ioat3.2 adds xor offload support for up to 8 sources. It can also
perform an xor-zero-sum operation to validate whether all given sources
sum to zero, without writing to a destination. Xor descriptors differ
from memcpy in that one operation may require multiple descriptors
depending on the number of sources. When the number of sources exceeds
5 an extended descriptor is needed. These descriptors need to be
accounted for when updating the DMA_COUNT register.
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
drivers/dma/ioat/dma_v2.c | 2
drivers/dma/ioat/dma_v2.h | 3 +
drivers/dma/ioat/dma_v3.c | 218 ++++++++++++++++++++++++++++++++++++++++++
drivers/dma/ioat/registers.h | 2
4 files changed, 222 insertions(+), 3 deletions(-)
diff --git a/drivers/dma/ioat/dma_v2.c b/drivers/dma/ioat/dma_v2.c
index 7408bef..56fa6d6 100644
--- a/drivers/dma/ioat/dma_v2.c
+++ b/drivers/dma/ioat/dma_v2.c
@@ -48,7 +48,7 @@ module_param(ioat_ring_max_alloc_order, int, 0644);
MODULE_PARM_DESC(ioat_ring_max_alloc_order,
"ioat2+: upper limit for dynamic ring resizing (default: n=16)");
-static void __ioat2_issue_pending(struct ioat2_dma_chan *ioat)
+void __ioat2_issue_pending(struct ioat2_dma_chan *ioat)
{
void * __iomem reg_base = ioat->base.reg_base;
diff --git a/drivers/dma/ioat/dma_v2.h b/drivers/dma/ioat/dma_v2.h
index a216b3f..47afcfd 100644
--- a/drivers/dma/ioat/dma_v2.h
+++ b/drivers/dma/ioat/dma_v2.h
@@ -127,6 +127,7 @@ static inline u16 ioat2_xferlen_to_descs(struct ioat2_dma_chan *ioat, size_t len
* @raw: hardware raw (un-typed) descriptor
* @txd: the generic software descriptor for all engines
* @len: total transaction length for unmap
+ * @result: asynchronous result of validate operations
* @id: identifier for debug
*/
@@ -143,6 +144,7 @@ struct ioat_ring_ent {
};
size_t len;
struct dma_async_tx_descriptor txd;
+ enum sum_check_flags *result;
#ifdef DEBUG
int id;
#endif
@@ -182,5 +184,6 @@ enum dma_status ioat2_is_complete(struct dma_chan *c, dma_cookie_t cookie,
void __ioat2_restart_chan(struct ioat2_dma_chan *ioat);
dma_cookie_t ioat2_inject_fence(struct ioat2_dma_chan *ioat);
bool reshape_ring(struct ioat2_dma_chan *ioat, int order);
+void __ioat2_issue_pending(struct ioat2_dma_chan *ioat);
extern struct kobj_type ioat2_ktype;
#endif /* IOATDMA_V2_H */
diff --git a/drivers/dma/ioat/dma_v3.c b/drivers/dma/ioat/dma_v3.c
index 0913d11..957c205 100644
--- a/drivers/dma/ioat/dma_v3.c
+++ b/drivers/dma/ioat/dma_v3.c
@@ -64,8 +64,33 @@
#include "dma.h"
#include "dma_v2.h"
+/* ioat hardware assumes at least two sources for raid operations */
+#define src_cnt_to_sw(x) ((x) + 2)
+#define src_cnt_to_hw(x) ((x) - 2)
+
+/* provide a lookup table for setting the source address in the base or
+ * extended descriptor of an xor descriptor
+ */
+static const u8 xor_idx_to_desc __read_mostly = 0xd0;
+static const u8 xor_idx_to_field[] __read_mostly = { 1, 4, 5, 6, 7, 0, 1, 2 };
+
+static dma_addr_t xor_get_src(struct ioat_raw_descriptor *descs[2], int idx)
+{
+ struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1];
+
+ return raw->field[xor_idx_to_field[idx]];
+}
+
+static void xor_set_src(struct ioat_raw_descriptor *descs[2],
+ dma_addr_t addr, u32 offset, int idx)
+{
+ struct ioat_raw_descriptor *raw = descs[xor_idx_to_desc >> idx & 1];
+
+ raw->field[xor_idx_to_field[idx]] = addr + offset;
+}
+
static void ioat3_dma_unmap(struct ioat2_dma_chan *ioat,
- struct ioat_ring_ent *desc)
+ struct ioat_ring_ent *desc, int idx)
{
struct ioat_chan_common *chan = &ioat->base;
struct pci_dev *pdev = chan->device->pdev;
@@ -86,13 +111,71 @@ static void ioat3_dma_unmap(struct ioat2_dma_chan *ioat,
PCI_DMA_FROMDEVICE, flags, 1);
break;
}
+ case IOAT_OP_XOR_VAL:
+ case IOAT_OP_XOR: {
+ struct ioat_xor_descriptor *xor = desc->xor;
+ struct ioat_ring_ent *ext;
+ struct ioat_xor_ext_descriptor *xor_ex = NULL;
+ int src_cnt = src_cnt_to_sw(xor->ctl_f.src_cnt);
+ struct ioat_raw_descriptor *descs[2];
+ int i;
+
+ if (src_cnt > 5) {
+ ext = ioat2_get_ring_ent(ioat, idx + 1);
+ xor_ex = ext->xor_ex;
+ }
+
+ if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
+ descs[0] = (struct ioat_raw_descriptor *) xor;
+ descs[1] = (struct ioat_raw_descriptor *) xor_ex;
+ for (i = 0; i < src_cnt; i++) {
+ dma_addr_t src = xor_get_src(descs, i);
+
+ ioat_unmap(pdev, src - offset, len,
+ PCI_DMA_TODEVICE, flags, 0);
+ }
+
+ /* dest is a source in xor validate operations */
+ if (xor->ctl_f.op == IOAT_OP_XOR_VAL) {
+ ioat_unmap(pdev, xor->dst_addr - offset, len,
+ PCI_DMA_TODEVICE, flags, 1);
+ break;
+ }
+ }
+
+ if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP))
+ ioat_unmap(pdev, xor->dst_addr - offset, len,
+ PCI_DMA_FROMDEVICE, flags, 1);
+ break;
+ }
default:
dev_err(&pdev->dev, "%s: unknown op type: %#x\n",
__func__, desc->hw->ctl_f.op);
}
}
+static bool desc_has_ext(struct ioat_ring_ent *desc)
+{
+ struct ioat_dma_descriptor *hw = desc->hw;
+
+ if (hw->ctl_f.op == IOAT_OP_XOR ||
+ hw->ctl_f.op == IOAT_OP_XOR_VAL) {
+ struct ioat_xor_descriptor *xor = desc->xor;
+ if (src_cnt_to_sw(xor->ctl_f.src_cnt) > 5)
+ return true;
+ }
+
+ return false;
+}
+
+/**
+ * __cleanup - reclaim used descriptors
+ * @ioat: channel (ring) to clean
+ *
+ * The difference from the dma_v2.c __cleanup() is that this routine
+ * handles extended descriptors and dma-unmapping raid operations.
+ */
static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
{
struct ioat_chan_common *chan = &ioat->base;
@@ -114,7 +197,7 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
tx = &desc->txd;
if (tx->cookie) {
chan->completed_cookie = tx->cookie;
- ioat3_dma_unmap(ioat, desc);
+ ioat3_dma_unmap(ioat, desc, ioat->tail + i);
tx->cookie = 0;
if (tx->callback) {
tx->callback(tx->callback_param);
@@ -124,6 +207,12 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
if (tx->phys == phys_complete)
seen_current = true;
+
+ /* skip extended descriptors */
+ if (desc_has_ext(desc)) {
+ BUG_ON(i + 1 >= active);
+ i++;
+ }
}
ioat->tail += i;
BUG_ON(!seen_current); /* no active descs have written a completion? */
@@ -309,6 +398,121 @@ ioat3_prep_memset_lock(struct dma_chan *c, dma_addr_t dest, int value,
return &desc->txd;
}
+static struct dma_async_tx_descriptor *
+__ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result,
+ dma_addr_t dest, dma_addr_t *src, unsigned int src_cnt,
+ size_t len, unsigned long flags)
+{
+ struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
+ struct ioat_ring_ent *compl_desc;
+ struct ioat_ring_ent *desc;
+ struct ioat_ring_ent *ext;
+ size_t total_len = len;
+ struct ioat_xor_descriptor *xor;
+ struct ioat_xor_ext_descriptor *xor_ex = NULL;
+ struct ioat_dma_descriptor *hw;
+ u32 offset = 0;
+ int num_descs;
+ int with_ext;
+ int i;
+ u16 idx;
+ u8 op = result ? IOAT_OP_XOR_VAL : IOAT_OP_XOR;
+
+ BUG_ON(src_cnt < 2);
+
+ num_descs = ioat2_xferlen_to_descs(ioat, len);
+ /* we need 2x the number of descriptors to cover greater than 5
+ * sources
+ */
+ if (src_cnt > 5) {
+ with_ext = 1;
+ num_descs *= 2;
+ } else
+ with_ext = 0;
+
+ /* completion writes from the raid engine may pass completion
+ * writes from the legacy engine, so we need one extra null
+ * (legacy) descriptor to ensure all completion writes arrive in
+ * order.
+ */
+ if (likely(num_descs) &&
+ ioat2_alloc_and_lock(&idx, ioat, num_descs+1) == 0)
+ /* pass */;
+ else
+ return NULL;
+ for (i = 0; i < num_descs; i += 1 + with_ext) {
+ struct ioat_raw_descriptor *descs[2];
+ size_t xfer_size = min_t(size_t, len, 1 << ioat->xfercap_log);
+ int s;
+
+ desc = ioat2_get_ring_ent(ioat, idx + i);
+ xor = desc->xor;
+
+ /* save a branch by unconditionally retrieving the
+ * extended descriptor xor_set_src() knows to not write
+ * to it in the single descriptor case
+ */
+ ext = ioat2_get_ring_ent(ioat, idx + i + 1);
+ xor_ex = ext->xor_ex;
+
+ descs[0] = (struct ioat_raw_descriptor *) xor;
+ descs[1] = (struct ioat_raw_descriptor *) xor_ex;
+ for (s = 0; s < src_cnt; s++)
+ xor_set_src(descs, src[s], offset, s);
+ xor->size = xfer_size;
+ xor->dst_addr = dest + offset;
+ xor->ctl = 0;
+ xor->ctl_f.op = op;
+ xor->ctl_f.src_cnt = src_cnt_to_hw(src_cnt);
+
+ len -= xfer_size;
+ offset += xfer_size;
+ dump_desc_dbg(ioat, desc);
+ }
+
+ /* last xor descriptor carries the unmap parameters and fence bit */
+ desc->txd.flags = flags;
+ desc->len = total_len;
+ if (result)
+ desc->result = result;
+ xor->ctl_f.fence = !!(flags & DMA_PREP_FENCE);
+
+ /* completion descriptor carries interrupt bit */
+ compl_desc = ioat2_get_ring_ent(ioat, idx + i);
+ compl_desc->txd.flags = flags & DMA_PREP_INTERRUPT;
+ hw = compl_desc->hw;
+ hw->ctl = 0;
+ hw->ctl_f.null = 1;
+ hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT);
+ hw->ctl_f.compl_write = 1;
+ hw->size = NULL_DESC_BUFFER_SIZE;
+ dump_desc_dbg(ioat, compl_desc);
+
+ /* we leave the channel locked to ensure in order submission */
+ return &desc->txd;
+}
+
+static struct dma_async_tx_descriptor *
+ioat3_prep_xor(struct dma_chan *chan, dma_addr_t dest, dma_addr_t *src,
+ unsigned int src_cnt, size_t len, unsigned long flags)
+{
+ return __ioat3_prep_xor_lock(chan, NULL, dest, src, src_cnt, len, flags);
+}
+
+struct dma_async_tx_descriptor *
+ioat3_prep_xor_val(struct dma_chan *chan, dma_addr_t *src,
+ unsigned int src_cnt, size_t len,
+ enum sum_check_flags *result, unsigned long flags)
+{
+ /* the cleanup routine only sets bits on validate failure, it
+ * does not clear bits on validate success... so clear it here
+ */
+ *result = 0;
+
+ return __ioat3_prep_xor_lock(chan, result, src[0], &src[1],
+ src_cnt - 1, len, flags);
+}
+
int __devinit ioat3_dma_probe(struct ioatdma_device *device, int dca)
{
struct pci_dev *pdev = device->pdev;
@@ -333,6 +537,16 @@ int __devinit ioat3_dma_probe(struct ioatdma_device *device, int dca)
dma_cap_set(DMA_MEMSET, dma->cap_mask);
dma->device_prep_dma_memset = ioat3_prep_memset_lock;
}
+ if (cap & IOAT_CAP_XOR) {
+ dma->max_xor = 8;
+ dma->xor_align = 2;
+
+ dma_cap_set(DMA_XOR, dma->cap_mask);
+ dma->device_prep_dma_xor = ioat3_prep_xor;
+
+ dma_cap_set(DMA_XOR_VAL, dma->cap_mask);
+ dma->device_prep_dma_xor_val = ioat3_prep_xor_val;
+ }
/* -= IOAT ver.3 workarounds =- */
/* Write CHANERRMSK_INT with 3E07h to mask out the errors
diff --git a/drivers/dma/ioat/registers.h b/drivers/dma/ioat/registers.h
index 97d26ea..63038e1 100644
--- a/drivers/dma/ioat/registers.h
+++ b/drivers/dma/ioat/registers.h
@@ -243,6 +243,8 @@
#define IOAT_CHANERR_XOR_Q_ERR 0x20000
#define IOAT_CHANERR_DESCRIPTOR_COUNT_ERR 0x40000
+#define IOAT_CHANERR_HANDLE_MASK (IOAT_CHANERR_XOR_P_OR_CRC_ERR | IOAT_CHANERR_XOR_Q_ERR)
+
#define IOAT_CHANERR_MASK_OFFSET 0x2C /* 32-bit Channel Error Register */
#endif /* _IOAT_REGISTERS_H_ */
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
