[dm-devel] dm-integrity stalls with iMX6q CAAM

[Gilad Ben-Yossef]

On Wed, Oct 17, 2018 at 4:35 PM Horia Geanta <horia.geanta at nxp.com> wrote:
>
> On 9/21/2018 3:06 PM, Neil Armstrong wrote:
> > Hi,
> >
> > I recently configured dm-crypt + dm-integrity on an iMX6q platform with CAAM Hash functions enabled using the following command lines :
> >
> > Linux 4.14.71
> >
> > cryptsetup luksFormat /dev/mmcblk1p3 --cipher aes-xts-plain64 --type luks2 --integrity hmac-sha256 --sector-size 512 --use-urandom
> > cryptsetup open /dev/mmcblk1p3 root
> > mkfs.ext4 -v -F /dev/mapper/root
> >
> > luksFormat and open finishes correctly, luksDump and status reports correct dm and luks properties.
> >
> > but when trying to access the /dev/mapper/root (mkfs, sha256sum or whatever), it blocks without any warning or errors reported by the kernel.
> > I can see a few (~170) interrupts on the jr0 interrupt then nothing.
> >
> > Is there a particular issue with hmac-sha256 with the linux CAAM driver ?
> >
> > When disabling the CAAM hash functions or using another cipher & integrity pair, it works like a charm.
> >
> The issue you are seeing is most probably related to CAAM driver not
> implementing correctly the crypto requests "backlogging".
> See commit c0403ec0bb5a ("Revert "dm crypt: fix deadlock when async crypto
> algorithm returns -EBUSY"") for a detailed description.
>
> Historically, there have been several CAAM backlogging implementations proposed
> - but all were rejected:

hmmm... obviously the right solution is to implement backlogging but
you might want to consider changing the driver to
return -ENOSPC or -ENOSYS instead in the meantime. It should cause
dm-crypt to fail rather than hang... which is not necessarily better,
but at least more obvious by the user what is going on?

Gilad

-- 
Gilad Ben-Yossef
Chief Coffee Drinker

values of β will give rise to dom!