On 3/7/22 5:17 AM, Etienne Carriere wrote:
Hello Sean,
Thanks for the feedback.
Sorry I missed your post end Feb.
On Fri, 25 Feb 2022 at 07:33, Sean Anderson <sean...@gmail.com> wrote:
Hi Etienne,
On 2/21/22 3:22 AM, Etienne Carriere wrote:
Implements SCMI APIs to retrieve the number exposed SCMI clocks using
SCMI_PROTOCOL_ATTRIBUTES messages and the names of the clocks using
SCMI_CLOCK_ATTRIBUTES messages.
This change updates sandbox SCMI clock test driver to manage these
2 new message IDs.
Cc: Lukasz Majewski <lu...@denx.de>
Cc: Sean Anderson <sean...@gmail.com>
Cc: Clement Leger <clement.le...@bootlin.com>
Cc: Patrick Delaunay <patrick.delau...@foss.st.com>
Reviewed-by: Patrick Delaunay <patrick.delau...@foss.st.com>
Signed-off-by: Gabriel Fernandez <gabriel.fernan...@st.com>
Signed-off-by: Etienne Carriere <etienne.carri...@linaro.org>
---
Changes since v1:
- Remove Series-links tag and apply R-b tag.
---
drivers/clk/clk_scmi.c | 90 ++++++++++++++++++++++
drivers/firmware/scmi/sandbox-scmi_agent.c | 53 +++++++++++++
include/scmi_protocols.h | 43 +++++++++++
3 files changed, 186 insertions(+)
diff --git a/drivers/clk/clk_scmi.c b/drivers/clk/clk_scmi.c
index 42fbab0d21..57022685e2 100644
--- a/drivers/clk/clk_scmi.c
+++ b/drivers/clk/clk_scmi.c
@@ -11,6 +11,52 @@
#include <scmi_agent.h>
#include <scmi_protocols.h>
#include <asm/types.h>
+#include <linux/clk-provider.h>
+
+static int scmi_clk_get_num_clock(struct udevice *dev, size_t *num_clocks)
+{
+ struct scmi_clk_protocol_attr_out out;
+ struct scmi_msg msg = {
+ .protocol_id = SCMI_PROTOCOL_ID_CLOCK,
+ .message_id = SCMI_PROTOCOL_ATTRIBUTES,
+ .out_msg = (u8 *)&out,
+ .out_msg_sz = sizeof(out),
+ };
+ int ret;
+
+ ret = devm_scmi_process_msg(dev, &msg);
+ if (ret)
+ return ret;
+
+ *num_clocks = out.attributes & SCMI_CLK_PROTO_ATTR_COUNT_MASK;
+
+ return 0;
+}
+
+static int scmi_clk_get_attibute(struct udevice *dev, int clkid, char **name)
nit: scmi_clk_get_attribute
right!
+{
+ struct scmi_clk_attribute_in in = {
+ .clock_id = clkid,
+ };
+ struct scmi_clk_attribute_out out;
+ struct scmi_msg msg = {
+ .protocol_id = SCMI_PROTOCOL_ID_CLOCK,
+ .message_id = SCMI_CLOCK_ATTRIBUTES,
+ .in_msg = (u8 *)&in,
+ .in_msg_sz = sizeof(in),
+ .out_msg = (u8 *)&out,
+ .out_msg_sz = sizeof(out),
+ };
+ int ret;
+
+ ret = devm_scmi_process_msg(dev, &msg);
+ if (ret)
+ return ret;
+
+ *name = out.clock_name;
+
+ return 0;
+}
static int scmi_clk_gate(struct clk *clk, int enable)
{
@@ -88,6 +134,49 @@ static ulong scmi_clk_set_rate(struct clk *clk, ulong rate)
return scmi_clk_get_rate(clk);
}
+static int scmi_clk_probe(struct udevice *dev)
+{
+ struct clk *clk;
+ size_t num_clocks, i;
+ int ret;
+
+ if (!CONFIG_IS_ENABLED(CLK_CCF))
+ return 0;
+
+ /* register CCF children: CLK UCLASS, no probed again */
+ if (device_get_uclass_id(dev->parent) == UCLASS_CLK)
+ return 0;
+
+ ret = scmi_clk_get_num_clock(dev, &num_clocks);
+ if (ret)
+ return ret;
+
+ for (i = 0; i < num_clocks; i++) {
+ char *name;
+
+ if (!scmi_clk_get_attibute(dev, i, &name)) {
+ char *clock_name = strdup(name);
+
+ clk = kzalloc(sizeof(*clk), GFP_KERNEL);
+ if (!clk || !clock_name)
+ ret = -ENOMEM;
+ else
+ ret = clk_register(clk, dev->driver->name,
+ clock_name, dev->name);
+
+ if (ret) {
+ free(clk);
+ free(clock_name);
+ return ret;
+ }
+
+ clk_dm(i, clk);
+ }
+ }
+
+ return 0;
+}
+
So why not call scmi_clk_get_num_clock during probe() and then check against
it in xlate? I don't really see why we need CCF. This also means you don't
have a driver with copies of itself as children, and is lighter on memory.
There is no specific need for CCF for the scmi clock driver itself.
The goal of these changes is rather to allow platforms that do already
enable CCF (for an u-boot native clock driver)
to be able to also embed SCMI clocks support, for clocks controlled
from an external firmware.
Sure, but you can still use non-CCF clocks in a CCF clock.
--Sean
