linux-kernel - Re: [PATCH platform-next v2 2/2] [PATCH platform-next 2/2] platform/mellanox: mlxreg-hotplug: Add support for handling interrupt storm

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <066ef64c-c6e4-f070-9851-ba9358b0970e@linux.intel.com>
Date: Tue, 23 Sep 2025 13:54:57 +0300 (EEST)
From: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
To: Ciju Rajan K <crajank@...dia.com>
cc: hdegoede@...hat.com, tglx@...utronix.de, 
    Andy Shevchenko <andriy.shevchenko@...ux.intel.com>, 
    LKML <linux-kernel@...r.kernel.org>, christophe.jaillet@...adoo.fr, 
    platform-driver-x86@...r.kernel.org, vadimp@...dia.com
Subject: Re: [PATCH platform-next v2 2/2] [PATCH platform-next 2/2]
 platform/mellanox: mlxreg-hotplug: Add support for handling interrupt
 storm

On Tue, 23 Sep 2025, Ciju Rajan K wrote:

> In case of broken hardware, it is possible that broken device will
> flood interrupt handler with false events. For example, if fan or
> power supply has damaged presence pin, it will cause permanent
> generation of plugged in / plugged out events. As a result, interrupt
> handler will consume a lot of CPU resources and will keep raising
> "UDEV" events to the user space.
> 
> This patch provides a mechanism to detect device causing interrupt
> flooding and mask interrupt for this specific device, to isolate
> from interrupt handling flow. Use the following criteria: if the
> specific interrupt was generated 'N' times during 'T' seconds,
> such device is to be considered as broken and will be closed for
> getting interrupts. User will be notified through the log error
> and will be instructed to replace broken device.
> 
> Reviewed-by: Vadim Pasternak <vadimp@...dia.com>
> Signed-off-by: Ciju Rajan K <crajank@...dia.com>
> ---
>  drivers/platform/mellanox/mlxreg-hotplug.c | 32 ++++++++++++++++++++--
>  1 file changed, 30 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/platform/mellanox/mlxreg-hotplug.c b/drivers/platform/mellanox/mlxreg-hotplug.c
> index d246772aafd6..ae0115ea1fd1 100644
> --- a/drivers/platform/mellanox/mlxreg-hotplug.c
> +++ b/drivers/platform/mellanox/mlxreg-hotplug.c
> @@ -11,6 +11,7 @@
>  #include <linux/hwmon-sysfs.h>
>  #include <linux/i2c.h>
>  #include <linux/interrupt.h>
> +#include <linux/jiffies.h>
>  #include <linux/module.h>
>  #include <linux/platform_data/mlxreg.h>
>  #include <linux/platform_device.h>
> @@ -30,6 +31,11 @@
>  #define MLXREG_HOTPLUG_ATTRS_MAX	128
>  #define MLXREG_HOTPLUG_NOT_ASSERT	3
>  
> +/* Interrupt storm definitions */
> +#define MLXREG_HOTPLUG_WM_COUNTER	100
> +/* Time window in milliseconds */
> +#define MLXREG_HOTPLUG_WM_WINDOW_MS	3000
> +
>  /**
>   * struct mlxreg_hotplug_priv_data - platform private data:
>   * @irq: platform device interrupt number;
> @@ -366,11 +372,33 @@ mlxreg_hotplug_work_helper(struct mlxreg_hotplug_priv_data *priv,
>  	for_each_set_bit(bit, &asserted, 8) {
>  		int pos;
>  
> +		/* Skip already marked storming bit. */
> +		if (item->storming_bits & BIT(bit))
> +			continue;
> +
>  		pos = mlxreg_hotplug_item_label_index_get(item->mask, bit);
>  		if (pos < 0)
>  			goto out;
>  
>  		data = item->data + pos;
> +
> +		/* Interrupt storm handling logic. */
> +		if (data->wmark_cntr == 0)
> +			data->wmark_window = jiffies +
> +				msecs_to_jiffies(MLXREG_HOTPLUG_WM_WINDOW_MS);

Please use braces for multi-line if blocks.

> +
> +		if (data->wmark_cntr >= MLXREG_HOTPLUG_WM_COUNTER - 1) {
> +			if (time_after(data->wmark_window, jiffies)) {
> +				dev_err(priv->dev,
> +					"Storming bit %d (label: %s) - interrupt masked permanently. Replace broken HW.",
> +					bit, data->label);
> +				/* Mark bit as storming. */
> +				item->storming_bits |= BIT(bit);
> +				continue;
> +			}
> +			data->wmark_cntr = 0;
> +		}
> +		data->wmark_cntr++;

I think this should be in else block to allow recalculation of the time 
window when the counter wraps.

>  		if (regval & BIT(bit)) {
>  			if (item->inversed)
>  				mlxreg_hotplug_device_destroy(priv, data, item->kind);
> @@ -390,9 +418,9 @@ mlxreg_hotplug_work_helper(struct mlxreg_hotplug_priv_data *priv,
>  	if (ret)
>  		goto out;
>  
> -	/* Unmask event. */
> +	/* Unmask event, exclude storming bits. */
>  	ret = regmap_write(priv->regmap, item->reg + MLXREG_HOTPLUG_MASK_OFF,
> -			   item->mask);
> +			   item->mask & ~item->storming_bits);
>  
>   out:
>  	if (ret)
> 

-- 
 i.