When a zone device page is split (via huge pmd folio split). The driver
callback for folio_split is invoked to let the device driver know that the
folio size has been split into a smaller order.
Provide a default implementation for drivers that do not provide this
callback that copies the pgmap and mapping fields for the split folios.
Update the HMM test driver to handle the split.
Link: https://lkml.kernel.org/r/20251001065707.920170-11-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Enhance the hmm test driver (lib/test_hmm) with support for THP pages.
A new pool of free_folios() has now been added to the dmirror device,
which can be allocated when a request for a THP zone device private page
is made.
Add compound page awareness to the allocation function during normal
migration and fault based migration. These routines also copy
folio_nr_pages() when moving data between system memory and device memory.
args.src and args.dst used to hold migration entries are now dynamically
allocated (as they need to hold HPAGE_PMD_NR entries or more).
Split and migrate support will be added in future patches in this series.
Link: https://lkml.kernel.org/r/20251001065707.920170-10-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Implement CPU fault handling for zone device THP entries through
do_huge_pmd_device_private(), enabling transparent migration of
device-private large pages back to system memory on CPU access.
When the CPU accesses a zone device THP entry, the fault handler calls the
device driver's migrate_to_ram() callback to migrate the entire large page
back to system memory.
Link: https://lkml.kernel.org/r/20251001065707.920170-9-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
MIGRATE_VMA_SELECT_COMPOUND will be used to select THP pages during
migrate_vma_setup() and MIGRATE_PFN_COMPOUND will make migrating device
pages as compound pages during device pfn migration.
migrate_device code paths go through the collect, setup and finalize
phases of migration.
The entries in src and dst arrays passed to these functions still remain
at a PAGE_SIZE granularity. When a compound page is passed, the first
entry has the PFN along with MIGRATE_PFN_COMPOUND and other flags set
(MIGRATE_PFN_MIGRATE, MIGRATE_PFN_VALID), the remaining entries
(HPAGE_PMD_NR - 1) are filled with 0's. This representation allows for
the compound page to be split into smaller page sizes.
migrate_vma_collect_hole(), migrate_vma_collect_pmd() are now THP page
aware. Two new helper functions migrate_vma_collect_huge_pmd() and
migrate_vma_insert_huge_pmd_page() have been added.
migrate_vma_collect_huge_pmd() can collect THP pages, but if for some
reason this fails, there is fallback support to split the folio and
migrate it.
migrate_vma_insert_huge_pmd_page() closely follows the logic of
migrate_vma_insert_page()
Support for splitting pages as needed for migration will follow in later
patches in this series.
Link: https://lkml.kernel.org/r/20251001065707.920170-8-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Extend migrate_vma_collect_pmd() to handle partially mapped large folios
that require splitting before migration can proceed.
During PTE walk in the collection phase, if a large folio is only
partially mapped in the migration range, it must be split to ensure the
folio is correctly migrated.
[matthew.brost@intel.com: handle partially mapped folios during split]
Link: https://lkml.kernel.org/r/20251120230825.181072-1-matthew.brost@intel.com
Link: https://lkml.kernel.org/r/20251001065707.920170-7-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Add support for splitting device-private THP folios, enabling fallback
to smaller page sizes when large page allocation or migration fails.
Key changes:
- split_huge_pmd(): Handle device-private PMD entries during splitting
- Preserve RMAP_EXCLUSIVE semantics for anonymous exclusive folios
- Skip RMP_USE_SHARED_ZEROPAGE for device-private entries as they
don't support shared zero page semantics
Link: https://lkml.kernel.org/r/20251001065707.920170-6-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Add device-private THP support to reverse mapping infrastructure, enabling
proper handling during migration and walk operations.
The key changes are:
- add_migration_pmd()/remove_migration_pmd(): Handle device-private
entries during folio migration and splitting
- page_vma_mapped_walk(): Recognize device-private THP entries during
VMA traversal operations
This change supports folio splitting and migration operations on
device-private entries.
[balbirs@nvidia.com: fix override of entry in remove_migration_pmd]
Link: https://lkml.kernel.org/r/20251114012153.2634497-2-balbirs@nvidia.com
[balbirs@nvidia.com: follow pattern used in remove_migration_pte()]
Link: https://lkml.kernel.org/r/20251115002835.3515194-1-balbirs@nvidia.com
Link: https://lkml.kernel.org/r/20251001065707.920170-5-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Reviewed-by: SeongJae Park <sj@kernel.org>
Acked-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Extend core huge page management functions to handle device-private THP
entries. This enables proper handling of large device-private folios in
fundamental MM operations.
The following functions have been updated:
- copy_huge_pmd(): Handle device-private entries during fork/clone
- zap_huge_pmd(): Properly free device-private THP during munmap
- change_huge_pmd(): Support protection changes on device-private THP
- __pte_offset_map(): Add device-private entry awareness
Link: https://lkml.kernel.org/r/20251001065707.920170-4-balbirs@nvidia.com
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Acked-by: Zi Yan <ziy@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Change page_free to folio_free to make the folio support for
zone device-private more consistent. The PCI P2PDMA callback
has also been updated and changed to folio_free() as a result.
For drivers that do not support folios (yet), the folio is
converted back into page via &folio->page and the page is used
as is, in the current callback implementation.
Link: https://lkml.kernel.org/r/20251001065707.920170-3-balbirs@nvidia.com
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm: support device-private THP", v7.
This patch series introduces support for Transparent Huge Page (THP)
migration in zone device-private memory. The implementation enables
efficient migration of large folios between system memory and
device-private memory
Background
Current zone device-private memory implementation only supports PAGE_SIZE
granularity, leading to:
- Increased TLB pressure
- Inefficient migration between CPU and device memory
This series extends the existing zone device-private infrastructure to
support THP, leading to:
- Reduced page table overhead
- Improved memory bandwidth utilization
- Seamless fallback to base pages when needed
In my local testing (using lib/test_hmm) and a throughput test, the series
shows a 350% improvement in data transfer throughput and a 80% improvement
in latency
These patches build on the earlier posts by Ralph Campbell [1]
Two new flags are added in vma_migration to select and mark compound
pages. migrate_vma_setup(), migrate_vma_pages() and
migrate_vma_finalize() support migration of these pages when
MIGRATE_VMA_SELECT_COMPOUND is passed in as arguments.
The series also adds zone device awareness to (m)THP pages along with
fault handling of large zone device private pages. page vma walk and the
rmap code is also zone device aware. Support has also been added for
folios that might need to be split in the middle of migration (when the
src and dst do not agree on MIGRATE_PFN_COMPOUND), that occurs when src
side of the migration can migrate large pages, but the destination has not
been able to allocate large pages. The code supported and used
folio_split() when migrating THP pages, this is used when
MIGRATE_VMA_SELECT_COMPOUND is not passed as an argument to
migrate_vma_setup().
The test infrastructure lib/test_hmm.c has been enhanced to support THP
migration. A new ioctl to emulate failure of large page allocations has
been added to test the folio split code path. hmm-tests.c has new test
cases for huge page migration and to test the folio split path. A new
throughput test has been added as well.
The nouveau dmem code has been enhanced to use the new THP migration
capability.
mTHP support:
The patches hard code, HPAGE_PMD_NR in a few places, but the code has been
kept generic to support various order sizes. With additional refactoring
of the code support of different order sizes should be possible.
The future plan is to post enhancements to support mTHP with a rough
design as follows:
1. Add the notion of allowable thp orders to the HMM based test driver
2. For non PMD based THP paths in migrate_device.c, check to see if
a suitable order is found and supported by the driver
3. Iterate across orders to check the highest supported order for migration
4. Migrate and finalize
The mTHP patches can be built on top of this series, the key design
elements that need to be worked out are infrastructure and driver support
for multiple ordered pages and their migration.
HMM support for large folios was added in 10b9feee2d ("mm/hmm:
populate PFNs from PMD swap entry").
This patch (of 16)
Add routines to support allocation of large order zone device folios and
helper functions for zone device folios, to check if a folio is device
private and helpers for setting zone device data.
When large folios are used, the existing page_free() callback in pgmap is
called when the folio is freed, this is true for both PAGE_SIZE and higher
order pages.
Zone device private large folios do not support deferred split and scan
like normal THP folios.
Link: https://lkml.kernel.org/r/20251001065707.920170-1-balbirs@nvidia.com
Link: https://lkml.kernel.org/r/20251001065707.920170-2-balbirs@nvidia.com
Link: https://lore.kernel.org/linux-mm/20201106005147.20113-1-rcampbell@nvidia.com/ [1]
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
For hugetlbs, gmap puds have the present bit set. For normal puds (which
point to ptes), the bit is not set. This is in contrast to the normal
userspace puds, which always have the bit set for present pmds.
This causes issues when ___pte_offset_map() is modified to only check for
the present bit.
The solution to the problem is simply to always set the present bit for
present gmap pmds.
Link: https://lkml.kernel.org/r/20251028130150.57379-2-imbrenda@linux.ibm.com
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Link: https://lore.kernel.org/lkml/20251017144924.10034-1-borntraeger@linux.ibm.com/
Tested-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Acked-by: Christian Borntraeger <borntraeger@linux.ibm.com>
Acked-by: Balbir Singh <balbirs@nvidia.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: "Huang, Ying" <ying.huang@linux.alibaba.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: Mathew Brost <matthew.brost@intel.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
After recovering from a PCI error through reset, affected devices are in
D0_uninitialized state and need to be brought into D0_active state by
re-initializing their Config Space registers (PCIe r7.0 sec 5.3.1.1).
To facilitate that, the PCI core provides pci_restore_state() and
pci_save_state() helpers. Document rules governing their usage.
As Bjorn notes, so far no file in "Documentation/ includes anything about
the idea of a driver using pci_save_state() to capture the state it wants
to restore after an error", even though it is a common pattern in drivers.
So that's obviously a gap that should be closed.
Reported-by: Bjorn Helgaas <helgaas@kernel.org>
Closes: https://lore.kernel.org/r/20251113161556.GA2284238@bhelgaas/
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Link: https://patch.msgid.link/077596ba70202be0e43fdad3bb9b93d356cbe4ec.1763746079.git.lukas@wunner.de
In 2009, commit c82f63e411 ("PCI: check saved state before restore")
changed the behavior of pci_restore_state() such that it became necessary
to call pci_save_state() afterwards, lest recovery from subsequent PCI
errors fails.
The commit has just been reverted and so all the pci_save_state() after
pci_restore_state() calls that have accumulated in the tree are now
superfluous. Drop them.
Two drivers chose a different approach to achieve the same result:
drivers/scsi/ipr.c and drivers/net/ethernet/intel/e1000e/netdev.c set the
pci_dev's "state_saved" flag to true before calling pci_restore_state().
Drop this as well.
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Dave Jiang <dave.jiang@intel.com>
Acked-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com> # qat
Link: https://patch.msgid.link/c2b28cc4defa1b743cf1dedee23c455be98b397a.1760274044.git.lukas@wunner.de
When the PCI core gained power management support in 2002, it introduced
pci_save_state() and pci_restore_state() helpers to restore Config Space
after a D3hot or D3cold transition, which implies a Soft or Fundamental
Reset (PCIe r7.0 sec 5.8):
https://git.kernel.org/tglx/history/c/a5287abe398b
In 2006, EEH and AER were introduced to recover from errors by performing
a reset. Because errors can occur at any time, drivers began calling
pci_save_state() on probe to ensure recoverability.
In 2009, recoverability was foiled by commit c82f63e411 ("PCI: check
saved state before restore"): It amended pci_restore_state() to bail out
if the "state_saved" flag has been cleared. The flag is cleared by
pci_restore_state() itself, hence a saved state is now allowed to be
restored only once and is then invalidated. That doesn't seem to make
sense because the saved state should be good enough to be reused.
Soon after, drivers began to work around this behavior by calling
pci_save_state() immediately after pci_restore_state(), see e.g. commit
b94f2d775a ("igb: call pci_save_state after pci_restore_state").
Hilariously, two drivers even set the "saved_state" flag to true before
invoking pci_restore_state(), see ipr_reset_restore_cfg_space() and
e1000_io_slot_reset().
Despite these workarounds, recoverability at all times is not guaranteed:
E.g. when a PCIe port goes through a runtime suspend and resume cycle,
the "saved_state" flag is cleared by:
pci_pm_runtime_resume()
pci_pm_default_resume_early()
pci_restore_state()
... and hence on a subsequent AER event, the port's Config Space cannot be
restored. Riana reports a recovery failure of a GPU-integrated PCIe
switch and has root-caused it to the behavior of pci_restore_state().
Another workaround would be necessary, namely calling pci_save_state() in
pcie_port_device_runtime_resume().
The motivation of commit c82f63e411 was to prevent restoring state if
pci_save_state() hasn't been called before. But that can be achieved by
saving state already on device addition, after Config Space has been
initialized. A desirable side effect is that devices become recoverable
even if no driver gets bound. This renders the commit unnecessary, so
revert it.
Reported-by: Riana Tauro <riana.tauro@intel.com> # off-list
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tested-by: Riana Tauro <riana.tauro@intel.com>
Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Link: https://patch.msgid.link/9e34ce61c5404e99ffdd29205122c6fb334b38aa.1763483367.git.lukas@wunner.de
The state_saved flag tells the PCI core whether a driver assumes
responsibility to save Config Space and put the device into a low power
state on suspend.
The flag is currently initialized to false on enumeration, even though it
already is false (because struct pci_dev is zeroed by kzalloc()) and even
though it is set to false before commencing the suspend sequence (the only
code path where it's relevant).
The flag is also set to false in pci_pm_thaw(), i.e. on resume, when it's
no longer relevant.
Drop these two superfluous flag assignments for simplicity.
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Link: https://patch.msgid.link/fd167945bd7852e1ca08cd4b202130659eea2c2f.1763483367.git.lukas@wunner.de
When a PCI device is suspended, it is normally the PCI core's job to save
Config Space and put the device into a low power state. However drivers
are allowed to assume these responsibilities. When they do, the PCI core
can tell by looking at the state_saved flag in struct pci_dev: The flag
is cleared before commencing the suspend sequence and it is set when
pci_save_state() is called. If the PCI core finds the flag set late in
the suspend sequence, it refrains from calling pci_save_state() itself.
But there are two corner cases where the PCI core neglects to clear the
flag before commencing the suspend sequence:
* If a driver has legacy PCI PM callbacks, pci_legacy_suspend() neglects
to clear the flag. The (stale) flag is subsequently queried by
pci_legacy_suspend() itself and pci_legacy_suspend_late().
* If a device has no driver or its driver has no PCI PM callbacks,
pci_pm_freeze() neglects to clear the flag. The (stale) flag is
subsequently queried by pci_pm_freeze_noirq().
The flag may be set prior to suspend if the device went through error
recovery: Drivers commonly invoke pci_restore_state() + pci_save_state()
to restore Config Space after reset.
The flag may also be set if drivers call pci_save_state() on probe to
allow for recovery from subsequent errors.
The result is that pci_legacy_suspend_late() and pci_pm_freeze_noirq()
don't call pci_save_state() and so the state that will be restored on
resume is the one recorded on last error recovery or on probe, not the one
that the device had on suspend. If the two states happen to be identical,
there's no problem.
Reinstate clearing the flag in pci_legacy_suspend() and pci_pm_freeze().
The two functions used to do that until commit 4b77b0a2ba ("PCI: Clear
saved_state after the state has been restored") deemed it unnecessary
because it assumed that it's sufficient to clear the flag on resume in
pci_restore_state(). The commit seemingly did not take into account that
pci_save_state() and pci_restore_state() are not only used by power
management code, but also for error recovery.
Devices without driver or whose driver has no PCI PM callbacks may be in
runtime suspend when pci_pm_freeze() is called. Their state has already
been saved, so don't clear the flag to skip a pointless pci_save_state()
in pci_pm_freeze_noirq().
None of the drivers with legacy PCI PM callbacks seem to use runtime PM,
so clear the flag unconditionally in their case.
Fixes: 4b77b0a2ba ("PCI: Clear saved_state after the state has been restored")
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Cc: stable@vger.kernel.org # v2.6.32+
Link: https://patch.msgid.link/094f2aad64418710daf0940112abe5a0afdc6bce.1763483367.git.lukas@wunner.de
Ensure that the DMA address of the framebuffer flush page is not larger
than its hardware register.
On GPUs older than Hopper, the register for the address can hold up to a
40-bit address (right-shifted by 8 so that it fits in the 32-bit
register), and on Hopper and later it can be 52 bits (64-bit register
where bits 52-63 must be zero).
Recently it was discovered that under certain conditions, the flush page
could be allocated outside this range. Although this bug was fixed, we
can ensure that any future changes to this code don't accidentally
generate an invalid page address.
Signed-off-by: Timur Tabi <ttabi@nvidia.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20251113230323.1271726-2-ttabi@nvidia.com
The flush page DMA address is stored in a special register that is not
associated with the GPU's standard DMA range. For example, on Turing,
the GPU's MMU can handle 47-bit addresses, but the flush page address
register is limited to 40 bits.
At the point during device initialization when the flush page is
allocated, the DMA mask is still at its default of 32 bits. So even
though it's unlikely that the flush page could exist above a 40-bit
address, the dma_map_page() call could fail, e.g. if IOMMU is disabled
and the address is above 32 bits. The simplest way to achieve all
constraints is to allocate the page in the DMA32 zone. Since the flush
page is literally just a page, this is an acceptable limitation. The
alternative is to temporarily set the DMA mask to 40 (or 52 for Hopper
and later) bits, but that could have unforseen side effects.
In situations where the flush page is allocated above 32 bits and IOMMU
is disabled, you will get an error like this:
nouveau 0000:65:00.0: DMA addr 0x0000000107c56000+4096 overflow (mask ffffffff, bus limit 0).
Fixes: 5728d06419 ("drm/nouveau/fb: handle sysmem flush page from common code")
Signed-off-by: Timur Tabi <ttabi@nvidia.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20251113230323.1271726-1-ttabi@nvidia.com
L1 PM Substates for RC mode require support in the dw-rockchip driver
including proper handling of the CLKREQ# sideband signal. It is mostly
handled by hardware, but software still needs to set the clkreq fields
in the PCIE_CLIENT_POWER_CON register to match the hardware implementation.
For more details, see section '18.6.6.4 L1 Substate' in the RK3568 TRM 1.1
Part 2, or section '11.6.6.4 L1 Substate' in the RK3588 TRM 1.0 Part2.
[bhelgaas: set pci->l1ss_support so DWC core preserves L1SS Capability bits;
drop corresponding code here, include updates from
https://lore.kernel.org/r/aRRG8wv13HxOCqgA@ryzen]
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Link: https://patch.msgid.link/1761187883-150120-1-git-send-email-shawn.lin@rock-chips.com
Link: https://patch.msgid.link/20251118214312.2598220-4-helgaas@kernel.org
The DWC core clears the L1 Substates Supported bits unless the driver sets
the "dw_pcie.l1ss_support" flag.
The tegra194 init_host_aspm() sets "dw_pcie.l1ss_support" if the platform
has the "supports-clkreq" DT property. If "supports-clkreq" is absent,
"dw_pcie.l1ss_support" is not set, and the DWC core will clear the L1
Substates Supported bits.
The tegra194 code to clear the L1 Substates Supported bits is unnecessary,
so remove it.
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Link: https://patch.msgid.link/20251118214312.2598220-3-helgaas@kernel.org
L1 PM Substates require the CLKREQ# signal and may also require
device-specific support. If CLKREQ# is not supported or driver support is
lacking, enabling L1.1 or L1.2 may cause errors when accessing devices,
e.g.,
nvme nvme0: controller is down; will reset: CSTS=0xffffffff, PCI_STATUS=0x10
If the kernel is built with CONFIG_PCIEASPM_POWER_SUPERSAVE=y or users
enable L1.x via sysfs, users may trip over these errors even if L1
Substates haven't been enabled by firmware or the driver.
To prevent such errors, disable advertising the L1 PM Substates unless the
driver sets "dw_pcie.l1ss_support" to indicate that it knows CLKREQ# is
present and any device-specific configuration has been done.
Set "dw_pcie.l1ss_support" in tegra194 (if DT includes the
"supports-clkreq' property) and qcom (for cfg_2_7_0, cfg_1_9_0, cfg_1_34_0,
and cfg_sc8280xp controllers) so they can continue to use L1 Substates.
Based on Niklas's patch:
https://patch.msgid.link/20251017163252.598812-2-cassel@kernel.org
[bhelgaas: drop hiding for endpoints]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Link: https://patch.msgid.link/20251118214312.2598220-2-helgaas@kernel.org
As per DesignWare Cores PCI Express Controller Databook, section 5.50,
SII: Debug Signals, cxpl_debug_info[63:0]:
[5:0] smlh_ltssm_state: LTSSM current state. Encoding is same as the
dedicated smlh_ltssm_state output.
The mask should be 6 bits, from 0 to 5. Hence, fix the mask definition.
Fixes: 23fe5bd4be ("PCI: keystone: Cleanup ks_pcie_link_up()")
Signed-off-by: Shawn Lin <shawn.lin@rock-chips.com>
[mani: reworded description]
Signed-off-by: Manivannan Sadhasivam <mani@kernel.org>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Link: https://patch.msgid.link/1763122140-203068-1-git-send-email-shawn.lin@rock-chips.com
Add a new test case that triggers the HW deactivation emulation path
when trapping ICV_DIR_EL1. This is obviously tied to the way KVM
works now, but the test follows the expected architectural behaviour.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-50-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Add a new test case that inject a Group-0 interrupt together
with a bunch of Group-1 interrupts, Ack/EOI the G1 interrupts,
and only then enable G0, expecting to get the G0 interrupt.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-49-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Add a new test case that makes an interrupt pending on a vcpu,
activates it, do the priority drop, and then get *another* vcpu
to do the deactivation.
Special care is taken not to trigger an exit in the process, so
that we are sure that the active interrupt is in an LR. Joy.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-48-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
When EOImode==1, perform the deactivation in the order of activation,
just to make things a bit worse for KVM. Yes, I'm nasty.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-47-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Good news: our GIC emulation is not completely broken, and we can
activate as many interrupts as we want.
Bump the test to cover all the SGIs, all the allowed PPIs, and
31 SPIs. Yes, 31, because we have 31 available priorities, and the
test is not happy with having two interrupts with the same priority.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-46-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
The PPI injection API is clear that you can't inject the timer PPIs
from userspace, since they are controlled by the timers themselves.
Add an exclusion list for this purpose.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-45-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
The architecture is pretty clear that changing the configuration of
an enable interrupt is not OK. It doesn't really matter here, but
doing the right thing is not more expensive.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-44-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
No, 0 is not a spurious INTID. Never been, never was.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-43-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Make sure G0 is disabled at the point of initialising the GIC.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-42-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Being able to set the group of an interrupt is pretty useful.
Add such a helper.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-41-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Since we can't decide to trap the DIR register on a per-vcpu basis,
always trap the second page of the GIC CPU interface. Yes, this is
costly. On the bright side, no sane SW should use EOImode==1 on
GICv2...
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-40-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Add the plumbing of GICv2 interrupt deactivation via GICV_DIR.
This requires adding a new device so that we can easily decode
the DIR address.
The deactivation itself is very similar to the GICv3 version.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-39-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Similarly to the GICv3 version, handle the EOIcount-driven deactivation
by walking the overflow list.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-38-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
FEAT_NV2 is pretty terrible for anything that tries to enforce immediate
effects, and writing to ICH_HCR_EL2 in the hope to disable a maintenance
interrupt is vain. This only hits memory, and the guest hasn't cleared
anything -- the MI will fire.
For example, running the vgic_irq test under NV results in about 800
maintenance interrupts being actually handled by the L1 guest,
when none were expected.
As a cheap workaround, read back ICH_MISR_EL2 after writing 0 to
ICH_HCR_EL2. This is very cheap on real HW, and causes a trap to
the host in NV, giving it the opportunity to retire the pending MI.
With this, the above test runs to completion without any MI being
actually handled.
Yes, this is really poor...
Tested-by: Fuad Tabba <tabba@google.com>
Reviewed-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-37-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Pretty much like the rest of the LR handling, deactivation of an
L2 interrupt gets reflected in the L1 LRs, and therefore must be
propagated into the L1 shadow state if the interrupt is HW-bound.
Instead of directly handling the active state (which looks a bit
off as it ignores locking and L1->L0 HW propagation), use the new
deactivation primitive to perform the deactivation and deal with
the required maintenance.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-36-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
The current approach to nested GICv3 support is to not do anything
while L2 is running, wait a transition from L2 to L1 to resync
LRs, VMCR and HCR, and only then evaluate the state to decide
whether to generate a maintenance interrupt.
This doesn't provide a good quality of emulation, and it would be
far preferable to find out early that we need to perform a switch.
Move the LRs/VMCR and HCR resync into vgic_v3_sync_nested(), so
that we have most of the state available. As we turning the vgic
off at this stage to avoid a screaming host MI, add a new helper
vgic_v3_flush_nested() that switches the vgic on again. The MI can
then be directly injected as required.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-35-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
CPUs lacking TDIR always trap ICV_DIR_EL1, no matter what, since
we have ICH_HCR_EL2.TC set permanently. For these CPUs, it is
useless to use a broadcast kick on SPI injection, as the sole
purpose of this is to set TDIR.
We can therefore skip this on these CPUs, which are challenged
enough not to be burdened by extra IPIs. As a consequence,
permanently set the TDIR bit in the shadow state to notify the
fast-path emulation code of the exit reason.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-34-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Even when we have either an LR overflow or SPIs in flight, it is
extremely likely that the interrupt being deactivated is still in
the LRs, and that going all the way back to the the generic trap
handling code is a waste of time.
Instead, try and deactivate in place when possible, and only if
this fails, perform a full exit.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-33-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
SPIs are specially annpying, as they can be activated on a CPU and
deactivated on another. WHich means that when an SPI is in flight
anywhere, all CPUs need to have their TDIR trap bit set.
This translates into broadcasting an IPI across all CPUs to make sure
they set their trap bit, The number of in-flight SPIs is kept in
an atomic variable so that CPUs can turn the trap bit off as soon
as possible.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-32-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Now that we are ready to handle deactivation through ICV_DIR_EL1,
set the trap bit if we have active interrupts outside of the LRs.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-31-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
The GICv2 SGIs require additional handling for deactivation, as they
are effectively multiple interrrupts muxed into one. Make sure we
check for the source CPU when deactivating.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-30-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Deactivation via ICV_DIR_EL1 is both relatively straightforward
(we have the interrupt that needs deactivation) and really awkward.
The main issue is that the interrupt may either be in an LR on
another CPU, or ourside of any LR.
In the former case, we process the deactivation is if ot was
a write to GICD_CACTIVERn, which is already implemented as a big
hammer IPI'ing all vcpus. In the latter case, we just perform
a normal deactivation, similar to what we do for EOImode==0.
Another annoying aspect is that we need to tell the CPU owning
the interrupt that its ap_list needs laudering. We use a brand new
vcpu request to that effect.
Note that this doesn't address deactivation via the GICV MMIO view,
which will be taken care of in a later change.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-29-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Now that we can identify interrupts that have not made it into the LRs,
it becomes relatively easy to use EOIcount to walk the overflow list.
What is a bit odd is that we compute a fake LR for the original
state of the interrupt, clear the active bit, and feed into the existing
logic for processing. In a way, this is what would have happened if
the interrupt was in an LR.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-28-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Add the maintenance interrupt to force an exit when the guest
enables/disables individual groups, so that we can resort the
ap_list accordingly.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-27-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Interrupts in the ap_list that cannot be acted upon because they
are not enabled, or that their group is not enabled, shouldn't
make it into the LRs if we are space-constrained.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-26-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Having established that pending interrupts should have priority
to be moved into the LRs over the active interrupts, implement this
in the ap_list sorting.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-25-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
Make the internal crystal ball global, so that implementation-specific
code can use it.
Tested-by: Fuad Tabba <tabba@google.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Tested-by: Mark Brown <broonie@kernel.org>
Link: https://msgid.link/20251120172540.2267180-24-maz@kernel.org
Signed-off-by: Oliver Upton <oupton@kernel.org>
