[RFC PATCH 00/14] fsi: Fixes and Coldfire coprocessor offload

[RFC PATCH 00/14] fsi: Fixes and Coldfire coprocessor offload

[Benjamin Herrenschmidt]

This series implements support for offloading the FSI protocol bitbanging
to the ColdFire secondary core of the Aspeed SoCs. The result increases
FSI performance by a factor of 4, and on systems that don't support async
FSI clock, provide much more regular and continuous clocking which helps
reliability.

Patch 1 may go a different route and was already posted a few weeks ago,
I included it for completeness.

Patches 2..9 add some infrastructure to the FSI core to control some
of the FSI protocol delays and adjustements/fixes to the existing GPIO
bitbanging master. They are "mechanical" dependencies

Patch 10 moves some protocol definitions to a common place where the
new master driver can find them

Patch 11 is the DT binding for the new driver with comes with patch 12

Finally patch 13 and 14 update the Romulus and Palmetto board device-trees
to use the new driver.

There's another dependency on the Aspeed GPIO driver changes for handling
with GPIO lines ownership and handshaking. The patches have been submitted
and can be found for reference there:

        https://github.com/ozbenh/linux-ast/commits/gpio

Finally, the driver needs a machine specific firmware file. The firwmare
is open source and available at:

        https://github.com/ozbenh/cf-fsi

I will submit it to linux-firmware if there's enough popular demand ;-)

[RFC PATCH 01/14] devres: Add devm_of_iomap()

[Benjamin Herrenschmidt]

There are still quite a few cases where a device might want
to get to a different node of the device-tree, obtain the
resources and map them.

We have of_iomap() and of_io_request_and_map() but they both
have shortcomings, such as not returning the size of the
resource found (which can be useful) and not being "managed".

This adds a devm_of_iomap() that provides all of these and
should probably replace uses of the above in most drivers.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
 include/linux/device.h |  4 ++++
 lib/devres.c           | 36 ++++++++++++++++++++++++++++++++++++
 2 files changed, 40 insertions(+)

diff --git a/include/linux/device.h b/include/linux/device.h
index 477956990f5e..96249d790374 100644
--- a/include/linux/device.h
+++ b/include/linux/device.h
@@ -688,6 +688,10 @@ extern void devm_free_pages(struct device *dev, unsigned long addr);
 
 void __iomem *devm_ioremap_resource(struct device *dev, struct resource *res);
 
+void __iomem *devm_of_iomap(struct device *dev,
+			    struct device_node *node, int index,
+			    resource_size_t *size);
+
 /* allows to add/remove a custom action to devres stack */
 int devm_add_action(struct device *dev, void (*action)(void *), void *data);
 void devm_remove_action(struct device *dev, void (*action)(void *), void *data);
diff --git a/lib/devres.c b/lib/devres.c
index 5bec1120b392..faccf1a037d0 100644
--- a/lib/devres.c
+++ b/lib/devres.c
@@ -4,6 +4,7 @@
 #include <linux/io.h>
 #include <linux/gfp.h>
 #include <linux/export.h>
+#include <linux/of_address.h>
 
 enum devm_ioremap_type {
 	DEVM_IOREMAP = 0,
@@ -162,6 +163,41 @@ void __iomem *devm_ioremap_resource(struct device *dev, struct resource *res)
 }
 EXPORT_SYMBOL(devm_ioremap_resource);
 
+/*
+ * devm_of_iomap - Requests a resource and maps the memory mapped IO
+ *		   for a given device_node managed by a given device
+ *
+ * Checks that a resource is a valid memory region, requests the memory
+ * region and ioremaps it. All operations are managed and will be undone
+ * on driver detach of the device.
+ *
+ * This is to be used when a device requests/maps resources described
+ * by other device tree nodes (children or otherwise).
+ *
+ * @dev:	The device "managing" the resource
+ * @node:       The device-tree node where the resource resides
+ * @index:	index of the MMIO range in the "reg" property
+ * @size:	Returns the size of the resource (pass NULL if not needed)
+ * Returns a pointer to the requested and mapped memory or an ERR_PTR() encoded
+ * error code on failure. Usage example:
+ *
+ *	base = devm_of_iomap(&pdev->dev, node, 0, NULL);
+ *	if (IS_ERR(base))
+ *		return PTR_ERR(base);
+ */
+void __iomem *devm_of_iomap(struct device *dev, struct device_node *node, int index,
+			    resource_size_t *size)
+{
+	struct resource res;
+
+	if (of_address_to_resource(node, index, &res))
+		return IOMEM_ERR_PTR(-EINVAL);
+	if (size)
+		*size = resource_size(&res);
+	return devm_ioremap_resource(dev, &res);
+}
+EXPORT_SYMBOL(devm_of_iomap);
+
 #ifdef CONFIG_HAS_IOPORT_MAP
 /*
  * Generic iomap devres
-- 
2.17.1

[RFC PATCH 02/14] fsi: Move code around to avoid forward declaration

[Benjamin Herrenschmidt]

Move fsi_slave_set_smode() and its helpers to before it's
first user and remove the corresponding forward declaration.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
 drivers/fsi/fsi-core.c | 94 +++++++++++++++++++++---------------------
 1 file changed, 46 insertions(+), 48 deletions(-)

diff --git a/drivers/fsi/fsi-core.c b/drivers/fsi/fsi-core.c
index 565218872635..2f6f9b8c75e4 100644
--- a/drivers/fsi/fsi-core.c
+++ b/drivers/fsi/fsi-core.c
@@ -215,7 +215,52 @@ static int fsi_slave_report_and_clear_errors(struct fsi_slave *slave)
 			&irq, sizeof(irq));
 }
 
-static int fsi_slave_set_smode(struct fsi_master *master, int link, int id);
+/* Encode slave local bus echo delay */
+static inline uint32_t fsi_smode_echodly(int x)
+{
+	return (x & FSI_SMODE_ED_MASK) << FSI_SMODE_ED_SHIFT;
+}
+
+/* Encode slave local bus send delay */
+static inline uint32_t fsi_smode_senddly(int x)
+{
+	return (x & FSI_SMODE_SD_MASK) << FSI_SMODE_SD_SHIFT;
+}
+
+/* Encode slave local bus clock rate ratio */
+static inline uint32_t fsi_smode_lbcrr(int x)
+{
+	return (x & FSI_SMODE_LBCRR_MASK) << FSI_SMODE_LBCRR_SHIFT;
+}
+
+/* Encode slave ID */
+static inline uint32_t fsi_smode_sid(int x)
+{
+	return (x & FSI_SMODE_SID_MASK) << FSI_SMODE_SID_SHIFT;
+}
+
+static uint32_t fsi_slave_smode(int id)
+{
+	return FSI_SMODE_WSC | FSI_SMODE_ECRC
+		| fsi_smode_sid(id)
+		| fsi_smode_echodly(0xf) | fsi_smode_senddly(0xf)
+		| fsi_smode_lbcrr(0x8);
+}
+
+static int fsi_slave_set_smode(struct fsi_master *master, int link, int id)
+{
+	uint32_t smode;
+	__be32 data;
+
+	/* set our smode register with the slave ID field to 0; this enables
+	 * extended slave addressing
+	 */
+	smode = fsi_slave_smode(id);
+	data = cpu_to_be32(smode);
+
+	return fsi_master_write(master, link, id, FSI_SLAVE_BASE + FSI_SMODE,
+			&data, sizeof(data));
+}
 
 static int fsi_slave_handle_error(struct fsi_slave *slave, bool write,
 				  uint32_t addr, size_t size)
@@ -569,53 +614,6 @@ static const struct bin_attribute fsi_slave_term_attr = {
 	.write = fsi_slave_sysfs_term_write,
 };
 
-/* Encode slave local bus echo delay */
-static inline uint32_t fsi_smode_echodly(int x)
-{
-	return (x & FSI_SMODE_ED_MASK) << FSI_SMODE_ED_SHIFT;
-}
-
-/* Encode slave local bus send delay */
-static inline uint32_t fsi_smode_senddly(int x)
-{
-	return (x & FSI_SMODE_SD_MASK) << FSI_SMODE_SD_SHIFT;
-}
-
-/* Encode slave local bus clock rate ratio */
-static inline uint32_t fsi_smode_lbcrr(int x)
-{
-	return (x & FSI_SMODE_LBCRR_MASK) << FSI_SMODE_LBCRR_SHIFT;
-}
-
-/* Encode slave ID */
-static inline uint32_t fsi_smode_sid(int x)
-{
-	return (x & FSI_SMODE_SID_MASK) << FSI_SMODE_SID_SHIFT;
-}
-
-static uint32_t fsi_slave_smode(int id)
-{
-	return FSI_SMODE_WSC | FSI_SMODE_ECRC
-		| fsi_smode_sid(id)
-		| fsi_smode_echodly(0xf) | fsi_smode_senddly(0xf)
-		| fsi_smode_lbcrr(0x8);
-}
-
-static int fsi_slave_set_smode(struct fsi_master *master, int link, int id)
-{
-	uint32_t smode;
-	__be32 data;
-
-	/* set our smode register with the slave ID field to 0; this enables
-	 * extended slave addressing
-	 */
-	smode = fsi_slave_smode(id);
-	data = cpu_to_be32(smode);
-
-	return fsi_master_write(master, link, id, FSI_SLAVE_BASE + FSI_SMODE,
-			&data, sizeof(data));
-}
-
 static void fsi_slave_release(struct device *dev)
 {
 	struct fsi_slave *slave = to_fsi_slave(dev);
-- 
2.17.1

Re: [RFC PATCH 01/14] devres: Add devm_of_iomap()

[Linus Walleij]

On Fri, Jun 22, 2018 at 6:31 AM Benjamin Herrenschmidt
<benh@kernel.crashing.org> wrote:

> There are still quite a few cases where a device might want
> to get to a different node of the device-tree, obtain the
> resources and map them.
>
> We have of_iomap() and of_io_request_and_map() but they both
> have shortcomings, such as not returning the size of the
> resource found (which can be useful) and not being "managed".
>
> This adds a devm_of_iomap() that provides all of these and
> should probably replace uses of the above in most drivers.
>
> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

Ugh I just feel I have seen homecooked solutions to this problem
a few times :/

I wonder if it is easy to find these cases and replace them with
this neat function...

Thanks for doing this.
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

Yours,
Linus Walleij

Re: [RFC PATCH 01/14] devres: Add devm_of_iomap()

[Andy Shevchenko]

On Fri, Jun 29, 2018 at 12:14 PM, Linus Walleij
<linus.walleij@linaro.org> wrote:

> I wonder if it is easy to find these cases and replace them with
> this neat function...

Would be reasonable easy by using coccinelle.

-- 
With Best Regards,
Andy Shevchenko

Re: [RFC PATCH 01/14] devres: Add devm_of_iomap()

[Benjamin Herrenschmidt]

On Fri, 2018-06-29 at 23:27 +0300, Andy Shevchenko wrote:
> On Fri, Jun 29, 2018 at 12:14 PM, Linus Walleij
> <linus.walleij@linaro.org> wrote:
> 
> > I wonder if it is easy to find these cases and replace them with
> > this neat function...
> 
> Would be reasonable easy by using coccinelle.

For the obvious ones yes. A lot of the existing users of of_iomap
however don't do the request_region, and while they probably should and
should use the new accessor, this can't be done blindly without
testing, because there are many old things around that have broken
memory region tracking and that will fail..

I plan to do a sweep through some of my old powermac/powerpc stuff one
of these days and do some conversions.

Cheers,
Ben.