On 2017/2/21 3:00, Marc Zyngier wrote:

On 10/02/17 07:10, Hanjun Guo wrote:

quoted

Hi Marc,

On 2017/1/24 19:32, Marc Zyngier wrote:

quoted

On 24/01/17 10:57, Mark Rutland wrote:

quoted

On Tue, Jan 24, 2017 at 06:39:51PM +0800, Hanjun Guo wrote:

quoted

From: Hanjun Guo <redacted>

Add hisilicon timer specific erratum fixes.

Signed-off-by: Hanjun Guo <redacted>
---
 drivers/clocksource/arm_arch_timer.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
index 80d6f76..3e62a09 100644
--- a/drivers/clocksource/arm_arch_timer.c
+++ b/drivers/clocksource/arm_arch_timer.c

@@ -1156,10 +1156,32 @@ struct gtdt_arch_timer_fixup {
 	void *context;
 };
 
+#ifdef CONFIG_HISILICON_ERRATUM_161010101
+static void __init hisi_erratum_workaroud_enable(void *context)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ool_workarounds); i++) {
+		if (!strcmp(context, ool_workarounds[i].id)) {
+			timer_unstable_counter_workaround = &ool_workarounds[i];
+			static_branch_enable(&arch_timer_read_ool_enabled);
+			pr_info("arch_timer: Enabling workaround for %s\n",
+				timer_unstable_counter_workaround->id);
+			break;
+		}
+	}
+}
+#endif
+
 /* note: this needs to be updated according to the doc of OEM ID
  * and TABLE ID for different board.
  */
 static struct gtdt_arch_timer_fixup arch_timer_quirks[] __initdata = {
+#ifdef CONFIG_HISILICON_ERRATUM_161010101
+	{"HISI  ", "HIP05   ", 0, &hisi_erratum_workaroud_enable, "hisilicon,erratum-161010101"},
+	{"HISI  ", "HIP06   ", 0, &hisi_erratum_workaroud_enable, "hisilicon,erratum-161010101"},
+	{"HISI  ", "HIP07   ", 0, &hisi_erratum_workaroud_enable, "hisilicon,erratum-161010101"},
+#endif
 };

NAK. This duplicates logic unnecessarily (for enabling the workaround),
and (ab)uses the id, which was intended to be specific to DT (since it
is a DT property name).

Agreed, that's properly revolting.

quoted

We should split the matching from the particular workaround (and
enabling thereof), so that we can go straight from ACPI match to
workaround (without having to use the DT id in this manner), and don't
have to duplicate the logic to enable the workaround.

I believe Marc is looking at some rework in this area which may enable
this, so please wait for that to appear.

Yeah, I'm implementing a semi-flexible way to add new quirk types, and
the last thing I want to see is two (or more) tables describing the same
thing.

Kindly ping, if you have patches in hand, I can test against our platforms,
thank you very much.

I've just pushed out a branch based on arm64/for-next/core, plus the
HiSi timer workaround:

https://git.kernel.org/cgit/linux/kernel/git/maz/arm-platforms.git/log/?h=timers/errata-rework

I'll post the patches for review once the merge window is open, but
hopefully that should get you going.

Thanks,

Test this branch and could work fine for Hisilicon board, great work !

Still need time to review the details, but this should not seem very difficult. :)

Thanks
Ding

	M.