Skip to contentSkip to conversation

i3c: dw: Disable IBI IRQ depends on hot-join and SIR enabling

PATCH Applied Dylan Hung 4 participants 8 messages Jan 19, 2024 – May 6, 2024

Subsystems (2)
lore
Share & tools
lore

Dylan Hung

author
#0 lore raw

PATCH 1/1 i3c: dw: Disable IBI IRQ depends on hot-join and SIR enabling

Disable IBI IRQ signal and status only when hot-join and SIR enabling of all target devices attached to the bus are disabled.

Fixes e389b1d72a62 ("i3c: dw: Add support for in-band interrupts")
---
 drivers/i3c/master/dw-i3c-master.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/i3c/master/dw-i3c-master.c +3 -1
index ef5751e91cc9..276153e10f5a 100644
@@ -1163,8 +1163,10 @@ static void dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master,
 		global = reg == 0xffffffff;
 		reg &= ~BIT(idx);
 	} else {
-		global = reg == 0;
+		bool hj_rejected = !!(readl(master->regs + DEVICE_CTRL) & DEV_CTRL_HOT_JOIN_NACK);
+
 		reg |= BIT(idx);
+		global = (reg == 0xffffffff) && hj_rejected;
 	}
 	writel(reg, master->regs + IBI_SIR_REQ_REJECT);
 
Signature · 7 lines
2.25.1


-- 
linux-i3c mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-i3c
maintainer Applied
#1 lore raw
Quoted 6 lines · Dylan Hung

Disable IBI IRQ signal and status only when hot-join and SIR enabling of all target devices attached to the bus are disabled.

Fixes e389b1d72a62 ("i3c: dw: Add support for in-band interrupts")

Applied, thanks!

[1/1] i3c: dw: Disable IBI IRQ depends on hot-join and SIR enabling

      https://git.kernel.org/abelloni/c/10201396ef64

Best regards,

Signature · 3 lines
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#2 lore raw
In reply to Dylan Hung

Fri, Jan 19, 2024 at 01:45:47PM +0800, Dylan Hung kirjoitti:

Quoted 6 lines

Disable IBI IRQ signal and status only when hot-join and SIR enabling of all target devices attached to the bus are disabled.

Fixes e389b1d72a62 ("i3c: dw: Add support for in-band interrupts")

Tag block is not supposed to have blank lines.

...

drivers/i3c/master/dw-i3c-master.c static void dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master, · line 1166

Quoted 6 lines
drivers/i3c/master/dw-i3c-master.c +1 -1
@@ -1163,8 +1163,10 @@ static void dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master,
 		global = reg == 0xffffffff;
 		reg &= ~BIT(idx);
 	} else {
-		global = reg == 0;
+		bool hj_rejected = !!(readl(master->regs + DEVICE_CTRL) & DEV_CTRL_HOT_JOIN_NACK);

'!!()' is redundant.

drivers/i3c/master/dw-i3c-master.c static void dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master, · line 1170

Quoted 4 lines
drivers/i3c/master/dw-i3c-master.c +2 -0
+
 		reg |= BIT(idx);
+		global = (reg == 0xffffffff) && hj_rejected;
 	}

Moreover, you can refactor a bit both branches, like

	bool hj_rejected = true;
	...
	if (...) {
		...
	} else {
		hj_rejected = readl(master->regs + DEVICE_CTRL) & DEV_CTRL_HOT_JOIN_NACK;
		...
	}
	global = (reg == 0xffffffff) && hj_rejected;
Signature · 2 lines
With Best Regards,
Andy Shevchenko
#3 lore raw
In reply to Dylan Hung

Hi Dylan,

Just a question on a prior patch you sent:

Quoted 4 lines

Disable IBI IRQ signal and status only when hot-join and SIR enabling of all target devices attached to the bus are disabled.

Fixes e389b1d72a62 ("i3c: dw: Add support for in-band interrupts")

[...]

Quoted 13 lines
drivers/i3c/master/dw-i3c-master.c +3 -1
@@ -1163,8 +1163,10 @@ static void dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master,
                 global = reg == 0xffffffff;
                 reg &= ~BIT(idx);
         } else {
-               global = reg == 0;
+               bool hj_rejected = !!(readl(master->regs + DEVICE_CTRL) & DEV_CTRL_HOT_JOIN_NACK);
+
                 reg |= BIT(idx);
+               global = (reg == 0xffffffff) && hj_rejected;
         }
         writel(reg, master->regs + IBI_SIR_REQ_REJECT);
Quoted 1 line

My interpretation of this change is that we keep the "global" IBI irq enabled if hot-join-nack is set (ie, always, because we don't support hot join, and configure the hardware to nack all hot join requests).

However, we never enable the hot-join NACK interrupt - IBI_QUEUE_CTRL bit 0 is never set. So I can't see how we would ever get an interrupt for the hot join NACK case anyway. Because of that, this patch is just keeping the IBI threshold interrupt always enabled for no reason.

Or is something else happening here? Is there another cause for the IBI threshold IRQs?

Cheers,

Jeremy

author
#4 lore raw

Hi Jeremy,

Quoted 24 lines

-----Original Message-----
From: Jeremy Kerr <[email protected]>
Sent: Wednesday, May 1, 2024 2:22 PM
To: Dylan Hung <[email protected]>; [email protected]; [email protected]; u.kleine- [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; linux- [email protected]
Cc: BMC-SW <[email protected]>
Subject: Re: [PATCH] i3c: dw: Disable IBI IRQ depends on hot-join and SIR enabling

Hi Dylan,

Just a question on a prior patch you sent:

Quoted 4 lines

Disable IBI IRQ signal and status only when hot-join and SIR enabling of all target devices attached to the bus are disabled.

Fixes e389b1d72a62 ("i3c: dw: Add support for in-band interrupts")

[...]

Quoted 17 lines
Quoted 14 lines
drivers/i3c/master/dw-i3c-master.c +4 -1
@@ -1163,8 +1163,10 @@ static void
 dw_i3c_master_set_sir_enabled(struct dw_i3c_master *master,
                 global = reg == 0xffffffff;
                 reg &= ~BIT(idx);
         } else {
-               global = reg == 0;
+               bool hj_rejected = !!(readl(master->regs +
+DEVICE_CTRL) & DEV_CTRL_HOT_JOIN_NACK);
+
                 reg |= BIT(idx);
+               global = (reg == 0xffffffff) && hj_rejected;
         }
Quoted 2 lines

        writel(reg, master->regs + IBI_SIR_REQ_REJECT);

```
	if (global)
		dw_i3c_master_enable_sir_signal(master, enable);
```
Quoted 4 lines

My interpretation of this change is that we keep the "global" IBI irq enabled if hot-join-nack is set (ie, always, because we don't support hot join, and configure the hardware to nack all hot join requests).

I would like to clarify the control logic, incorporating the principle of disabling the SIR interrupt signal:

Case 1:
When DEV_CTRL_HOT_JOIN_NACK is set, indicating hj_rejected is true, it signifies the controller's non-receptiveness to the hot-join event. Consequently, we can safely disable the SIR interrupt signal if none of the target devices request SIR (reg == 0xffffffff).

Case 2:
When DEV_CTRL_HOT_JOIN_NACK is unset, indicating hj_rejected is false, this indicates the controller's readiness to engage with the hot-join event. Therefore, it's imperative to keep the SIR interrupt signal enabled, even if not all target devices request SIR. In this case, global is false and enable is false.

Here's a summary table to illustrate the logic:

| enable | reg==0xffffffff? | hj_rejected? | global | result                            |
|---         |---                          |---                   |---         |---                                  |
| false     | true                      | true                | true     | disable SIR interrupt |
| false     | false                     | true                | false     | keep SIR interrupt    |
| false     | true                      | false               | false     | keep SIR interrupt    |
| false     | false                     | false               | false     | keep SIR interrupt    |
Quoted 5 lines

However, we never enable the hot-join NACK interrupt - IBI_QUEUE_CTRL bit 0 is never set. So I can't see how we would ever get an interrupt for the hot join NACK case anyway. Because of that, this patch is just keeping the IBI threshold interrupt always enabled for no reason.

Billy recently submitted a change to implement the hot-join enabling/disabling. Therefore, it is timely to consider the hot-join functionality.
https://patchwork.kernel.org/project/linux-i3c/patch/[email protected]/

Quoted 7 lines

Or is something else happening here? Is there another cause for the IBI threshold IRQs?

Cheers,

Jeremy

#5 lore raw

Hi Dylan,

Thanks for the response! I have a couple of follow-up things though:

Quoted 13 lines
Quoted 4 lines

My interpretation of this change is that we keep the "global" IBI irq enabled if hot-join-nack is set (ie, always, because we don't support hot join, and configure the hardware to nack all hot join requests).

I would like to clarify the control logic, incorporating the principle of disabling the SIR interrupt signal:

Case 1:
When DEV_CTRL_HOT_JOIN_NACK is set, indicating hj_rejected is true, it signifies the controller's non-receptiveness to the hot-join event. Consequently, we can safely disable the SIR interrupt signal if none of the target devices request SIR (reg == 0xffffffff).

Quoted 6 lines

Case 2:
When DEV_CTRL_HOT_JOIN_NACK is unset, indicating hj_rejected is false, this indicates the controller's readiness to engage with the hot-join event. Therefore, it's imperative to keep the SIR interrupt signal enabled, even if not all target devices request SIR. In this case, global is false and enable is false.

Yep, I see what you're doing there, but it looks like the correct state would never be set if we're not enabling/disabling the IBIs separately; with this code, we would only ever enable the SIR for the HJ if we *also* happen to enable IBIs.

The initial state would be to have all SIRs masked.

Quoted 2 lines

Billy recently submitted a change to implement the hot-join enabling/disabling. Therefore, it is timely to consider the hot-join functionality.
https://patchwork.kernel.org/project/linux-i3c/patch/[email protected]/

Yep, I saw that, excellent! It's next on my list to take a look at.

It's just a little unusual that we're enabling the HJ interrupt before actually having the HJ support though.

Cheers,

Jeremy

author
#6 lore raw

Hi Jeremy,

Quoted 38 lines

-----Original Message-----
From: Jeremy Kerr <[email protected]>
Sent: Monday, May 6, 2024 1:10 PM
To: Dylan Hung <[email protected]>; [email protected]; [email protected]; u.kleine- [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; linux- [email protected]
Cc: BMC-SW <[email protected]>
Subject: Re: [PATCH] i3c: dw: Disable IBI IRQ depends on hot-join and SIR enabling

Hi Dylan,

Thanks for the response! I have a couple of follow-up things though:

Quoted 3 lines
Quoted 3 lines

My interpretation of this change is that we keep the "global" IBI irq enabled if hot-join-nack is set (ie, always, because we don't support hot join, and configure the hardware to nack all hot join

requests).

Quoted 10 lines
Quoted 1 line

I would like to clarify the control logic, incorporating the principle of disabling the SIR interrupt signal:

Case 1:
When DEV_CTRL_HOT_JOIN_NACK is set, indicating hj_rejected is true, it signifies the controller's non-receptiveness to the hot-join event. Consequently, we can safely disable the SIR interrupt signal if none of the target devices request SIR (reg == 0xffffffff).

Quoted 6 lines

Case 2:
When DEV_CTRL_HOT_JOIN_NACK is unset, indicating hj_rejected is false, this indicates the controller's readiness to engage with the hot-join event. Therefore, it's imperative to keep the SIR interrupt signal enabled, even if not all target devices request SIR. In this case, global is false and enable is false.

Quoted 4 lines

Yep, I see what you're doing there, but it looks like the correct state would never be set if we're not enabling/disabling the IBIs separately; with this code, we would only ever enable the SIR for the HJ if we
*also* happen to enable IBIs.

When we want to enable the SIR, the current code doesn't check whether the hot-join is enable or not. The modification I made pertains to disabling the SIR interrupt. It doesn't alter the logic for enabling the SIR.

```
	reg = readl(master->regs + IBI_SIR_REQ_REJECT);
	if (enable) {
		global = reg == 0xffffffff;
		reg &= ~BIT(idx);
	}
```
Quoted 3 lines

The initial state would be to have all SIRs masked.

Yes, indeed. The "global" variable is also true because "reg == 0xffffffff" is true. Therefore, the INTR_IBI_THLD_STAT bit will be set in the following code.

Quoted 15 lines
Quoted 1 line

Billy recently submitted a change to implement the hot-join

enabling/disabling. Therefore, it is timely to consider the hot-join functionality.

Quoted 2 lines

https://patchwork.kernel.org/project/linux-i3c/patch/20240429073624.25
[email protected]/

Yep, I saw that, excellent! It's next on my list to take a look at.

It's just a little unusual that we're enabling the HJ interrupt before actually having the HJ support though.

Cheers,

Jeremy

#7 lore raw

Hi Dylan,

Quoted 7 lines
Quoted 2 lines

The initial state would be to have all SIRs masked.

Yes, indeed. The "global" variable is also true because "reg == 0xffffffff" is true.
Therefore, the INTR_IBI_THLD_STAT bit will be set in the following code.

That's mainly my point - none of this code is ever run unless the ->enable_ibi or ->disable_ibi controller callback is invoked.

So we'll end up with the HJ interrupt only being enabled if some i3c device driver enables IBIs, which is a bit of a weird side-effect.

It probably makes more sense when the rest of the HJ code is added, but not so much as a standalone patch.

Cheers,

Jeremy