Re: [PATCH 0/2] iommu/dmar: expose fault counters via sysfs

[Lu Baolu]

Hi,

On 10/15/19 11:11 PM, Yuri Volchkov wrote:
> For health monitoring, it can be useful to know if iommu is behaving as
> expected. DMAR faults can be an indicator that a device:
>   - has been misconfigured, or
>   - has experienced a hardware hiccup and replacement should
>     be considered, or
>   - has been issuing faults due to malicious activity
>
> Currently the only way to check if there were any DMAR faults on the
> host is to scan the dmesg output. However this approach is not very
> elegant. The information we are looking for can be wrapped out of the
> buffer, or masked (since it is a rate-limited print) by another
> device.
>
> The series adds counters for DMAR faults and exposes them via sysfs.

We now have an iommu API named iommu_register_fault_handler() to
register callbacks for dmar faults. How about monitoring the dmar
fault through this api so that your code could be generic and vendor
agnostic?

Best regards,
Baolu

> Yuri Volchkov (2):
>    iommu/dmar: collect fault statistics
>    iommu/dmar: catch early fault occurrences
>
>   drivers/iommu/dmar.c        | 182 ++++++++++++++++++++++++++++++++----
>   drivers/iommu/intel-iommu.c |   1 +
>   drivers/pci/pci-sysfs.c     |  20 ++++
>   include/linux/intel-iommu.h |   4 +
>   include/linux/pci.h         |  11 +++
>   5 files changed, 201 insertions(+), 17 deletions(-)