.h index 0eeb2a1..9e9eb21 100644 --- a/include/linux/mlx4/qp.h +++ b/include/linux/mlx4/qp.h @@ -303,6 +303,7 @@ struct mlx4_wqe_data_seg { enum { MLX4_INLINE_ALIGN = 64, + MLX4_INLINE_SEG = 1 << 31, }; struct mlx4_wqe_inline_seg { -- 2.7.4