Re: [RFC v3 00/10] KVM PCIe/MSI passthrough on ARM/ARM64 and IOVA reserved regions

From: Don Dutile
Date: Sat Dec 10 2016 - 21:06:12 EST


On 12/08/2016 04:36 AM, Auger Eric wrote:
Hi,

On 15/11/2016 14:09, Eric Auger wrote:
Following LPC discussions, we now report reserved regions through
iommu-group sysfs reserved_regions attribute file.


While I am respinning this series into v4, here is a tentative summary
of technical topics for which no consensus was reached at this point.

1) Shall we report the usable IOVA range instead of reserved IOVA
ranges. Not discussed at/after LPC.
x I currently report reserved regions. Alex expressed the need to
report the full usable IOVA range instead (x86 min-max range
minus MSI APIC window). I think this is meaningful for ARM
too where arm-smmu might not support the full 64b range.
x Any objection we report the usable IOVA regions instead?

2) Shall the kernel check collision with MSI window* when userspace
calls VFIO_IOMMU_MAP_DMA?
Joerg/Will No; Alex yes
*for IOVA regions consumed downstream to the IOMMU: everyone says NO

3) RMRR reporting in the iommu group sysfs? Joerg: yes; Don: no
Um, I'm missing context, but the only thing I recall saying no to wrt RMRR
is that _any_ device that has an RMRR cannot be assigned to a guest.
Or, are you saying, RMRR's should be exposed in the guest os? if so, then
you have my 'no' there.

My current series does not expose them in iommu group sysfs.
I understand we can expose the RMRR regions in the iomm group sysfs
without necessarily supporting RMRR requiring device assignment.
This sentence doesn't make sense to me.
Can you try re-wording it?
I can't tell what RMRR has to do w/device assignment, other than what I said above.
Exposing RMRR's in sysfs is not an issue in general.

We can also add this support later.

Thanks

Eric



Reserved regions are populated through the IOMMU get_resv_region callback
(former get_dm_regions), now implemented by amd-iommu, intel-iommu and
arm-smmu.

The intel-iommu reports the [FEE0_0000h - FEF0_000h] MSI window as an
IOMMU_RESV_NOMAP reserved region.

arm-smmu reports the MSI window (arbitrarily located at 0x8000000 and
1MB large) and the PCI host bridge windows.

The series integrates a not officially posted patch from Robin:
"iommu/dma: Allow MSI-only cookies".

This series currently does not address IRQ safety assessment.

Best Regards

Eric

Git: complete series available at
https://github.com/eauger/linux/tree/v4.9-rc5-reserved-rfc-v3

History:
RFC v2 -> v3:
- switch to an iommu-group sysfs API
- use new dummy allocator provided by Robin
- dummy allocator initialized by vfio-iommu-type1 after enumerating
the reserved regions
- at the moment ARM MSI base address/size is left unchanged compared
to v2
- we currently report reserved regions and not usable IOVA regions as
requested by Alex

RFC v1 -> v2:
- fix intel_add_reserved_regions
- add mutex lock/unlock in vfio_iommu_type1


Eric Auger (10):
iommu/dma: Allow MSI-only cookies
iommu: Rename iommu_dm_regions into iommu_resv_regions
iommu: Add new reserved IOMMU attributes
iommu: iommu_alloc_resv_region
iommu: Do not map reserved regions
iommu: iommu_get_group_resv_regions
iommu: Implement reserved_regions iommu-group sysfs file
iommu/vt-d: Implement reserved region get/put callbacks
iommu/arm-smmu: Implement reserved region get/put callbacks
vfio/type1: Get MSI cookie

drivers/iommu/amd_iommu.c | 20 +++---
drivers/iommu/arm-smmu.c | 52 +++++++++++++++
drivers/iommu/dma-iommu.c | 116 ++++++++++++++++++++++++++-------
drivers/iommu/intel-iommu.c | 50 ++++++++++----
drivers/iommu/iommu.c | 141 ++++++++++++++++++++++++++++++++++++----
drivers/vfio/vfio_iommu_type1.c | 26 ++++++++
include/linux/dma-iommu.h | 7 ++
include/linux/iommu.h | 49 ++++++++++----
8 files changed, 391 insertions(+), 70 deletions(-)