Hi Simon,
On Tue, Feb 4, 2020 at 8:19 AM Simon Glass sjg@chromium.org wrote:
Hi Bin,
On Mon, 3 Feb 2020 at 10:05, Bin Meng bmeng.cn@gmail.com wrote:
Hi Simon,
On Sun, Dec 22, 2019 at 2:16 AM Simon Glass sjg@chromium.org wrote:
At present driver model supports the IRQ uclass but there is no way to request a particular interrupt for a driver.
Add a mechanism, similar to clock and reset, to read the interrupts required by a device from the device tree and to request those interrupts.
U-Boot itself does not have interrupt-driven handlers, so just provide a means to read and clear an interrupt. This can be useful to handle peripherals which must use an interrupt to determine when data is available, for example.
Bring over the basic binding file as well, from Linux v5.4. Note that the older binding is not supported in U-Boot; the newer 'special form' must be used.
Add a simple test of the new functionality.
Signed-off-by: Simon Glass sjg@chromium.org
Changes in v2: None
arch/sandbox/dts/test.dts | 5 +- .../interrupt-controller/interrupts.txt | 131 ++++++++++++++++++ drivers/misc/irq-uclass.c | 116 ++++++++++++++++ drivers/misc/irq_sandbox.c | 41 ++++++ include/irq.h | 115 +++++++++++++++ test/dm/irq.c | 31 +++++ 6 files changed, 438 insertions(+), 1 deletion(-) create mode 100644 doc/device-tree-bindings/interrupt-controller/interrupts.txt
[..]
diff --git a/drivers/misc/irq-uclass.c b/drivers/misc/irq-uclass.c index 38498a66a4..7d65192858 100644 --- a/drivers/misc/irq-uclass.c +++ b/drivers/misc/irq-uclass.c @@ -4,8 +4,11 @@
- Written by Simon Glass sjg@chromium.org
*/
+#define LOG_CATEGORY UCLASS_IRQ
#include <common.h> #include <dm.h> +#include <dt-structs.h> #include <irq.h> #include <dm/device-internal.h>
@@ -49,6 +52,119 @@ int irq_restore_polarities(struct udevice *dev) return ops->restore_polarities(dev); }
+int irq_read_and_clear(struct irq *irq) +{
const struct irq_ops *ops = irq_get_ops(irq->dev);if (!ops->read_and_clear)return -ENOSYS;return ops->read_and_clear(irq);+}
+#if CONFIG_IS_ENABLED(OF_PLATDATA) +int irq_get_by_index_platdata(struct udevice *dev, int index,
struct phandle_1_arg *cells, struct irq *irq)+{
int ret;if (index != 0)return -ENOSYS;ret = uclass_get_device(UCLASS_IRQ, 0, &irq->dev);if (ret)return ret;irq->id = cells[0].arg[0];return 0;+} +#else +static int irq_of_xlate_default(struct irq *irq,
struct ofnode_phandle_args *args)+{
log_debug("(irq=%p)\n", irq);if (args->args_count > 1) {log_debug("Invaild args_count: %d\n", args->args_count);return -EINVAL;}if (args->args_count)irq->id = args->args[0];elseirq->id = 0;return 0;+}
+static int irq_get_by_index_tail(int ret, ofnode node,
It's odd to pass the return value of some other functions to this function and check it here. Can we please remove ret, and check its value at where we get that?
This pattern is similar to how it is done in DM core. It routes all return paths through a single function where you can do error checking, etc. So I think this is better than the alternative. The only strange thing here is that this function is only used one. I did this since I suspected there would be others coming. But I could inline if it you like..
If that's the way how other DM codes do, I am okay with that.
[..]
+static int sandbox_irq_of_xlate(struct irq *irq,
struct ofnode_phandle_args *args)+{
irq->id = args->args[0];return 0;+}
This
...?
I must have been working too long yesterday ...
Regards, Bin