Re: [PATCH] timer: bcmbca: Add Broadcom BCMBCA timer support

[William Zhang]

On 08/04/2022 01:25 PM, William Zhang wrote:

Hi Florian,

On 08/03/2022 08:21 PM, Florian Fainelli wrote:

On 8/2/2022 5:07 PM, William Zhang wrote:

Hi Rafal,

On 08/01/2022 10:26 PM, Rafał Miłecki wrote:

On 2.08.2022 00:03, William Zhang wrote:

This driver supports the peripheral block timer found on the Broadcom
BCA SoCs. It is 30-bit up-count timer running at 50MHz and can be used
as the system clock source such as on BCM63138.
Signed-off-by: William Zhang <william.zh...@broadcom.com>

(...)

+static const struct udevice_id bcmbca_timer_ids[] = {
+    { .compatible = "brcm,bcm-timers" },
+    { }
+};
+
+U_BOOT_DRIVER(bcmbca_timer) = {
+    .name = "bcmbca_timer",
+    .id = UCLASS_TIMER,
+    .of_match = bcmbca_timer_ids,
+    .priv_auto = sizeof(struct bcmbca_timer_priv),
+    .probe = bcmbca_timer_probe,
+    .ops = &bcmbca_timer_ops,
+    .flags = DM_FLAG_PRE_RELOC,
+};

That "brcm,bcm-timers" seems like a really wide bidding. Is that exact
timer block guaranteed to be present on all Broadcom devices? Does it
exist e.g. on Northstar SoCs? Or old MIPS SoCs like BCM4706?

Agree I will make change to use brcm,bcmbca-timers.

Why not change to use a compatible string that uses the first chip in which this timer block was introduced, was that 6345 or later?

I don't mind to change to 6345. That chip should support the same timer block. Looks u-boot has a lot of Broadcom MIPS based chips that uses 6345 in compatible string name for various blocks.

It seems that even across BCMBCA devices this block may differ and may
need different bindings. Most SoCs have 4 CTL and 4 CNT registers but
some have only 3 + 3 (BCM6838 BCM60333 BCM63268).

This timer driver is intended for the clock source for u-boot only so only need the first channel. Although it could work for the old mips based dsl chip, BCMBCA chip family support is only for the new ARM based   chips which is stated in the BCMBA introduction patch.

Finally could we have that binding actually documented?

I don't see u-boot has binding document.  I believe it generally use linux binding document.  I have no plan to upstream this driver to linux. But I can put the binding info in the driver if that is the right way to do in u-boot.

Out of curiosity, why does u-boot require this timer as opposed to using the Cortex-A9 architected timers? Is it necessary to boot strap the A9 timers?

Cortex-A9 has an arch timer called global timer.  But I don't see a generic global timer driver in the u-boot. There is one from STMicroelectronics sti-timer.c which is a a9 global timer but it is implemented under STI_TIMER and default enabled for ARCH_STI. I guess we can use that too when we have the clk driver(iproc clock driver) ported to u-boot. Just curious, does anyone in this list know why sti timer is not put under global timer config and file name?  I don't see anything in that driver is sti specific. CC'ed driver maintainer Patrice as well.

Since there is already global timer support in u-boot, I decided to use that for BCM63138. So I withdrew this patch and please discard. I will also send a patch to move the sti a9 global timer to generic arm a9 global timer driver if Patrice agrees or I don't hear from him back next week.

smime.p7s
Description: S/MIME Cryptographic Signature