[PATCH v3 0/5] PCI: functions/pci-epf-test: Add DMA data transfer

[PATCH v3 0/5] PCI: functions/pci-epf-test: Add DMA data transfer

[Kishon Vijay Abraham I]

Patch series uses dma engine APIs in pci-epf-test to transfer data using
DMA. It also adds an option "-d" in pcitest for the user to indicate
whether DMA has to be used for data transfer. This also prints
throughput information for data transfer.

Changes from v2:
*) Removed use of "bool" in uapi/linux/pcitest.h as that seems to fail
   for certain configs (header should be self contained). It was failing
   for "x86_64 allmodconfig".
*) Modified function comments added before DMA transfer functions.

Changes from v1:
*) Fixed some of the function names from pci_epf_* to pci_epf_test_*
since the DMA support is now been moved to pci-epf-test.c

Kishon Vijay Abraham I (5):
  PCI: endpoint: functions/pci-epf-test: Add DMA support to transfer
    data
  PCI: endpoint: functions/pci-epf-test: Print throughput information
  misc: pci_endpoint_test: Use streaming DMA APIs for buffer allocation
  tools: PCI: Add 'd' command line option to support DMA
  misc: pci_endpoint_test: Add support to get DMA option from userspace

 drivers/misc/pci_endpoint_test.c              | 165 ++++++++++--
 drivers/pci/endpoint/functions/pci-epf-test.c | 253 +++++++++++++++++-
 include/uapi/linux/pcitest.h                  |   7 +
 tools/pci/pcitest.c                           |  23 +-
 4 files changed, 417 insertions(+), 31 deletions(-)

-- 
2.17.1

[PATCH v3 1/5] PCI: endpoint: functions/pci-epf-test: Add DMA support to transfer data

[Kishon Vijay Abraham I]

Use dmaengine API and add support for transferring data using DMA.

Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Tested-by: Alan Mikhak <alan.mikhak@sifive.com>
---
 drivers/pci/endpoint/functions/pci-epf-test.c | 205 +++++++++++++++++-
 1 file changed, 202 insertions(+), 3 deletions(-)

diff --git a/drivers/pci/endpoint/functions/pci-epf-test.c b/drivers/pci/endpoint/functions/pci-epf-test.c
index bddff15052cc..c9d9d1780e66 100644
--- a/drivers/pci/endpoint/functions/pci-epf-test.c
+++ b/drivers/pci/endpoint/functions/pci-epf-test.c
@@ -8,6 +8,7 @@
 
 #include <linux/crc32.h>
 #include <linux/delay.h>
+#include <linux/dmaengine.h>
 #include <linux/io.h>
 #include <linux/module.h>
 #include <linux/slab.h>
@@ -39,6 +40,8 @@
 #define STATUS_SRC_ADDR_INVALID		BIT(7)
 #define STATUS_DST_ADDR_INVALID		BIT(8)
 
+#define FLAG_USE_DMA			BIT(0)
+
 #define TIMER_RESOLUTION		1
 
 static struct workqueue_struct *kpcitest_workqueue;
@@ -48,6 +51,9 @@ struct pci_epf_test {
 	struct pci_epf		*epf;
 	enum pci_barno		test_reg_bar;
 	struct delayed_work	cmd_handler;
+	struct dma_chan		*dma_chan;
+	struct completion	transfer_complete;
+	bool			dma_supported;
 	const struct pci_epc_features *epc_features;
 };
 
@@ -61,6 +67,7 @@ struct pci_epf_test_reg {
 	u32	checksum;
 	u32	irq_type;
 	u32	irq_number;
+	u32	flags;
 } __packed;
 
 static struct pci_epf_header test_header = {
@@ -72,9 +79,121 @@ static struct pci_epf_header test_header = {
 
 static size_t bar_size[] = { 512, 512, 1024, 16384, 131072, 1048576 };
 
+static void pci_epf_test_dma_callback(void *param)
+{
+	struct pci_epf_test *epf_test = param;
+
+	complete(&epf_test->transfer_complete);
+}
+
+/**
+ * pci_epf_test_data_transfer() - Function that uses dmaengine API to transfer
+ *				  data between PCIe EP and remote PCIe RC
+ * @epf_test: the EPF test device that performs the data transfer operation
+ * @dma_dst: The destination address of the data transfer. It can be a physical
+ *	     address given by pci_epc_mem_alloc_addr or DMA mapping APIs.
+ * @dma_src: The source address of the data transfer. It can be a physical
+ *	     address given by pci_epc_mem_alloc_addr or DMA mapping APIs.
+ * @len: The size of the data transfer
+ *
+ * Function that uses dmaengine API to transfer data between PCIe EP and remote
+ * PCIe RC. The source and destination address can be a physical address given
+ * by pci_epc_mem_alloc_addr or the one obtained using DMA mapping APIs.
+ *
+ * The function returns '0' on success and negative value on failure.
+ */
+static int pci_epf_test_data_transfer(struct pci_epf_test *epf_test,
+				      dma_addr_t dma_dst, dma_addr_t dma_src,
+				      size_t len)
+{
+	enum dma_ctrl_flags flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT;
+	struct dma_chan *chan = epf_test->dma_chan;
+	struct pci_epf *epf = epf_test->epf;
+	struct dma_async_tx_descriptor *tx;
+	struct device *dev = &epf->dev;
+	dma_cookie_t cookie;
+	int ret;
+
+	if (IS_ERR_OR_NULL(chan)) {
+		dev_err(dev, "Invalid DMA memcpy channel\n");
+		return -EINVAL;
+	}
+
+	tx = dmaengine_prep_dma_memcpy(chan, dma_dst, dma_src, len, flags);
+	if (!tx) {
+		dev_err(dev, "Failed to prepare DMA memcpy\n");
+		return -EIO;
+	}
+
+	tx->callback = pci_epf_test_dma_callback;
+	tx->callback_param = epf_test;
+	cookie = tx->tx_submit(tx);
+	reinit_completion(&epf_test->transfer_complete);
+
+	ret = dma_submit_error(cookie);
+	if (ret) {
+		dev_err(dev, "Failed to do DMA tx_submit %d\n", cookie);
+		return -EIO;
+	}
+
+	dma_async_issue_pending(chan);
+	ret = wait_for_completion_interruptible(&epf_test->transfer_complete);
+	if (ret < 0) {
+		dmaengine_terminate_sync(chan);
+		dev_err(dev, "DMA wait_for_completion_timeout\n");
+		return -ETIMEDOUT;
+	}
+
+	return 0;
+}
+
+/**
+ * pci_epf_test_init_dma_chan() - Function to initialize EPF test DMA channel
+ * @epf_test: the EPF test device that performs data transfer operation
+ *
+ * Function to initialize EPF test DMA channel.
+ */
+static int pci_epf_test_init_dma_chan(struct pci_epf_test *epf_test)
+{
+	struct pci_epf *epf = epf_test->epf;
+	struct device *dev = &epf->dev;
+	struct dma_chan *dma_chan;
+	dma_cap_mask_t mask;
+	int ret;
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_MEMCPY, mask);
+
+	dma_chan = dma_request_chan_by_mask(&mask);
+	if (IS_ERR(dma_chan)) {
+		ret = PTR_ERR(dma_chan);
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get DMA channel\n");
+		return ret;
+	}
+	init_completion(&epf_test->transfer_complete);
+
+	epf_test->dma_chan = dma_chan;
+
+	return 0;
+}
+
+/**
+ * pci_epf_test_clean_dma_chan() - Function to cleanup EPF test DMA channel
+ * @epf: the EPF test device that performs data transfer operation
+ *
+ * Helper to cleanup EPF test DMA channel.
+ */
+static void pci_epf_test_clean_dma_chan(struct pci_epf_test *epf_test)
+{
+	dma_release_channel(epf_test->dma_chan);
+	epf_test->dma_chan = NULL;
+}
+
 static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 {
 	int ret;
+	bool use_dma;
 	void __iomem *src_addr;
 	void __iomem *dst_addr;
 	phys_addr_t src_phys_addr;
@@ -117,8 +236,23 @@ static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 		goto err_dst_addr;
 	}
 
-	memcpy(dst_addr, src_addr, reg->size);
+	use_dma = !!(reg->flags & FLAG_USE_DMA);
+	if (use_dma) {
+		if (!epf_test->dma_supported) {
+			dev_err(dev, "Cannot transfer data using DMA\n");
+			ret = -EINVAL;
+			goto err_map_addr;
+		}
 
+		ret = pci_epf_test_data_transfer(epf_test, dst_phys_addr,
+						 src_phys_addr, reg->size);
+		if (ret)
+			dev_err(dev, "Data transfer failed\n");
+	} else {
+		memcpy(dst_addr, src_addr, reg->size);
+	}
+
+err_map_addr:
 	pci_epc_unmap_addr(epc, epf->func_no, dst_phys_addr);
 
 err_dst_addr:
@@ -140,10 +274,13 @@ static int pci_epf_test_read(struct pci_epf_test *epf_test)
 	void __iomem *src_addr;
 	void *buf;
 	u32 crc32;
+	bool use_dma;
 	phys_addr_t phys_addr;
+	phys_addr_t dst_phys_addr;
 	struct pci_epf *epf = epf_test->epf;
 	struct device *dev = &epf->dev;
 	struct pci_epc *epc = epf->epc;
+	struct device *dma_dev = epf->epc->dev.parent;
 	enum pci_barno test_reg_bar = epf_test->test_reg_bar;
 	struct pci_epf_test_reg *reg = epf_test->reg[test_reg_bar];
 
@@ -169,12 +306,38 @@ static int pci_epf_test_read(struct pci_epf_test *epf_test)
 		goto err_map_addr;
 	}
 
-	memcpy_fromio(buf, src_addr, reg->size);
+	use_dma = !!(reg->flags & FLAG_USE_DMA);
+	if (use_dma) {
+		if (!epf_test->dma_supported) {
+			dev_err(dev, "Cannot transfer data using DMA\n");
+			ret = -EINVAL;
+			goto err_map_addr;
+		}
+
+		dst_phys_addr = dma_map_single(dma_dev, buf, reg->size,
+					       DMA_FROM_DEVICE);
+		if (dma_mapping_error(dma_dev, dst_phys_addr)) {
+			dev_err(dev, "Failed to map destination buffer addr\n");
+			ret = -ENOMEM;
+			goto err_dma_map;
+		}
+
+		ret = pci_epf_test_data_transfer(epf_test, dst_phys_addr,
+						 phys_addr, reg->size);
+		if (ret)
+			dev_err(dev, "Data transfer failed\n");
+
+		dma_unmap_single(dma_dev, dst_phys_addr, reg->size,
+				 DMA_FROM_DEVICE);
+	} else {
+		memcpy_fromio(buf, src_addr, reg->size);
+	}
 
 	crc32 = crc32_le(~0, buf, reg->size);
 	if (crc32 != reg->checksum)
 		ret = -EIO;
 
+err_dma_map:
 	kfree(buf);
 
 err_map_addr:
@@ -192,10 +355,13 @@ static int pci_epf_test_write(struct pci_epf_test *epf_test)
 	int ret;
 	void __iomem *dst_addr;
 	void *buf;
+	bool use_dma;
 	phys_addr_t phys_addr;
+	phys_addr_t src_phys_addr;
 	struct pci_epf *epf = epf_test->epf;
 	struct device *dev = &epf->dev;
 	struct pci_epc *epc = epf->epc;
+	struct device *dma_dev = epf->epc->dev.parent;
 	enum pci_barno test_reg_bar = epf_test->test_reg_bar;
 	struct pci_epf_test_reg *reg = epf_test->reg[test_reg_bar];
 
@@ -224,7 +390,32 @@ static int pci_epf_test_write(struct pci_epf_test *epf_test)
 	get_random_bytes(buf, reg->size);
 	reg->checksum = crc32_le(~0, buf, reg->size);
 
-	memcpy_toio(dst_addr, buf, reg->size);
+	use_dma = !!(reg->flags & FLAG_USE_DMA);
+	if (use_dma) {
+		if (!epf_test->dma_supported) {
+			dev_err(dev, "Cannot transfer data using DMA\n");
+			ret = -EINVAL;
+			goto err_map_addr;
+		}
+
+		src_phys_addr = dma_map_single(dma_dev, buf, reg->size,
+					       DMA_TO_DEVICE);
+		if (dma_mapping_error(dma_dev, src_phys_addr)) {
+			dev_err(dev, "Failed to map source buffer addr\n");
+			ret = -ENOMEM;
+			goto err_dma_map;
+		}
+
+		ret = pci_epf_test_data_transfer(epf_test, phys_addr,
+						 src_phys_addr, reg->size);
+		if (ret)
+			dev_err(dev, "Data transfer failed\n");
+
+		dma_unmap_single(dma_dev, src_phys_addr, reg->size,
+				 DMA_TO_DEVICE);
+	} else {
+		memcpy_toio(dst_addr, buf, reg->size);
+	}
 
 	/*
 	 * wait 1ms inorder for the write to complete. Without this delay L3
@@ -232,6 +423,7 @@ static int pci_epf_test_write(struct pci_epf_test *epf_test)
 	 */
 	usleep_range(1000, 2000);
 
+err_dma_map:
 	kfree(buf);
 
 err_map_addr:
@@ -380,6 +572,7 @@ static void pci_epf_test_unbind(struct pci_epf *epf)
 	int bar;
 
 	cancel_delayed_work(&epf_test->cmd_handler);
+	pci_epf_test_clean_dma_chan(epf_test);
 	pci_epc_stop(epc);
 	for (bar = 0; bar < PCI_STD_NUM_BARS; bar++) {
 		epf_bar = &epf->bar[bar];
@@ -550,6 +743,12 @@ static int pci_epf_test_bind(struct pci_epf *epf)
 		}
 	}
 
+	epf_test->dma_supported = true;
+
+	ret = pci_epf_test_init_dma_chan(epf_test);
+	if (ret)
+		epf_test->dma_supported = false;
+
 	if (linkup_notifier) {
 		epf->nb.notifier_call = pci_epf_test_notifier;
 		pci_epc_register_notifier(epc, &epf->nb);
-- 
2.17.1

[PATCH v3 2/5] PCI: endpoint: functions/pci-epf-test: Print throughput information

[Kishon Vijay Abraham I]

Print throughput information in KB/s after every completed transfer,
including information on whether DMA is used or not.

Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Tested-by: Alan Mikhak <alan.mikhak@sifive.com>
---
 drivers/pci/endpoint/functions/pci-epf-test.c | 48 +++++++++++++++++++
 1 file changed, 48 insertions(+)

diff --git a/drivers/pci/endpoint/functions/pci-epf-test.c b/drivers/pci/endpoint/functions/pci-epf-test.c
index c9d9d1780e66..108e4b40e67a 100644
--- a/drivers/pci/endpoint/functions/pci-epf-test.c
+++ b/drivers/pci/endpoint/functions/pci-epf-test.c
@@ -190,6 +190,36 @@ static void pci_epf_test_clean_dma_chan(struct pci_epf_test *epf_test)
 	epf_test->dma_chan = NULL;
 }
 
+static void pci_epf_test_print_rate(const char *ops, u64 size,
+				    struct timespec64 *start,
+				    struct timespec64 *end, bool dma)
+{
+	struct timespec64 ts;
+	u64 rate, ns;
+
+	ts = timespec64_sub(*end, *start);
+
+	/* convert both size (stored in 'rate') and time in terms of 'ns' */
+	ns = timespec64_to_ns(&ts);
+	rate = size * NSEC_PER_SEC;
+
+	/* Divide both size (stored in 'rate') and ns by a common factor */
+	while (ns > UINT_MAX) {
+		rate >>= 1;
+		ns >>= 1;
+	}
+
+	if (!ns)
+		return;
+
+	/* calculate the rate */
+	do_div(rate, (uint32_t)ns);
+
+	pr_info("\n%s => Size: %llu bytes\t DMA: %s\t Time: %llu.%09u seconds\t"
+		"Rate: %llu KB/s\n", ops, size, dma ? "YES" : "NO",
+		(u64)ts.tv_sec, (u32)ts.tv_nsec, rate / 1024);
+}
+
 static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 {
 	int ret;
@@ -198,6 +228,7 @@ static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 	void __iomem *dst_addr;
 	phys_addr_t src_phys_addr;
 	phys_addr_t dst_phys_addr;
+	struct timespec64 start, end;
 	struct pci_epf *epf = epf_test->epf;
 	struct device *dev = &epf->dev;
 	struct pci_epc *epc = epf->epc;
@@ -236,6 +267,7 @@ static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 		goto err_dst_addr;
 	}
 
+	ktime_get_ts64(&start);
 	use_dma = !!(reg->flags & FLAG_USE_DMA);
 	if (use_dma) {
 		if (!epf_test->dma_supported) {
@@ -251,6 +283,8 @@ static int pci_epf_test_copy(struct pci_epf_test *epf_test)
 	} else {
 		memcpy(dst_addr, src_addr, reg->size);
 	}
+	ktime_get_ts64(&end);
+	pci_epf_test_print_rate("COPY", reg->size, &start, &end, use_dma);
 
 err_map_addr:
 	pci_epc_unmap_addr(epc, epf->func_no, dst_phys_addr);
@@ -277,6 +311,7 @@ static int pci_epf_test_read(struct pci_epf_test *epf_test)
 	bool use_dma;
 	phys_addr_t phys_addr;
 	phys_addr_t dst_phys_addr;
+	struct timespec64 start, end;
 	struct pci_epf *epf = epf_test->epf;
 	struct device *dev = &epf->dev;
 	struct pci_epc *epc = epf->epc;
@@ -322,17 +357,23 @@ static int pci_epf_test_read(struct pci_epf_test *epf_test)
 			goto err_dma_map;
 		}
 
+		ktime_get_ts64(&start);
 		ret = pci_epf_test_data_transfer(epf_test, dst_phys_addr,
 						 phys_addr, reg->size);
 		if (ret)
 			dev_err(dev, "Data transfer failed\n");
+		ktime_get_ts64(&end);
 
 		dma_unmap_single(dma_dev, dst_phys_addr, reg->size,
 				 DMA_FROM_DEVICE);
 	} else {
+		ktime_get_ts64(&start);
 		memcpy_fromio(buf, src_addr, reg->size);
+		ktime_get_ts64(&end);
 	}
 
+	pci_epf_test_print_rate("READ", reg->size, &start, &end, use_dma);
+
 	crc32 = crc32_le(~0, buf, reg->size);
 	if (crc32 != reg->checksum)
 		ret = -EIO;
@@ -358,6 +399,7 @@ static int pci_epf_test_write(struct pci_epf_test *epf_test)
 	bool use_dma;
 	phys_addr_t phys_addr;
 	phys_addr_t src_phys_addr;
+	struct timespec64 start, end;
 	struct pci_epf *epf = epf_test->epf;
 	struct device *dev = &epf->dev;
 	struct pci_epc *epc = epf->epc;
@@ -406,17 +448,23 @@ static int pci_epf_test_write(struct pci_epf_test *epf_test)
 			goto err_dma_map;
 		}
 
+		ktime_get_ts64(&start);
 		ret = pci_epf_test_data_transfer(epf_test, phys_addr,
 						 src_phys_addr, reg->size);
 		if (ret)
 			dev_err(dev, "Data transfer failed\n");
+		ktime_get_ts64(&end);
 
 		dma_unmap_single(dma_dev, src_phys_addr, reg->size,
 				 DMA_TO_DEVICE);
 	} else {
+		ktime_get_ts64(&start);
 		memcpy_toio(dst_addr, buf, reg->size);
+		ktime_get_ts64(&end);
 	}
 
+	pci_epf_test_print_rate("WRITE", reg->size, &start, &end, use_dma);
+
 	/*
 	 * wait 1ms inorder for the write to complete. Without this delay L3
 	 * error in observed in the host system.
-- 
2.17.1

[PATCH v3 3/5] misc: pci_endpoint_test: Use streaming DMA APIs for buffer allocation

[Kishon Vijay Abraham I]

Use streaming DMA APIs (dma_map_single/dma_unmap_single) for buffers
transmitted/received by the endpoint device instead of allocating
a coherent memory. Also add default_data to set the alignment to
4KB since dma_map_single might not return a 4KB aligned address.

Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Tested-by: Alan Mikhak <alan.mikhak@sifive.com>
---
 drivers/misc/pci_endpoint_test.c | 100 ++++++++++++++++++++++++-------
 1 file changed, 79 insertions(+), 21 deletions(-)

diff --git a/drivers/misc/pci_endpoint_test.c b/drivers/misc/pci_endpoint_test.c
index a5e317073d95..5998df1c84e9 100644
--- a/drivers/misc/pci_endpoint_test.c
+++ b/drivers/misc/pci_endpoint_test.c
@@ -341,14 +341,22 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 		goto err;
 	}
 
-	orig_src_addr = dma_alloc_coherent(dev, size + alignment,
-					   &orig_src_phys_addr, GFP_KERNEL);
+	orig_src_addr = kzalloc(size + alignment, GFP_KERNEL);
 	if (!orig_src_addr) {
 		dev_err(dev, "Failed to allocate source buffer\n");
 		ret = false;
 		goto err;
 	}
 
+	get_random_bytes(orig_src_addr, size + alignment);
+	orig_src_phys_addr = dma_map_single(dev, orig_src_addr,
+					    size + alignment, DMA_TO_DEVICE);
+	if (dma_mapping_error(dev, orig_src_phys_addr)) {
+		dev_err(dev, "failed to map source buffer address\n");
+		ret = false;
+		goto err_src_phys_addr;
+	}
+
 	if (alignment && !IS_ALIGNED(orig_src_phys_addr, alignment)) {
 		src_phys_addr = PTR_ALIGN(orig_src_phys_addr, alignment);
 		offset = src_phys_addr - orig_src_phys_addr;
@@ -364,15 +372,21 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_UPPER_SRC_ADDR,
 				 upper_32_bits(src_phys_addr));
 
-	get_random_bytes(src_addr, size);
 	src_crc32 = crc32_le(~0, src_addr, size);
 
-	orig_dst_addr = dma_alloc_coherent(dev, size + alignment,
-					   &orig_dst_phys_addr, GFP_KERNEL);
+	orig_dst_addr = kzalloc(size + alignment, GFP_KERNEL);
 	if (!orig_dst_addr) {
 		dev_err(dev, "Failed to allocate destination address\n");
 		ret = false;
-		goto err_orig_src_addr;
+		goto err_dst_addr;
+	}
+
+	orig_dst_phys_addr = dma_map_single(dev, orig_dst_addr,
+					    size + alignment, DMA_FROM_DEVICE);
+	if (dma_mapping_error(dev, orig_dst_phys_addr)) {
+		dev_err(dev, "failed to map destination buffer address\n");
+		ret = false;
+		goto err_dst_phys_addr;
 	}
 
 	if (alignment && !IS_ALIGNED(orig_dst_phys_addr, alignment)) {
@@ -399,16 +413,22 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 
 	wait_for_completion(&test->irq_raised);
 
+	dma_unmap_single(dev, orig_dst_phys_addr, size + alignment,
+			 DMA_FROM_DEVICE);
+
 	dst_crc32 = crc32_le(~0, dst_addr, size);
 	if (dst_crc32 == src_crc32)
 		ret = true;
 
-	dma_free_coherent(dev, size + alignment, orig_dst_addr,
-			  orig_dst_phys_addr);
+err_dst_phys_addr:
+	kfree(orig_dst_addr);
 
-err_orig_src_addr:
-	dma_free_coherent(dev, size + alignment, orig_src_addr,
-			  orig_src_phys_addr);
+err_dst_addr:
+	dma_unmap_single(dev, orig_src_phys_addr, size + alignment,
+			 DMA_TO_DEVICE);
+
+err_src_phys_addr:
+	kfree(orig_src_addr);
 
 err:
 	return ret;
@@ -436,14 +456,23 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 		goto err;
 	}
 
-	orig_addr = dma_alloc_coherent(dev, size + alignment, &orig_phys_addr,
-				       GFP_KERNEL);
+	orig_addr = kzalloc(size + alignment, GFP_KERNEL);
 	if (!orig_addr) {
 		dev_err(dev, "Failed to allocate address\n");
 		ret = false;
 		goto err;
 	}
 
+	get_random_bytes(orig_addr, size + alignment);
+
+	orig_phys_addr = dma_map_single(dev, orig_addr, size + alignment,
+					DMA_TO_DEVICE);
+	if (dma_mapping_error(dev, orig_phys_addr)) {
+		dev_err(dev, "failed to map source buffer address\n");
+		ret = false;
+		goto err_phys_addr;
+	}
+
 	if (alignment && !IS_ALIGNED(orig_phys_addr, alignment)) {
 		phys_addr =  PTR_ALIGN(orig_phys_addr, alignment);
 		offset = phys_addr - orig_phys_addr;
@@ -453,8 +482,6 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 		addr = orig_addr;
 	}
 
-	get_random_bytes(addr, size);
-
 	crc32 = crc32_le(~0, addr, size);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_CHECKSUM,
 				 crc32);
@@ -477,7 +504,11 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 	if (reg & STATUS_READ_SUCCESS)
 		ret = true;
 
-	dma_free_coherent(dev, size + alignment, orig_addr, orig_phys_addr);
+	dma_unmap_single(dev, orig_phys_addr, size + alignment,
+			 DMA_TO_DEVICE);
+
+err_phys_addr:
+	kfree(orig_addr);
 
 err:
 	return ret;
@@ -504,14 +535,21 @@ static bool pci_endpoint_test_read(struct pci_endpoint_test *test, size_t size)
 		goto err;
 	}
 
-	orig_addr = dma_alloc_coherent(dev, size + alignment, &orig_phys_addr,
-				       GFP_KERNEL);
+	orig_addr = kzalloc(size + alignment, GFP_KERNEL);
 	if (!orig_addr) {
 		dev_err(dev, "Failed to allocate destination address\n");
 		ret = false;
 		goto err;
 	}
 
+	orig_phys_addr = dma_map_single(dev, orig_addr, size + alignment,
+					DMA_FROM_DEVICE);
+	if (dma_mapping_error(dev, orig_phys_addr)) {
+		dev_err(dev, "failed to map source buffer address\n");
+		ret = false;
+		goto err_phys_addr;
+	}
+
 	if (alignment && !IS_ALIGNED(orig_phys_addr, alignment)) {
 		phys_addr = PTR_ALIGN(orig_phys_addr, alignment);
 		offset = phys_addr - orig_phys_addr;
@@ -535,11 +573,15 @@ static bool pci_endpoint_test_read(struct pci_endpoint_test *test, size_t size)
 
 	wait_for_completion(&test->irq_raised);
 
+	dma_unmap_single(dev, orig_phys_addr, size + alignment,
+			 DMA_FROM_DEVICE);
+
 	crc32 = crc32_le(~0, addr, size);
 	if (crc32 == pci_endpoint_test_readl(test, PCI_ENDPOINT_TEST_CHECKSUM))
 		ret = true;
 
-	dma_free_coherent(dev, size + alignment, orig_addr, orig_phys_addr);
+err_phys_addr:
+	kfree(orig_addr);
 err:
 	return ret;
 }
@@ -667,6 +709,12 @@ static int pci_endpoint_test_probe(struct pci_dev *pdev,
 	init_completion(&test->irq_raised);
 	mutex_init(&test->mutex);
 
+	if ((dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48)) != 0) &&
+	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0) {
+		dev_err(dev, "Cannot set DMA mask\n");
+		return -EINVAL;
+	}
+
 	err = pci_enable_device(pdev);
 	if (err) {
 		dev_err(dev, "Cannot enable PCI device\n");
@@ -783,6 +831,12 @@ static void pci_endpoint_test_remove(struct pci_dev *pdev)
 	pci_disable_device(pdev);
 }
 
+static const struct pci_endpoint_test_data default_data = {
+	.test_reg_bar = BAR_0,
+	.alignment = SZ_4K,
+	.irq_type = IRQ_TYPE_MSI,
+};
+
 static const struct pci_endpoint_test_data am654_data = {
 	.test_reg_bar = BAR_2,
 	.alignment = SZ_64K,
@@ -790,8 +844,12 @@ static const struct pci_endpoint_test_data am654_data = {
 };
 
 static const struct pci_device_id pci_endpoint_test_tbl[] = {
-	{ PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_DRA74x) },
-	{ PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_DRA72x) },
+	{ PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_DRA74x),
+	  .driver_data = (kernel_ulong_t)&default_data,
+	},
+	{ PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_DRA72x),
+	  .driver_data = (kernel_ulong_t)&default_data,
+	},
 	{ PCI_DEVICE(PCI_VENDOR_ID_FREESCALE, 0x81c0) },
 	{ PCI_DEVICE_DATA(SYNOPSYS, EDDA, NULL) },
 	{ PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_AM654),
-- 
2.17.1

[PATCH v3 4/5] tools: PCI: Add 'd' command line option to support DMA

[Kishon Vijay Abraham I]

Add a new command line option 'd' to use DMA for data transfers.
It should be used with read, write or copy commands.

Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Tested-by: Alan Mikhak <alan.mikhak@sifive.com>
---
 include/uapi/linux/pcitest.h |  7 +++++++
 tools/pci/pcitest.c          | 23 +++++++++++++++++++----
 2 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/include/uapi/linux/pcitest.h b/include/uapi/linux/pcitest.h
index cbf422e56696..8b868761f8b4 100644
--- a/include/uapi/linux/pcitest.h
+++ b/include/uapi/linux/pcitest.h
@@ -20,4 +20,11 @@
 #define PCITEST_SET_IRQTYPE	_IOW('P', 0x8, int)
 #define PCITEST_GET_IRQTYPE	_IO('P', 0x9)
 
+#define PCITEST_FLAGS_USE_DMA	0x00000001
+
+struct pci_endpoint_test_xfer_param {
+	unsigned long size;
+	unsigned char flags;
+};
+
 #endif /* __UAPI_LINUX_PCITEST_H */
diff --git a/tools/pci/pcitest.c b/tools/pci/pcitest.c
index 32b7c6f9043d..5e3b6368c5e0 100644
--- a/tools/pci/pcitest.c
+++ b/tools/pci/pcitest.c
@@ -34,10 +34,12 @@ struct pci_test {
 	bool		write;
 	bool		copy;
 	unsigned long	size;
+	bool		use_dma;
 };
 
 static int run_test(struct pci_test *test)
 {
+	struct pci_endpoint_test_xfer_param param;
 	int ret = -EINVAL;
 	int fd;
 
@@ -102,7 +104,10 @@ static int run_test(struct pci_test *test)
 	}
 
 	if (test->write) {
-		ret = ioctl(fd, PCITEST_WRITE, test->size);
+		param.size = test->size;
+		if (test->use_dma)
+			param.flags = PCITEST_FLAGS_USE_DMA;
+		ret = ioctl(fd, PCITEST_WRITE, &param);
 		fprintf(stdout, "WRITE (%7ld bytes):\t\t", test->size);
 		if (ret < 0)
 			fprintf(stdout, "TEST FAILED\n");
@@ -111,7 +116,10 @@ static int run_test(struct pci_test *test)
 	}
 
 	if (test->read) {
-		ret = ioctl(fd, PCITEST_READ, test->size);
+		param.size = test->size;
+		if (test->use_dma)
+			param.flags = PCITEST_FLAGS_USE_DMA;
+		ret = ioctl(fd, PCITEST_READ, &param);
 		fprintf(stdout, "READ (%7ld bytes):\t\t", test->size);
 		if (ret < 0)
 			fprintf(stdout, "TEST FAILED\n");
@@ -120,7 +128,10 @@ static int run_test(struct pci_test *test)
 	}
 
 	if (test->copy) {
-		ret = ioctl(fd, PCITEST_COPY, test->size);
+		param.size = test->size;
+		if (test->use_dma)
+			param.flags = PCITEST_FLAGS_USE_DMA;
+		ret = ioctl(fd, PCITEST_COPY, &param);
 		fprintf(stdout, "COPY (%7ld bytes):\t\t", test->size);
 		if (ret < 0)
 			fprintf(stdout, "TEST FAILED\n");
@@ -153,7 +164,7 @@ int main(int argc, char **argv)
 	/* set default endpoint device */
 	test->device = "/dev/pci-endpoint-test.0";
 
-	while ((c = getopt(argc, argv, "D:b:m:x:i:Ilhrwcs:")) != EOF)
+	while ((c = getopt(argc, argv, "D:b:m:x:i:dIlhrwcs:")) != EOF)
 	switch (c) {
 	case 'D':
 		test->device = optarg;
@@ -197,6 +208,9 @@ int main(int argc, char **argv)
 	case 's':
 		test->size = strtoul(optarg, NULL, 0);
 		continue;
+	case 'd':
+		test->use_dma = true;
+		continue;
 	case 'h':
 	default:
 usage:
@@ -209,6 +223,7 @@ int main(int argc, char **argv)
 			"\t-x <msix num>	\tMSI-X test (msix number between 1..2048)\n"
 			"\t-i <irq type>	\tSet IRQ type (0 - Legacy, 1 - MSI, 2 - MSI-X)\n"
 			"\t-I			Get current IRQ type configured\n"
+			"\t-d			Use DMA\n"
 			"\t-l			Legacy IRQ test\n"
 			"\t-r			Read buffer test\n"
 			"\t-w			Write buffer test\n"
-- 
2.17.1

[PATCH v3 5/5] misc: pci_endpoint_test: Add support to get DMA option from userspace

[Kishon Vijay Abraham I]

'pcitest' utility now uses '-d' option to allow the user to test DMA.
Add support to parse option to use DMA from userspace application.

Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Tested-by: Alan Mikhak <alan.mikhak@sifive.com>
---
 drivers/misc/pci_endpoint_test.c | 65 ++++++++++++++++++++++++++++++--
 1 file changed, 62 insertions(+), 3 deletions(-)

diff --git a/drivers/misc/pci_endpoint_test.c b/drivers/misc/pci_endpoint_test.c
index 5998df1c84e9..8682867ac14a 100644
--- a/drivers/misc/pci_endpoint_test.c
+++ b/drivers/misc/pci_endpoint_test.c
@@ -17,6 +17,7 @@
 #include <linux/mutex.h>
 #include <linux/random.h>
 #include <linux/slab.h>
+#include <linux/uaccess.h>
 #include <linux/pci.h>
 #include <linux/pci_ids.h>
 
@@ -52,6 +53,7 @@
 #define STATUS_SRC_ADDR_INVALID			BIT(7)
 #define STATUS_DST_ADDR_INVALID			BIT(8)
 
+#define PCI_ENDPOINT_TEST_STATUS		0x8
 #define PCI_ENDPOINT_TEST_LOWER_SRC_ADDR	0x0c
 #define PCI_ENDPOINT_TEST_UPPER_SRC_ADDR	0x10
 
@@ -64,6 +66,9 @@
 #define PCI_ENDPOINT_TEST_IRQ_TYPE		0x24
 #define PCI_ENDPOINT_TEST_IRQ_NUMBER		0x28
 
+#define PCI_ENDPOINT_TEST_FLAGS			0x2c
+#define FLAG_USE_DMA				BIT(0)
+
 #define PCI_DEVICE_ID_TI_AM654			0xb00c
 
 #define is_am654_pci_dev(pdev)		\
@@ -315,11 +320,16 @@ static bool pci_endpoint_test_msi_irq(struct pci_endpoint_test *test,
 	return false;
 }
 
-static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
+static bool pci_endpoint_test_copy(struct pci_endpoint_test *test,
+				   unsigned long arg)
 {
+	struct pci_endpoint_test_xfer_param param;
 	bool ret = false;
 	void *src_addr;
 	void *dst_addr;
+	u32 flags = 0;
+	bool use_dma;
+	size_t size;
 	dma_addr_t src_phys_addr;
 	dma_addr_t dst_phys_addr;
 	struct pci_dev *pdev = test->pdev;
@@ -332,10 +342,22 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 	size_t alignment = test->alignment;
 	u32 src_crc32;
 	u32 dst_crc32;
+	int err;
+
+	err = copy_from_user(&param, (void __user *)arg, sizeof(param));
+	if (err) {
+		dev_err(dev, "Failed to get transfer param\n");
+		return false;
+	}
 
+	size = param.size;
 	if (size > SIZE_MAX - alignment)
 		goto err;
 
+	use_dma = !!(param.flags & PCITEST_FLAGS_USE_DMA);
+	if (use_dma)
+		flags |= FLAG_USE_DMA;
+
 	if (irq_type < IRQ_TYPE_LEGACY || irq_type > IRQ_TYPE_MSIX) {
 		dev_err(dev, "Invalid IRQ type option\n");
 		goto err;
@@ -406,6 +428,7 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_SIZE,
 				 size);
 
+	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_FLAGS, flags);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_TYPE, irq_type);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_NUMBER, 1);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_COMMAND,
@@ -434,9 +457,13 @@ static bool pci_endpoint_test_copy(struct pci_endpoint_test *test, size_t size)
 	return ret;
 }
 
-static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
+static bool pci_endpoint_test_write(struct pci_endpoint_test *test,
+				    unsigned long arg)
 {
+	struct pci_endpoint_test_xfer_param param;
 	bool ret = false;
+	u32 flags = 0;
+	bool use_dma;
 	u32 reg;
 	void *addr;
 	dma_addr_t phys_addr;
@@ -446,11 +473,24 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 	dma_addr_t orig_phys_addr;
 	size_t offset;
 	size_t alignment = test->alignment;
+	size_t size;
 	u32 crc32;
+	int err;
 
+	err = copy_from_user(&param, (void __user *)arg, sizeof(param));
+	if (err != 0) {
+		dev_err(dev, "Failed to get transfer param\n");
+		return false;
+	}
+
+	size = param.size;
 	if (size > SIZE_MAX - alignment)
 		goto err;
 
+	use_dma = !!(param.flags & PCITEST_FLAGS_USE_DMA);
+	if (use_dma)
+		flags |= FLAG_USE_DMA;
+
 	if (irq_type < IRQ_TYPE_LEGACY || irq_type > IRQ_TYPE_MSIX) {
 		dev_err(dev, "Invalid IRQ type option\n");
 		goto err;
@@ -493,6 +533,7 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_SIZE, size);
 
+	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_FLAGS, flags);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_TYPE, irq_type);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_NUMBER, 1);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_COMMAND,
@@ -514,9 +555,14 @@ static bool pci_endpoint_test_write(struct pci_endpoint_test *test, size_t size)
 	return ret;
 }
 
-static bool pci_endpoint_test_read(struct pci_endpoint_test *test, size_t size)
+static bool pci_endpoint_test_read(struct pci_endpoint_test *test,
+				   unsigned long arg)
 {
+	struct pci_endpoint_test_xfer_param param;
 	bool ret = false;
+	u32 flags = 0;
+	bool use_dma;
+	size_t size;
 	void *addr;
 	dma_addr_t phys_addr;
 	struct pci_dev *pdev = test->pdev;
@@ -526,10 +572,22 @@ static bool pci_endpoint_test_read(struct pci_endpoint_test *test, size_t size)
 	size_t offset;
 	size_t alignment = test->alignment;
 	u32 crc32;
+	int err;
 
+	err = copy_from_user(&param, (void __user *)arg, sizeof(param));
+	if (err) {
+		dev_err(dev, "Failed to get transfer param\n");
+		return false;
+	}
+
+	size = param.size;
 	if (size > SIZE_MAX - alignment)
 		goto err;
 
+	use_dma = !!(param.flags & PCITEST_FLAGS_USE_DMA);
+	if (use_dma)
+		flags |= FLAG_USE_DMA;
+
 	if (irq_type < IRQ_TYPE_LEGACY || irq_type > IRQ_TYPE_MSIX) {
 		dev_err(dev, "Invalid IRQ type option\n");
 		goto err;
@@ -566,6 +624,7 @@ static bool pci_endpoint_test_read(struct pci_endpoint_test *test, size_t size)
 
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_SIZE, size);
 
+	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_FLAGS, flags);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_TYPE, irq_type);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_IRQ_NUMBER, 1);
 	pci_endpoint_test_writel(test, PCI_ENDPOINT_TEST_COMMAND,
-- 
2.17.1

Re: [PATCH v3 0/5] PCI: functions/pci-epf-test: Add DMA data transfer

[Lorenzo Pieralisi]

On Mon, Mar 16, 2020 at 04:54:19PM +0530, Kishon Vijay Abraham I wrote:
> Patch series uses dma engine APIs in pci-epf-test to transfer data using
> DMA. It also adds an option "-d" in pcitest for the user to indicate
> whether DMA has to be used for data transfer. This also prints
> throughput information for data transfer.
> 
> Changes from v2:
> *) Removed use of "bool" in uapi/linux/pcitest.h as that seems to fail
>    for certain configs (header should be self contained). It was failing
>    for "x86_64 allmodconfig".
> *) Modified function comments added before DMA transfer functions.
> 
> Changes from v1:
> *) Fixed some of the function names from pci_epf_* to pci_epf_test_*
> since the DMA support is now been moved to pci-epf-test.c
> 
> Kishon Vijay Abraham I (5):
>   PCI: endpoint: functions/pci-epf-test: Add DMA support to transfer
>     data
>   PCI: endpoint: functions/pci-epf-test: Print throughput information
>   misc: pci_endpoint_test: Use streaming DMA APIs for buffer allocation
>   tools: PCI: Add 'd' command line option to support DMA
>   misc: pci_endpoint_test: Add support to get DMA option from userspace
> 
>  drivers/misc/pci_endpoint_test.c              | 165 ++++++++++--
>  drivers/pci/endpoint/functions/pci-epf-test.c | 253 +++++++++++++++++-
>  include/uapi/linux/pcitest.h                  |   7 +
>  tools/pci/pcitest.c                           |  23 +-
>  4 files changed, 417 insertions(+), 31 deletions(-)

Applied to pci/endpoint for v5.7, thanks.

Lorenzo
> 
> -- 
> 2.17.1
>