linux-kernel - Re: [PATCH] intel_th: core: fix null pointer dereference in intel_th

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: <61a6cca8379529c49a031d51f90d2da6@linux.intel.com>
Date: Fri, 26 Sep 2025 19:19:11 +0300
From: alex@....works
To: David Arcari <darcari@...hat.com>
Cc: linux-kernel@...r.kernel.org, Jerry Hoemann <jerry.hoemann@....com>,
 alexander.shishkin@...ux.intel.com, alex@....works
Subject: Re: [PATCH] intel_th: core: fix null pointer dereference in
 intel_th_irq

On 2025-08-25 20:45, David Arcari wrote:
> In certain cases intel_th_irq can reference a null entry in
> the th->thdev array.  This results in the splat shown below.
> The problem is that intel_th_output_enable() can modify the
> thdev[] array at the same time intel_th_irq is referencing
> the same array.  This can be fixed by disabling interrupts
> during the call to intel_th_output_enable().

Hi David,

Thank you for the bug report and rootcausing! Can you please also
detail the sequence of actions by which this is reproduced, so
that I can test my fix and not bother you with a back-and-forth
over-email debugging and also add it to our regression testing?
Doesn't have to be a shell script (although I wouldn't say no
to that), plain english would work in a pinch. If you have the
time, I'm also curious about your use case for intel_th.

This has eluded our testing for about 10 years, so I'm very
interested in the reproducer.

> BUG: kernel NULL pointer dereference, address: 0000000000000304
> Oops: Oops: 0000 [#1] SMP NOPTI
> RIP: 0010:intel_th_irq+0x26/0x70 [intel_th]

Yes, this is absolutely a bug.

> @@ -715,7 +715,9 @@ intel_th_subdevice_alloc(struct intel_th *th,
>  int intel_th_output_enable(struct intel_th *th, unsigned int otype)
>  {
>  	struct intel_th_device *thdev;
> -	int src = 0, dst = 0;
> +	int src = 0, dst = 0, ret = 0;
> +
> +	disable_irq(th->irq);
> 
>  	for (src = 0, dst = 0; dst <= th->num_thdevs; src++, dst++) {
>  		for (; src < ARRAY_SIZE(intel_th_subdevices); src++) {

[...]

> @@ -750,16 +752,19 @@ int intel_th_output_enable(struct intel_th *th, 
> unsigned int otype)
>  			goto found;
>  	}
> 
> +nodev:
> +	enable_irq(th->irq);
>  	return -ENODEV;
> 
>  found:
>  	thdev = intel_th_subdevice_alloc(th, &intel_th_subdevices[src]);
>  	if (IS_ERR(thdev))
> -		return PTR_ERR(thdev);
> -
> -	th->thdev[th->num_thdevs++] = thdev;
> +		ret = PTR_ERR(thdev);
> +	else
> +		th->thdev[th->num_thdevs++] = thdev;
> 
> -	return 0;
> +	enable_irq(th->irq);
> +	return ret;
>  }
>  EXPORT_SYMBOL_GPL(intel_th_output_enable);

This is indeed a possible fix, but I believe a little bit of
serialization can be employed here.

Lastly, my apologies for tardiness.

Thanks!
--
Alex