*/
#include <linux/pci.h>
+#include <linux/gfp.h>
#include <linux/dmaengine.h>
#include <linux/dma-mapping.h>
+#include <linux/prefetch.h>
#include "registers.h"
#include "hw.h"
#include "dma.h"
/* provide a lookup table for setting the source address in the base or
* extended descriptor of an xor or pq 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 const u8 pq_idx_to_desc __read_mostly = 0xf8;
-static const u8 pq_idx_to_field[] __read_mostly = { 1, 4, 5, 0, 1, 2, 4, 5 };
+static const u8 xor_idx_to_desc = 0xe0;
+static const u8 xor_idx_to_field[] = { 1, 4, 5, 6, 7, 0, 1, 2 };
+static const u8 pq_idx_to_desc = 0xf8;
+static const u8 pq_idx_to_field[] = { 1, 4, 5, 0, 1, 2, 4, 5 };
static dma_addr_t xor_get_src(struct ioat_raw_descriptor *descs[2], int idx)
{
* 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)
+static void __cleanup(struct ioat2_dma_chan *ioat, dma_addr_t phys_complete)
{
struct ioat_chan_common *chan = &ioat->base;
struct ioat_ring_ent *desc;
bool seen_current = false;
+ int idx = ioat->tail, i;
u16 active;
- int i;
dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n",
__func__, ioat->head, ioat->tail, ioat->issued);
for (i = 0; i < active && !seen_current; i++) {
struct dma_async_tx_descriptor *tx;
- prefetch(ioat2_get_ring_ent(ioat, ioat->tail + i + 1));
- desc = ioat2_get_ring_ent(ioat, ioat->tail + i);
+ smp_read_barrier_depends();
+ prefetch(ioat2_get_ring_ent(ioat, idx + i + 1));
+ desc = ioat2_get_ring_ent(ioat, idx + i);
dump_desc_dbg(ioat, desc);
tx = &desc->txd;
if (tx->cookie) {
chan->completed_cookie = tx->cookie;
- ioat3_dma_unmap(ioat, desc, ioat->tail + i);
+ ioat3_dma_unmap(ioat, desc, idx + i);
tx->cookie = 0;
if (tx->callback) {
tx->callback(tx->callback_param);
i++;
}
}
- ioat->tail += i;
+ smp_mb(); /* finish all descriptor reads before incrementing tail */
+ ioat->tail = idx + i;
BUG_ON(active && !seen_current); /* no active descs have written a completion? */
chan->last_completion = phys_complete;
- active = ioat2_ring_active(ioat);
- if (active == 0) {
+ if (active - i == 0) {
dev_dbg(to_dev(chan), "%s: cancel completion timeout\n",
__func__);
clear_bit(IOAT_COMPLETION_PENDING, &chan->state);
mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
}
/* 5 microsecond delay per pending descriptor */
- writew(min((5 * active), IOAT_INTRDELAY_MASK),
+ writew(min((5 * (active - i)), IOAT_INTRDELAY_MASK),
chan->device->reg_base + IOAT_INTRDELAY_OFFSET);
}
-/* try to cleanup, but yield (via spin_trylock) to incoming submissions
- * with the expectation that we will immediately poll again shortly
- */
-static void ioat3_cleanup_poll(struct ioat2_dma_chan *ioat)
-{
- struct ioat_chan_common *chan = &ioat->base;
- unsigned long phys_complete;
-
- prefetch(chan->completion);
-
- if (!spin_trylock_bh(&chan->cleanup_lock))
- return;
-
- if (!ioat_cleanup_preamble(chan, &phys_complete)) {
- spin_unlock_bh(&chan->cleanup_lock);
- return;
- }
-
- if (!spin_trylock_bh(&ioat->ring_lock)) {
- spin_unlock_bh(&chan->cleanup_lock);
- return;
- }
-
- __cleanup(ioat, phys_complete);
-
- spin_unlock_bh(&ioat->ring_lock);
- spin_unlock_bh(&chan->cleanup_lock);
-}
-
-/* run cleanup now because we already delayed the interrupt via INTRDELAY */
-static void ioat3_cleanup_sync(struct ioat2_dma_chan *ioat)
+static void ioat3_cleanup(struct ioat2_dma_chan *ioat)
{
struct ioat_chan_common *chan = &ioat->base;
- unsigned long phys_complete;
-
- prefetch(chan->completion);
+ dma_addr_t phys_complete;
spin_lock_bh(&chan->cleanup_lock);
- if (!ioat_cleanup_preamble(chan, &phys_complete)) {
- spin_unlock_bh(&chan->cleanup_lock);
- return;
- }
- spin_lock_bh(&ioat->ring_lock);
-
- __cleanup(ioat, phys_complete);
-
- spin_unlock_bh(&ioat->ring_lock);
+ if (ioat_cleanup_preamble(chan, &phys_complete))
+ __cleanup(ioat, phys_complete);
spin_unlock_bh(&chan->cleanup_lock);
}
{
struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
- ioat3_cleanup_sync(ioat);
+ ioat3_cleanup(ioat);
writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
}
static void ioat3_restart_channel(struct ioat2_dma_chan *ioat)
{
struct ioat_chan_common *chan = &ioat->base;
- unsigned long phys_complete;
+ dma_addr_t phys_complete;
ioat2_quiesce(chan, 0);
if (ioat_cleanup_preamble(chan, &phys_complete))
struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
struct ioat_chan_common *chan = &ioat->base;
- spin_lock_bh(&chan->cleanup_lock);
if (test_bit(IOAT_COMPLETION_PENDING, &chan->state)) {
- unsigned long phys_complete;
+ dma_addr_t phys_complete;
u64 status;
- spin_lock_bh(&ioat->ring_lock);
status = ioat_chansts(chan);
/* when halted due to errors check for channel
chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET);
dev_err(to_dev(chan), "%s: Channel halted (%x)\n",
__func__, chanerr);
- BUG_ON(is_ioat_bug(chanerr));
+ if (test_bit(IOAT_RUN, &chan->state))
+ BUG_ON(is_ioat_bug(chanerr));
+ else /* we never got off the ground */
+ return;
}
/* if we haven't made progress and we have already
* acknowledged a pending completion once, then be more
* forceful with a restart
*/
+ spin_lock_bh(&chan->cleanup_lock);
if (ioat_cleanup_preamble(chan, &phys_complete))
__cleanup(ioat, phys_complete);
- else if (test_bit(IOAT_COMPLETION_ACK, &chan->state))
+ else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) {
+ spin_lock_bh(&ioat->prep_lock);
ioat3_restart_channel(ioat);
- else {
+ spin_unlock_bh(&ioat->prep_lock);
+ } else {
set_bit(IOAT_COMPLETION_ACK, &chan->state);
mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
}
- spin_unlock_bh(&ioat->ring_lock);
+ spin_unlock_bh(&chan->cleanup_lock);
} else {
u16 active;
/* if the ring is idle, empty, and oversized try to step
* down the size
*/
- spin_lock_bh(&ioat->ring_lock);
+ spin_lock_bh(&chan->cleanup_lock);
+ spin_lock_bh(&ioat->prep_lock);
active = ioat2_ring_active(ioat);
if (active == 0 && ioat->alloc_order > ioat_get_alloc_order())
reshape_ring(ioat, ioat->alloc_order-1);
- spin_unlock_bh(&ioat->ring_lock);
+ spin_unlock_bh(&ioat->prep_lock);
+ spin_unlock_bh(&chan->cleanup_lock);
/* keep shrinking until we get back to our minimum
* default size
if (ioat->alloc_order > ioat_get_alloc_order())
mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
}
- spin_unlock_bh(&chan->cleanup_lock);
}
static enum dma_status
if (ioat_tx_status(c, cookie, txstate) == DMA_SUCCESS)
return DMA_SUCCESS;
- ioat3_cleanup_poll(ioat);
+ ioat3_cleanup(ioat);
return ioat_tx_status(c, cookie, txstate);
}
struct ioat_ring_ent *desc;
size_t total_len = len;
struct ioat_fill_descriptor *fill;
- int num_descs;
u64 src_data = (0x0101010101010101ULL) * (value & 0xff);
- u16 idx;
- int i;
+ int num_descs, idx, i;
num_descs = ioat2_xferlen_to_descs(ioat, len);
- if (likely(num_descs) &&
- ioat2_alloc_and_lock(&idx, ioat, num_descs) == 0)
- /* pass */;
+ if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs) == 0)
+ idx = ioat->head;
else
return NULL;
i = 0;
struct ioat_xor_descriptor *xor;
struct ioat_xor_ext_descriptor *xor_ex = NULL;
struct ioat_dma_descriptor *hw;
+ int num_descs, with_ext, idx, i;
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);
* (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 */;
+ if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs+1) == 0)
+ idx = ioat->head;
else
return NULL;
i = 0;
struct ioat_pq_ext_descriptor *pq_ex = NULL;
struct ioat_dma_descriptor *hw;
u32 offset = 0;
- int num_descs;
- int with_ext;
- int i, s;
- u16 idx;
u8 op = result ? IOAT_OP_PQ_VAL : IOAT_OP_PQ;
+ int i, s, idx, with_ext, num_descs;
dev_dbg(to_dev(chan), "%s\n", __func__);
/* the engine requires at least two sources (we provide
* order.
*/
if (likely(num_descs) &&
- ioat2_alloc_and_lock(&idx, ioat, num_descs+1) == 0)
- /* pass */;
+ ioat2_check_space_lock(ioat, num_descs+1) == 0)
+ idx = ioat->head;
else
return NULL;
i = 0;
struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
struct ioat_ring_ent *desc;
struct ioat_dma_descriptor *hw;
- u16 idx;
- if (ioat2_alloc_and_lock(&idx, ioat, 1) == 0)
- desc = ioat2_get_ring_ent(ioat, idx);
+ if (ioat2_check_space_lock(ioat, 1) == 0)
+ desc = ioat2_get_ring_ent(ioat, ioat->head);
else
return NULL;
if (cap & IOAT_CAP_XOR) {
is_raid_device = true;
dma->max_xor = 8;
- dma->xor_align = 2;
+ dma->xor_align = 6;
dma_cap_set(DMA_XOR, dma->cap_mask);
dma->device_prep_dma_xor = ioat3_prep_xor;
if (cap & IOAT_CAP_PQ) {
is_raid_device = true;
dma_set_maxpq(dma, 8, 0);
- dma->pq_align = 2;
+ dma->pq_align = 6;
dma_cap_set(DMA_PQ, dma->cap_mask);
dma->device_prep_dma_pq = ioat3_prep_pq;
if (!(cap & IOAT_CAP_XOR)) {
dma->max_xor = 8;
- dma->xor_align = 2;
+ dma->xor_align = 6;
dma_cap_set(DMA_XOR, dma->cap_mask);
dma->device_prep_dma_xor = ioat3_prep_pqxor;