Asynchronous raid6 acceleration (part 2 of 3)

index 4496bc6..ce45f3f 100644
@@ -61,17 +61,18 @@ static dma_cookie_t
 iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
 	struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
 {
-	BUG_ON(desc->async_tx.cookie < 0);
-	if (desc->async_tx.cookie > 0) {
-		cookie = desc->async_tx.cookie;
-		desc->async_tx.cookie = 0;
+	struct dma_async_tx_descriptor *tx = &desc->async_tx;
+
+	BUG_ON(tx->cookie < 0);
+	if (tx->cookie > 0) {
+		cookie = tx->cookie;
+		tx->cookie = 0;
 
 		/* call the callback (must not sleep or submit new
 		 * operations to this channel)
 		 */
-		if (desc->async_tx.callback)
-			desc->async_tx.callback(
-				desc->async_tx.callback_param);
+		if (tx->callback)
+			tx->callback(tx->callback_param);
 
 		/* unmap dma addresses
 		 * (unmap_single vs unmap_page?)

@@ -81,7 +82,7 @@ iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
 			struct device *dev =
 				&iop_chan->device->pdev->dev;
 			u32 len = unmap->unmap_len;
-			enum dma_ctrl_flags flags = desc->async_tx.flags;
+			enum dma_ctrl_flags flags = tx->flags;
 			u32 src_cnt;
 			dma_addr_t addr;
 			dma_addr_t dest;

@@ -115,7 +116,7 @@ iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
 	}
 
 	/* run dependent operations */
-	dma_run_dependencies(&desc->async_tx);
+	dma_run_dependencies(tx);
 
 	return cookie;
 }

index ce45f3f..9c752bd 100644
@@ -288,7 +288,12 @@ static void iop_adma_tasklet(unsigned long data)
 {
 	struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data;
 
-	spin_lock(&iop_chan->lock);
+	/* lockdep will flag depedency submissions as potentially
+	 * recursive locking, this is not the case as a dependency
+	 * submission will never recurse a channels submit routine.
+	 * There are checks in async_tx.c to prevent this.
+	 */
+	spin_lock_nested(&iop_chan->lock, SINGLE_DEPTH_NESTING);
 	__iop_adma_slot_cleanup(iop_chan);
 	spin_unlock(&iop_chan->lock);
 }

index 26eefea..1a8c727 100644
@@ -187,11 +187,74 @@ union iop3xx_desc {
 	void *ptr;
 };
 
+/* No support for p+q operations */
+static inline int
+iop_chan_pq_slot_count(size_t len, int src_cnt, int *slots_per_op)
+{
+	BUG();
+	return 0;
+}
+
+static inline void
+iop_desc_init_pq(struct iop_adma_desc_slot *desc, int src_cnt,
+		  unsigned long flags)
+{
+	BUG();
+}
+
+static inline void
+iop_desc_set_pq_addr(struct iop_adma_desc_slot *desc, dma_addr_t *addr)
+{
+	BUG();
+}
+
+static inline void
+iop_desc_set_pq_src_addr(struct iop_adma_desc_slot *desc, int src_idx,
+			 dma_addr_t addr, unsigned char coef)
+{
+	BUG();
+}
+
+static inline int
+iop_chan_pq_zero_sum_slot_count(size_t len, int src_cnt, int *slots_per_op)
+{
+	BUG();
+	return 0;
+}
+
+static inline void
+iop_desc_init_pq_zero_sum(struct iop_adma_desc_slot *desc, int src_cnt,
+			  unsigned long flags)
+{
+	BUG();
+}
+
+static inline void
+iop_desc_set_pq_zero_sum_byte_count(struct iop_adma_desc_slot *desc, u32 len)
+{
+	BUG();
+}
+
+#define iop_desc_set_pq_zero_sum_src_addr iop_desc_set_pq_src_addr
+
+static inline void
+iop_desc_set_pq_zero_sum_addr(struct iop_adma_desc_slot *desc, int pq_idx,
+			      dma_addr_t *src)
+{
+	BUG();
+}
+
 static inline int iop_adma_get_max_xor(void)
 {
 	return 32;
 }
 
+static inline int iop_adma_get_max_pq(void)
+{
+	BUG();
+	return 0;
+}
+
 static inline u32 iop_chan_get_current_descriptor(struct iop_adma_chan *chan)
 {
 	int id = chan->device->id;

@@ -332,6 +395,11 @@ static inline int iop_chan_zero_sum_slot_count(size_t len, int src_cnt,
 	return slot_cnt;
 }
 
+static inline int iop_desc_is_pq(struct iop_adma_desc_slot *desc)
+{
+	return 0;
+}
+
 static inline u32 iop_desc_get_dest_addr(struct iop_adma_desc_slot *desc,
 					struct iop_adma_chan *chan)
 {

@@ -349,6 +417,14 @@ static inline u32 iop_desc_get_dest_addr(struct iop_adma_desc_slot *desc,
 	return 0;
 }
 
+
+static inline u32 iop_desc_get_qdest_addr(struct iop_adma_desc_slot *desc,
+					  struct iop_adma_chan *chan)
+{
+	BUG();
+	return 0;
+}
+
 static inline u32 iop_desc_get_byte_count(struct iop_adma_desc_slot *desc,
 					struct iop_adma_chan *chan)
 {

index 385c6e8..bbe8a04 100644
@@ -106,6 +106,7 @@ struct iop_adma_desc_slot {
 	union {
 		u32 *xor_check_result;
 		u32 *crc32_result;
+		u32 *pq_check_result;
 	};
 };
 

156 lines not on this page · raw message

255 lines not on this page · raw message

index 5a0f4fe..f4c59e5 100644
@@ -31,6 +31,7 @@
 #include <linux/platform_device.h>
 #include <linux/memory.h>
 #include <linux/ioport.h>
+#include <linux/raid/pq.h>
 
 #include <mach/adma.h>
 

@@ -1267,6 +1268,170 @@ out:
 	return err;
 }
 
+#ifdef CONFIG_MD_RAID6_PQ
+static int __devinit
+iop_adma_pq_zero_sum_self_test(struct iop_adma_device *device)
+{
+	/* combined sources, software pq results, and extra hw pq results */
+	struct page *pq[IOP_ADMA_NUM_SRC_TEST+2+2];
+	/* ptr to the extra hw pq buffers defined above */
+	struct page **pq_hw = &pq[IOP_ADMA_NUM_SRC_TEST+2];
+	/* address conversion buffers (dma_map / page_address) */
+	void *pq_sw[IOP_ADMA_NUM_SRC_TEST+2];
+	dma_addr_t pq_src[IOP_ADMA_NUM_SRC_TEST];
+	dma_addr_t pq_dest[2];
+
+	int i;
+	struct dma_async_tx_descriptor *tx;
+	struct dma_chan *dma_chan;
+	dma_cookie_t cookie;
+	u32 zero_sum_result;
+	int err = 0;
+	struct device *dev;
+
+	dev_dbg(device->common.dev, "%s\n", __func__);
+
+	for (i = 0; i < ARRAY_SIZE(pq); i++) {
+		pq[i] = alloc_page(GFP_KERNEL);
+		if (!pq[i]) {
+			while (i--)
+				__free_page(pq[i]);
+			return -ENOMEM;
+		}
+	}
+
+	/* Fill in src buffers */
+	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
+		pq_sw[i] = page_address(pq[i]);
+		memset(pq_sw[i], 0x11111111 * (1<<i), PAGE_SIZE);
+	}
+	pq_sw[i] = page_address(pq[i]);
+	pq_sw[i+1] = page_address(pq[i+1]);
+
+	dma_chan = container_of(device->common.channels.next,
+				struct dma_chan,
+				device_node);
+	if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
+		err = -ENODEV;
+		goto out;
+	}
+
+	dev = dma_chan->device->dev;
+
+	/* initialize the dests */
+	memset(page_address(pq_hw[0]), 0 , PAGE_SIZE);
+	memset(page_address(pq_hw[1]), 0 , PAGE_SIZE);
+
+	/* test pq */
+	pq_dest[0] = dma_map_page(dev, pq_hw[0], 0, PAGE_SIZE, DMA_FROM_DEVICE);
+	pq_dest[1] = dma_map_page(dev, pq_hw[1], 0, PAGE_SIZE, DMA_FROM_DEVICE);
+	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
+		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
+					 DMA_TO_DEVICE);
+
+	tx = iop_adma_prep_dma_pq(dma_chan, pq_dest, pq_src,
+				  IOP_ADMA_NUM_SRC_TEST, (u8 *)raid6_gfexp,
+				  PAGE_SIZE,
+				  DMA_PREP_INTERRUPT |
+				  DMA_CTRL_ACK);
+
+	cookie = iop_adma_tx_submit(tx);
+	iop_adma_issue_pending(dma_chan);
+	msleep(8);
+
+	if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
+		DMA_SUCCESS) {
+		dev_err(dev, "Self-test pq timed out, disabling\n");
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+	raid6_call.gen_syndrome(IOP_ADMA_NUM_SRC_TEST+2, PAGE_SIZE, pq_sw);
+
+	if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST],
+		   page_address(pq_hw[0]), PAGE_SIZE) != 0) {
+		dev_err(dev, "Self-test p failed compare, disabling\n");
+		err = -ENODEV;
+		goto free_resources;
+	}
+	if (memcmp(pq_sw[IOP_ADMA_NUM_SRC_TEST+1],
+		   page_address(pq_hw[1]), PAGE_SIZE) != 0) {
+		dev_err(dev, "Self-test q failed compare, disabling\n");
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+	/* test correct zero sum using the software generated pq values */
+	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
+		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
+					 DMA_TO_DEVICE);
+
+	zero_sum_result = ~0;
+	tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
+				      pq_src, IOP_ADMA_NUM_SRC_TEST,
+				      raid6_gfexp, PAGE_SIZE, &zero_sum_result,
+				      DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
+
+	cookie = iop_adma_tx_submit(tx);
+	iop_adma_issue_pending(dma_chan);
+	msleep(8);
+
+	if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
+		DMA_SUCCESS) {
+		dev_err(dev, "Self-test pq-zero-sum timed out, disabling\n");
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+	if (zero_sum_result != 0) {
+		dev_err(dev, "Self-test pq-zero-sum failed to validate: %x\n",
+			zero_sum_result);
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+	/* test incorrect zero sum */
+	i = IOP_ADMA_NUM_SRC_TEST;
+	memset(pq_sw[i] + 100, 0, 100);
+	memset(pq_sw[i+1] + 200, 0, 200);
+	for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 2; i++)
+		pq_src[i] = dma_map_page(dev, pq[i], 0, PAGE_SIZE,
+					 DMA_TO_DEVICE);
+
+	zero_sum_result = 0;
+	tx = iop_adma_prep_dma_pq_val(dma_chan, &pq_src[IOP_ADMA_NUM_SRC_TEST],
+				      pq_src, IOP_ADMA_NUM_SRC_TEST,
+				      raid6_gfexp, PAGE_SIZE, &zero_sum_result,
+				      DMA_PREP_INTERRUPT|DMA_CTRL_ACK);
+
+	cookie = iop_adma_tx_submit(tx);
+	iop_adma_issue_pending(dma_chan);
+	msleep(8);
+
+	if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
+		DMA_SUCCESS) {
+		dev_err(dev, "Self-test !pq-zero-sum timed out, disabling\n");
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+	if (zero_sum_result != (SUM_CHECK_P_RESULT | SUM_CHECK_Q_RESULT)) {
+		dev_err(dev, "Self-test !pq-zero-sum failed to validate: %x\n",
+			zero_sum_result);
+		err = -ENODEV;
+		goto free_resources;
+	}
+
+free_resources:
+	iop_adma_free_chan_resources(dma_chan);
+out:
+	i = ARRAY_SIZE(pq);
+	while (i--)
+		__free_page(pq[i]);
+	return err;
+}
+#endif
+
 static int __devexit iop_adma_remove(struct platform_device *dev)
 {
 	struct iop_adma_device *device = platform_get_drvdata(dev);

@@ -1417,13 +1582,28 @@ static int __devinit iop_adma_probe(struct platform_device *pdev)
 	}
 
 	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
-		dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
+	    dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
 		ret = iop_adma_xor_val_self_test(adev);
 		dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
 		if (ret)
 			goto err_free_iop_chan;
 	}
 
+	if (dma_has_cap(DMA_PQ, dma_dev->cap_mask) &&
+	    dma_has_cap(DMA_PQ_VAL, dma_dev->cap_mask)) {
+		#ifdef CONFIG_MD_RAID6_PQ
+		ret = iop_adma_pq_zero_sum_self_test(adev);
+		dev_dbg(&pdev->dev, "pq self test returned %d\n", ret);
+		#else
+		/* can not test raid6, so do not publish capability */
+		dma_cap_clear(DMA_PQ, dma_dev->cap_mask);
+		dma_cap_clear(DMA_PQ_VAL, dma_dev->cap_mask);
+		ret = 0;
+		#endif
+		if (ret)
+			goto err_free_iop_chan;
+	}
+
 	dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
 	  "( %s%s%s%s%s%s%s%s%s%s)\n",
 	  dma_has_cap(DMA_PQ, dma_dev->cap_mask) ? "pq " : "",