Re: [PATCH 05/13] media: mgb4: protect driver against spectre

[Mauro Carvalho Chehab <mchehab+huawei@kernel.org>]

Em Wed, 16 Oct 2024 13:59:18 +0200
Martin Tůma <tumic@gpxsee.org> escreveu:

> On 16. 10. 24 12:22 odp., Mauro Carvalho Chehab wrote:
> > Frequency range is set from sysfs via frequency_range_store(),
> > being vulnerable to spectre, as reported by smatch:
> > 
> > 	drivers/media/pci/mgb4/mgb4_cmt.c:231 mgb4_cmt_set_vin_freq_range() warn: potential spectre issue 'cmt_vals_in' [r]
> > 	drivers/media/pci/mgb4/mgb4_cmt.c:238 mgb4_cmt_set_vin_freq_range() warn: possible spectre second half.  'reg_set'
> > 
> > Fix it.
> > 
> > Fixes: 0ab13674a9bd ("media: pci: mgb4: Added Digiteq Automotive MGB4 driver")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
> > ---
> >   drivers/media/pci/mgb4/mgb4_cmt.c | 2 ++
> >   1 file changed, 2 insertions(+)
> > 
> > diff --git a/drivers/media/pci/mgb4/mgb4_cmt.c b/drivers/media/pci/mgb4/mgb4_cmt.c
> > index 70dc78ef193c..a25b68403bc6 100644
> > --- a/drivers/media/pci/mgb4/mgb4_cmt.c
> > +++ b/drivers/media/pci/mgb4/mgb4_cmt.c
> > @@ -227,6 +227,8 @@ void mgb4_cmt_set_vin_freq_range(struct mgb4_vin_dev *vindev,
> >   	u32 config;
> >   	size_t i;
> >   
> > +	freq_range = array_index_nospec(freq_range, ARRAY_SIZE(cmt_vals_in));
> > +
> >   	addr = cmt_addrs_in[vindev->config->id];
> >   	reg_set = cmt_vals_in[freq_range];
> >     
> 
> I still do not fully understand the exact vulnerability here, but the 
> patch should definitely not do any harm, so I'm ok with it even if it's 
> real purpose would only be to silence the smatch warning :-)

With Spectre, just checking if freq_range is between 0 and the
size of the array is not enough, as malicious code could use CPU
speculative logic to retrieve data from memory outside the limits
of the array.

As freq_range is specified by the user via sysfs attribute
frequency_range, it is subject to Spectre v1 attack as described
at Documentation/admin-guide/hw-vuln/spectre.rst. 

Silencing smatch is a plus.

> 
> Reviewed-by: Martin Tůma <martin.tuma@digiteqautomotive.com>

Thanks!

Thanks,
Mauro