Re: [RESEND PATCH 1/1] usb: dwc2: add multiple clock handling

From: Frank Wang
Date: Mon Feb 06 2017 - 22:18:45 EST


Hi Heiko, John and Greg,

On 2017/2/7 8:06, Heiko Stuebner wrote:
Hi Frank,

Am Sonntag, 5. Februar 2017, 10:51:01 CET schrieb Frank Wang:
Originally, dwc2 just handle one clock named otg, however, it may have
two or more clock need to manage for some new SoCs, so this adds
change clk to clk's array of dwc2_hsotg to handle more clocks operation.

Signed-off-by: Frank Wang <frank.wang@xxxxxxxxxxxxxx>
---
drivers/usb/dwc2/core.h | 5 ++++-
drivers/usb/dwc2/platform.c | 39 ++++++++++++++++++++++++++-------------
2 files changed, 30 insertions(+), 14 deletions(-)

diff --git a/drivers/usb/dwc2/core.h b/drivers/usb/dwc2/core.h
index 1a7e830..d10a466 100644
--- a/drivers/usb/dwc2/core.h
+++ b/drivers/usb/dwc2/core.h
@@ -121,6 +121,9 @@ static inline void dwc2_writel(u32 value, void __iomem
*addr) /* Maximum number of Endpoints/HostChannels */
#define MAX_EPS_CHANNELS 16

+/* Maximum number of dwc2 clocks */
+#define DWC2_MAX_CLKS 3
why 3 clocks?

Showing from the chapter 2.4 of dwc otg databook v3.10, it seems there should be five clocks, am I right?
hclk/pmu_hclk/utmi_clk/ulpi_clk/utmifs_clk48

I.e. the binding currently only specifies the "otg" clock, so you should
definitly amend it to also specifiy this "pmu" clock - in a separate patch
before this one of course :-) .
"pmu" also looks like a good name for that clock-binding and these new clocks
of course should be optional in the binding.

No problem, I will amend the binding when the implementation scheme is clear.

And ideally also just specify this mysterious third clock as well while you're
at it.

+
/* dwc2-hsotg declarations */
static const char * const dwc2_hsotg_supply_names[] = {
"vusb_d", /* digital USB supply, 1.2V */
@@ -913,7 +916,7 @@ struct dwc2_hsotg {
spinlock_t lock;
void *priv;
int irq;
- struct clk *clk;
+ struct clk *clks[DWC2_MAX_CLKS];
struct reset_control *reset;

unsigned int queuing_high_bandwidth:1;
[...]

@@ -123,17 +123,20 @@ static int dwc2_get_dr_mode(struct dwc2_hsotg *hsotg)
static int __dwc2_lowlevel_hw_enable(struct dwc2_hsotg *hsotg)
{
struct platform_device *pdev = to_platform_device(hsotg->dev);
- int ret;
+ int clk, ret;

ret = regulator_bulk_enable(ARRAY_SIZE(hsotg->supplies),
hsotg->supplies);
if (ret)
return ret;

- if (hsotg->clk) {
- ret = clk_prepare_enable(hsotg->clk);
- if (ret)
+ for (clk = 0; clk < DWC2_MAX_CLKS && hsotg->clks[clk]; clk++) {
+ ret = clk_prepare_enable(hsotg->clks[clk]);
+ if (ret) {
+ while (--clk >= 0)
+ clk_disable_unprepare(hsotg->clks[clk]);
return ret;
+ }
}

if (hsotg->uphy) {
@@ -168,7 +171,7 @@ int dwc2_lowlevel_hw_enable(struct dwc2_hsotg *hsotg)
static int __dwc2_lowlevel_hw_disable(struct dwc2_hsotg *hsotg)
{
struct platform_device *pdev = to_platform_device(hsotg->dev);
- int ret = 0;
+ int clk, ret = 0;

if (hsotg->uphy) {
usb_phy_shutdown(hsotg->uphy);
@@ -182,8 +185,9 @@ static int __dwc2_lowlevel_hw_disable(struct dwc2_hsotg
*hsotg) if (ret)
return ret;

- if (hsotg->clk)
- clk_disable_unprepare(hsotg->clk);
+ for (clk = DWC2_MAX_CLKS - 1; clk >= 0; clk--)
+ if (hsotg->clks[clk])
+ clk_disable_unprepare(hsotg->clks[clk]);
ret = regulator_bulk_disable(ARRAY_SIZE(hsotg->supplies),
hsotg->supplies);
@@ -209,7 +213,7 @@ int dwc2_lowlevel_hw_disable(struct dwc2_hsotg *hsotg)

static int dwc2_lowlevel_hw_init(struct dwc2_hsotg *hsotg)
{
- int i, ret;
+ int i, clk, ret;

hsotg->reset = devm_reset_control_get_optional(hsotg->dev, "dwc2");
if (IS_ERR(hsotg->reset)) {
@@ -282,11 +286,20 @@ static int dwc2_lowlevel_hw_init(struct dwc2_hsotg
*hsotg) hsotg->phyif = GUSBCFG_PHYIF8;
}

- /* Clock */
- hsotg->clk = devm_clk_get(hsotg->dev, "otg");
- if (IS_ERR(hsotg->clk)) {
- hsotg->clk = NULL;
- dev_dbg(hsotg->dev, "cannot get otg clock\n");
+ /* Clocks */
+ for (clk = 0; clk < DWC2_MAX_CLKS; clk++) {
+ hsotg->clks[clk] = of_clk_get(hsotg->dev->of_node, clk);
+ if (IS_ERR(hsotg->clks[clk])) {
+ ret = PTR_ERR(hsotg->clks[clk]);
+ if (ret == -EPROBE_DEFER) {
+ while (--clk >= 0)
+ clk_put(hsotg->clks[clk]);
+ return ret;
+ }
+
+ hsotg->clks[clk] = NULL;
+ break;
+ }
I guess this depends on the feelings of the usb-people, but for me it might
make more sense to not carry the clocks in an anonymous array but instead as
named members in struct dwc2_hsotg - aka clk_otg, clk_pmu, clk_??? .

Because the otg clocks is actually marked as required in the binding while our
two new clocks are optional and some future code might actually want to
control these separate clocks to save some power somewhere.


Sidenote: I don't really understand why the driver allows the otg clock to be
missing, as it is a required property in the binding.

Yes, if there are only two clocks need to control, separate member in struct dwc2_hsotg is really better,
but for more clocks, the operations of each clock (get/prepare/unprepare) may become tedious

How about keeping the original 'otg' clk and adding a new clk array member for optional clocks in struct dwc2_hsotg?
because excepting 'otg' clk, the others are all optional clocks.

Of course, if we only consider hclk and pmu_hclk, I guess the separate member scheme is still better.

John and Greg, would you like to give some comments please?


BR.
Frank

Heiko