RE: [PATCH] i801_smbus: clear SMBALERT status bit and disable SMBALERT interrupt

From: Xu, Lingyan (NSB - CN/Hangzhou)
Date: Mon Sep 02 2019 - 22:31:52 EST


Hi Jean,
Thanks a lot for your comments. And, yes, it is dangerous that clear all interrupt bit here based my local test. And about the interrupt flood, I will show you in attached file. And I agree with you that add SMBALERT interrupt handler if possible, but I have no idea about what action is need in this handler because that it seams that i801 can not clear salve chip's status bit directly.


Best Regards!
Lingyan xu

-----Original Message-----
From: Jean Delvare <jdelvare@xxxxxxx>
Sent: 2019年8月28日 21:58
To: Xu, Lingyan (NSB - CN/Hangzhou) <lingyan.xu@xxxxxxxxxxxxxxx>
Cc: Adamski, Krzysztof (Nokia - PL/Wroclaw) <krzysztof.adamski@xxxxxxxxx>; Wiebe, Wladislav (Nokia - DE/Ulm) <wladislav.wiebe@xxxxxxxxx>; linux-i2c@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
Subject: Re: [PATCH] i801_smbus: clear SMBALERT status bit and disable SMBALERT interrupt

Hi Lingyan,

On Mon, 12 Aug 2019 10:40:34 +0800, lingyxu wrote:
> From: Lingyan Xu <lingyan.xu@xxxxxxxxxxxxxxx>
>
> In current i801 driver, SMBALERT interrupt is allowed (Slave Command
> Register bit2 is 0).
> But these is no handler for SMBALERT interrupt in i801_isr, if there
> is SMBALERT interrupt asserted and deasserted,
> i801 will have an irq flood for the related status bit is setted.
>
> So SMBALERT interrupt handler is needed, and also, SMBALERT interrupt
> will be generated from time to time if slave chip have some fault.
> So disable SMBALERT interrupt is also needed.
>
> About the solution,
> please see http://www.farnell.com/datasheets/1581967.pdf
> Page632 P640 for more.
>
> Signed-off-by: Lingyan Xu <lingyan.xu@xxxxxxxxxxxxxxx>
> ---
> drivers/i2c/busses/i2c-i801.c | 7 ++++++-
> 1 files changed, 6 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/i2c/busses/i2c-i801.c
> b/drivers/i2c/busses/i2c-i801.c index f295693..033bafe 100644
> --- a/drivers/i2c/busses/i2c-i801.c
> +++ b/drivers/i2c/busses/i2c-i801.c
> @@ -661,9 +661,11 @@ static irqreturn_t i801_isr(int irq, void *dev_id)
> * Clear irq sources and report transaction result.
> * ->status must be cleared before the next transaction is started.
> */
> +
> + outb_p(status, SMBHSTSTS(priv));
> +
> status &= SMBHSTSTS_INTR | STATUS_ERROR_FLAGS;
> if (status) {
> - outb_p(status, SMBHSTSTS(priv));
> priv->status = status;
> wake_up(&priv->waitq);
> }

Looks scary. Writing the whole value of SMBHSTSTS back to itself without selecting which bits you write is dangerous. Specifically, writing back SMBHSTSTS_BYTE_DONE, SMBHSTSTS_INUSE_STS and SMBHSTSTS_HOST_BUSY could have unexpected consequences. I would feel much better if you would just explicitly add SMBHSTSTS_SMBALERT_STS to the list.

> @@ -1810,6 +1812,9 @@ static int i801_probe(struct pci_dev *dev, const struct pci_device_id *id)
> /* Default timeout in interrupt mode: 200 ms */
> priv->adapter.timeout = HZ / 5;
>
> + /* Disable SMBALERT interrupt */
> + outb_p(inb_p(SMBSLVCMD(priv)) | BIT(2), SMBSLVCMD(priv));

Please give SMBSLVCMD's BIT(2) a name and define it after SMBSLVCMD_HST_NTFY_INTREN.

Also it is mandatory to restore the value of SMBSLVCMD before returning the control back to the BIOS. Currently this is only being done when the FEATURE_HOST_NOTIFY bit is set because that's the only case where we change the value of that register, but if we change it unconditionally then it must be saved and restored unconditionally too.

> +
> if (dev->irq == IRQ_NOTCONNECTED)
> priv->features &= ~FEATURE_IRQ;
>

That being said, if you see this interrupt flood, it means that at least one device on your SMBus would benefit from SMBus Alert being supported. The infrastructure is already there as we added support in a few I2C bus drivers already. So maybe instead of silencing the interrupts, we could add proper SMBus Alert support to the i2c-i801 driver?

Did you figure out which device is raising the SMBus Alert and why?

--
Jean Delvare
SUSE L3 Support

Attachment: rootcause.PNG
Description: rootcause.PNG