#include <linux/dmaengine.h>
#include <linux/delay.h>
#include <linux/dma-mapping.h>
+#include <linux/workqueue.h>
+#include <linux/i7300_idle.h>
#include "ioatdma.h"
#include "ioatdma_registers.h"
#include "ioatdma_hw.h"
-#define INITIAL_IOAT_DESC_COUNT 128
-
#define to_ioat_chan(chan) container_of(chan, struct ioat_dma_chan, common)
#define to_ioatdma_device(dev) container_of(dev, struct ioatdma_device, common)
#define to_ioat_desc(lh) container_of(lh, struct ioat_desc_sw, node)
#define tx_to_ioat_desc(tx) container_of(tx, struct ioat_desc_sw, async_tx)
+#define chan_num(ch) ((int)((ch)->reg_base - (ch)->device->reg_base) / 0x80)
+static int ioat_pending_level = 4;
+module_param(ioat_pending_level, int, 0644);
+MODULE_PARM_DESC(ioat_pending_level,
+ "high-water mark for pushing ioat descriptors (default: 4)");
+
+#define RESET_DELAY msecs_to_jiffies(100)
+#define WATCHDOG_DELAY round_jiffies(msecs_to_jiffies(2000))
+static void ioat_dma_chan_reset_part2(struct work_struct *work);
+static void ioat_dma_chan_watchdog(struct work_struct *work);
+
+/*
+ * workaround for IOAT ver.3.0 null descriptor issue
+ * (channel returns error when size is 0)
+ */
+#define NULL_DESC_BUFFER_SIZE 1
+
/* internal functions */
static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat_chan);
static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan);
+static struct ioat_desc_sw *
+ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan);
+static struct ioat_desc_sw *
+ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan);
+
+static inline struct ioat_dma_chan *ioat_lookup_chan_by_index(
+ struct ioatdma_device *device,
+ int index)
+{
+ return device->idx[index];
+}
+
+/**
+ * ioat_dma_do_interrupt - handler used for single vector interrupt mode
+ * @irq: interrupt id
+ * @data: interrupt data
+ */
+static irqreturn_t ioat_dma_do_interrupt(int irq, void *data)
+{
+ struct ioatdma_device *instance = data;
+ struct ioat_dma_chan *ioat_chan;
+ unsigned long attnstatus;
+ int bit;
+ u8 intrctrl;
+
+ intrctrl = readb(instance->reg_base + IOAT_INTRCTRL_OFFSET);
+
+ if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN))
+ return IRQ_NONE;
+
+ if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) {
+ writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
+ return IRQ_NONE;
+ }
+
+ attnstatus = readl(instance->reg_base + IOAT_ATTNSTATUS_OFFSET);
+ for_each_bit(bit, &attnstatus, BITS_PER_LONG) {
+ ioat_chan = ioat_lookup_chan_by_index(instance, bit);
+ tasklet_schedule(&ioat_chan->cleanup_task);
+ }
+
+ writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
+ return IRQ_HANDLED;
+}
+
+/**
+ * ioat_dma_do_interrupt_msix - handler used for vector-per-channel interrupt mode
+ * @irq: interrupt id
+ * @data: interrupt data
+ */
+static irqreturn_t ioat_dma_do_interrupt_msix(int irq, void *data)
+{
+ struct ioat_dma_chan *ioat_chan = data;
+
+ tasklet_schedule(&ioat_chan->cleanup_task);
+
+ return IRQ_HANDLED;
+}
+
+static void ioat_dma_cleanup_tasklet(unsigned long data);
+
+/**
+ * ioat_dma_enumerate_channels - find and initialize the device's channels
+ * @device: the device to be enumerated
+ */
static int ioat_dma_enumerate_channels(struct ioatdma_device *device)
{
u8 xfercap_scale;
int i;
struct ioat_dma_chan *ioat_chan;
+ /*
+ * IOAT ver.3 workarounds
+ */
+ if (device->version == IOAT_VER_3_0) {
+ u32 chan_err_mask;
+ u16 dev_id;
+ u32 dmauncerrsts;
+
+ /*
+ * Write CHANERRMSK_INT with 3E07h to mask out the errors
+ * that can cause stability issues for IOAT ver.3
+ */
+ chan_err_mask = 0x3E07;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_CHANERRMASK_INT_OFFSET,
+ chan_err_mask);
+
+ /*
+ * Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit
+ * (workaround for spurious config parity error after restart)
+ */
+ pci_read_config_word(device->pdev,
+ IOAT_PCI_DEVICE_ID_OFFSET,
+ &dev_id);
+ if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) {
+ dmauncerrsts = 0x10;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_DMAUNCERRSTS_OFFSET,
+ dmauncerrsts);
+ }
+ }
+
device->common.chancnt = readb(device->reg_base + IOAT_CHANCNT_OFFSET);
xfercap_scale = readb(device->reg_base + IOAT_XFERCAP_OFFSET);
xfercap = (xfercap_scale == 0 ? -1 : (1UL << xfercap_scale));
+#ifdef CONFIG_I7300_IDLE_IOAT_CHANNEL
+ if (i7300_idle_platform_probe(NULL, NULL) == 0) {
+ device->common.chancnt--;
+ }
+#endif
for (i = 0; i < device->common.chancnt; i++) {
ioat_chan = kzalloc(sizeof(*ioat_chan), GFP_KERNEL);
if (!ioat_chan) {
ioat_chan->device = device;
ioat_chan->reg_base = device->reg_base + (0x80 * (i + 1));
ioat_chan->xfercap = xfercap;
+ ioat_chan->desccount = 0;
+ INIT_DELAYED_WORK(&ioat_chan->work, ioat_dma_chan_reset_part2);
+ if (ioat_chan->device->version != IOAT_VER_1_2) {
+ writel(IOAT_DCACTRL_CMPL_WRITE_ENABLE
+ | IOAT_DMA_DCA_ANY_CPU,
+ ioat_chan->reg_base + IOAT_DCACTRL_OFFSET);
+ }
spin_lock_init(&ioat_chan->cleanup_lock);
spin_lock_init(&ioat_chan->desc_lock);
INIT_LIST_HEAD(&ioat_chan->free_desc);
ioat_chan->common.device = &device->common;
list_add_tail(&ioat_chan->common.device_node,
&device->common.channels);
+ device->idx[i] = ioat_chan;
+ tasklet_init(&ioat_chan->cleanup_task,
+ ioat_dma_cleanup_tasklet,
+ (unsigned long) ioat_chan);
+ tasklet_disable(&ioat_chan->cleanup_task);
}
return device->common.chancnt;
}
-static void ioat_set_src(dma_addr_t addr,
- struct dma_async_tx_descriptor *tx,
- int index)
+/**
+ * ioat_dma_memcpy_issue_pending - push potentially unrecognized appended
+ * descriptors to hw
+ * @chan: DMA channel handle
+ */
+static inline void __ioat1_dma_memcpy_issue_pending(
+ struct ioat_dma_chan *ioat_chan)
{
- struct ioat_desc_sw *iter, *desc = tx_to_ioat_desc(tx);
- struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
+ ioat_chan->pending = 0;
+ writeb(IOAT_CHANCMD_APPEND, ioat_chan->reg_base + IOAT1_CHANCMD_OFFSET);
+}
- pci_unmap_addr_set(desc, src, addr);
+static void ioat1_dma_memcpy_issue_pending(struct dma_chan *chan)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
- list_for_each_entry(iter, &desc->async_tx.tx_list, node) {
- iter->hw->src_addr = addr;
- addr += ioat_chan->xfercap;
+ if (ioat_chan->pending > 0) {
+ spin_lock_bh(&ioat_chan->desc_lock);
+ __ioat1_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
}
+}
+static inline void __ioat2_dma_memcpy_issue_pending(
+ struct ioat_dma_chan *ioat_chan)
+{
+ ioat_chan->pending = 0;
+ writew(ioat_chan->dmacount,
+ ioat_chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET);
}
-static void ioat_set_dest(dma_addr_t addr,
- struct dma_async_tx_descriptor *tx,
- int index)
+static void ioat2_dma_memcpy_issue_pending(struct dma_chan *chan)
{
- struct ioat_desc_sw *iter, *desc = tx_to_ioat_desc(tx);
- struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+
+ if (ioat_chan->pending > 0) {
+ spin_lock_bh(&ioat_chan->desc_lock);
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ }
+}
+
+
+/**
+ * ioat_dma_chan_reset_part2 - reinit the channel after a reset
+ */
+static void ioat_dma_chan_reset_part2(struct work_struct *work)
+{
+ struct ioat_dma_chan *ioat_chan =
+ container_of(work, struct ioat_dma_chan, work.work);
+ struct ioat_desc_sw *desc;
+
+ spin_lock_bh(&ioat_chan->cleanup_lock);
+ spin_lock_bh(&ioat_chan->desc_lock);
+
+ ioat_chan->completion_virt->low = 0;
+ ioat_chan->completion_virt->high = 0;
+ ioat_chan->pending = 0;
+
+ /*
+ * count the descriptors waiting, and be sure to do it
+ * right for both the CB1 line and the CB2 ring
+ */
+ ioat_chan->dmacount = 0;
+ if (ioat_chan->used_desc.prev) {
+ desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ do {
+ ioat_chan->dmacount++;
+ desc = to_ioat_desc(desc->node.next);
+ } while (&desc->node != ioat_chan->used_desc.next);
+ }
+
+ /*
+ * write the new starting descriptor address
+ * this puts channel engine into ARMED state
+ */
+ desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH);
+
+ writeb(IOAT_CHANCMD_START, ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ break;
+ case IOAT_VER_2_0:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH);
+
+ /* tell the engine to go with what's left to be done */
+ writew(ioat_chan->dmacount,
+ ioat_chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET);
+
+ break;
+ }
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d reset - %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+}
+
+/**
+ * ioat_dma_reset_channel - restart a channel
+ * @ioat_chan: IOAT DMA channel handle
+ */
+static void ioat_dma_reset_channel(struct ioat_dma_chan *ioat_chan)
+{
+ u32 chansts, chanerr;
+
+ if (!ioat_chan->used_desc.prev)
+ return;
+
+ chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ chansts = (ioat_chan->completion_virt->low
+ & IOAT_CHANSTS_DMA_TRANSFER_STATUS);
+ if (chanerr) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d, CHANSTS = 0x%08x CHANERR = 0x%04x, clearing\n",
+ chan_num(ioat_chan), chansts, chanerr);
+ writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
+ }
+
+ /*
+ * whack it upside the head with a reset
+ * and wait for things to settle out.
+ * force the pending count to a really big negative
+ * to make sure no one forces an issue_pending
+ * while we're waiting.
+ */
+
+ spin_lock_bh(&ioat_chan->desc_lock);
+ ioat_chan->pending = INT_MIN;
+ writeb(IOAT_CHANCMD_RESET,
+ ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ spin_unlock_bh(&ioat_chan->desc_lock);
+
+ /* schedule the 2nd half instead of sleeping a long time */
+ schedule_delayed_work(&ioat_chan->work, RESET_DELAY);
+}
+
+/**
+ * ioat_dma_chan_watchdog - watch for stuck channels
+ */
+static void ioat_dma_chan_watchdog(struct work_struct *work)
+{
+ struct ioatdma_device *device =
+ container_of(work, struct ioatdma_device, work.work);
+ struct ioat_dma_chan *ioat_chan;
+ int i;
+
+ union {
+ u64 full;
+ struct {
+ u32 low;
+ u32 high;
+ };
+ } completion_hw;
+ unsigned long compl_desc_addr_hw;
+
+ for (i = 0; i < device->common.chancnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+
+ if (ioat_chan->device->version == IOAT_VER_1_2
+ /* have we started processing anything yet */
+ && ioat_chan->last_completion
+ /* have we completed any since last watchdog cycle? */
+ && (ioat_chan->last_completion ==
+ ioat_chan->watchdog_completion)
+ /* has TCP stuck on one cookie since last watchdog? */
+ && (ioat_chan->watchdog_tcp_cookie ==
+ ioat_chan->watchdog_last_tcp_cookie)
+ && (ioat_chan->watchdog_tcp_cookie !=
+ ioat_chan->completed_cookie)
+ /* is there something in the chain to be processed? */
+ /* CB1 chain always has at least the last one processed */
+ && (ioat_chan->used_desc.prev != ioat_chan->used_desc.next)
+ && ioat_chan->pending == 0) {
+
+ /*
+ * check CHANSTS register for completed
+ * descriptor address.
+ * if it is different than completion writeback,
+ * it is not zero
+ * and it has changed since the last watchdog
+ * we can assume that channel
+ * is still working correctly
+ * and the problem is in completion writeback.
+ * update completion writeback
+ * with actual CHANSTS value
+ * else
+ * try resetting the channel
+ */
+
+ completion_hw.low = readl(ioat_chan->reg_base +
+ IOAT_CHANSTS_OFFSET_LOW(ioat_chan->device->version));
+ completion_hw.high = readl(ioat_chan->reg_base +
+ IOAT_CHANSTS_OFFSET_HIGH(ioat_chan->device->version));
+#if (BITS_PER_LONG == 64)
+ compl_desc_addr_hw =
+ completion_hw.full
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+#else
+ compl_desc_addr_hw =
+ completion_hw.low & IOAT_LOW_COMPLETION_MASK;
+#endif
- pci_unmap_addr_set(desc, dst, addr);
+ if ((compl_desc_addr_hw != 0)
+ && (compl_desc_addr_hw != ioat_chan->watchdog_completion)
+ && (compl_desc_addr_hw != ioat_chan->last_compl_desc_addr_hw)) {
+ ioat_chan->last_compl_desc_addr_hw = compl_desc_addr_hw;
+ ioat_chan->completion_virt->low = completion_hw.low;
+ ioat_chan->completion_virt->high = completion_hw.high;
+ } else {
+ ioat_dma_reset_channel(ioat_chan);
+ ioat_chan->watchdog_completion = 0;
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ }
- list_for_each_entry(iter, &desc->async_tx.tx_list, node) {
- iter->hw->dst_addr = addr;
- addr += ioat_chan->xfercap;
+ /*
+ * for version 2.0 if there are descriptors yet to be processed
+ * and the last completed hasn't changed since the last watchdog
+ * if they haven't hit the pending level
+ * issue the pending to push them through
+ * else
+ * try resetting the channel
+ */
+ } else if (ioat_chan->device->version == IOAT_VER_2_0
+ && ioat_chan->used_desc.prev
+ && ioat_chan->last_completion
+ && ioat_chan->last_completion == ioat_chan->watchdog_completion) {
+
+ if (ioat_chan->pending < ioat_pending_level)
+ ioat2_dma_memcpy_issue_pending(&ioat_chan->common);
+ else {
+ ioat_dma_reset_channel(ioat_chan);
+ ioat_chan->watchdog_completion = 0;
+ }
+ } else {
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ ioat_chan->watchdog_completion
+ = ioat_chan->last_completion;
+ }
+
+ ioat_chan->watchdog_last_tcp_cookie =
+ ioat_chan->watchdog_tcp_cookie;
}
+
+ schedule_delayed_work(&device->work, WATCHDOG_DELAY);
}
-static dma_cookie_t ioat_tx_submit(struct dma_async_tx_descriptor *tx)
+static dma_cookie_t ioat1_tx_submit(struct dma_async_tx_descriptor *tx)
{
struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
- struct ioat_desc_sw *desc = tx_to_ioat_desc(tx);
- int append = 0;
+ struct ioat_desc_sw *first = tx_to_ioat_desc(tx);
+ struct ioat_desc_sw *prev, *new;
+ struct ioat_dma_descriptor *hw;
dma_cookie_t cookie;
- struct ioat_desc_sw *group_start;
+ LIST_HEAD(new_chain);
+ u32 copy;
+ size_t len;
+ dma_addr_t src, dst;
+ unsigned long orig_flags;
+ unsigned int desc_count = 0;
+
+ /* src and dest and len are stored in the initial descriptor */
+ len = first->len;
+ src = first->src;
+ dst = first->dst;
+ orig_flags = first->async_tx.flags;
+ new = first;
- group_start = list_entry(desc->async_tx.tx_list.next,
- struct ioat_desc_sw, node);
spin_lock_bh(&ioat_chan->desc_lock);
+ prev = to_ioat_desc(ioat_chan->used_desc.prev);
+ prefetch(prev->hw);
+ do {
+ copy = min_t(size_t, len, ioat_chan->xfercap);
+
+ async_tx_ack(&new->async_tx);
+
+ hw = new->hw;
+ hw->size = copy;
+ hw->ctl = 0;
+ hw->src_addr = src;
+ hw->dst_addr = dst;
+ hw->next = 0;
+
+ /* chain together the physical address list for the HW */
+ wmb();
+ prev->hw->next = (u64) new->async_tx.phys;
+
+ len -= copy;
+ dst += copy;
+ src += copy;
+
+ list_add_tail(&new->node, &new_chain);
+ desc_count++;
+ prev = new;
+ } while (len && (new = ioat1_dma_get_next_descriptor(ioat_chan)));
+
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "tx submit failed\n");
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ return -ENOMEM;
+ }
+
+ hw->ctl = IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ if (first->async_tx.callback) {
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_INT_GN;
+ if (first != new) {
+ /* move callback into to last desc */
+ new->async_tx.callback = first->async_tx.callback;
+ new->async_tx.callback_param
+ = first->async_tx.callback_param;
+ first->async_tx.callback = NULL;
+ first->async_tx.callback_param = NULL;
+ }
+ }
+
+ new->tx_cnt = desc_count;
+ new->async_tx.flags = orig_flags; /* client is in control of this ack */
+
+ /* store the original values for use in later cleanup */
+ if (new != first) {
+ new->src = first->src;
+ new->dst = first->dst;
+ new->len = first->len;
+ }
+
/* cookie incr and addition to used_list must be atomic */
cookie = ioat_chan->common.cookie;
cookie++;
if (cookie < 0)
cookie = 1;
- ioat_chan->common.cookie = desc->async_tx.cookie = cookie;
+ ioat_chan->common.cookie = new->async_tx.cookie = cookie;
/* write address into NextDescriptor field of last desc in chain */
to_ioat_desc(ioat_chan->used_desc.prev)->hw->next =
- group_start->async_tx.phys;
- list_splice_init(&desc->async_tx.tx_list, ioat_chan->used_desc.prev);
+ first->async_tx.phys;
+ list_splice_tail(&new_chain, &ioat_chan->used_desc);
- ioat_chan->pending += desc->tx_cnt;
- if (ioat_chan->pending >= 4) {
- append = 1;
- ioat_chan->pending = 0;
- }
+ ioat_chan->dmacount += desc_count;
+ ioat_chan->pending += desc_count;
+ if (ioat_chan->pending >= ioat_pending_level)
+ __ioat1_dma_memcpy_issue_pending(ioat_chan);
spin_unlock_bh(&ioat_chan->desc_lock);
- if (append)
- writeb(IOAT_CHANCMD_APPEND,
- ioat_chan->reg_base + IOAT_CHANCMD_OFFSET);
+ return cookie;
+}
+
+static dma_cookie_t ioat2_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(tx->chan);
+ struct ioat_desc_sw *first = tx_to_ioat_desc(tx);
+ struct ioat_desc_sw *new;
+ struct ioat_dma_descriptor *hw;
+ dma_cookie_t cookie;
+ u32 copy;
+ size_t len;
+ dma_addr_t src, dst;
+ unsigned long orig_flags;
+ unsigned int desc_count = 0;
+
+ /* src and dest and len are stored in the initial descriptor */
+ len = first->len;
+ src = first->src;
+ dst = first->dst;
+ orig_flags = first->async_tx.flags;
+ new = first;
+
+ /*
+ * ioat_chan->desc_lock is still in force in version 2 path
+ * it gets unlocked at end of this function
+ */
+ do {
+ copy = min_t(size_t, len, ioat_chan->xfercap);
+
+ async_tx_ack(&new->async_tx);
+
+ hw = new->hw;
+ hw->size = copy;
+ hw->ctl = 0;
+ hw->src_addr = src;
+ hw->dst_addr = dst;
+
+ len -= copy;
+ dst += copy;
+ src += copy;
+ desc_count++;
+ } while (len && (new = ioat2_dma_get_next_descriptor(ioat_chan)));
+
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "tx submit failed\n");
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ return -ENOMEM;
+ }
+
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ if (first->async_tx.callback) {
+ hw->ctl |= IOAT_DMA_DESCRIPTOR_CTL_INT_GN;
+ if (first != new) {
+ /* move callback into to last desc */
+ new->async_tx.callback = first->async_tx.callback;
+ new->async_tx.callback_param
+ = first->async_tx.callback_param;
+ first->async_tx.callback = NULL;
+ first->async_tx.callback_param = NULL;
+ }
+ }
+
+ new->tx_cnt = desc_count;
+ new->async_tx.flags = orig_flags; /* client is in control of this ack */
+
+ /* store the original values for use in later cleanup */
+ if (new != first) {
+ new->src = first->src;
+ new->dst = first->dst;
+ new->len = first->len;
+ }
+
+ /* cookie incr and addition to used_list must be atomic */
+ cookie = ioat_chan->common.cookie;
+ cookie++;
+ if (cookie < 0)
+ cookie = 1;
+ ioat_chan->common.cookie = new->async_tx.cookie = cookie;
+
+ ioat_chan->dmacount += desc_count;
+ ioat_chan->pending += desc_count;
+ if (ioat_chan->pending >= ioat_pending_level)
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ spin_unlock_bh(&ioat_chan->desc_lock);
return cookie;
}
+/**
+ * ioat_dma_alloc_descriptor - allocate and return a sw and hw descriptor pair
+ * @ioat_chan: the channel supplying the memory pool for the descriptors
+ * @flags: allocation flags
+ */
static struct ioat_desc_sw *ioat_dma_alloc_descriptor(
struct ioat_dma_chan *ioat_chan,
gfp_t flags)
memset(desc, 0, sizeof(*desc));
dma_async_tx_descriptor_init(&desc_sw->async_tx, &ioat_chan->common);
- desc_sw->async_tx.tx_set_src = ioat_set_src;
- desc_sw->async_tx.tx_set_dest = ioat_set_dest;
- desc_sw->async_tx.tx_submit = ioat_tx_submit;
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ desc_sw->async_tx.tx_submit = ioat1_tx_submit;
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ desc_sw->async_tx.tx_submit = ioat2_tx_submit;
+ break;
+ }
INIT_LIST_HEAD(&desc_sw->async_tx.tx_list);
+
desc_sw->hw = desc;
desc_sw->async_tx.phys = phys;
return desc_sw;
}
-/* returns the actual number of allocated descriptors */
+static int ioat_initial_desc_count = 256;
+module_param(ioat_initial_desc_count, int, 0644);
+MODULE_PARM_DESC(ioat_initial_desc_count,
+ "initial descriptors per channel (default: 256)");
+
+/**
+ * ioat2_dma_massage_chan_desc - link the descriptors into a circle
+ * @ioat_chan: the channel to be massaged
+ */
+static void ioat2_dma_massage_chan_desc(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *desc, *_desc;
+
+ /* setup used_desc */
+ ioat_chan->used_desc.next = ioat_chan->free_desc.next;
+ ioat_chan->used_desc.prev = NULL;
+
+ /* pull free_desc out of the circle so that every node is a hw
+ * descriptor, but leave it pointing to the list
+ */
+ ioat_chan->free_desc.prev->next = ioat_chan->free_desc.next;
+ ioat_chan->free_desc.next->prev = ioat_chan->free_desc.prev;
+
+ /* circle link the hw descriptors */
+ desc = to_ioat_desc(ioat_chan->free_desc.next);
+ desc->hw->next = to_ioat_desc(desc->node.next)->async_tx.phys;
+ list_for_each_entry_safe(desc, _desc, ioat_chan->free_desc.next, node) {
+ desc->hw->next = to_ioat_desc(desc->node.next)->async_tx.phys;
+ }
+}
+
+/**
+ * ioat_dma_alloc_chan_resources - returns the number of allocated descriptors
+ * @chan: the channel to be filled out
+ */
static int ioat_dma_alloc_chan_resources(struct dma_chan *chan)
{
struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
- struct ioat_desc_sw *desc = NULL;
+ struct ioat_desc_sw *desc;
u16 chanctrl;
u32 chanerr;
int i;
/* have we already been set up? */
if (!list_empty(&ioat_chan->free_desc))
- return INITIAL_IOAT_DESC_COUNT;
+ return ioat_chan->desccount;
/* Setup register to interrupt and write completion status on error */
chanctrl = IOAT_CHANCTRL_ERR_INT_EN |
chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
if (chanerr) {
dev_err(&ioat_chan->device->pdev->dev,
- "ioatdma: CHANERR = %x, clearing\n", chanerr);
+ "CHANERR = %x, clearing\n", chanerr);
writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
}
/* Allocate descriptors */
- for (i = 0; i < INITIAL_IOAT_DESC_COUNT; i++) {
+ for (i = 0; i < ioat_initial_desc_count; i++) {
desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL);
if (!desc) {
dev_err(&ioat_chan->device->pdev->dev,
- "ioatdma: Only %d initial descriptors\n", i);
+ "Only %d initial descriptors\n", i);
break;
}
list_add_tail(&desc->node, &tmp_list);
}
spin_lock_bh(&ioat_chan->desc_lock);
+ ioat_chan->desccount = i;
list_splice(&tmp_list, &ioat_chan->free_desc);
+ if (ioat_chan->device->version != IOAT_VER_1_2)
+ ioat2_dma_massage_chan_desc(ioat_chan);
spin_unlock_bh(&ioat_chan->desc_lock);
/* allocate a completion writeback area */
writel(((u64) ioat_chan->completion_addr) >> 32,
ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH);
- ioat_dma_start_null_desc(ioat_chan);
- return i;
+ tasklet_enable(&ioat_chan->cleanup_task);
+ ioat_dma_start_null_desc(ioat_chan); /* give chain to dma device */
+ return ioat_chan->desccount;
}
+/**
+ * ioat_dma_free_chan_resources - release all the descriptors
+ * @chan: the channel to be cleaned
+ */
static void ioat_dma_free_chan_resources(struct dma_chan *chan)
{
struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
struct ioat_desc_sw *desc, *_desc;
int in_use_descs = 0;
+ /* Before freeing channel resources first check
+ * if they have been previously allocated for this channel.
+ */
+ if (ioat_chan->desccount == 0)
+ return;
+
+ tasklet_disable(&ioat_chan->cleanup_task);
ioat_dma_memcpy_cleanup(ioat_chan);
- writeb(IOAT_CHANCMD_RESET, ioat_chan->reg_base + IOAT_CHANCMD_OFFSET);
+ /* Delay 100ms after reset to allow internal DMA logic to quiesce
+ * before removing DMA descriptor resources.
+ */
+ writeb(IOAT_CHANCMD_RESET,
+ ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ mdelay(100);
spin_lock_bh(&ioat_chan->desc_lock);
- list_for_each_entry_safe(desc, _desc, &ioat_chan->used_desc, node) {
- in_use_descs++;
- list_del(&desc->node);
- pci_pool_free(ioatdma_device->dma_pool, desc->hw,
- desc->async_tx.phys);
- kfree(desc);
- }
- list_for_each_entry_safe(desc, _desc, &ioat_chan->free_desc, node) {
- list_del(&desc->node);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->used_desc, node) {
+ in_use_descs++;
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->free_desc, node) {
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ list_for_each_entry_safe(desc, _desc,
+ ioat_chan->free_desc.next, node) {
+ list_del(&desc->node);
+ pci_pool_free(ioatdma_device->dma_pool, desc->hw,
+ desc->async_tx.phys);
+ kfree(desc);
+ }
+ desc = to_ioat_desc(ioat_chan->free_desc.next);
pci_pool_free(ioatdma_device->dma_pool, desc->hw,
desc->async_tx.phys);
kfree(desc);
+ INIT_LIST_HEAD(&ioat_chan->free_desc);
+ INIT_LIST_HEAD(&ioat_chan->used_desc);
+ break;
}
spin_unlock_bh(&ioat_chan->desc_lock);
/* one is ok since we left it on there on purpose */
if (in_use_descs > 1)
dev_err(&ioat_chan->device->pdev->dev,
- "ioatdma: Freeing %d in use descriptors!\n",
+ "Freeing %d in use descriptors!\n",
in_use_descs - 1);
ioat_chan->last_completion = ioat_chan->completion_addr = 0;
+ ioat_chan->pending = 0;
+ ioat_chan->dmacount = 0;
+ ioat_chan->desccount = 0;
+ ioat_chan->watchdog_completion = 0;
+ ioat_chan->last_compl_desc_addr_hw = 0;
+ ioat_chan->watchdog_tcp_cookie =
+ ioat_chan->watchdog_last_tcp_cookie = 0;
}
-static struct dma_async_tx_descriptor *ioat_dma_prep_memcpy(
- struct dma_chan *chan,
- size_t len,
- int int_en)
+/**
+ * ioat_dma_get_next_descriptor - return the next available descriptor
+ * @ioat_chan: IOAT DMA channel handle
+ *
+ * Gets the next descriptor from the chain, and must be called with the
+ * channel's desc_lock held. Allocates more descriptors if the channel
+ * has run out.
+ */
+static struct ioat_desc_sw *
+ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan)
{
- struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
- struct ioat_desc_sw *first, *prev, *new;
- LIST_HEAD(new_chain);
- u32 copy;
- size_t orig_len;
- int desc_count = 0;
-
- if (!len)
- return NULL;
+ struct ioat_desc_sw *new;
- orig_len = len;
+ if (!list_empty(&ioat_chan->free_desc)) {
+ new = to_ioat_desc(ioat_chan->free_desc.next);
+ list_del(&new->node);
+ } else {
+ /* try to get another desc */
+ new = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
+ if (!new) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "alloc failed\n");
+ return NULL;
+ }
+ }
- first = NULL;
- prev = NULL;
+ prefetch(new->hw);
+ return new;
+}
- spin_lock_bh(&ioat_chan->desc_lock);
- while (len) {
- if (!list_empty(&ioat_chan->free_desc)) {
- new = to_ioat_desc(ioat_chan->free_desc.next);
- list_del(&new->node);
- } else {
- /* try to get another desc */
- new = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
- /* will this ever happen? */
- /* TODO add upper limit on these */
- BUG_ON(!new);
+static struct ioat_desc_sw *
+ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat_chan)
+{
+ struct ioat_desc_sw *new;
+
+ /*
+ * used.prev points to where to start processing
+ * used.next points to next free descriptor
+ * if used.prev == NULL, there are none waiting to be processed
+ * if used.next == used.prev.prev, there is only one free descriptor,
+ * and we need to use it to as a noop descriptor before
+ * linking in a new set of descriptors, since the device
+ * has probably already read the pointer to it
+ */
+ if (ioat_chan->used_desc.prev &&
+ ioat_chan->used_desc.next == ioat_chan->used_desc.prev->prev) {
+
+ struct ioat_desc_sw *desc;
+ struct ioat_desc_sw *noop_desc;
+ int i;
+
+ /* set up the noop descriptor */
+ noop_desc = to_ioat_desc(ioat_chan->used_desc.next);
+ /* set size to non-zero value (channel returns error when size is 0) */
+ noop_desc->hw->size = NULL_DESC_BUFFER_SIZE;
+ noop_desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL;
+ noop_desc->hw->src_addr = 0;
+ noop_desc->hw->dst_addr = 0;
+
+ ioat_chan->used_desc.next = ioat_chan->used_desc.next->next;
+ ioat_chan->pending++;
+ ioat_chan->dmacount++;
+
+ /* try to get a few more descriptors */
+ for (i = 16; i; i--) {
+ desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_ATOMIC);
+ if (!desc) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "alloc failed\n");
+ break;
+ }
+ list_add_tail(&desc->node, ioat_chan->used_desc.next);
+
+ desc->hw->next
+ = to_ioat_desc(desc->node.next)->async_tx.phys;
+ to_ioat_desc(desc->node.prev)->hw->next
+ = desc->async_tx.phys;
+ ioat_chan->desccount++;
}
- copy = min((u32) len, ioat_chan->xfercap);
+ ioat_chan->used_desc.next = noop_desc->node.next;
+ }
+ new = to_ioat_desc(ioat_chan->used_desc.next);
+ prefetch(new);
+ ioat_chan->used_desc.next = new->node.next;
- new->hw->size = copy;
- new->hw->ctl = 0;
- new->async_tx.cookie = 0;
- new->async_tx.ack = 1;
+ if (ioat_chan->used_desc.prev == NULL)
+ ioat_chan->used_desc.prev = &new->node;
- /* chain together the physical address list for the HW */
- if (!first)
- first = new;
- else
- prev->hw->next = (u64) new->async_tx.phys;
+ prefetch(new->hw);
+ return new;
+}
- prev = new;
- len -= copy;
- list_add_tail(&new->node, &new_chain);
- desc_count++;
- }
+static struct ioat_desc_sw *ioat_dma_get_next_descriptor(
+ struct ioat_dma_chan *ioat_chan)
+{
+ if (!ioat_chan)
+ return NULL;
- list_splice(&new_chain, &new->async_tx.tx_list);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ return ioat1_dma_get_next_descriptor(ioat_chan);
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ return ioat2_dma_get_next_descriptor(ioat_chan);
+ }
+ return NULL;
+}
- new->hw->ctl = IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
- new->hw->next = 0;
- new->tx_cnt = desc_count;
- new->async_tx.ack = 0; /* client is in control of this ack */
- new->async_tx.cookie = -EBUSY;
+static struct dma_async_tx_descriptor *ioat1_dma_prep_memcpy(
+ struct dma_chan *chan,
+ dma_addr_t dma_dest,
+ dma_addr_t dma_src,
+ size_t len,
+ unsigned long flags)
+{
+ struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioat_desc_sw *new;
- pci_unmap_len_set(new, len, orig_len);
+ spin_lock_bh(&ioat_chan->desc_lock);
+ new = ioat_dma_get_next_descriptor(ioat_chan);
spin_unlock_bh(&ioat_chan->desc_lock);
- return new ? &new->async_tx : NULL;
+ if (new) {
+ new->len = len;
+ new->dst = dma_dest;
+ new->src = dma_src;
+ new->async_tx.flags = flags;
+ return &new->async_tx;
+ } else {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+ return NULL;
+ }
}
-/**
- * ioat_dma_memcpy_issue_pending - push potentially unrecognized appended
- * descriptors to hw
- * @chan: DMA channel handle
- */
-static void ioat_dma_memcpy_issue_pending(struct dma_chan *chan)
+static struct dma_async_tx_descriptor *ioat2_dma_prep_memcpy(
+ struct dma_chan *chan,
+ dma_addr_t dma_dest,
+ dma_addr_t dma_src,
+ size_t len,
+ unsigned long flags)
{
struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
+ struct ioat_desc_sw *new;
- if (ioat_chan->pending != 0) {
- ioat_chan->pending = 0;
- writeb(IOAT_CHANCMD_APPEND,
- ioat_chan->reg_base + IOAT_CHANCMD_OFFSET);
+ spin_lock_bh(&ioat_chan->desc_lock);
+ new = ioat2_dma_get_next_descriptor(ioat_chan);
+
+ /*
+ * leave ioat_chan->desc_lock set in ioat 2 path
+ * it will get unlocked at end of tx_submit
+ */
+
+ if (new) {
+ new->len = len;
+ new->dst = dma_dest;
+ new->src = dma_src;
+ new->async_tx.flags = flags;
+ return &new->async_tx;
+ } else {
+ spin_unlock_bh(&ioat_chan->desc_lock);
+ dev_err(&ioat_chan->device->pdev->dev,
+ "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n",
+ chan_num(ioat_chan), ioat_chan->dmacount, ioat_chan->desccount);
+ return NULL;
}
}
+static void ioat_dma_cleanup_tasklet(unsigned long data)
+{
+ struct ioat_dma_chan *chan = (void *)data;
+ ioat_dma_memcpy_cleanup(chan);
+ writew(IOAT_CHANCTRL_INT_DISABLE,
+ chan->reg_base + IOAT_CHANCTRL_OFFSET);
+}
+
+static void
+ioat_dma_unmap(struct ioat_dma_chan *ioat_chan, struct ioat_desc_sw *desc)
+{
+ /*
+ * yes we are unmapping both _page and _single
+ * alloc'd regions with unmap_page. Is this
+ * *really* that bad?
+ */
+ if (!(desc->async_tx.flags & DMA_COMPL_SKIP_DEST_UNMAP))
+ pci_unmap_page(ioat_chan->device->pdev,
+ pci_unmap_addr(desc, dst),
+ pci_unmap_len(desc, len),
+ PCI_DMA_FROMDEVICE);
+
+ if (!(desc->async_tx.flags & DMA_COMPL_SKIP_SRC_UNMAP))
+ pci_unmap_page(ioat_chan->device->pdev,
+ pci_unmap_addr(desc, src),
+ pci_unmap_len(desc, len),
+ PCI_DMA_TODEVICE);
+}
+
+/**
+ * ioat_dma_memcpy_cleanup - cleanup up finished descriptors
+ * @chan: ioat channel to be cleaned up
+ */
static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan)
{
unsigned long phys_complete;
struct ioat_desc_sw *desc, *_desc;
dma_cookie_t cookie = 0;
+ unsigned long desc_phys;
+ struct ioat_desc_sw *latest_desc;
prefetch(ioat_chan->completion_virt);
- if (!spin_trylock(&ioat_chan->cleanup_lock))
+ if (!spin_trylock_bh(&ioat_chan->cleanup_lock))
return;
/* The completion writeback can happen at any time,
#if (BITS_PER_LONG == 64)
phys_complete =
- ioat_chan->completion_virt->full & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+ ioat_chan->completion_virt->full
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
#else
- phys_complete = ioat_chan->completion_virt->low & IOAT_LOW_COMPLETION_MASK;
+ phys_complete =
+ ioat_chan->completion_virt->low & IOAT_LOW_COMPLETION_MASK;
#endif
- if ((ioat_chan->completion_virt->full & IOAT_CHANSTS_DMA_TRANSFER_STATUS) ==
+ if ((ioat_chan->completion_virt->full
+ & IOAT_CHANSTS_DMA_TRANSFER_STATUS) ==
IOAT_CHANSTS_DMA_TRANSFER_STATUS_HALTED) {
dev_err(&ioat_chan->device->pdev->dev,
- "ioatdma: Channel halted, chanerr = %x\n",
+ "Channel halted, chanerr = %x\n",
readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET));
/* TODO do something to salvage the situation */
}
if (phys_complete == ioat_chan->last_completion) {
- spin_unlock(&ioat_chan->cleanup_lock);
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+ /*
+ * perhaps we're stuck so hard that the watchdog can't go off?
+ * try to catch it after 2 seconds
+ */
+ if (ioat_chan->device->version != IOAT_VER_3_0) {
+ if (time_after(jiffies,
+ ioat_chan->last_completion_time + HZ*WATCHDOG_DELAY)) {
+ ioat_dma_chan_watchdog(&(ioat_chan->device->work.work));
+ ioat_chan->last_completion_time = jiffies;
+ }
+ }
return;
}
+ ioat_chan->last_completion_time = jiffies;
- spin_lock_bh(&ioat_chan->desc_lock);
- list_for_each_entry_safe(desc, _desc, &ioat_chan->used_desc, node) {
-
- /*
- * Incoming DMA requests may use multiple descriptors, due to
- * exceeding xfercap, perhaps. If so, only the last one will
- * have a cookie, and require unmapping.
- */
- if (desc->async_tx.cookie) {
- cookie = desc->async_tx.cookie;
+ cookie = 0;
+ if (!spin_trylock_bh(&ioat_chan->desc_lock)) {
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
+ return;
+ }
- /*
- * yes we are unmapping both _page and _single alloc'd
- * regions with unmap_page. Is this *really* that bad?
- */
- pci_unmap_page(ioat_chan->device->pdev,
- pci_unmap_addr(desc, dst),
- pci_unmap_len(desc, len),
- PCI_DMA_FROMDEVICE);
- pci_unmap_page(ioat_chan->device->pdev,
- pci_unmap_addr(desc, src),
- pci_unmap_len(desc, len),
- PCI_DMA_TODEVICE);
- }
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ list_for_each_entry_safe(desc, _desc,
+ &ioat_chan->used_desc, node) {
- if (desc->async_tx.phys != phys_complete) {
/*
- * a completed entry, but not the last, so cleanup
- * if the client is done with the descriptor
+ * Incoming DMA requests may use multiple descriptors,
+ * due to exceeding xfercap, perhaps. If so, only the
+ * last one will have a cookie, and require unmapping.
*/
- if (desc->async_tx.ack) {
- list_del(&desc->node);
- list_add_tail(&desc->node,
- &ioat_chan->free_desc);
- } else
+ if (desc->async_tx.cookie) {
+ cookie = desc->async_tx.cookie;
+ ioat_dma_unmap(ioat_chan, desc);
+ if (desc->async_tx.callback) {
+ desc->async_tx.callback(desc->async_tx.callback_param);
+ desc->async_tx.callback = NULL;
+ }
+ }
+
+ if (desc->async_tx.phys != phys_complete) {
+ /*
+ * a completed entry, but not the last, so clean
+ * up if the client is done with the descriptor
+ */
+ if (async_tx_test_ack(&desc->async_tx)) {
+ list_del(&desc->node);
+ list_add_tail(&desc->node,
+ &ioat_chan->free_desc);
+ } else
+ desc->async_tx.cookie = 0;
+ } else {
+ /*
+ * last used desc. Do not remove, so we can
+ * append from it, but don't look at it next
+ * time, either
+ */
desc->async_tx.cookie = 0;
- } else {
- /*
- * last used desc. Do not remove, so we can append from
- * it, but don't look at it next time, either
- */
- desc->async_tx.cookie = 0;
- /* TODO check status bits? */
+ /* TODO check status bits? */
+ break;
+ }
+ }
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ /* has some other thread has already cleaned up? */
+ if (ioat_chan->used_desc.prev == NULL)
break;
+
+ /* work backwards to find latest finished desc */
+ desc = to_ioat_desc(ioat_chan->used_desc.next);
+ latest_desc = NULL;
+ do {
+ desc = to_ioat_desc(desc->node.prev);
+ desc_phys = (unsigned long)desc->async_tx.phys
+ & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR;
+ if (desc_phys == phys_complete) {
+ latest_desc = desc;
+ break;
+ }
+ } while (&desc->node != ioat_chan->used_desc.prev);
+
+ if (latest_desc != NULL) {
+
+ /* work forwards to clear finished descriptors */
+ for (desc = to_ioat_desc(ioat_chan->used_desc.prev);
+ &desc->node != latest_desc->node.next &&
+ &desc->node != ioat_chan->used_desc.next;
+ desc = to_ioat_desc(desc->node.next)) {
+ if (desc->async_tx.cookie) {
+ cookie = desc->async_tx.cookie;
+ desc->async_tx.cookie = 0;
+ ioat_dma_unmap(ioat_chan, desc);
+ if (desc->async_tx.callback) {
+ desc->async_tx.callback(desc->async_tx.callback_param);
+ desc->async_tx.callback = NULL;
+ }
+ }
+ }
+
+ /* move used.prev up beyond those that are finished */
+ if (&desc->node == ioat_chan->used_desc.next)
+ ioat_chan->used_desc.prev = NULL;
+ else
+ ioat_chan->used_desc.prev = &desc->node;
}
+ break;
}
spin_unlock_bh(&ioat_chan->desc_lock);
if (cookie != 0)
ioat_chan->completed_cookie = cookie;
- spin_unlock(&ioat_chan->cleanup_lock);
-}
-
-static void ioat_dma_dependency_added(struct dma_chan *chan)
-{
- struct ioat_dma_chan *ioat_chan = to_ioat_chan(chan);
- spin_lock_bh(&ioat_chan->desc_lock);
- if (ioat_chan->pending == 0) {
- spin_unlock_bh(&ioat_chan->desc_lock);
- ioat_dma_memcpy_cleanup(ioat_chan);
- } else
- spin_unlock_bh(&ioat_chan->desc_lock);
+ spin_unlock_bh(&ioat_chan->cleanup_lock);
}
/**
last_used = chan->cookie;
last_complete = ioat_chan->completed_cookie;
+ ioat_chan->watchdog_tcp_cookie = cookie;
if (done)
*done = last_complete;
return dma_async_is_complete(cookie, last_complete, last_used);
}
-/* PCI API */
-
-static irqreturn_t ioat_do_interrupt(int irq, void *data)
-{
- struct ioatdma_device *instance = data;
- unsigned long attnstatus;
- u8 intrctrl;
-
- intrctrl = readb(instance->reg_base + IOAT_INTRCTRL_OFFSET);
-
- if (!(intrctrl & IOAT_INTRCTRL_MASTER_INT_EN))
- return IRQ_NONE;
-
- if (!(intrctrl & IOAT_INTRCTRL_INT_STATUS)) {
- writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
- return IRQ_NONE;
- }
-
- attnstatus = readl(instance->reg_base + IOAT_ATTNSTATUS_OFFSET);
-
- printk(KERN_ERR "ioatdma: interrupt! status %lx\n", attnstatus);
-
- writeb(intrctrl, instance->reg_base + IOAT_INTRCTRL_OFFSET);
- return IRQ_HANDLED;
-}
-
static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat_chan)
{
struct ioat_desc_sw *desc;
spin_lock_bh(&ioat_chan->desc_lock);
- if (!list_empty(&ioat_chan->free_desc)) {
- desc = to_ioat_desc(ioat_chan->free_desc.next);
- list_del(&desc->node);
- } else {
- /* try to get another desc */
+ desc = ioat_dma_get_next_descriptor(ioat_chan);
+
+ if (!desc) {
+ dev_err(&ioat_chan->device->pdev->dev,
+ "Unable to start null desc - get next desc failed\n");
spin_unlock_bh(&ioat_chan->desc_lock);
- desc = ioat_dma_alloc_descriptor(ioat_chan, GFP_KERNEL);
- spin_lock_bh(&ioat_chan->desc_lock);
- /* will this ever happen? */
- BUG_ON(!desc);
+ return;
}
- desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL;
- desc->hw->next = 0;
- desc->async_tx.ack = 1;
-
- list_add_tail(&desc->node, &ioat_chan->used_desc);
+ desc->hw->ctl = IOAT_DMA_DESCRIPTOR_NUL
+ | IOAT_DMA_DESCRIPTOR_CTL_INT_GN
+ | IOAT_DMA_DESCRIPTOR_CTL_CP_STS;
+ /* set size to non-zero value (channel returns error when size is 0) */
+ desc->hw->size = NULL_DESC_BUFFER_SIZE;
+ desc->hw->src_addr = 0;
+ desc->hw->dst_addr = 0;
+ async_tx_ack(&desc->async_tx);
+ switch (ioat_chan->device->version) {
+ case IOAT_VER_1_2:
+ desc->hw->next = 0;
+ list_add_tail(&desc->node, &ioat_chan->used_desc);
+
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH);
+
+ writeb(IOAT_CHANCMD_START, ioat_chan->reg_base
+ + IOAT_CHANCMD_OFFSET(ioat_chan->device->version));
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW);
+ writel(((u64) desc->async_tx.phys) >> 32,
+ ioat_chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH);
+
+ ioat_chan->dmacount++;
+ __ioat2_dma_memcpy_issue_pending(ioat_chan);
+ break;
+ }
spin_unlock_bh(&ioat_chan->desc_lock);
-
- writel(((u64) desc->async_tx.phys) & 0x00000000FFFFFFFF,
- ioat_chan->reg_base + IOAT_CHAINADDR_OFFSET_LOW);
- writel(((u64) desc->async_tx.phys) >> 32,
- ioat_chan->reg_base + IOAT_CHAINADDR_OFFSET_HIGH);
-
- writeb(IOAT_CHANCMD_START, ioat_chan->reg_base + IOAT_CHANCMD_OFFSET);
}
/*
*/
#define IOAT_TEST_SIZE 2000
-static int ioat_self_test(struct ioatdma_device *device)
+static void ioat_dma_test_callback(void *dma_async_param)
+{
+ struct completion *cmp = dma_async_param;
+
+ complete(cmp);
+}
+
+/**
+ * ioat_dma_self_test - Perform a IOAT transaction to verify the HW works.
+ * @device: device to be tested
+ */
+static int ioat_dma_self_test(struct ioatdma_device *device)
{
int i;
u8 *src;
u8 *dest;
struct dma_chan *dma_chan;
struct dma_async_tx_descriptor *tx;
- dma_addr_t addr;
+ dma_addr_t dma_dest, dma_src;
dma_cookie_t cookie;
int err = 0;
+ struct completion cmp;
src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL);
if (!src)
dma_chan = container_of(device->common.channels.next,
struct dma_chan,
device_node);
- if (ioat_dma_alloc_chan_resources(dma_chan) < 1) {
+ if (device->common.device_alloc_chan_resources(dma_chan) < 1) {
dev_err(&device->pdev->dev,
"selftest cannot allocate chan resource\n");
err = -ENODEV;
goto out;
}
- tx = ioat_dma_prep_memcpy(dma_chan, IOAT_TEST_SIZE, 0);
+ dma_src = dma_map_single(dma_chan->device->dev, src, IOAT_TEST_SIZE,
+ DMA_TO_DEVICE);
+ dma_dest = dma_map_single(dma_chan->device->dev, dest, IOAT_TEST_SIZE,
+ DMA_FROM_DEVICE);
+ tx = device->common.device_prep_dma_memcpy(dma_chan, dma_dest, dma_src,
+ IOAT_TEST_SIZE, 0);
+ if (!tx) {
+ dev_err(&device->pdev->dev,
+ "Self-test prep failed, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+
async_tx_ack(tx);
- addr = dma_map_single(dma_chan->device->dev, src, IOAT_TEST_SIZE,
- DMA_TO_DEVICE);
- ioat_set_src(addr, tx, 0);
- addr = dma_map_single(dma_chan->device->dev, dest, IOAT_TEST_SIZE,
- DMA_FROM_DEVICE);
- ioat_set_dest(addr, tx, 0);
- cookie = ioat_tx_submit(tx);
- ioat_dma_memcpy_issue_pending(dma_chan);
- msleep(1);
-
- if (ioat_dma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
+ init_completion(&cmp);
+ tx->callback = ioat_dma_test_callback;
+ tx->callback_param = &cmp;
+ cookie = tx->tx_submit(tx);
+ if (cookie < 0) {
+ dev_err(&device->pdev->dev,
+ "Self-test setup failed, disabling\n");
+ err = -ENODEV;
+ goto free_resources;
+ }
+ device->common.device_issue_pending(dma_chan);
+
+ wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000));
+
+ if (device->common.device_is_tx_complete(dma_chan, cookie, NULL, NULL)
+ != DMA_SUCCESS) {
dev_err(&device->pdev->dev,
- "ioatdma: Self-test copy timed out, disabling\n");
+ "Self-test copy timed out, disabling\n");
err = -ENODEV;
goto free_resources;
}
if (memcmp(src, dest, IOAT_TEST_SIZE)) {
dev_err(&device->pdev->dev,
- "ioatdma: Self-test copy failed compare, disabling\n");
+ "Self-test copy failed compare, disabling\n");
err = -ENODEV;
goto free_resources;
}
free_resources:
- ioat_dma_free_chan_resources(dma_chan);
+ device->common.device_free_chan_resources(dma_chan);
out:
kfree(src);
kfree(dest);
return err;
}
+static char ioat_interrupt_style[32] = "msix";
+module_param_string(ioat_interrupt_style, ioat_interrupt_style,
+ sizeof(ioat_interrupt_style), 0644);
+MODULE_PARM_DESC(ioat_interrupt_style,
+ "set ioat interrupt style: msix (default), "
+ "msix-single-vector, msi, intx)");
+
+/**
+ * ioat_dma_setup_interrupts - setup interrupt handler
+ * @device: ioat device
+ */
+static int ioat_dma_setup_interrupts(struct ioatdma_device *device)
+{
+ struct ioat_dma_chan *ioat_chan;
+ int err, i, j, msixcnt;
+ u8 intrctrl = 0;
+
+ if (!strcmp(ioat_interrupt_style, "msix"))
+ goto msix;
+ if (!strcmp(ioat_interrupt_style, "msix-single-vector"))
+ goto msix_single_vector;
+ if (!strcmp(ioat_interrupt_style, "msi"))
+ goto msi;
+ if (!strcmp(ioat_interrupt_style, "intx"))
+ goto intx;
+ dev_err(&device->pdev->dev, "invalid ioat_interrupt_style %s\n",
+ ioat_interrupt_style);
+ goto err_no_irq;
+
+msix:
+ /* The number of MSI-X vectors should equal the number of channels */
+ msixcnt = device->common.chancnt;
+ for (i = 0; i < msixcnt; i++)
+ device->msix_entries[i].entry = i;
+
+ err = pci_enable_msix(device->pdev, device->msix_entries, msixcnt);
+ if (err < 0)
+ goto msi;
+ if (err > 0)
+ goto msix_single_vector;
+
+ for (i = 0; i < msixcnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+ err = request_irq(device->msix_entries[i].vector,
+ ioat_dma_do_interrupt_msix,
+ 0, "ioat-msix", ioat_chan);
+ if (err) {
+ for (j = 0; j < i; j++) {
+ ioat_chan =
+ ioat_lookup_chan_by_index(device, j);
+ free_irq(device->msix_entries[j].vector,
+ ioat_chan);
+ }
+ goto msix_single_vector;
+ }
+ }
+ intrctrl |= IOAT_INTRCTRL_MSIX_VECTOR_CONTROL;
+ device->irq_mode = msix_multi_vector;
+ goto done;
+
+msix_single_vector:
+ device->msix_entries[0].entry = 0;
+ err = pci_enable_msix(device->pdev, device->msix_entries, 1);
+ if (err)
+ goto msi;
+
+ err = request_irq(device->msix_entries[0].vector, ioat_dma_do_interrupt,
+ 0, "ioat-msix", device);
+ if (err) {
+ pci_disable_msix(device->pdev);
+ goto msi;
+ }
+ device->irq_mode = msix_single_vector;
+ goto done;
+
+msi:
+ err = pci_enable_msi(device->pdev);
+ if (err)
+ goto intx;
+
+ err = request_irq(device->pdev->irq, ioat_dma_do_interrupt,
+ 0, "ioat-msi", device);
+ if (err) {
+ pci_disable_msi(device->pdev);
+ goto intx;
+ }
+ /*
+ * CB 1.2 devices need a bit set in configuration space to enable MSI
+ */
+ if (device->version == IOAT_VER_1_2) {
+ u32 dmactrl;
+ pci_read_config_dword(device->pdev,
+ IOAT_PCI_DMACTRL_OFFSET, &dmactrl);
+ dmactrl |= IOAT_PCI_DMACTRL_MSI_EN;
+ pci_write_config_dword(device->pdev,
+ IOAT_PCI_DMACTRL_OFFSET, dmactrl);
+ }
+ device->irq_mode = msi;
+ goto done;
+
+intx:
+ err = request_irq(device->pdev->irq, ioat_dma_do_interrupt,
+ IRQF_SHARED, "ioat-intx", device);
+ if (err)
+ goto err_no_irq;
+ device->irq_mode = intx;
+
+done:
+ intrctrl |= IOAT_INTRCTRL_MASTER_INT_EN;
+ writeb(intrctrl, device->reg_base + IOAT_INTRCTRL_OFFSET);
+ return 0;
+
+err_no_irq:
+ /* Disable all interrupt generation */
+ writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET);
+ dev_err(&device->pdev->dev, "no usable interrupts\n");
+ device->irq_mode = none;
+ return -1;
+}
+
+/**
+ * ioat_dma_remove_interrupts - remove whatever interrupts were set
+ * @device: ioat device
+ */
+static void ioat_dma_remove_interrupts(struct ioatdma_device *device)
+{
+ struct ioat_dma_chan *ioat_chan;
+ int i;
+
+ /* Disable all interrupt generation */
+ writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET);
+
+ switch (device->irq_mode) {
+ case msix_multi_vector:
+ for (i = 0; i < device->common.chancnt; i++) {
+ ioat_chan = ioat_lookup_chan_by_index(device, i);
+ free_irq(device->msix_entries[i].vector, ioat_chan);
+ }
+ pci_disable_msix(device->pdev);
+ break;
+ case msix_single_vector:
+ free_irq(device->msix_entries[0].vector, device);
+ pci_disable_msix(device->pdev);
+ break;
+ case msi:
+ free_irq(device->pdev->irq, device);
+ pci_disable_msi(device->pdev);
+ break;
+ case intx:
+ free_irq(device->pdev->irq, device);
+ break;
+ case none:
+ dev_warn(&device->pdev->dev,
+ "call to %s without interrupts setup\n", __func__);
+ }
+ device->irq_mode = none;
+}
+
struct ioatdma_device *ioat_dma_probe(struct pci_dev *pdev,
void __iomem *iobase)
{
INIT_LIST_HEAD(&device->common.channels);
ioat_dma_enumerate_channels(device);
- dma_cap_set(DMA_MEMCPY, device->common.cap_mask);
device->common.device_alloc_chan_resources =
ioat_dma_alloc_chan_resources;
device->common.device_free_chan_resources =
ioat_dma_free_chan_resources;
- device->common.device_prep_dma_memcpy = ioat_dma_prep_memcpy;
- device->common.device_is_tx_complete = ioat_dma_is_complete;
- device->common.device_issue_pending = ioat_dma_memcpy_issue_pending;
- device->common.device_dependency_added = ioat_dma_dependency_added;
device->common.dev = &pdev->dev;
- printk(KERN_INFO "ioatdma: Intel(R) I/OAT DMA Engine found,"
- " %d channels, device version 0x%02x\n",
- device->common.chancnt, device->version);
- pci_set_drvdata(pdev, device);
- err = request_irq(pdev->irq, &ioat_do_interrupt, IRQF_SHARED, "ioat",
- device);
- if (err)
- goto err_irq;
+ dma_cap_set(DMA_MEMCPY, device->common.cap_mask);
+ device->common.device_is_tx_complete = ioat_dma_is_complete;
+ switch (device->version) {
+ case IOAT_VER_1_2:
+ device->common.device_prep_dma_memcpy = ioat1_dma_prep_memcpy;
+ device->common.device_issue_pending =
+ ioat1_dma_memcpy_issue_pending;
+ break;
+ case IOAT_VER_2_0:
+ case IOAT_VER_3_0:
+ device->common.device_prep_dma_memcpy = ioat2_dma_prep_memcpy;
+ device->common.device_issue_pending =
+ ioat2_dma_memcpy_issue_pending;
+ break;
+ }
+
+ dev_err(&device->pdev->dev,
+ "Intel(R) I/OAT DMA Engine found,"
+ " %d channels, device version 0x%02x, driver version %s\n",
+ device->common.chancnt, device->version, IOAT_DMA_VERSION);
- writeb(IOAT_INTRCTRL_MASTER_INT_EN,
- device->reg_base + IOAT_INTRCTRL_OFFSET);
- pci_set_master(pdev);
+ err = ioat_dma_setup_interrupts(device);
+ if (err)
+ goto err_setup_interrupts;
- err = ioat_self_test(device);
+ err = ioat_dma_self_test(device);
if (err)
goto err_self_test;
+ ioat_set_tcp_copy_break(device);
+
dma_async_device_register(&device->common);
+ if (device->version != IOAT_VER_3_0) {
+ INIT_DELAYED_WORK(&device->work, ioat_dma_chan_watchdog);
+ schedule_delayed_work(&device->work,
+ WATCHDOG_DELAY);
+ }
+
return device;
err_self_test:
- free_irq(device->pdev->irq, device);
-err_irq:
+ ioat_dma_remove_interrupts(device);
+err_setup_interrupts:
pci_pool_destroy(device->completion_pool);
err_completion_pool:
pci_pool_destroy(device->dma_pool);
err_dma_pool:
kfree(device);
err_kzalloc:
- iounmap(iobase);
- printk(KERN_ERR
- "ioatdma: Intel(R) I/OAT DMA Engine initialization failed\n");
+ dev_err(&pdev->dev,
+ "Intel(R) I/OAT DMA Engine initialization failed\n");
return NULL;
}
struct dma_chan *chan, *_chan;
struct ioat_dma_chan *ioat_chan;
- dma_async_device_unregister(&device->common);
+ ioat_dma_remove_interrupts(device);
- free_irq(device->pdev->irq, device);
+ dma_async_device_unregister(&device->common);
pci_pool_destroy(device->dma_pool);
pci_pool_destroy(device->completion_pool);
+ iounmap(device->reg_base);
+ pci_release_regions(device->pdev);
+ pci_disable_device(device->pdev);
+
+ if (device->version != IOAT_VER_3_0) {
+ cancel_delayed_work(&device->work);
+ }
+
list_for_each_entry_safe(chan, _chan,
&device->common.channels, device_node) {
ioat_chan = to_ioat_chan(chan);
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff