linux-kernel - [PATCH v4 0/1] Workaround for partial huge page unmaps in Panthor

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20251214202426.3186811-1-adrian.larumbe@collabora.com>
Date: Sun, 14 Dec 2025 20:24:24 +0000
From: Adrián Larumbe <adrian.larumbe@...labora.com>
To: linux-kernel@...r.kernel.org
Cc: dri-devel@...ts.freedesktop.org,
	Steven Price <steven.price@....com>,
	Boris Brezillon <boris.brezillon@...labora.com>,
	kernel@...labora.com,
	Adrián Larumbe <adrian.larumbe@...labora.com>
Subject: [PATCH v4 0/1] Workaround for partial huge page unmaps in Panthor

This is v4 of [1]. This patch is a workaround for performing partial unmaps of a VM region backed
by huge pages. Since these are now disallowed, the patch makes sure unmaps are done on a backing
page-granularity, and then regions untouched by the VM_BIND unmap operation are restored.

A patch series with IGT tests to validate this functionality is found at [2].

Changelog:
v4:
 - Added VM lock for expanded unmap region.
 - No longer pass the original VMA when calculating expanded unmap address and ranges for
 calculating BO offsets, as this can be obtained from the map va op's
 - Added more comments and renamed the unmap boundary calculation function
 - Calculate prev and next map offsets, sizes and addresses in the block prelude for
 the sake of clarity.
 - Addressed some minor style nits.
 - Rebased the patch onto the latest drm-misc.

v3:
 - Reworked address logic so that prev and next gpuava_op's va's are used in the calculations
   instead of those of the original unmap vma.
 - Got rid of the return struct from get_map_unmap_intervals() and now reckon panthor_vm_map_pages()
   arguments by fiddlign with the gpuva's respective gem object offsets.
 - Use folio_size() instead of folio_order() because the latter implies page sizes from the
   CPU's MMU perspective, rather than that of the GPU.

v2:
 - Fixed bug caused by confusion between semantics of gpu_va prev and next ops boundaries
   and those of the original vma object.
 - Coalesce all unmap operations into a single one.
 - Refactored and simplified code.

[1] https://lore.kernel.org/dri-devel/20251213190835.2444075-1-adrian.larumbe@collabora.com/T/#t
[2] https://lore.kernel.org/igt-dev/20251213190205.2435793-1-adrian.larumbe@collabora.com/T/#

Adrián Larumbe (1):
  drm/panthor: Support partial unmaps of huge pages

 drivers/gpu/drm/panthor/panthor_mmu.c | 99 ++++++++++++++++++++++++---
 1 file changed, 91 insertions(+), 8 deletions(-)


base-commit: d8684ae1cdcf848d21e00bc0e0de821d694a207b
prerequisite-patch-id: 3b0f61bfc22a616a205ff7c15d546d2049fd53de
--
2.51.2