/* CRQB needs alignment on a 1KB boundary. Size == 1KB
* CRPB needs alignment on a 256B boundary. Size == 256B
- * SG count of 176 leads to MV_PORT_PRIV_DMA_SZ == 4KB
* ePRD (SG) entries need alignment on a 16B boundary. Size == 16B
*/
MV_CRQB_Q_SZ = (32 * MV_MAX_Q_DEPTH),
MV_CRPB_Q_SZ = (8 * MV_MAX_Q_DEPTH),
- MV_MAX_SG_CT = 176,
+ MV_MAX_SG_CT = 256,
MV_SG_TBL_SZ = (16 * MV_MAX_SG_CT),
- MV_PORT_PRIV_DMA_SZ = (MV_CRQB_Q_SZ + MV_CRPB_Q_SZ + MV_SG_TBL_SZ),
MV_PORTS_PER_HC = 4,
/* == (port / MV_PORTS_PER_HC) to determine HC from 0-7 port */
dma_addr_t crqb_dma;
struct mv_crpb *crpb;
dma_addr_t crpb_dma;
- struct mv_sg *sg_tbl;
- dma_addr_t sg_tbl_dma;
+ struct mv_sg *sg_tbl[MV_MAX_Q_DEPTH];
+ dma_addr_t sg_tbl_dma[MV_MAX_Q_DEPTH];
unsigned int req_idx;
unsigned int resp_idx;
u32 irq_cause_ofs;
u32 irq_mask_ofs;
u32 unmask_all_irqs;
+ /*
+ * These consistent DMA memory pools give us guaranteed
+ * alignment for hardware-accessed data structures,
+ * and less memory waste in accomplishing the alignment.
+ */
+ struct dma_pool *crqb_pool;
+ struct dma_pool *crpb_pool;
+ struct dma_pool *sg_tbl_pool;
};
struct mv_hw_ops {
static void mv_post_int_cmd(struct ata_queued_cmd *qc);
static void mv_eh_freeze(struct ata_port *ap);
static void mv_eh_thaw(struct ata_port *ap);
+static void mv6_dev_config(struct ata_device *dev);
static int mv_init_one(struct pci_dev *pdev, const struct pci_device_id *ent);
static void mv5_phy_errata(struct mv_host_priv *hpriv, void __iomem *mmio,
void __iomem *port_mmio, int want_ncq);
static int __mv_stop_dma(struct ata_port *ap);
+/* .sg_tablesize is (MV_MAX_SG_CT / 2) in the structures below
+ * because we have to allow room for worst case splitting of
+ * PRDs for 64K boundaries in mv_fill_sg().
+ */
static struct scsi_host_template mv5_sht = {
.module = THIS_MODULE,
.name = DRV_NAME,
};
static const struct ata_port_operations mv6_ops = {
+ .dev_config = mv6_dev_config,
.tf_load = ata_tf_load,
.tf_read = ata_tf_read,
.check_status = ata_check_status,
return -EINVAL;
}
+static void mv6_dev_config(struct ata_device *adev)
+{
+ /*
+ * We don't have hob_nsect when doing NCQ commands on Gen-II.
+ * See mv_qc_prep() for more info.
+ */
+ if (adev->flags & ATA_DFLAG_NCQ)
+ if (adev->max_sectors > ATA_MAX_SECTORS)
+ adev->max_sectors = ATA_MAX_SECTORS;
+}
+
static void mv_edma_cfg(struct mv_port_priv *pp, struct mv_host_priv *hpriv,
void __iomem *port_mmio, int want_ncq)
{
writelfl(cfg, port_mmio + EDMA_CFG_OFS);
}
+static void mv_port_free_dma_mem(struct ata_port *ap)
+{
+ struct mv_host_priv *hpriv = ap->host->private_data;
+ struct mv_port_priv *pp = ap->private_data;
+ int tag;
+
+ if (pp->crqb) {
+ dma_pool_free(hpriv->crqb_pool, pp->crqb, pp->crqb_dma);
+ pp->crqb = NULL;
+ }
+ if (pp->crpb) {
+ dma_pool_free(hpriv->crpb_pool, pp->crpb, pp->crpb_dma);
+ pp->crpb = NULL;
+ }
+ /*
+ * For GEN_I, there's no NCQ, so we have only a single sg_tbl.
+ * For later hardware, we have one unique sg_tbl per NCQ tag.
+ */
+ for (tag = 0; tag < MV_MAX_Q_DEPTH; ++tag) {
+ if (pp->sg_tbl[tag]) {
+ if (tag == 0 || !IS_GEN_I(hpriv))
+ dma_pool_free(hpriv->sg_tbl_pool,
+ pp->sg_tbl[tag],
+ pp->sg_tbl_dma[tag]);
+ pp->sg_tbl[tag] = NULL;
+ }
+ }
+}
+
/**
* mv_port_start - Port specific init/start routine.
* @ap: ATA channel to manipulate
struct mv_host_priv *hpriv = ap->host->private_data;
struct mv_port_priv *pp;
void __iomem *port_mmio = mv_ap_base(ap);
- void *mem;
- dma_addr_t mem_dma;
unsigned long flags;
- int rc;
+ int tag, rc;
pp = devm_kzalloc(dev, sizeof(*pp), GFP_KERNEL);
if (!pp)
return -ENOMEM;
-
- mem = dmam_alloc_coherent(dev, MV_PORT_PRIV_DMA_SZ, &mem_dma,
- GFP_KERNEL);
- if (!mem)
- return -ENOMEM;
- memset(mem, 0, MV_PORT_PRIV_DMA_SZ);
+ ap->private_data = pp;
rc = ata_pad_alloc(ap, dev);
if (rc)
return rc;
- /* First item in chunk of DMA memory:
- * 32-slot command request table (CRQB), 32 bytes each in size
- */
- pp->crqb = mem;
- pp->crqb_dma = mem_dma;
- mem += MV_CRQB_Q_SZ;
- mem_dma += MV_CRQB_Q_SZ;
+ pp->crqb = dma_pool_alloc(hpriv->crqb_pool, GFP_KERNEL, &pp->crqb_dma);
+ if (!pp->crqb)
+ return -ENOMEM;
+ memset(pp->crqb, 0, MV_CRQB_Q_SZ);
- /* Second item:
- * 32-slot command response table (CRPB), 8 bytes each in size
- */
- pp->crpb = mem;
- pp->crpb_dma = mem_dma;
- mem += MV_CRPB_Q_SZ;
- mem_dma += MV_CRPB_Q_SZ;
+ pp->crpb = dma_pool_alloc(hpriv->crpb_pool, GFP_KERNEL, &pp->crpb_dma);
+ if (!pp->crpb)
+ goto out_port_free_dma_mem;
+ memset(pp->crpb, 0, MV_CRPB_Q_SZ);
- /* Third item:
- * Table of scatter-gather descriptors (ePRD), 16 bytes each
+ /*
+ * For GEN_I, there's no NCQ, so we only allocate a single sg_tbl.
+ * For later hardware, we need one unique sg_tbl per NCQ tag.
*/
- pp->sg_tbl = mem;
- pp->sg_tbl_dma = mem_dma;
+ for (tag = 0; tag < MV_MAX_Q_DEPTH; ++tag) {
+ if (tag == 0 || !IS_GEN_I(hpriv)) {
+ pp->sg_tbl[tag] = dma_pool_alloc(hpriv->sg_tbl_pool,
+ GFP_KERNEL, &pp->sg_tbl_dma[tag]);
+ if (!pp->sg_tbl[tag])
+ goto out_port_free_dma_mem;
+ } else {
+ pp->sg_tbl[tag] = pp->sg_tbl[0];
+ pp->sg_tbl_dma[tag] = pp->sg_tbl_dma[0];
+ }
+ }
spin_lock_irqsave(&ap->host->lock, flags);
mv_edma_cfg(pp, hpriv, port_mmio, 0);
-
mv_set_edma_ptrs(port_mmio, hpriv, pp);
spin_unlock_irqrestore(&ap->host->lock, flags);
* we'll be unable to send non-data, PIO, etc due to restricted access
* to shadow regs.
*/
- ap->private_data = pp;
return 0;
+
+out_port_free_dma_mem:
+ mv_port_free_dma_mem(ap);
+ return -ENOMEM;
}
/**
static void mv_port_stop(struct ata_port *ap)
{
mv_stop_dma(ap);
+ mv_port_free_dma_mem(ap);
}
/**
struct mv_sg *mv_sg, *last_sg = NULL;
unsigned int si;
- mv_sg = pp->sg_tbl;
+ mv_sg = pp->sg_tbl[qc->tag];
for_each_sg(qc->sg, sg, qc->n_elem, si) {
dma_addr_t addr = sg_dma_address(sg);
u32 sg_len = sg_dma_len(sg);
in_index = pp->req_idx & MV_MAX_Q_DEPTH_MASK;
pp->crqb[in_index].sg_addr =
- cpu_to_le32(pp->sg_tbl_dma & 0xffffffff);
+ cpu_to_le32(pp->sg_tbl_dma[qc->tag] & 0xffffffff);
pp->crqb[in_index].sg_addr_hi =
- cpu_to_le32((pp->sg_tbl_dma >> 16) >> 16);
+ cpu_to_le32((pp->sg_tbl_dma[qc->tag] >> 16) >> 16);
pp->crqb[in_index].ctrl_flags = cpu_to_le16(flags);
cw = &pp->crqb[in_index].ata_cmd[0];
in_index = pp->req_idx & MV_MAX_Q_DEPTH_MASK;
crqb = (struct mv_crqb_iie *) &pp->crqb[in_index];
- crqb->addr = cpu_to_le32(pp->sg_tbl_dma & 0xffffffff);
- crqb->addr_hi = cpu_to_le32((pp->sg_tbl_dma >> 16) >> 16);
+ crqb->addr = cpu_to_le32(pp->sg_tbl_dma[qc->tag] & 0xffffffff);
+ crqb->addr_hi = cpu_to_le32((pp->sg_tbl_dma[qc->tag] >> 16) >> 16);
crqb->flags = cpu_to_le32(flags);
tf = &qc->tf;
qc = ata_qc_from_tag(ap, tag);
- /* lower 8 bits of status are EDMA_ERR_IRQ_CAUSE_OFS
- * bits (WARNING: might not necessarily be associated
- * with this command), which -should- be clear
- * if all is well
+ /* For non-NCQ mode, the lower 8 bits of status
+ * are from EDMA_ERR_IRQ_CAUSE_OFS,
+ * which should be zero if all went well.
*/
status = le16_to_cpu(pp->crpb[out_index].flags);
- if (unlikely(status & 0xff)) {
+ if ((status & 0xff) && !(pp->pp_flags & MV_PP_FLAG_NCQ_EN)) {
mv_err_intr(ap, qc);
return;
}
scc_s, (MV_HP_FLAG_MSI & hpriv->hp_flags) ? "MSI" : "INTx");
}
+static int mv_create_dma_pools(struct mv_host_priv *hpriv, struct device *dev)
+{
+ hpriv->crqb_pool = dmam_pool_create("crqb_q", dev, MV_CRQB_Q_SZ,
+ MV_CRQB_Q_SZ, 0);
+ if (!hpriv->crqb_pool)
+ return -ENOMEM;
+
+ hpriv->crpb_pool = dmam_pool_create("crpb_q", dev, MV_CRPB_Q_SZ,
+ MV_CRPB_Q_SZ, 0);
+ if (!hpriv->crpb_pool)
+ return -ENOMEM;
+
+ hpriv->sg_tbl_pool = dmam_pool_create("sg_tbl", dev, MV_SG_TBL_SZ,
+ MV_SG_TBL_SZ, 0);
+ if (!hpriv->sg_tbl_pool)
+ return -ENOMEM;
+
+ return 0;
+}
+
/**
* mv_init_one - handle a positive probe of a Marvell host
* @pdev: PCI device found
if (rc)
return rc;
+ rc = mv_create_dma_pools(hpriv, &pdev->dev);
+ if (rc)
+ return rc;
+
/* initialize adapter */
rc = mv_init_host(host, board_idx);
if (rc)