+
+const struct ata_port_operations ata_bmdma_port_ops = {
+ .inherits = &ata_sff_port_ops,
+
+ .error_handler = ata_bmdma_error_handler,
+ .post_internal_cmd = ata_bmdma_post_internal_cmd,
+
+ .qc_prep = ata_bmdma_qc_prep,
+ .qc_issue = ata_bmdma_qc_issue,
+
+ .bmdma_setup = ata_bmdma_setup,
+ .bmdma_start = ata_bmdma_start,
+ .bmdma_stop = ata_bmdma_stop,
+ .bmdma_status = ata_bmdma_status,
+
+ .port_start = ata_bmdma_port_start,
+};
+EXPORT_SYMBOL_GPL(ata_bmdma_port_ops);
+
+const struct ata_port_operations ata_bmdma32_port_ops = {
+ .inherits = &ata_bmdma_port_ops,
+
+ .sff_data_xfer = ata_sff_data_xfer32,
+ .port_start = ata_bmdma_port_start32,
+};
+EXPORT_SYMBOL_GPL(ata_bmdma32_port_ops);
+
+/**
+ * ata_bmdma_fill_sg - Fill PCI IDE PRD table
+ * @qc: Metadata associated with taskfile to be transferred
+ *
+ * Fill PCI IDE PRD (scatter-gather) table with segments
+ * associated with the current disk command.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ *
+ */
+static void ata_bmdma_fill_sg(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ struct ata_bmdma_prd *prd = ap->bmdma_prd;
+ struct scatterlist *sg;
+ unsigned int si, pi;
+
+ pi = 0;
+ for_each_sg(qc->sg, sg, qc->n_elem, si) {
+ u32 addr, offset;
+ u32 sg_len, len;
+
+ /* determine if physical DMA addr spans 64K boundary.
+ * Note h/w doesn't support 64-bit, so we unconditionally
+ * truncate dma_addr_t to u32.
+ */
+ addr = (u32) sg_dma_address(sg);
+ sg_len = sg_dma_len(sg);
+
+ while (sg_len) {
+ offset = addr & 0xffff;
+ len = sg_len;
+ if ((offset + sg_len) > 0x10000)
+ len = 0x10000 - offset;
+
+ prd[pi].addr = cpu_to_le32(addr);
+ prd[pi].flags_len = cpu_to_le32(len & 0xffff);
+ VPRINTK("PRD[%u] = (0x%X, 0x%X)\n", pi, addr, len);
+
+ pi++;
+ sg_len -= len;
+ addr += len;
+ }
+ }
+
+ prd[pi - 1].flags_len |= cpu_to_le32(ATA_PRD_EOT);
+}
+
+/**
+ * ata_bmdma_fill_sg_dumb - Fill PCI IDE PRD table
+ * @qc: Metadata associated with taskfile to be transferred
+ *
+ * Fill PCI IDE PRD (scatter-gather) table with segments
+ * associated with the current disk command. Perform the fill
+ * so that we avoid writing any length 64K records for
+ * controllers that don't follow the spec.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ *
+ */
+static void ata_bmdma_fill_sg_dumb(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ struct ata_bmdma_prd *prd = ap->bmdma_prd;
+ struct scatterlist *sg;
+ unsigned int si, pi;
+
+ pi = 0;
+ for_each_sg(qc->sg, sg, qc->n_elem, si) {
+ u32 addr, offset;
+ u32 sg_len, len, blen;
+
+ /* determine if physical DMA addr spans 64K boundary.
+ * Note h/w doesn't support 64-bit, so we unconditionally
+ * truncate dma_addr_t to u32.
+ */
+ addr = (u32) sg_dma_address(sg);
+ sg_len = sg_dma_len(sg);
+
+ while (sg_len) {
+ offset = addr & 0xffff;
+ len = sg_len;
+ if ((offset + sg_len) > 0x10000)
+ len = 0x10000 - offset;
+
+ blen = len & 0xffff;
+ prd[pi].addr = cpu_to_le32(addr);
+ if (blen == 0) {
+ /* Some PATA chipsets like the CS5530 can't
+ cope with 0x0000 meaning 64K as the spec
+ says */
+ prd[pi].flags_len = cpu_to_le32(0x8000);
+ blen = 0x8000;
+ prd[++pi].addr = cpu_to_le32(addr + 0x8000);
+ }
+ prd[pi].flags_len = cpu_to_le32(blen);
+ VPRINTK("PRD[%u] = (0x%X, 0x%X)\n", pi, addr, len);
+
+ pi++;
+ sg_len -= len;
+ addr += len;
+ }
+ }
+
+ prd[pi - 1].flags_len |= cpu_to_le32(ATA_PRD_EOT);
+}
+
+/**
+ * ata_bmdma_qc_prep - Prepare taskfile for submission
+ * @qc: Metadata associated with taskfile to be prepared
+ *
+ * Prepare ATA taskfile for submission.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+void ata_bmdma_qc_prep(struct ata_queued_cmd *qc)
+{
+ if (!(qc->flags & ATA_QCFLAG_DMAMAP))
+ return;
+
+ ata_bmdma_fill_sg(qc);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_qc_prep);
+
+/**
+ * ata_bmdma_dumb_qc_prep - Prepare taskfile for submission
+ * @qc: Metadata associated with taskfile to be prepared
+ *
+ * Prepare ATA taskfile for submission.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+void ata_bmdma_dumb_qc_prep(struct ata_queued_cmd *qc)
+{
+ if (!(qc->flags & ATA_QCFLAG_DMAMAP))
+ return;
+
+ ata_bmdma_fill_sg_dumb(qc);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_dumb_qc_prep);
+
+/**
+ * ata_bmdma_qc_issue - issue taskfile to a BMDMA controller
+ * @qc: command to issue to device
+ *
+ * This function issues a PIO, NODATA or DMA command to a
+ * SFF/BMDMA controller. PIO and NODATA are handled by
+ * ata_sff_qc_issue().
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ *
+ * RETURNS:
+ * Zero on success, AC_ERR_* mask on failure
+ */
+unsigned int ata_bmdma_qc_issue(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+
+ /* see ata_dma_blacklisted() */
+ BUG_ON((ap->flags & ATA_FLAG_PIO_POLLING) &&
+ qc->tf.protocol == ATAPI_PROT_DMA);
+
+ /* defer PIO handling to sff_qc_issue */
+ if (!ata_is_dma(qc->tf.protocol))
+ return ata_sff_qc_issue(qc);
+
+ /* select the device */
+ ata_dev_select(ap, qc->dev->devno, 1, 0);
+
+ /* start the command */
+ switch (qc->tf.protocol) {
+ case ATA_PROT_DMA:
+ WARN_ON_ONCE(qc->tf.flags & ATA_TFLAG_POLLING);
+
+ ap->ops->sff_tf_load(ap, &qc->tf); /* load tf registers */
+ ap->ops->bmdma_setup(qc); /* set up bmdma */
+ ap->ops->bmdma_start(qc); /* initiate bmdma */
+ ap->hsm_task_state = HSM_ST_LAST;
+ break;
+
+ case ATAPI_PROT_DMA:
+ WARN_ON_ONCE(qc->tf.flags & ATA_TFLAG_POLLING);
+
+ ap->ops->sff_tf_load(ap, &qc->tf); /* load tf registers */
+ ap->ops->bmdma_setup(qc); /* set up bmdma */
+ ap->hsm_task_state = HSM_ST_FIRST;
+
+ /* send cdb by polling if no cdb interrupt */
+ if (!(qc->dev->flags & ATA_DFLAG_CDB_INTR))
+ ata_sff_queue_pio_task(ap, 0);
+ break;
+
+ default:
+ WARN_ON(1);
+ return AC_ERR_SYSTEM;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_qc_issue);
+
+/**
+ * ata_bmdma_error_handler - Stock error handler for BMDMA controller
+ * @ap: port to handle error for
+ *
+ * Stock error handler for BMDMA controller. It can handle both
+ * PATA and SATA controllers. Most BMDMA controllers should be
+ * able to use this EH as-is or with some added handling before
+ * and after.
+ *
+ * LOCKING:
+ * Kernel thread context (may sleep)
+ */
+void ata_bmdma_error_handler(struct ata_port *ap)
+{
+ struct ata_queued_cmd *qc;
+ unsigned long flags;
+ bool thaw = false;
+
+ qc = __ata_qc_from_tag(ap, ap->link.active_tag);
+ if (qc && !(qc->flags & ATA_QCFLAG_FAILED))
+ qc = NULL;
+
+ /* reset PIO HSM and stop DMA engine */
+ spin_lock_irqsave(ap->lock, flags);
+
+ if (qc && ata_is_dma(qc->tf.protocol)) {
+ u8 host_stat;
+
+ host_stat = ap->ops->bmdma_status(ap);
+
+ /* BMDMA controllers indicate host bus error by
+ * setting DMA_ERR bit and timing out. As it wasn't
+ * really a timeout event, adjust error mask and
+ * cancel frozen state.
+ */
+ if (qc->err_mask == AC_ERR_TIMEOUT && (host_stat & ATA_DMA_ERR)) {
+ qc->err_mask = AC_ERR_HOST_BUS;
+ thaw = true;
+ }
+
+ ap->ops->bmdma_stop(qc);
+
+ /* if we're gonna thaw, make sure IRQ is clear */
+ if (thaw) {
+ ap->ops->sff_check_status(ap);
+ ap->ops->sff_irq_clear(ap);
+ }
+ }
+
+ spin_unlock_irqrestore(ap->lock, flags);
+
+ if (thaw)
+ ata_eh_thaw_port(ap);
+
+ ata_sff_error_handler(ap);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_error_handler);
+
+/**
+ * ata_bmdma_post_internal_cmd - Stock post_internal_cmd for BMDMA
+ * @qc: internal command to clean up
+ *
+ * LOCKING:
+ * Kernel thread context (may sleep)
+ */
+void ata_bmdma_post_internal_cmd(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ unsigned long flags;
+
+ if (ata_is_dma(qc->tf.protocol)) {
+ spin_lock_irqsave(ap->lock, flags);
+ ap->ops->bmdma_stop(qc);
+ spin_unlock_irqrestore(ap->lock, flags);
+ }
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_post_internal_cmd);
+
+/**
+ * ata_bmdma_setup - Set up PCI IDE BMDMA transaction
+ * @qc: Info associated with this ATA transaction.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+void ata_bmdma_setup(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ unsigned int rw = (qc->tf.flags & ATA_TFLAG_WRITE);
+ u8 dmactl;
+
+ /* load PRD table addr. */
+ mb(); /* make sure PRD table writes are visible to controller */
+ iowrite32(ap->bmdma_prd_dma, ap->ioaddr.bmdma_addr + ATA_DMA_TABLE_OFS);
+
+ /* specify data direction, triple-check start bit is clear */
+ dmactl = ioread8(ap->ioaddr.bmdma_addr + ATA_DMA_CMD);
+ dmactl &= ~(ATA_DMA_WR | ATA_DMA_START);
+ if (!rw)
+ dmactl |= ATA_DMA_WR;
+ iowrite8(dmactl, ap->ioaddr.bmdma_addr + ATA_DMA_CMD);
+
+ /* issue r/w command */
+ ap->ops->sff_exec_command(ap, &qc->tf);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_setup);
+
+/**
+ * ata_bmdma_start - Start a PCI IDE BMDMA transaction
+ * @qc: Info associated with this ATA transaction.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+void ata_bmdma_start(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ u8 dmactl;
+
+ /* start host DMA transaction */
+ dmactl = ioread8(ap->ioaddr.bmdma_addr + ATA_DMA_CMD);
+ iowrite8(dmactl | ATA_DMA_START, ap->ioaddr.bmdma_addr + ATA_DMA_CMD);
+
+ /* Strictly, one may wish to issue an ioread8() here, to
+ * flush the mmio write. However, control also passes
+ * to the hardware at this point, and it will interrupt
+ * us when we are to resume control. So, in effect,
+ * we don't care when the mmio write flushes.
+ * Further, a read of the DMA status register _immediately_
+ * following the write may not be what certain flaky hardware
+ * is expected, so I think it is best to not add a readb()
+ * without first all the MMIO ATA cards/mobos.
+ * Or maybe I'm just being paranoid.
+ *
+ * FIXME: The posting of this write means I/O starts are
+ * unneccessarily delayed for MMIO
+ */
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_start);
+
+/**
+ * ata_bmdma_stop - Stop PCI IDE BMDMA transfer
+ * @qc: Command we are ending DMA for
+ *
+ * Clears the ATA_DMA_START flag in the dma control register
+ *
+ * May be used as the bmdma_stop() entry in ata_port_operations.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+void ata_bmdma_stop(struct ata_queued_cmd *qc)
+{
+ struct ata_port *ap = qc->ap;
+ void __iomem *mmio = ap->ioaddr.bmdma_addr;
+
+ /* clear start/stop bit */
+ iowrite8(ioread8(mmio + ATA_DMA_CMD) & ~ATA_DMA_START,
+ mmio + ATA_DMA_CMD);
+
+ /* one-PIO-cycle guaranteed wait, per spec, for HDMA1:0 transition */
+ ata_sff_dma_pause(ap);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_stop);
+
+/**
+ * ata_bmdma_status - Read PCI IDE BMDMA status
+ * @ap: Port associated with this ATA transaction.
+ *
+ * Read and return BMDMA status register.
+ *
+ * May be used as the bmdma_status() entry in ata_port_operations.
+ *
+ * LOCKING:
+ * spin_lock_irqsave(host lock)
+ */
+u8 ata_bmdma_status(struct ata_port *ap)
+{
+ return ioread8(ap->ioaddr.bmdma_addr + ATA_DMA_STATUS);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_status);
+
+
+/**
+ * ata_bmdma_port_start - Set port up for bmdma.
+ * @ap: Port to initialize
+ *
+ * Called just after data structures for each port are
+ * initialized. Allocates space for PRD table.
+ *
+ * May be used as the port_start() entry in ata_port_operations.
+ *
+ * LOCKING:
+ * Inherited from caller.
+ */
+int ata_bmdma_port_start(struct ata_port *ap)
+{
+ if (ap->mwdma_mask || ap->udma_mask) {
+ ap->bmdma_prd =
+ dmam_alloc_coherent(ap->host->dev, ATA_PRD_TBL_SZ,
+ &ap->bmdma_prd_dma, GFP_KERNEL);
+ if (!ap->bmdma_prd)
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_port_start);
+
+/**
+ * ata_bmdma_port_start32 - Set port up for dma.
+ * @ap: Port to initialize
+ *
+ * Called just after data structures for each port are
+ * initialized. Enables 32bit PIO and allocates space for PRD
+ * table.
+ *
+ * May be used as the port_start() entry in ata_port_operations for
+ * devices that are capable of 32bit PIO.
+ *
+ * LOCKING:
+ * Inherited from caller.
+ */
+int ata_bmdma_port_start32(struct ata_port *ap)
+{
+ ap->pflags |= ATA_PFLAG_PIO32 | ATA_PFLAG_PIO32CHANGE;
+ return ata_bmdma_port_start(ap);
+}
+EXPORT_SYMBOL_GPL(ata_bmdma_port_start32);
+
+#ifdef CONFIG_PCI
+
+/**
+ * ata_pci_bmdma_clear_simplex - attempt to kick device out of simplex
+ * @pdev: PCI device
+ *
+ * Some PCI ATA devices report simplex mode but in fact can be told to
+ * enter non simplex mode. This implements the necessary logic to
+ * perform the task on such devices. Calling it on other devices will
+ * have -undefined- behaviour.
+ */
+int ata_pci_bmdma_clear_simplex(struct pci_dev *pdev)
+{
+ unsigned long bmdma = pci_resource_start(pdev, 4);
+ u8 simplex;
+
+ if (bmdma == 0)
+ return -ENOENT;
+
+ simplex = inb(bmdma + 0x02);
+ outb(simplex & 0x60, bmdma + 0x02);
+ simplex = inb(bmdma + 0x02);
+ if (simplex & 0x80)
+ return -EOPNOTSUPP;
+ return 0;
+}
+EXPORT_SYMBOL_GPL(ata_pci_bmdma_clear_simplex);
+
+static void ata_bmdma_nodma(struct ata_host *host, const char *reason)
+{
+ int i;
+
+ dev_printk(KERN_ERR, host->dev, "BMDMA: %s, falling back to PIO\n",
+ reason);
+
+ for (i = 0; i < 2; i++) {
+ host->ports[i]->mwdma_mask = 0;
+ host->ports[i]->udma_mask = 0;
+ }
+}
+
+/**
+ * ata_pci_bmdma_init - acquire PCI BMDMA resources and init ATA host
+ * @host: target ATA host
+ *
+ * Acquire PCI BMDMA resources and initialize @host accordingly.
+ *
+ * LOCKING:
+ * Inherited from calling layer (may sleep).
+ */
+void ata_pci_bmdma_init(struct ata_host *host)
+{
+ struct device *gdev = host->dev;
+ struct pci_dev *pdev = to_pci_dev(gdev);
+ int i, rc;
+
+ /* No BAR4 allocation: No DMA */
+ if (pci_resource_start(pdev, 4) == 0) {
+ ata_bmdma_nodma(host, "BAR4 is zero");
+ return;
+ }
+
+ /*
+ * Some controllers require BMDMA region to be initialized
+ * even if DMA is not in use to clear IRQ status via
+ * ->sff_irq_clear method. Try to initialize bmdma_addr
+ * regardless of dma masks.
+ */
+ rc = pci_set_dma_mask(pdev, ATA_DMA_MASK);
+ if (rc)
+ ata_bmdma_nodma(host, "failed to set dma mask");
+ if (!rc) {
+ rc = pci_set_consistent_dma_mask(pdev, ATA_DMA_MASK);
+ if (rc)
+ ata_bmdma_nodma(host,
+ "failed to set consistent dma mask");
+ }
+
+ /* request and iomap DMA region */
+ rc = pcim_iomap_regions(pdev, 1 << 4, dev_driver_string(gdev));
+ if (rc) {
+ ata_bmdma_nodma(host, "failed to request/iomap BAR4");
+ return;
+ }
+ host->iomap = pcim_iomap_table(pdev);
+
+ for (i = 0; i < 2; i++) {
+ struct ata_port *ap = host->ports[i];
+ void __iomem *bmdma = host->iomap[4] + 8 * i;
+
+ if (ata_port_is_dummy(ap))
+ continue;
+
+ ap->ioaddr.bmdma_addr = bmdma;
+ if ((!(ap->flags & ATA_FLAG_IGN_SIMPLEX)) &&
+ (ioread8(bmdma + 2) & 0x80))
+ host->flags |= ATA_HOST_SIMPLEX;
+
+ ata_port_desc(ap, "bmdma 0x%llx",
+ (unsigned long long)pci_resource_start(pdev, 4) + 8 * i);
+ }
+}
+EXPORT_SYMBOL_GPL(ata_pci_bmdma_init);
+
+#endif /* CONFIG_PCI */
+
+/**
+ * ata_sff_port_init - Initialize SFF/BMDMA ATA port
+ * @ap: Port to initialize
+ *
+ * Called on port allocation to initialize SFF/BMDMA specific
+ * fields.
+ *
+ * LOCKING:
+ * None.
+ */
+void ata_sff_port_init(struct ata_port *ap)
+{
+ INIT_DELAYED_WORK(&ap->sff_pio_task, ata_sff_pio_task);
+ ap->ctl = ATA_DEVCTL_OBS;
+ ap->last_ctl = 0xFF;
+}
+
+int __init ata_sff_init(void)
+{
+ /*
+ * FIXME: In UP case, there is only one workqueue thread and if you
+ * have more than one PIO device, latency is bloody awful, with
+ * occasional multi-second "hiccups" as one PIO device waits for
+ * another. It's an ugly wart that users DO occasionally complain
+ * about; luckily most users have at most one PIO polled device.
+ */
+ ata_sff_wq = create_workqueue("ata_sff");
+ if (!ata_sff_wq)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void __exit ata_sff_exit(void)
+{
+ destroy_workqueue(ata_sff_wq);
+}