iommu: Don't reserve 0-length IOVA region
authorAshish Mhetre <amhetre@nvidia.com>
Tue, 5 Dec 2023 06:56:56 +0000 (12:26 +0530)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 25 Jan 2024 23:35:54 +0000 (15:35 -0800)
commit98b8a550da83cc392a14298c4b3eaaf0332ae6ad
treee05c5452f3811096f2de709ffe341916269c9fbf
parentfd83ff901d69f6ecc2381c3dda412cca14b984bb
iommu: Don't reserve 0-length IOVA region

[ Upstream commit bb57f6705960bebeb832142ce9abf43220c3eab1 ]

When the bootloader/firmware doesn't setup the framebuffers, their
address and size are 0 in "iommu-addresses" property. If IOVA region is
reserved with 0 length, then it ends up corrupting the IOVA rbtree with
an entry which has pfn_hi < pfn_lo.
If we intend to use display driver in kernel without framebuffer then
it's causing the display IOMMU mappings to fail as entire valid IOVA
space is reserved when address and length are passed as 0.
An ideal solution would be firmware removing the "iommu-addresses"
property and corresponding "memory-region" if display is not present.
But the kernel should be able to handle this by checking for size of
IOVA region and skipping the IOVA reservation if size is 0. Also, add
a warning if firmware is requesting 0-length IOVA region reservation.

Fixes: a5bf3cfce8cb ("iommu: Implement of_iommu_get_resv_regions()")
Signed-off-by: Ashish Mhetre <amhetre@nvidia.com>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/20231205065656.9544-1-amhetre@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/iommu/of_iommu.c