diff mbox series

[v5,1/6] misc: Add Synopsys DesignWare xData IP driver

Message ID 02835da8fc8c9293fecbe666a8db3fb79276fdde.1613034397.git.gustavo.pimentel@synopsys.com
State New
Headers show
Series misc: Add Add Synopsys DesignWare xData IP driver | expand

Commit Message

Gustavo Pimentel Feb. 11, 2021, 9:08 a.m. UTC
Add Synopsys DesignWare xData IP driver. This driver enables/disables
the PCI traffic generator module pertain to the Synopsys DesignWare
prototype.

Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
---
 drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 394 insertions(+)
 create mode 100644 drivers/misc/dw-xdata-pcie.c

Comments

Greg KH Feb. 11, 2021, 9:30 a.m. UTC | #1
On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> +			  char *buf)
> +{
> +	struct pci_dev *pdev = to_pci_dev(dev);
> +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> +	u64 rate;
> +
> +	mutex_lock(&dw->mutex);
> +	dw_xdata_perf(dw, &rate, true);
> +	mutex_unlock(&dw->mutex);
> +
> +	return sysfs_emit(buf, "%llu MB/s\n", rate);

Do not put units in a sysfs file, that should be in the documentation,
otherwise this forces userspace to "parse" the units which is a mess.

Same for the other sysfs file.

And why do you need a lock for this show function?

thanks,

greg k-h
Gustavo Pimentel Feb. 11, 2021, 9:50 a.m. UTC | #2
On Thu, Feb 11, 2021 at 9:30:16, Greg Kroah-Hartman 
<gregkh@linuxfoundation.org> wrote:

> On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> > +			  char *buf)
> > +{
> > +	struct pci_dev *pdev = to_pci_dev(dev);
> > +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> > +	u64 rate;
> > +
> > +	mutex_lock(&dw->mutex);
> > +	dw_xdata_perf(dw, &rate, true);
> > +	mutex_unlock(&dw->mutex);
> > +
> > +	return sysfs_emit(buf, "%llu MB/s\n", rate);
> 
> Do not put units in a sysfs file, that should be in the documentation,
> otherwise this forces userspace to "parse" the units which is a mess.

Okay.

> 
> Same for the other sysfs file.
> 
> And why do you need a lock for this show function?

Maybe I understood it wrongly, please correct me in that case. The 
dw_xdata_perf() is called on the write_show() and read_show(), to avoid a 
possible race condition between those calls, I have added this mutex.
Thanks.

-Gustavo

> 
> thanks,
> 
> greg k-h
Greg KH Feb. 11, 2021, 9:59 a.m. UTC | #3
On Thu, Feb 11, 2021 at 09:50:33AM +0000, Gustavo Pimentel wrote:
> On Thu, Feb 11, 2021 at 9:30:16, Greg Kroah-Hartman 
> <gregkh@linuxfoundation.org> wrote:
> 
> > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> > > +			  char *buf)
> > > +{
> > > +	struct pci_dev *pdev = to_pci_dev(dev);
> > > +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> > > +	u64 rate;
> > > +
> > > +	mutex_lock(&dw->mutex);
> > > +	dw_xdata_perf(dw, &rate, true);
> > > +	mutex_unlock(&dw->mutex);
> > > +
> > > +	return sysfs_emit(buf, "%llu MB/s\n", rate);
> > 
> > Do not put units in a sysfs file, that should be in the documentation,
> > otherwise this forces userspace to "parse" the units which is a mess.
> 
> Okay.
> 
> > 
> > Same for the other sysfs file.
> > 
> > And why do you need a lock for this show function?
> 
> Maybe I understood it wrongly, please correct me in that case. The 
> dw_xdata_perf() is called on the write_show() and read_show(), to avoid a 
> possible race condition between those calls, I have added this mutex.

What race?  If the value changes with a write right after a read, what
does it matter?

What exactly are you trying to protect with this lock?

thanks,

greg k-h
Gustavo Pimentel Feb. 11, 2021, 10:21 a.m. UTC | #4
On Thu, Feb 11, 2021 at 9:59:26, Greg Kroah-Hartman 
<gregkh@linuxfoundation.org> wrote:

> On Thu, Feb 11, 2021 at 09:50:33AM +0000, Gustavo Pimentel wrote:
> > On Thu, Feb 11, 2021 at 9:30:16, Greg Kroah-Hartman 
> > <gregkh@linuxfoundation.org> wrote:
> > 
> > > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > > +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> > > > +			  char *buf)
> > > > +{
> > > > +	struct pci_dev *pdev = to_pci_dev(dev);
> > > > +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> > > > +	u64 rate;
> > > > +
> > > > +	mutex_lock(&dw->mutex);
> > > > +	dw_xdata_perf(dw, &rate, true);
> > > > +	mutex_unlock(&dw->mutex);
> > > > +
> > > > +	return sysfs_emit(buf, "%llu MB/s\n", rate);
> > > 
> > > Do not put units in a sysfs file, that should be in the documentation,
> > > otherwise this forces userspace to "parse" the units which is a mess.
> > 
> > Okay.
> > 
> > > 
> > > Same for the other sysfs file.
> > > 
> > > And why do you need a lock for this show function?
> > 
> > Maybe I understood it wrongly, please correct me in that case. The 
> > dw_xdata_perf() is called on the write_show() and read_show(), to avoid a 
> > possible race condition between those calls, I have added this mutex.
> 
> What race?  If the value changes with a write right after a read, what
> does it matter?
> 
> What exactly are you trying to protect with this lock?

The write_store() does a procedure to enable the traffic on the write 
direction, however, the write_show() does a different procedure to 
calculate the link throughput speed, which uses a different set of 
registers on the HW.

Similar happens on the read_store() (which enable the traffic on the read 
direction) and on the read_show()

To summarize write_store() follows the same approach of read_store() and 
the write_show() of the read_show(). I added the mutex on those functions 
for instance to avoid while during the write_show() call the possibility 
of been called the read_show() messing up the link throughput speed 
calculation.
Or while during the write_store() call to be called the read_store or 
even the write_show() for the same reasons.

This is the reason why I added those mutexes, maybe this isn't necessary 
and it's overkill. Please advise me if a different approach can be done.

-Gustavo

> 
> thanks,
> 
> greg k-h
Greg KH Feb. 11, 2021, 10:33 a.m. UTC | #5
On Thu, Feb 11, 2021 at 10:21:07AM +0000, Gustavo Pimentel wrote:
> On Thu, Feb 11, 2021 at 9:59:26, Greg Kroah-Hartman 
> <gregkh@linuxfoundation.org> wrote:
> 
> > On Thu, Feb 11, 2021 at 09:50:33AM +0000, Gustavo Pimentel wrote:
> > > On Thu, Feb 11, 2021 at 9:30:16, Greg Kroah-Hartman 
> > > <gregkh@linuxfoundation.org> wrote:
> > > 
> > > > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > > > +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> > > > > +			  char *buf)
> > > > > +{
> > > > > +	struct pci_dev *pdev = to_pci_dev(dev);
> > > > > +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> > > > > +	u64 rate;
> > > > > +
> > > > > +	mutex_lock(&dw->mutex);
> > > > > +	dw_xdata_perf(dw, &rate, true);
> > > > > +	mutex_unlock(&dw->mutex);
> > > > > +
> > > > > +	return sysfs_emit(buf, "%llu MB/s\n", rate);
> > > > 
> > > > Do not put units in a sysfs file, that should be in the documentation,
> > > > otherwise this forces userspace to "parse" the units which is a mess.
> > > 
> > > Okay.
> > > 
> > > > 
> > > > Same for the other sysfs file.
> > > > 
> > > > And why do you need a lock for this show function?
> > > 
> > > Maybe I understood it wrongly, please correct me in that case. The 
> > > dw_xdata_perf() is called on the write_show() and read_show(), to avoid a 
> > > possible race condition between those calls, I have added this mutex.
> > 
> > What race?  If the value changes with a write right after a read, what
> > does it matter?
> > 
> > What exactly are you trying to protect with this lock?
> 
> The write_store() does a procedure to enable the traffic on the write 
> direction, however, the write_show() does a different procedure to 
> calculate the link throughput speed, which uses a different set of 
> registers on the HW.
> 
> Similar happens on the read_store() (which enable the traffic on the read 
> direction) and on the read_show()
> 
> To summarize write_store() follows the same approach of read_store() and 
> the write_show() of the read_show(). I added the mutex on those functions 
> for instance to avoid while during the write_show() call the possibility 
> of been called the read_show() messing up the link throughput speed 
> calculation.
> Or while during the write_store() call to be called the read_store or 
> even the write_show() for the same reasons.

If you need to protect these types of things, but the lock down in the
function that does this, not above it which forces people to audit
everything and manually try to determine what lock is doing what for
what.

Make it impossible to get wrong, as it is, you have to do extra work
here to keep things working properly, always a bad idea in an api.

thanks,

greg k-h
Gustavo Pimentel Feb. 11, 2021, 10:51 a.m. UTC | #6
On Thu, Feb 11, 2021 at 10:33:25, Greg Kroah-Hartman 
<gregkh@linuxfoundation.org> wrote:

> On Thu, Feb 11, 2021 at 10:21:07AM +0000, Gustavo Pimentel wrote:
> > On Thu, Feb 11, 2021 at 9:59:26, Greg Kroah-Hartman 
> > <gregkh@linuxfoundation.org> wrote:
> > 
> > > On Thu, Feb 11, 2021 at 09:50:33AM +0000, Gustavo Pimentel wrote:
> > > > On Thu, Feb 11, 2021 at 9:30:16, Greg Kroah-Hartman 
> > > > <gregkh@linuxfoundation.org> wrote:
> > > > 
> > > > > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > > > > +static ssize_t write_show(struct device *dev, struct device_attribute *attr,
> > > > > > +			  char *buf)
> > > > > > +{
> > > > > > +	struct pci_dev *pdev = to_pci_dev(dev);
> > > > > > +	struct dw_xdata *dw = pci_get_drvdata(pdev);
> > > > > > +	u64 rate;
> > > > > > +
> > > > > > +	mutex_lock(&dw->mutex);
> > > > > > +	dw_xdata_perf(dw, &rate, true);
> > > > > > +	mutex_unlock(&dw->mutex);
> > > > > > +
> > > > > > +	return sysfs_emit(buf, "%llu MB/s\n", rate);
> > > > > 
> > > > > Do not put units in a sysfs file, that should be in the documentation,
> > > > > otherwise this forces userspace to "parse" the units which is a mess.
> > > > 
> > > > Okay.
> > > > 
> > > > > 
> > > > > Same for the other sysfs file.
> > > > > 
> > > > > And why do you need a lock for this show function?
> > > > 
> > > > Maybe I understood it wrongly, please correct me in that case. The 
> > > > dw_xdata_perf() is called on the write_show() and read_show(), to avoid a 
> > > > possible race condition between those calls, I have added this mutex.
> > > 
> > > What race?  If the value changes with a write right after a read, what
> > > does it matter?
> > > 
> > > What exactly are you trying to protect with this lock?
> > 
> > The write_store() does a procedure to enable the traffic on the write 
> > direction, however, the write_show() does a different procedure to 
> > calculate the link throughput speed, which uses a different set of 
> > registers on the HW.
> > 
> > Similar happens on the read_store() (which enable the traffic on the read 
> > direction) and on the read_show()
> > 
> > To summarize write_store() follows the same approach of read_store() and 
> > the write_show() of the read_show(). I added the mutex on those functions 
> > for instance to avoid while during the write_show() call the possibility 
> > of been called the read_show() messing up the link throughput speed 
> > calculation.
> > Or while during the write_store() call to be called the read_store or 
> > even the write_show() for the same reasons.
> 
> If you need to protect these types of things, but the lock down in the
> function that does this, not above it which forces people to audit
> everything and manually try to determine what lock is doing what for
> what.
> 
> Make it impossible to get wrong, as it is, you have to do extra work
> here to keep things working properly, always a bad idea in an api.

I think I understood what you mean, I will *reduce* the mutex scope to 
the basic functions that are called by the sysfs *_store() and *_show().

-Gustavo

> 
> thanks,
> 
> greg k-h
Leon Romanovsky Feb. 11, 2021, 11:42 a.m. UTC | #7
On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> Add Synopsys DesignWare xData IP driver. This driver enables/disables
> the PCI traffic generator module pertain to the Synopsys DesignWare
> prototype.
>
> Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
> ---
>  drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 394 insertions(+)
>  create mode 100644 drivers/misc/dw-xdata-pcie.c

<...>

> +MODULE_LICENSE("GPL v2");

"GPL" and not "GPL v2".

Thanks

> +MODULE_DESCRIPTION("Synopsys DesignWare xData PCIe driver");
> +MODULE_AUTHOR("Gustavo Pimentel <gustavo.pimentel@synopsys.com>");
> +
> --
> 2.7.4
>
Greg KH Feb. 11, 2021, 12:32 p.m. UTC | #8
On Thu, Feb 11, 2021 at 01:42:43PM +0200, Leon Romanovsky wrote:
> On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > Add Synopsys DesignWare xData IP driver. This driver enables/disables
> > the PCI traffic generator module pertain to the Synopsys DesignWare
> > prototype.
> >
> > Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
> > ---
> >  drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 394 insertions(+)
> >  create mode 100644 drivers/misc/dw-xdata-pcie.c
> 
> <...>
> 
> > +MODULE_LICENSE("GPL v2");
> 
> "GPL" and not "GPL v2".

There is no difference, please go read module.h.
Leon Romanovsky Feb. 11, 2021, 1:50 p.m. UTC | #9
On Thu, Feb 11, 2021 at 01:32:13PM +0100, Greg Kroah-Hartman wrote:
> On Thu, Feb 11, 2021 at 01:42:43PM +0200, Leon Romanovsky wrote:
> > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > Add Synopsys DesignWare xData IP driver. This driver enables/disables
> > > the PCI traffic generator module pertain to the Synopsys DesignWare
> > > prototype.
> > >
> > > Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
> > > ---
> > >  drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
> > >  1 file changed, 394 insertions(+)
> > >  create mode 100644 drivers/misc/dw-xdata-pcie.c
> >
> > <...>
> >
> > > +MODULE_LICENSE("GPL v2");
> >
> > "GPL" and not "GPL v2".
>
> There is no difference, please go read module.h.

I read and this is why I said it.
Documentation/process/license-rules.rst: "It exists for historic reasons."

Historic, for me, means that new code is better do not use this.

Thanks
Greg KH Feb. 11, 2021, 2:02 p.m. UTC | #10
On Thu, Feb 11, 2021 at 03:50:19PM +0200, Leon Romanovsky wrote:
> On Thu, Feb 11, 2021 at 01:32:13PM +0100, Greg Kroah-Hartman wrote:
> > On Thu, Feb 11, 2021 at 01:42:43PM +0200, Leon Romanovsky wrote:
> > > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > > Add Synopsys DesignWare xData IP driver. This driver enables/disables
> > > > the PCI traffic generator module pertain to the Synopsys DesignWare
> > > > prototype.
> > > >
> > > > Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
> > > > ---
> > > >  drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
> > > >  1 file changed, 394 insertions(+)
> > > >  create mode 100644 drivers/misc/dw-xdata-pcie.c
> > >
> > > <...>
> > >
> > > > +MODULE_LICENSE("GPL v2");
> > >
> > > "GPL" and not "GPL v2".
> >
> > There is no difference, please go read module.h.
> 
> I read and this is why I said it.
> Documentation/process/license-rules.rst: "It exists for historic reasons."
> 
> Historic, for me, means that new code is better do not use this.

Nope, either is fine for new code, author gets to pick what they want.
Personally, I like the explicitness of "GPL v2" for a variety of
reasons.

thanks,

greg k-h
Leon Romanovsky Feb. 11, 2021, 2:13 p.m. UTC | #11
On Thu, Feb 11, 2021 at 03:02:03PM +0100, Greg Kroah-Hartman wrote:
> On Thu, Feb 11, 2021 at 03:50:19PM +0200, Leon Romanovsky wrote:
> > On Thu, Feb 11, 2021 at 01:32:13PM +0100, Greg Kroah-Hartman wrote:
> > > On Thu, Feb 11, 2021 at 01:42:43PM +0200, Leon Romanovsky wrote:
> > > > On Thu, Feb 11, 2021 at 10:08:38AM +0100, Gustavo Pimentel wrote:
> > > > > Add Synopsys DesignWare xData IP driver. This driver enables/disables
> > > > > the PCI traffic generator module pertain to the Synopsys DesignWare
> > > > > prototype.
> > > > >
> > > > > Signed-off-by: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
> > > > > ---
> > > > >  drivers/misc/dw-xdata-pcie.c | 394 +++++++++++++++++++++++++++++++++++++++++++
> > > > >  1 file changed, 394 insertions(+)
> > > > >  create mode 100644 drivers/misc/dw-xdata-pcie.c
> > > >
> > > > <...>
> > > >
> > > > > +MODULE_LICENSE("GPL v2");
> > > >
> > > > "GPL" and not "GPL v2".
> > >
> > > There is no difference, please go read module.h.
> >
> > I read and this is why I said it.
> > Documentation/process/license-rules.rst: "It exists for historic reasons."
> >
> > Historic, for me, means that new code is better do not use this.
>
> Nope, either is fine for new code, author gets to pick what they want.
> Personally, I like the explicitness of "GPL v2" for a variety of
> reasons.

Feel free to update the documentation.

Personally, I don't like two names for the same license. This is one of
the reasons why SPFX clearly marks the code license, It is why we have
"SPDX-License-Identifier: GPL-2.0" and not "SPDX-License-Identifier: GPL".

https://spdx.org/licenses/preview/
GNU General Public License v2.0 only 	        GPL-2.0-only
GNU General Public License v2.0 or later 	GPL-2.0-or-later

Thanks

>
> thanks,
>
> greg k-h
diff mbox series

Patch

diff --git a/drivers/misc/dw-xdata-pcie.c b/drivers/misc/dw-xdata-pcie.c
new file mode 100644
index 00000000..2e023ba
--- /dev/null
+++ b/drivers/misc/dw-xdata-pcie.c
@@ -0,0 +1,394 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2020 Synopsys, Inc. and/or its affiliates.
+ * Synopsys DesignWare xData driver
+ *
+ * Author: Gustavo Pimentel <gustavo.pimentel@synopsys.com>
+ */
+
+#include <linux/bitfield.h>
+#include <linux/pci-epf.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/bitops.h>
+#include <linux/mutex.h>
+#include <linux/delay.h>
+#include <linux/pci.h>
+
+#define DW_XDATA_DRIVER_NAME		"dw-xdata-pcie"
+
+#define DW_XDATA_EP_MEM_OFFSET		0x8000000
+
+struct dw_xdata_pcie_data {
+	/* xData registers location */
+	enum pci_barno			rg_bar;
+	off_t				rg_off;
+	size_t				rg_sz;
+};
+
+static const struct dw_xdata_pcie_data snps_edda_data = {
+	/* xData registers location */
+	.rg_bar				= BAR_0,
+	.rg_off				= 0x00000000,   /*   0 Kbytes */
+	.rg_sz				= 0x0000012c,   /* 300  bytes */
+};
+
+#define STATUS_DONE			BIT(0)
+
+#define CONTROL_DOORBELL		BIT(0)
+#define CONTROL_IS_WRITE		BIT(1)
+#define CONTROL_LENGTH(a)		FIELD_PREP(GENMASK(13, 2), a)
+#define CONTROL_PATTERN_INC		BIT(16)
+#define CONTROL_NO_ADDR_INC		BIT(18)
+
+#define XPERF_CONTROL_ENABLE		BIT(5)
+
+#define BURST_REPEAT			BIT(31)
+#define BURST_VALUE			0x1001
+
+#define PATTERN_VALUE			0x0
+
+struct dw_xdata_regs {
+	u32 addr_lsb;					/* 0x000 */
+	u32 addr_msb;					/* 0x004 */
+	u32 burst_cnt;					/* 0x008 */
+	u32 control;					/* 0x00c */
+	u32 pattern;					/* 0x010 */
+	u32 status;					/* 0x014 */
+	u32 RAM_addr;					/* 0x018 */
+	u32 RAM_port;					/* 0x01c */
+	u32 _reserved0[14];				/* 0x020..0x054 */
+	u32 perf_control;				/* 0x058 */
+	u32 _reserved1[41];				/* 0x05c..0x0fc */
+	u32 wr_cnt_lsb;					/* 0x100 */
+	u32 wr_cnt_msb;					/* 0x104 */
+	u32 rd_cnt_lsb;					/* 0x108 */
+	u32 rd_cnt_msb;					/* 0x10c */
+} __packed;
+
+struct dw_xdata_region {
+	phys_addr_t paddr;				/* physical address */
+	void __iomem *vaddr;				/* virtual address */
+	size_t sz;					/* size */
+};
+
+struct dw_xdata {
+	struct dw_xdata_region rg_region;		/* registers */
+	size_t max_wr_len;				/* max wr xfer len */
+	size_t max_rd_len;				/* max rd xfer len */
+	struct mutex mutex;
+	struct pci_dev *pdev;
+};
+
+static inline struct dw_xdata_regs __iomem *__dw_regs(struct dw_xdata *dw)
+{
+	return dw->rg_region.vaddr;
+}
+
+static void dw_xdata_stop(struct dw_xdata *dw)
+{
+	u32 burst = readl(&(__dw_regs(dw)->burst_cnt));
+
+	if (burst & BURST_REPEAT) {
+		burst &= ~(u32)BURST_REPEAT;
+		writel(burst, &(__dw_regs(dw)->burst_cnt));
+	}
+}
+
+static void dw_xdata_start(struct dw_xdata *dw, bool write)
+{
+	u32 control, status;
+
+	/* Stop first if xfer in progress */
+	dw_xdata_stop(dw);
+
+	/* Clear status register */
+	writel(0x0, &(__dw_regs(dw)->status));
+
+	/* Burst count register set for continuous until stopped */
+	writel(BURST_REPEAT | BURST_VALUE, &(__dw_regs(dw)->burst_cnt));
+
+	/* Pattern register */
+	writel(PATTERN_VALUE, &(__dw_regs(dw)->pattern));
+
+	/* Control register */
+	control = CONTROL_DOORBELL | CONTROL_PATTERN_INC | CONTROL_NO_ADDR_INC;
+	if (write) {
+		control |= CONTROL_IS_WRITE;
+		control |= CONTROL_LENGTH(dw->max_wr_len);
+	} else {
+		control |= CONTROL_LENGTH(dw->max_rd_len);
+	}
+	writel(control, &(__dw_regs(dw)->control));
+
+	/*
+	 * The xData HW block needs about 100 ms to initiate the traffic
+	 * generation according this HW block datasheet.
+	 */
+	usleep_range(100, 150);
+
+	status = readl(&(__dw_regs(dw)->status));
+	if (!(status & STATUS_DONE))
+		pci_dbg(dw->pdev, "xData: started %s direction\n",
+			write ? "write" : "read");
+}
+
+static void dw_xdata_perf_meas(struct dw_xdata *dw, u64 *data, bool write)
+{
+	if (write) {
+		*data = readl(&(__dw_regs(dw)->wr_cnt_msb));
+		*data <<= 32;
+		*data |= readl(&(__dw_regs(dw)->wr_cnt_lsb));
+	} else {
+		*data = readl(&(__dw_regs(dw)->rd_cnt_msb));
+		*data <<= 32;
+		*data |= readl(&(__dw_regs(dw)->rd_cnt_lsb));
+	}
+}
+
+static u64 dw_xdata_perf_diff(u64 *m1, u64 *m2, u64 time)
+{
+	u64 rate = (*m1 - *m2);
+
+	rate *= (1000 * 1000 * 1000);
+	rate >>= 20;
+	rate = DIV_ROUND_CLOSEST_ULL(rate, time);
+
+	return rate;
+}
+
+static void dw_xdata_perf(struct dw_xdata *dw, u64 *rate, bool write)
+{
+	u64 data[2], time[2], diff;
+
+	/* First acquisition of current count frames */
+	writel(0x0, &(__dw_regs(dw)->perf_control));
+	dw_xdata_perf_meas(dw, &data[0], write);
+	time[0] = jiffies;
+	writel((u32)XPERF_CONTROL_ENABLE, &(__dw_regs(dw)->perf_control));
+
+	/*
+	 * Wait 100ms between the 1st count frame acquisition and the 2nd
+	 * count frame acquisition, in order to calculate the speed later
+	 */
+	mdelay(100);
+
+	/* Second acquisition of current count frames */
+	writel(0x0, &(__dw_regs(dw)->perf_control));
+	dw_xdata_perf_meas(dw, &data[1], write);
+	time[1] = jiffies;
+	writel((u32)XPERF_CONTROL_ENABLE, &(__dw_regs(dw)->perf_control));
+
+	/*
+	 * Speed calculation
+	 *
+	 * rate = (2nd count frames - 1st count frames) / (time elapsed)
+	 */
+	diff = jiffies_to_nsecs(time[1] - time[0]);
+	*rate = dw_xdata_perf_diff(&data[1], &data[0], diff);
+
+	pci_dbg(dw->pdev, "xData: time=%llu us, %s=%llu MB/s\n",
+		diff, write ? "write" : "read", *rate);
+}
+
+static ssize_t write_show(struct device *dev, struct device_attribute *attr,
+			  char *buf)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+	u64 rate;
+
+	mutex_lock(&dw->mutex);
+	dw_xdata_perf(dw, &rate, true);
+	mutex_unlock(&dw->mutex);
+
+	return sysfs_emit(buf, "%llu MB/s\n", rate);
+}
+
+static ssize_t write_store(struct device *dev, struct device_attribute *attr,
+			   const char *buf, size_t size)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+
+	pci_dbg(pdev, "xData: requested write transfer\n");
+
+	mutex_lock(&dw->mutex);
+	dw_xdata_start(dw, true);
+	mutex_unlock(&dw->mutex);
+
+	return size;
+}
+
+static DEVICE_ATTR_RW(write);
+
+static ssize_t read_show(struct device *dev, struct device_attribute *attr,
+			 char *buf)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+	u64 rate;
+
+	mutex_lock(&dw->mutex);
+	dw_xdata_perf(dw, &rate, false);
+	mutex_unlock(&dw->mutex);
+
+	return sysfs_emit(buf, "%llu MB/s\n", rate);
+}
+
+static ssize_t read_store(struct device *dev, struct device_attribute *attr,
+			  const char *buf, size_t size)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+
+	pci_dbg(pdev, "xData: requested read transfer\n");
+
+	mutex_lock(&dw->mutex);
+	dw_xdata_start(dw, false);
+	mutex_unlock(&dw->mutex);
+
+	return size;
+}
+
+static DEVICE_ATTR_RW(read);
+
+static ssize_t stop_store(struct device *dev, struct device_attribute *attr,
+			  const char *buf, size_t size)
+{
+	struct pci_dev *pdev = to_pci_dev(dev);
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+
+	pci_dbg(pdev, "xData: requested stop any transfer\n");
+
+	mutex_lock(&dw->mutex);
+	dw_xdata_stop(dw);
+	mutex_unlock(&dw->mutex);
+
+	return size;
+}
+
+static DEVICE_ATTR_WO(stop);
+
+static struct attribute *default_attrs[] = {
+	&dev_attr_write.attr,
+	&dev_attr_read.attr,
+	&dev_attr_stop.attr,
+	NULL,
+};
+
+static const struct attribute_group xdata_attr_group = {
+	.attrs = default_attrs,
+	.name = DW_XDATA_DRIVER_NAME,
+};
+
+static int dw_xdata_pcie_probe(struct pci_dev *pdev,
+			       const struct pci_device_id *pid)
+{
+	const struct dw_xdata_pcie_data *pdata = (void *)pid->driver_data;
+	struct dw_xdata *dw;
+	u64 addr;
+	int err;
+
+	/* Enable PCI device */
+	err = pcim_enable_device(pdev);
+	if (err) {
+		pci_err(pdev, "enabling device failed\n");
+		return err;
+	}
+
+	/* Mapping PCI BAR regions */
+	err = pcim_iomap_regions(pdev, BIT(pdata->rg_bar), pci_name(pdev));
+	if (err) {
+		pci_err(pdev, "xData BAR I/O remapping failed\n");
+		return err;
+	}
+
+	pci_set_master(pdev);
+
+	/* Allocate memory */
+	dw = devm_kzalloc(&pdev->dev, sizeof(*dw), GFP_KERNEL);
+	if (!dw)
+		return -ENOMEM;
+
+	/* Data structure initialization */
+	mutex_init(&dw->mutex);
+
+	dw->rg_region.vaddr = pcim_iomap_table(pdev)[pdata->rg_bar];
+	if (!dw->rg_region.vaddr)
+		return -ENOMEM;
+
+	dw->rg_region.vaddr += pdata->rg_off;
+	dw->rg_region.paddr = pdev->resource[pdata->rg_bar].start;
+	dw->rg_region.paddr += pdata->rg_off;
+	dw->rg_region.sz = pdata->rg_sz;
+
+	dw->max_wr_len = pcie_get_mps(pdev);
+	dw->max_wr_len >>= 2;
+
+	dw->max_rd_len = pcie_get_readrq(pdev);
+	dw->max_rd_len >>= 2;
+
+	dw->pdev = pdev;
+
+	writel(0x0, &(__dw_regs(dw)->RAM_addr));
+	writel(0x0, &(__dw_regs(dw)->RAM_port));
+
+	addr = dw->rg_region.paddr + DW_XDATA_EP_MEM_OFFSET;
+	writel(lower_32_bits(addr), &(__dw_regs(dw)->addr_lsb));
+	writel(upper_32_bits(addr), &(__dw_regs(dw)->addr_msb));
+	pci_dbg(pdev, "xData: target address = 0x%.16llx\n", addr);
+
+	pci_dbg(pdev, "xData: wr_len=%zu, rd_len=%zu\n",
+		dw->max_wr_len * 4, dw->max_rd_len * 4);
+
+	/* Saving data structure reference */
+	pci_set_drvdata(pdev, dw);
+
+	/* Sysfs */
+	err = sysfs_create_group(&pdev->dev.kobj, &xdata_attr_group);
+	if (err)
+		return err;
+
+	err = sysfs_create_link(kernel_kobj, &pdev->dev.kobj,
+				DW_XDATA_DRIVER_NAME);
+	if (err)
+		return err;
+
+	return 0;
+}
+
+static void dw_xdata_pcie_remove(struct pci_dev *pdev)
+{
+	struct dw_xdata *dw = pci_get_drvdata(pdev);
+
+	if (dw) {
+		mutex_lock(&dw->mutex);
+		dw_xdata_stop(dw);
+		mutex_unlock(&dw->mutex);
+	}
+
+	sysfs_remove_link(kernel_kobj, DW_XDATA_DRIVER_NAME);
+	sysfs_remove_group(&pdev->dev.kobj, &xdata_attr_group);
+}
+
+static const struct pci_device_id dw_xdata_pcie_id_table[] = {
+	{ PCI_DEVICE_DATA(SYNOPSYS, EDDA, &snps_edda_data) },
+	{ }
+};
+MODULE_DEVICE_TABLE(pci, dw_xdata_pcie_id_table);
+
+static struct pci_driver dw_xdata_pcie_driver = {
+	.name		= DW_XDATA_DRIVER_NAME,
+	.id_table	= dw_xdata_pcie_id_table,
+	.probe		= dw_xdata_pcie_probe,
+	.remove		= dw_xdata_pcie_remove,
+};
+
+module_pci_driver(dw_xdata_pcie_driver);
+
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Synopsys DesignWare xData PCIe driver");
+MODULE_AUTHOR("Gustavo Pimentel <gustavo.pimentel@synopsys.com>");
+