Hi all,
The goal of this series is to enable userspace driver designs that use VFIO to export DMABUFs representing subsets of PCI device BARs, and "vend" those buffers from a primary process to other subordinate processes by fd. These processes then mmap() the buffers and their access to the device is isolated to the exported ranges. This is an improvement on sharing the VFIO device fd to subordinate processes, which would allow unfettered access.
This is achieved by enabling mmap() of vfio-pci DMABUFs, passed by fd to subordinate processes. Second, a new revocation mechanism is added to allow the primary process to forcibly revoke access to previously-shared BAR spans, even if the subordinate processes haven't cleanly exited.
(The related topic of safe delegation of iommufd control to the subordinate processes is not addressed here, and is follow-up work.)
The background/rationale is covered in more detail in the RFC cover letters.
Feedback from the RFCs requested that, instead of creating DMABUF-specific vm_ops and .fault paths, to go the whole way and migrate the existing VFIO PCI BAR mmap() to be backed by a DMABUF too, resulting in a common vm_ops and fault handler for mmap()s of both the VFIO device and explicitly-exported DMABUFs. This will help future iommufd emulation of VFIO Type1 peer-to-peer, making it easier to get a DMABUF for a VFIO BAR as a DMA target.
mmap() conversion to use DMABUF underneath has been done for vfio-pci, but not sub-drivers:
nvgrace-gpu's mmap() override path is unchanged; I kept this out of scope for now not least because I don't have a thorough test setup for this system. I would prefer to help the nvgrace-gpu maintainers enable BAR mmap() DMABUFs themselves.
Notes on patches ================
PCI/P2PDMA: Split pool-related cleanup out of pci_p2pdma_release() PCI/P2PDMA: Add CONFIG_PCI_P2PDMA_CORE
Later in the series, vfio-pci's mmap() is going to depend on pcim_p2pdma_provider() which depended on CONFIG_PCI_P2PDMA, which in turn depended on ZONE_DEVICE. That isn't available on 32-bit and some archs, because they lack MEMORY_HOTPLUG and friends. VFIO does _not_ require actual P2P to be present for basic mmap() functionality, only for the optional CONFIG_DMA_SHARED_BUFFER feature.
These split out p2pdma_core.c under CONFIG_PCI_P2PDMA_CORE (which currently contains pcim_p2pdma_provider()), and an optional CONFIG_PCI_P2PDMA which depends on ZONE_DEVICE etc. providing P2P functionality in the existing p2pdma.c. The first splits out pool cleanup from the release path, and the second does the refactor/code move to the new file.
vfio/pci: Add a helper to look up PFNs for DMABUFs vfio/pci: Add a helper to create a DMABUF for a BAR-map VMA
The first adds a DMABUF VMA fault handler helper to determine arbitrary-sized PFNs from ranges in DMABUF. The second refactors DMABUF export for use by the existing export feature, and adds a helper that creates a DMABUF corresponding to a VFIO BAR mmap() request.
vfio/pci: Convert BAR mmap() to use a DMABUF
The vfio-pci core mmap() creates a DMABUF with the helper above, and the vm_ops fault handler uses the other helper to resolve the fault. Because this depends on DMABUF structs/code, CONFIG_VFIO_PCI_CORE needs to depend on CONFIG_DMA_SHARED_BUFFER. The CONFIG_VFIO_PCI_DMABUF still conditionally enables the export support code.
NOTE: The user mmap()s a device fd, but the resulting VMA's vm_file becomes that of the DMABUF. The DMABUF takes ownership of the device file and put()s it on release, which maintains the existing behaviour of a VMA keeping the VFIO device open.
BAR zapping then happens via the existing vfio_pci_dma_buf_move() path, which now needs to unmap PTEs in the DMABUF's address_space.
vfio/pci: Provide a user-facing name for BAR mappings
There was a request for decent debug naming in /proc/<pid>/maps etc. comparable to the existing VFIO names: since the VMAs are DMABUFs, they have a "dmabuf:" prefix and can't be 100% identical to before. This is a user-visible change, but this patch at least now gives us extra info on the BDF & BAR being mapped.
vfio/pci: Clean up BAR zap and revocation
In general (see NOTE!) the vfio_pci_zap_bars() is now obsolete, since it unmaps PTEs in the VFIO device address_space which is now unused. This consolidates all calls (e.g. around reset) with the neighbouring vfio_pci_dma_buf_move()s into new functions, to revoke/unrevoke (making the steps clearer).
NOTE: Because drivers can use their own vm_ops and override .mmap, the core must conservatively assume an overridden .mmap might still add PTEs to the VFIO device address_space and therefore still does the zap. A new flag, zap_bars_on_revoke, enables the zap when .mmap is overridden. A driver that does not need the zap can clear this to opt-out, e.g. if the driver calls down to the common mmap (and so uses DMABUFs). hisi-acc-vfio-pci does just this, and thus sets the opt-out flag.
vfio/pci: Support mmap() of a VFIO DMABUF
Adds mmap() for a DMABUF fd exported from vfio-pci.
It was a goal to keep the VFIO device fd lifetime behaviour unchanged with respect to the DMABUFs. An application can close all device fds, and this will revoke/clean up all DMABUFs; then, no mappings or other access can be performed. When enabling mmap() of the DMABUFs, this means access through the VMA is also revoked. This complicates the fault handler because whilst the DMABUF exists, it has no guarantee that the corresponding VFIO device is still alive. Adds synchronisation ensuring the vdev is available before vdev->memory_lock is touched; this holds the device registration so that even if the buffer has been cleaned up, vdev hasn't been freed and so the lock can be safely taken.
vfio/pci: Permanently revoke a DMABUF on request
This is mostly a rename of `revoked` to an enum, `status`, and adding a third state for a buffer: usable, revoked temporary, revoked permanent. A new VFIO feature is added, VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE, which takes a DMABUF (exported from the same device) and permanently revokes it. Thus a userspace driver can guarantee any downstream consumers of a shared fd are prevented from accessing a BAR range, and that range can be reused. NOTE: This might block userspace, waiting on importers to detach.
The code doing revocation in vfio_pci_dma_buf_move() is moved, to a common function for use by ..._move() and this new feature.
NOTE: See changelog, by request v4 added a condition to the existing code to elide the unnecessary invalidation/sync on the un-revoke path.)
NOTE: Previous versions contained an additional feature patch, "vfio/pci: Add mmap() attributes to DMABUF feature". This has been dropped in v5 because:
- The mechanism simply set vma->vm_page_prot. This would be sufficient for arm64 and other architectures.
- However, (locally-run claude-opus-4-8) Sashiko flagged that, on x86, additional memtype handling is required to set up the PAT. Without this, the memtype is returned back to UC- by pfnmap_setup_cachemode() upon PTE creation.
Most other sources of userspace WC mappings create PTEs eagerly with e.g. io_remap_pfn_range() which memtype_reserve() WC for the range. Getting them with lazy-fault used by vfio-pci is more complicated (e.g. perhaps registering WC for BARs with PAT/MTRRs, and deciding how to deal with aliasing...). Since this feature is not critical for this series to be useful, I've decided for now to drop it in favour of a simpler series now and revisiting this separ*ately.
Testing =======
(The [RFC ONLY] userspace test program, for QEMU edu-plus, can be found in the GitHub branch below. It at least illustrates how the export, map, revoke, and close semantics interoperate.)
This code has been tested in mapping DMABUFs of single/multiple ranges from multiple BARs, aliasing mmap()s, aliasing ranges across DMABUFs, vm_pgoff > 0, revocation, shutdown/cleanup scenarios, and hugepage mappings. No regressions observed on the VFIO selftests, or on our internal vfio-pci applications. VFIO on i386 has been build-tested.
Dear Reviewers, ===============
I was grateful for the reviews and Reviewed-Bys on previous versions. Thanks; I've added some Reviewed-Bys/Acks. I have NOT included your tags where the patch has materially changed after your review (or where requested changes ended up more than super-trivial). I hope that's okay.
End ===
This is based on v7.2-rc3.
These commits are on GitHub for easier browsing, along with "[RFC ONLY] selftests: vfio: Add standalone vfio_dmabuf_mmap_test":
https://github.com/metamev/linux/compare/v7.2-rc3...dev/mev/vfio-dmabuf-mmap...
Thanks for reading,
Matt
================================================================================ Changelog:
v5:
- Rebased on 7.2-rc3
- Dropped the memattr/WC feature (see explanation above).
- "vfio/pci: Convert BAR mmap() to use a DMABUF": Fixed a potentially-nasty bug (which (locally-run) Sashiko found!) whereby the unmap_mapping_range() performed in cleanup was passed a range up from offset zero for the DMABUF size. Initially this was how all DMABUFs were created and an appropriate zap, but a new version kept the VFIO region index encoded in the offset -- for BAR > 0 the unmap span would then mismatch. Instead, pass size 0 to mean an "all" range. Because the goal is to shoot down everything relating to one DMABUF and the address_space can only contain things relating to that DMABUF, this is equivalent and has the bonus of never failing to match mappings...
Praan, Kevin, I kept your R-Bs on this fix.
- The revoke patch converts vfio_pci_dma_buf_cleanup()'s priv->vdev = NULL to a WRITE_ONCE, corresponding to the revoke function's READ_ONCE (performed to test that the VFIO and DMABUF are related).
- Clarified the VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE UAPI comments, documenting previously-missing error cases and their reasons.
v4: https://lore.kernel.org/all/20260701171245.90111-1-matt@ozlabs.org/ v3: https://lore.kernel.org/all/20260610154327.37758-1-matt@ozlabs.org/ v2: https://lore.kernel.org/all/20260527102319.100128-1-mattev@meta.com/ v1: https://lore.kernel.org/kvm/20260416131815.2729131-1-mattev@meta.com/ RFCv2: https://lore.kernel.org/kvm/20260312184613.3710705-1-mattev@meta.com/ RFCv1: https://lore.kernel.org/all/20260226202211.929005-1-mattev@meta.com/ Tech topic: https://lore.kernel.org/linux-iommu/20250918214425.2677057-1-amastro@fb.com/
Matt Evans (9): PCI/P2PDMA: Split pool-related cleanup out of pci_p2pdma_release() PCI/P2PDMA: Add CONFIG_PCI_P2PDMA_CORE vfio/pci: Add a helper to look up PFNs for DMABUFs vfio/pci: Add a helper to create a DMABUF for a BAR-map VMA vfio/pci: Convert BAR mmap() to use a DMABUF vfio/pci: Provide a user-facing name for BAR mappings vfio/pci: Clean up BAR zap and revocation vfio/pci: Support mmap() of a VFIO DMABUF vfio/pci: Permanently revoke a DMABUF on request
MAINTAINERS | 2 +- drivers/pci/Kconfig | 5 + drivers/pci/Makefile | 1 + drivers/pci/p2pdma.c | 113 +--- drivers/pci/p2pdma.h | 29 + drivers/pci/p2pdma_core.c | 122 +++++ drivers/vfio/pci/Kconfig | 5 +- drivers/vfio/pci/Makefile | 3 +- .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c | 8 + drivers/vfio/pci/vfio_pci_config.c | 30 +- drivers/vfio/pci/vfio_pci_core.c | 210 +++++-- drivers/vfio/pci/vfio_pci_dmabuf.c | 515 +++++++++++++++--- drivers/vfio/pci/vfio_pci_priv.h | 53 +- include/linux/pci-p2pdma.h | 24 +- include/linux/pci.h | 2 +- include/linux/vfio_pci_core.h | 1 + include/uapi/linux/vfio.h | 25 + 17 files changed, 875 insertions(+), 273 deletions(-) create mode 100644 drivers/pci/p2pdma.h create mode 100644 drivers/pci/p2pdma_core.c
Preparing for a refactor in a subsequent patch, split the pool-related release code into a new pci_p2pdma_release_pool() function.
This allows future compile-time selection of a null implementation for pci_p2pdma_release_pool(), when p2pdma.c is refactored into core- and P2P-related files.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com --- drivers/pci/p2pdma.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-)
diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c index b2d5266f8653..498bca257419 100644 --- a/drivers/pci/p2pdma.c +++ b/drivers/pci/p2pdma.c @@ -226,6 +226,17 @@ static const struct dev_pagemap_ops p2pdma_pgmap_ops = { .folio_free = p2pdma_folio_free, };
+static void pci_p2pdma_release_pool(struct pci_dev *pdev, + struct pci_p2pdma *p2pdma) +{ + if (!p2pdma->pool) + return; + + synchronize_rcu(); + gen_pool_destroy(p2pdma->pool); + sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); +} + static void pci_p2pdma_release(void *data) { struct pci_dev *pdev = data; @@ -237,15 +248,8 @@ static void pci_p2pdma_release(void *data)
/* Flush and disable pci_alloc_p2p_mem() */ pdev->p2pdma = NULL; - if (p2pdma->pool) - synchronize_rcu(); + pci_p2pdma_release_pool(pdev, p2pdma); xa_destroy(&p2pdma->map_types); - - if (!p2pdma->pool) - return; - - gen_pool_destroy(p2pdma->pool); - sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); }
/** @@ -946,8 +950,8 @@ void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) struct pci_p2pdma *p2pdma;
/* - * Pairs with synchronize_rcu() in pci_p2pdma_release() to - * ensure pdev->p2pdma is non-NULL for the duration of the + * Pairs with synchronize_rcu() in pci_p2pdma_release_pool() + * to ensure pdev->p2pdma is non-NULL for the duration of the * read-lock. */ rcu_read_lock();
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:47 AM
Preparing for a refactor in a subsequent patch, split the pool-related release code into a new pci_p2pdma_release_pool() function.
This allows future compile-time selection of a null implementation for pci_p2pdma_release_pool(), when p2pdma.c is refactored into core- and P2P-related files.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com
Reviewed-by: Kevin Tian kevin.tian@intel.com
Hi Leon,
Please see below...
On Wed, 15 Jul 2026 18:47:24 +0100 Matt Evans matt@ozlabs.org wrote:
Preparing for a refactor in a subsequent patch, split the pool-related release code into a new pci_p2pdma_release_pool() function.
This allows future compile-time selection of a null implementation for pci_p2pdma_release_pool(), when p2pdma.c is refactored into core- and P2P-related files.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com
drivers/pci/p2pdma.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-)
diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c index b2d5266f8653..498bca257419 100644 --- a/drivers/pci/p2pdma.c +++ b/drivers/pci/p2pdma.c @@ -226,6 +226,17 @@ static const struct dev_pagemap_ops p2pdma_pgmap_ops = { .folio_free = p2pdma_folio_free, }; +static void pci_p2pdma_release_pool(struct pci_dev *pdev,
struct pci_p2pdma *p2pdma)+{
- if (!p2pdma->pool)
return;- synchronize_rcu();
Sashiko notes[1] a high, preexisting issue here that looks like it was introduced via 372d6d1b8ae3 ("PCI/P2PDMA: Refactor to separate core P2P functionality from memory allocation"). That commit creates a two-tier scheme where the optional memory allocation capabilities, such as the pool, live above a core layer. Prior to that commit, this synchronize RCU call was unconditional. Making it conditional on the pool suggests it was only considered relevant to the optional layer.
However, map_types, which remains in the core layer, is RCU referenced. Was the synchronize_rcu() call here miscategorized into the optional tier? Should it instead have remained unconditional?
We might need a precursor Fixes: patch to this series that makes it unconditional in pci_p2pdma_release(), so that it remains there with this refactor rather than becoming part of this new pool release function. Thanks,
Alex
[1]https://lore.kernel.org/all/20260715180521.3CDD61F00A3A@smtp.kernel.org/
- gen_pool_destroy(p2pdma->pool);
- sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group);
+}
static void pci_p2pdma_release(void *data) { struct pci_dev *pdev = data; @@ -237,15 +248,8 @@ static void pci_p2pdma_release(void *data) /* Flush and disable pci_alloc_p2p_mem() */ pdev->p2pdma = NULL;
- if (p2pdma->pool)
synchronize_rcu();
- pci_p2pdma_release_pool(pdev, p2pdma); xa_destroy(&p2pdma->map_types);
- if (!p2pdma->pool)
return;- gen_pool_destroy(p2pdma->pool);
- sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group);
} /** @@ -946,8 +950,8 @@ void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) struct pci_p2pdma *p2pdma; /*
* Pairs with synchronize_rcu() in pci_p2pdma_release() to* ensure pdev->p2pdma is non-NULL for the duration of the
* Pairs with synchronize_rcu() in pci_p2pdma_release_pool()* to ensure pdev->p2pdma is non-NULL for the duration of the*/ rcu_read_lock();
- read-lock.
On Tue, Jul 28, 2026 at 04:33:43PM -0600, Alex Williamson wrote:
Hi Leon,
Please see below...
On Wed, 15 Jul 2026 18:47:24 +0100 Matt Evans matt@ozlabs.org wrote:
Preparing for a refactor in a subsequent patch, split the pool-related release code into a new pci_p2pdma_release_pool() function.
This allows future compile-time selection of a null implementation for pci_p2pdma_release_pool(), when p2pdma.c is refactored into core- and P2P-related files.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com
drivers/pci/p2pdma.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-)
diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c index b2d5266f8653..498bca257419 100644 --- a/drivers/pci/p2pdma.c +++ b/drivers/pci/p2pdma.c @@ -226,6 +226,17 @@ static const struct dev_pagemap_ops p2pdma_pgmap_ops = { .folio_free = p2pdma_folio_free, }; +static void pci_p2pdma_release_pool(struct pci_dev *pdev,
struct pci_p2pdma *p2pdma)+{
- if (!p2pdma->pool)
return;- synchronize_rcu();
Sashiko notes[1] a high, preexisting issue here that looks like it was introduced via 372d6d1b8ae3 ("PCI/P2PDMA: Refactor to separate core P2P functionality from memory allocation"). That commit creates a two-tier scheme where the optional memory allocation capabilities, such as the pool, live above a core layer. Prior to that commit, this synchronize RCU call was unconditional. Making it conditional on the pool suggests it was only considered relevant to the optional layer.
However, map_types, which remains in the core layer, is RCU referenced. Was the synchronize_rcu() call here miscategorized into the optional tier? Should it instead have remained unconditional?
I remember this concern when I wrote the patches and wanted to remove RCU entirely. I revisited the Sashiko report, but reached the same conclusion again.
The "bug" reported by Sashiko does not exist. The two-layer split ensures that p2p is bound to the driver's lifecycle. As a result, pdev->p2pdma is assigned and cleared only once during the lifetime of pdev.
In this case, the RCU primitives are effectively NOPs, since nothing will ever update that pointer.
I still needed to keep rcu_dereference() in pci_p2pdma_map_type() to satisfy static analyzers, which would otherwise complain about accessing an RCU-protected pointer without the proper annotations.
RCU is used only in the sysfs flow.
Thanks
We might need a precursor Fixes: patch to this series that makes it unconditional in pci_p2pdma_release(), so that it remains there with this refactor rather than becoming part of this new pool release function. Thanks,
Alex
[1]https://lore.kernel.org/all/20260715180521.3CDD61F00A3A@smtp.kernel.org/
- gen_pool_destroy(p2pdma->pool);
- sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group);
+}
static void pci_p2pdma_release(void *data) { struct pci_dev *pdev = data; @@ -237,15 +248,8 @@ static void pci_p2pdma_release(void *data) /* Flush and disable pci_alloc_p2p_mem() */ pdev->p2pdma = NULL;
- if (p2pdma->pool)
synchronize_rcu();
- pci_p2pdma_release_pool(pdev, p2pdma); xa_destroy(&p2pdma->map_types);
- if (!p2pdma->pool)
return;- gen_pool_destroy(p2pdma->pool);
- sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group);
} /** @@ -946,8 +950,8 @@ void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) struct pci_p2pdma *p2pdma; /*
* Pairs with synchronize_rcu() in pci_p2pdma_release() to* ensure pdev->p2pdma is non-NULL for the duration of the
* Pairs with synchronize_rcu() in pci_p2pdma_release_pool()* to ensure pdev->p2pdma is non-NULL for the duration of the*/ rcu_read_lock();
- read-lock.
On Wed, Jul 29, 2026 at 01:08:40PM +0300, Leon Romanovsky wrote:
I remember this concern when I wrote the patches and wanted to remove RCU entirely. I revisited the Sashiko report, but reached the same conclusion again.
The "bug" reported by Sashiko does not exist. The two-layer split ensures that p2p is bound to the driver's lifecycle. As a result, pdev->p2pdma is assigned and cleared only once during the lifetime of pdev.
In this case, the RCU primitives are effectively NOPs, since nothing will ever update that pointer.
This sounds right to me too.
The lowest layer has no need of RCU at all, it just remains RCU because the genalloc layer wants it as part of its locking scheme.
So long as the genalloc gets the synchronize_rcu on destruction there is no bug.
Jason
The P2PDMA code currently provides two features under the same CONFIG_PCI_P2PDMA option:
1. Locate providers via pcim_p2pdma_provider() 2. Manage actual P2P DMA
Some drivers (such as vfio-pci) depend on (1), without having a hard dependency on (2).
A future vfio-pci commit will rely on pcim_p2pdma_provider() always being present. If that depended on CONFIG_PCI_P2PDMA, it would make vfio-pci only available if CONFIG_ZONE_DEVICE is present (e.g. 64-bit systems), even when P2P is not needed.
To resolve this, introduce CONFIG_PCI_P2PDMA_CORE and refactor the basic provider functionality into a new p2pdma_core.c file. This is available even if the CONFIG_PCI_P2PDMA feature is disabled (or unavailable due to !CONFIG_ZONE_DEVICE), satisfying (1).
Then, when the original CONFIG_PCI_P2PDMA is set, drivers have access to the additional P2P features of (2). This still depends on CONFIG_ZONE_DEVICE.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com --- MAINTAINERS | 2 +- drivers/pci/Kconfig | 5 ++ drivers/pci/Makefile | 1 + drivers/pci/p2pdma.c | 111 +-------------------------------- drivers/pci/p2pdma.h | 29 +++++++++ drivers/pci/p2pdma_core.c | 122 +++++++++++++++++++++++++++++++++++++ include/linux/pci-p2pdma.h | 24 +++++--- include/linux/pci.h | 2 +- 8 files changed, 175 insertions(+), 121 deletions(-) create mode 100644 drivers/pci/p2pdma.h create mode 100644 drivers/pci/p2pdma_core.c
diff --git a/MAINTAINERS b/MAINTAINERS index 806bd2d80d15..192029942145 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -20867,7 +20867,7 @@ B: https://bugzilla.kernel.org C: irc://irc.oftc.net/linux-pci T: git git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git F: Documentation/driver-api/pci/p2pdma.rst -F: drivers/pci/p2pdma.c +F: drivers/pci/p2pdma* F: include/linux/pci-p2pdma.h
PCI POWER CONTROL diff --git a/drivers/pci/Kconfig b/drivers/pci/Kconfig index 0c7408509ba2..c991e92bcb85 100644 --- a/drivers/pci/Kconfig +++ b/drivers/pci/Kconfig @@ -211,6 +211,7 @@ config PCI_P2PDMA # requires 64bit # depends on 64BIT + select PCI_P2PDMA_CORE select GENERIC_ALLOCATOR select NEED_SG_DMA_FLAGS help @@ -226,6 +227,10 @@ config PCI_P2PDMA
If unsure, say N.
+config PCI_P2PDMA_CORE + default n + bool + config PCI_LABEL def_bool y if (DMI || ACPI) select NLS diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile index 41ebc3b9a518..0b32572d57a1 100644 --- a/drivers/pci/Makefile +++ b/drivers/pci/Makefile @@ -30,6 +30,7 @@ obj-$(CONFIG_PCI_SYSCALL) += syscall.o obj-$(CONFIG_PCI_STUB) += pci-stub.o obj-$(CONFIG_PCI_PF_STUB) += pci-pf-stub.o obj-$(CONFIG_PCI_ECAM) += ecam.o +obj-$(CONFIG_PCI_P2PDMA_CORE) += p2pdma_core.o obj-$(CONFIG_PCI_P2PDMA) += p2pdma.o obj-$(CONFIG_XEN_PCIDEV_FRONTEND) += xen-pcifront.o obj-$(CONFIG_VGA_ARB) += vgaarb.o diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c index 498bca257419..607580f58c26 100644 --- a/drivers/pci/p2pdma.c +++ b/drivers/pci/p2pdma.c @@ -21,12 +21,7 @@ #include <linux/seq_buf.h> #include <linux/xarray.h>
-struct pci_p2pdma { - struct gen_pool *pool; - bool p2pmem_published; - struct xarray map_types; - struct p2pdma_provider mem[PCI_STD_NUM_BARS]; -}; +#include "p2pdma.h"
struct pci_p2pdma_pagemap { struct dev_pagemap pgmap; @@ -226,8 +221,7 @@ static const struct dev_pagemap_ops p2pdma_pgmap_ops = { .folio_free = p2pdma_folio_free, };
-static void pci_p2pdma_release_pool(struct pci_dev *pdev, - struct pci_p2pdma *p2pdma) +void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma) { if (!p2pdma->pool) return; @@ -237,107 +231,6 @@ static void pci_p2pdma_release_pool(struct pci_dev *pdev, sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); }
-static void pci_p2pdma_release(void *data) -{ - struct pci_dev *pdev = data; - struct pci_p2pdma *p2pdma; - - p2pdma = rcu_dereference_protected(pdev->p2pdma, 1); - if (!p2pdma) - return; - - /* Flush and disable pci_alloc_p2p_mem() */ - pdev->p2pdma = NULL; - pci_p2pdma_release_pool(pdev, p2pdma); - xa_destroy(&p2pdma->map_types); -} - -/** - * pcim_p2pdma_init - Initialise peer-to-peer DMA providers - * @pdev: The PCI device to enable P2PDMA for - * - * This function initializes the peer-to-peer DMA infrastructure - * for a PCI device. It allocates and sets up the necessary data - * structures to support P2PDMA operations, including mapping type - * tracking. - */ -int pcim_p2pdma_init(struct pci_dev *pdev) -{ - struct pci_p2pdma *p2p; - int i, ret; - - if (pdev->non_mappable_bars) - return -EOPNOTSUPP; - - p2p = rcu_dereference_protected(pdev->p2pdma, 1); - if (p2p) - return 0; - - p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL); - if (!p2p) - return -ENOMEM; - - xa_init(&p2p->map_types); - /* - * Iterate over all standard PCI BARs and record only those that - * correspond to MMIO regions. Skip non-memory resources (e.g. I/O - * port BARs) since they cannot be used for peer-to-peer (P2P) - * transactions. - */ - for (i = 0; i < PCI_STD_NUM_BARS; i++) { - if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) - continue; - - p2p->mem[i].owner = &pdev->dev; - p2p->mem[i].bus_offset = - pci_bus_address(pdev, i) - pci_resource_start(pdev, i); - } - - ret = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev); - if (ret) - goto out_p2p; - - rcu_assign_pointer(pdev->p2pdma, p2p); - return 0; - -out_p2p: - devm_kfree(&pdev->dev, p2p); - return ret; -} -EXPORT_SYMBOL_GPL(pcim_p2pdma_init); - -/** - * pcim_p2pdma_provider - Get peer-to-peer DMA provider - * @pdev: The PCI device to enable P2PDMA for - * @bar: BAR index to get provider - * - * This function gets peer-to-peer DMA provider for a PCI device. The lifetime - * of the provider (and of course the MMIO) is bound to the lifetime of the - * driver. A driver calling this function must ensure that all references to the - * provider, and any DMA mappings created for any MMIO, are all cleaned up - * before the driver remove() completes. - * - * Since P2P is almost always shared with a second driver this means some system - * to notify, invalidate and revoke the MMIO's DMA must be in place to use this - * function. For example a revoke can be built using DMABUF. - */ -struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar) -{ - struct pci_p2pdma *p2p; - - if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM) || - pdev->non_mappable_bars) - return NULL; - - p2p = rcu_dereference_protected(pdev->p2pdma, 1); - if (WARN_ON(!p2p)) - /* Someone forgot to call to pcim_p2pdma_init() before */ - return NULL; - - return &p2p->mem[bar]; -} -EXPORT_SYMBOL_GPL(pcim_p2pdma_provider); - static int pci_p2pdma_setup_pool(struct pci_dev *pdev) { struct pci_p2pdma *p2pdma; diff --git a/drivers/pci/p2pdma.h b/drivers/pci/p2pdma.h new file mode 100644 index 000000000000..946383809981 --- /dev/null +++ b/drivers/pci/p2pdma.h @@ -0,0 +1,29 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * PCI peer-to-peer DMA support. + */ + +#ifndef _PCI_P2PDMA_H +#define _PCI_P2PDMA_H + +#include <linux/genalloc.h> +#include <linux/pci-p2pdma.h> +#include <linux/xarray.h> + +struct pci_p2pdma { + struct gen_pool *pool; + bool p2pmem_published; + struct xarray map_types; + struct p2pdma_provider mem[PCI_STD_NUM_BARS]; +}; + +#ifdef CONFIG_PCI_P2PDMA +void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma); +#else +static inline void pci_p2pdma_release_pool(struct pci_dev *pdev, + struct pci_p2pdma *p2pdma) +{ +} +#endif + +#endif diff --git a/drivers/pci/p2pdma_core.c b/drivers/pci/p2pdma_core.c new file mode 100644 index 000000000000..9e997f373b1e --- /dev/null +++ b/drivers/pci/p2pdma_core.c @@ -0,0 +1,122 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * PCI peer-to-peer DMA support core, providing a bare-bones + * pcim_p2pdma_provider() interface to drivers even if full P2PDMA + * isn't present. The full P2PDMA feature is in p2pdma.c (see + * CONFIG_PCI_P2PDMA). + * + * Copyright (c) 2016-2018, Logan Gunthorpe + * Copyright (c) 2016-2017, Microsemi Corporation + * Copyright (c) 2017, Christoph Hellwig + * Copyright (c) 2018, Eideticom Inc. + */ + +#define pr_fmt(fmt) "pci-p2pdma: " fmt +#include <linux/ctype.h> +#include <linux/genalloc.h> +#include <linux/memremap.h> +#include <linux/pci-p2pdma.h> +#include <linux/xarray.h> + +#include "p2pdma.h" + +static void pci_p2pdma_release(void *data) +{ + struct pci_dev *pdev = data; + struct pci_p2pdma *p2pdma; + + p2pdma = rcu_dereference_protected(pdev->p2pdma, 1); + if (!p2pdma) + return; + + /* Flush and disable pci_alloc_p2p_mem() */ + pdev->p2pdma = NULL; + pci_p2pdma_release_pool(pdev, p2pdma); + xa_destroy(&p2pdma->map_types); +} + +/** + * pcim_p2pdma_init - Initialise peer-to-peer DMA providers + * @pdev: The PCI device to enable P2PDMA for + * + * This function initializes the peer-to-peer DMA infrastructure + * for a PCI device. It allocates and sets up the necessary data + * structures to support P2PDMA operations, including mapping type + * tracking. + */ +int pcim_p2pdma_init(struct pci_dev *pdev) +{ + struct pci_p2pdma *p2p; + int i, ret; + + if (pdev->non_mappable_bars) + return -EOPNOTSUPP; + + p2p = rcu_dereference_protected(pdev->p2pdma, 1); + if (p2p) + return 0; + + p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL); + if (!p2p) + return -ENOMEM; + + xa_init(&p2p->map_types); + /* + * Iterate over all standard PCI BARs and record only those that + * correspond to MMIO regions. Skip non-memory resources (e.g. I/O + * port BARs) since they cannot be used for peer-to-peer (P2P) + * transactions. + */ + for (i = 0; i < PCI_STD_NUM_BARS; i++) { + if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM)) + continue; + + p2p->mem[i].owner = &pdev->dev; + p2p->mem[i].bus_offset = + pci_bus_address(pdev, i) - pci_resource_start(pdev, i); + } + + ret = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev); + if (ret) + goto out_p2p; + + rcu_assign_pointer(pdev->p2pdma, p2p); + return 0; + +out_p2p: + devm_kfree(&pdev->dev, p2p); + return ret; +} +EXPORT_SYMBOL_GPL(pcim_p2pdma_init); + +/** + * pcim_p2pdma_provider - Get peer-to-peer DMA provider + * @pdev: The PCI device to enable P2PDMA for + * @bar: BAR index to get provider + * + * This function gets peer-to-peer DMA provider for a PCI device. The lifetime + * of the provider (and of course the MMIO) is bound to the lifetime of the + * driver. A driver calling this function must ensure that all references to the + * provider, and any DMA mappings created for any MMIO, are all cleaned up + * before the driver remove() completes. + * + * Since P2P is almost always shared with a second driver this means some system + * to notify, invalidate and revoke the MMIO's DMA must be in place to use this + * function. For example a revoke can be built using DMABUF. + */ +struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar) +{ + struct pci_p2pdma *p2p; + + if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM) || + pdev->non_mappable_bars) + return NULL; + + p2p = rcu_dereference_protected(pdev->p2pdma, 1); + if (WARN_ON(!p2p)) + /* Someone forgot to call to pcim_p2pdma_init() before */ + return NULL; + + return &p2p->mem[bar]; +} +EXPORT_SYMBOL_GPL(pcim_p2pdma_provider); diff --git a/include/linux/pci-p2pdma.h b/include/linux/pci-p2pdma.h index 873de20a2247..4c42a7b2ee85 100644 --- a/include/linux/pci-p2pdma.h +++ b/include/linux/pci-p2pdma.h @@ -67,9 +67,22 @@ enum pci_p2pdma_map_type { PCI_P2PDMA_MAP_THRU_HOST_BRIDGE, };
-#ifdef CONFIG_PCI_P2PDMA +#ifdef CONFIG_PCI_P2PDMA_CORE int pcim_p2pdma_init(struct pci_dev *pdev); struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar); +#else +static inline int pcim_p2pdma_init(struct pci_dev *pdev) +{ + return -EOPNOTSUPP; +} +static inline struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, + int bar) +{ + return NULL; +} +#endif + +#ifdef CONFIG_PCI_P2PDMA int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, u64 offset); int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients, @@ -89,15 +102,6 @@ ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev, enum pci_p2pdma_map_type pci_p2pdma_map_type(struct p2pdma_provider *provider, struct device *dev); #else /* CONFIG_PCI_P2PDMA */ -static inline int pcim_p2pdma_init(struct pci_dev *pdev) -{ - return -EOPNOTSUPP; -} -static inline struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, - int bar) -{ - return NULL; -} static inline int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, u64 offset) { diff --git a/include/linux/pci.h b/include/linux/pci.h index 64b308b6e61c..8adbabbec5bf 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -556,7 +556,7 @@ struct pci_dev { u16 pasid_cap; /* PASID Capability offset */ u16 pasid_features; #endif -#ifdef CONFIG_PCI_P2PDMA +#ifdef CONFIG_PCI_P2PDMA_CORE struct pci_p2pdma __rcu *p2pdma; #endif #ifdef CONFIG_PCI_DOE
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:47 AM
The P2PDMA code currently provides two features under the same CONFIG_PCI_P2PDMA option:
- Locate providers via pcim_p2pdma_provider()
- Manage actual P2P DMA
Some drivers (such as vfio-pci) depend on (1), without having a hard dependency on (2).
A future vfio-pci commit will rely on pcim_p2pdma_provider() always being present. If that depended on CONFIG_PCI_P2PDMA, it would make vfio-pci only available if CONFIG_ZONE_DEVICE is present (e.g. 64-bit systems), even when P2P is not needed.
To resolve this, introduce CONFIG_PCI_P2PDMA_CORE and refactor the basic provider functionality into a new p2pdma_core.c file. This is available even if the CONFIG_PCI_P2PDMA feature is disabled (or unavailable due to !CONFIG_ZONE_DEVICE), satisfying (1).
Then, when the original CONFIG_PCI_P2PDMA is set, drivers have access to the additional P2P features of (2). This still depends on CONFIG_ZONE_DEVICE.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com
Reviewed-by: Kevin Tian kevin.tian@intel.com
Hi Leon, Jason, Logan,
I want to ask for your input on a small issue, please:
On 15/07/2026 18:47, Matt Evans wrote:
The P2PDMA code currently provides two features under the same CONFIG_PCI_P2PDMA option:
- Locate providers via pcim_p2pdma_provider()
- Manage actual P2P DMA
Some drivers (such as vfio-pci) depend on (1), without having a hard dependency on (2).
A future vfio-pci commit will rely on pcim_p2pdma_provider() always being present. If that depended on CONFIG_PCI_P2PDMA, it would make vfio-pci only available if CONFIG_ZONE_DEVICE is present (e.g. 64-bit systems), even when P2P is not needed.
(As a reminder, this exists so that vfio-pci can always export a DMABUF for mmap() even if !P2PDMA.)
The (in-kernel) users of pcim_p2pdma_provider() are vfio-pci, its nvgrace-gpu driver, and (since d6c58f4eb3d0 "RDMA/mlx5: Implement DMABUF export ops") the mlx5 driver.
The vfio-pci drivers gate an actual export on CONFIG_VFIO_PCI_DMABUF (.'. CONFIG_PCI_P2PDMA), but a comment (in a Claude review!) just flagged that mlx5 uses the success of pcim_p2pdma_provider() as a sign that P2PDMA is also supported:
drivers/infiniband/hw/mlx5/main.c:mlx5_ib_mmap_get_pfns():
*provider = pcim_p2pdma_provider(pdev, bar); /* If the kernel was not compiled with CONFIG_PCI_P2PDMA the * functionality is not supported. */ if (!*provider) return -EOPNOTSUPP;
This will return success even without CONFIG_PCI_P2PDMA, then UVERBS_METHOD_DMABUF_ALLOC won't fail early anymore. Even though the export's attach ultimately fails, moving the point of failure is ugly.
Just shoving an #ifdef in mlx5 isn't right, so would be grateful for your advice on the most tasteful resolution:
1. Imagining future drivers might use pcim_p2pdma_provider(), maybe p2pdma could export a `pcim_p2pdma_supported()` or similar which returns whether CONFIG_PCI_P2PDMA is present and P2P can ever work. A user of pcim_p2pdma_provider() is required to also check if P2P is overall supported.
Ugliness aside, a driver could use #ifdef CONFIG_PCI_P2PDMA to test whether P2P is expected to work, except that doesn't stop modules falling into the same trap. I think a function export is better.
2. And/or a test (such the function from #1, or an #ifdef) in the IB core DMABUF path, to avoid individual drivers from doing the test in their .mmap_get_pfns(). Other future drivers would be expected to do similar.
Any better ideas? Sorry I didn't catch this earlier in the review cycles, I know we've looked at this patch a few times now.
Many thanks,
Matt
To resolve this, introduce CONFIG_PCI_P2PDMA_CORE and refactor the basic provider functionality into a new p2pdma_core.c file. This is available even if the CONFIG_PCI_P2PDMA feature is disabled (or unavailable due to !CONFIG_ZONE_DEVICE), satisfying (1).
Then, when the original CONFIG_PCI_P2PDMA is set, drivers have access to the additional P2P features of (2). This still depends on CONFIG_ZONE_DEVICE.
Signed-off-by: Matt Evans matt@ozlabs.org Acked-by: Bjorn Helgaas bhelgaas@google.com Reviewed-by: Logan Gunthorpe logang@deltatee.com
MAINTAINERS | 2 +- drivers/pci/Kconfig | 5 ++ drivers/pci/Makefile | 1 + drivers/pci/p2pdma.c | 111 +-------------------------------- drivers/pci/p2pdma.h | 29 +++++++++ drivers/pci/p2pdma_core.c | 122 +++++++++++++++++++++++++++++++++++++ include/linux/pci-p2pdma.h | 24 +++++--- include/linux/pci.h | 2 +- 8 files changed, 175 insertions(+), 121 deletions(-) create mode 100644 drivers/pci/p2pdma.h create mode 100644 drivers/pci/p2pdma_core.c
diff --git a/MAINTAINERS b/MAINTAINERS index 806bd2d80d15..192029942145 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -20867,7 +20867,7 @@ B: https://bugzilla.kernel.org C: irc://irc.oftc.net/linux-pci T: git git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git F: Documentation/driver-api/pci/p2pdma.rst -F: drivers/pci/p2pdma.c +F: drivers/pci/p2pdma* F: include/linux/pci-p2pdma.h PCI POWER CONTROL diff --git a/drivers/pci/Kconfig b/drivers/pci/Kconfig index 0c7408509ba2..c991e92bcb85 100644 --- a/drivers/pci/Kconfig +++ b/drivers/pci/Kconfig @@ -211,6 +211,7 @@ config PCI_P2PDMA # requires 64bit # depends on 64BIT
- select PCI_P2PDMA_CORE select GENERIC_ALLOCATOR select NEED_SG_DMA_FLAGS help
@@ -226,6 +227,10 @@ config PCI_P2PDMA If unsure, say N. +config PCI_P2PDMA_CORE
- default n
- bool
config PCI_LABEL def_bool y if (DMI || ACPI) select NLS diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile index 41ebc3b9a518..0b32572d57a1 100644 --- a/drivers/pci/Makefile +++ b/drivers/pci/Makefile @@ -30,6 +30,7 @@ obj-$(CONFIG_PCI_SYSCALL) += syscall.o obj-$(CONFIG_PCI_STUB) += pci-stub.o obj-$(CONFIG_PCI_PF_STUB) += pci-pf-stub.o obj-$(CONFIG_PCI_ECAM) += ecam.o +obj-$(CONFIG_PCI_P2PDMA_CORE) += p2pdma_core.o obj-$(CONFIG_PCI_P2PDMA) += p2pdma.o obj-$(CONFIG_XEN_PCIDEV_FRONTEND) += xen-pcifront.o obj-$(CONFIG_VGA_ARB) += vgaarb.o diff --git a/drivers/pci/p2pdma.c b/drivers/pci/p2pdma.c index 498bca257419..607580f58c26 100644 --- a/drivers/pci/p2pdma.c +++ b/drivers/pci/p2pdma.c @@ -21,12 +21,7 @@ #include <linux/seq_buf.h> #include <linux/xarray.h> -struct pci_p2pdma {
- struct gen_pool *pool;
- bool p2pmem_published;
- struct xarray map_types;
- struct p2pdma_provider mem[PCI_STD_NUM_BARS];
-}; +#include "p2pdma.h" struct pci_p2pdma_pagemap { struct dev_pagemap pgmap; @@ -226,8 +221,7 @@ static const struct dev_pagemap_ops p2pdma_pgmap_ops = { .folio_free = p2pdma_folio_free, }; -static void pci_p2pdma_release_pool(struct pci_dev *pdev,
struct pci_p2pdma *p2pdma)+void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma) { if (!p2pdma->pool) return; @@ -237,107 +231,6 @@ static void pci_p2pdma_release_pool(struct pci_dev *pdev, sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); } -static void pci_p2pdma_release(void *data) -{
- struct pci_dev *pdev = data;
- struct pci_p2pdma *p2pdma;
- p2pdma = rcu_dereference_protected(pdev->p2pdma, 1);
- if (!p2pdma)
return;- /* Flush and disable pci_alloc_p2p_mem() */
- pdev->p2pdma = NULL;
- pci_p2pdma_release_pool(pdev, p2pdma);
- xa_destroy(&p2pdma->map_types);
-}
-/**
- pcim_p2pdma_init - Initialise peer-to-peer DMA providers
- @pdev: The PCI device to enable P2PDMA for
- This function initializes the peer-to-peer DMA infrastructure
- for a PCI device. It allocates and sets up the necessary data
- structures to support P2PDMA operations, including mapping type
- tracking.
- */
-int pcim_p2pdma_init(struct pci_dev *pdev) -{
- struct pci_p2pdma *p2p;
- int i, ret;
- if (pdev->non_mappable_bars)
return -EOPNOTSUPP;- p2p = rcu_dereference_protected(pdev->p2pdma, 1);
- if (p2p)
return 0;- p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL);
- if (!p2p)
return -ENOMEM;- xa_init(&p2p->map_types);
- /*
* Iterate over all standard PCI BARs and record only those that* correspond to MMIO regions. Skip non-memory resources (e.g. I/O* port BARs) since they cannot be used for peer-to-peer (P2P)* transactions.*/- for (i = 0; i < PCI_STD_NUM_BARS; i++) {
if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))continue;p2p->mem[i].owner = &pdev->dev;p2p->mem[i].bus_offset =pci_bus_address(pdev, i) - pci_resource_start(pdev, i);- }
- ret = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev);
- if (ret)
goto out_p2p;- rcu_assign_pointer(pdev->p2pdma, p2p);
- return 0;
-out_p2p:
- devm_kfree(&pdev->dev, p2p);
- return ret;
-} -EXPORT_SYMBOL_GPL(pcim_p2pdma_init);
-/**
- pcim_p2pdma_provider - Get peer-to-peer DMA provider
- @pdev: The PCI device to enable P2PDMA for
- @bar: BAR index to get provider
- This function gets peer-to-peer DMA provider for a PCI device. The lifetime
- of the provider (and of course the MMIO) is bound to the lifetime of the
- driver. A driver calling this function must ensure that all references to the
- provider, and any DMA mappings created for any MMIO, are all cleaned up
- before the driver remove() completes.
- Since P2P is almost always shared with a second driver this means some system
- to notify, invalidate and revoke the MMIO's DMA must be in place to use this
- function. For example a revoke can be built using DMABUF.
- */
-struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar) -{
- struct pci_p2pdma *p2p;
- if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM) ||
pdev->non_mappable_bars)return NULL;- p2p = rcu_dereference_protected(pdev->p2pdma, 1);
- if (WARN_ON(!p2p))
/* Someone forgot to call to pcim_p2pdma_init() before */return NULL;- return &p2p->mem[bar];
-} -EXPORT_SYMBOL_GPL(pcim_p2pdma_provider);
static int pci_p2pdma_setup_pool(struct pci_dev *pdev) { struct pci_p2pdma *p2pdma; diff --git a/drivers/pci/p2pdma.h b/drivers/pci/p2pdma.h new file mode 100644 index 000000000000..946383809981 --- /dev/null +++ b/drivers/pci/p2pdma.h @@ -0,0 +1,29 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/*
- PCI peer-to-peer DMA support.
- */
+#ifndef _PCI_P2PDMA_H +#define _PCI_P2PDMA_H
+#include <linux/genalloc.h> +#include <linux/pci-p2pdma.h> +#include <linux/xarray.h>
+struct pci_p2pdma {
- struct gen_pool *pool;
- bool p2pmem_published;
- struct xarray map_types;
- struct p2pdma_provider mem[PCI_STD_NUM_BARS];
+};
+#ifdef CONFIG_PCI_P2PDMA +void pci_p2pdma_release_pool(struct pci_dev *pdev, struct pci_p2pdma *p2pdma); +#else +static inline void pci_p2pdma_release_pool(struct pci_dev *pdev,
struct pci_p2pdma *p2pdma)+{ +} +#endif
+#endif diff --git a/drivers/pci/p2pdma_core.c b/drivers/pci/p2pdma_core.c new file mode 100644 index 000000000000..9e997f373b1e --- /dev/null +++ b/drivers/pci/p2pdma_core.c @@ -0,0 +1,122 @@ +// SPDX-License-Identifier: GPL-2.0 +/*
- PCI peer-to-peer DMA support core, providing a bare-bones
- pcim_p2pdma_provider() interface to drivers even if full P2PDMA
- isn't present. The full P2PDMA feature is in p2pdma.c (see
- CONFIG_PCI_P2PDMA).
- Copyright (c) 2016-2018, Logan Gunthorpe
- Copyright (c) 2016-2017, Microsemi Corporation
- Copyright (c) 2017, Christoph Hellwig
- Copyright (c) 2018, Eideticom Inc.
- */
+#define pr_fmt(fmt) "pci-p2pdma: " fmt +#include <linux/ctype.h> +#include <linux/genalloc.h> +#include <linux/memremap.h> +#include <linux/pci-p2pdma.h> +#include <linux/xarray.h>
+#include "p2pdma.h"
+static void pci_p2pdma_release(void *data) +{
- struct pci_dev *pdev = data;
- struct pci_p2pdma *p2pdma;
- p2pdma = rcu_dereference_protected(pdev->p2pdma, 1);
- if (!p2pdma)
return;- /* Flush and disable pci_alloc_p2p_mem() */
- pdev->p2pdma = NULL;
- pci_p2pdma_release_pool(pdev, p2pdma);
- xa_destroy(&p2pdma->map_types);
+}
+/**
- pcim_p2pdma_init - Initialise peer-to-peer DMA providers
- @pdev: The PCI device to enable P2PDMA for
- This function initializes the peer-to-peer DMA infrastructure
- for a PCI device. It allocates and sets up the necessary data
- structures to support P2PDMA operations, including mapping type
- tracking.
- */
+int pcim_p2pdma_init(struct pci_dev *pdev) +{
- struct pci_p2pdma *p2p;
- int i, ret;
- if (pdev->non_mappable_bars)
return -EOPNOTSUPP;- p2p = rcu_dereference_protected(pdev->p2pdma, 1);
- if (p2p)
return 0;- p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL);
- if (!p2p)
return -ENOMEM;- xa_init(&p2p->map_types);
- /*
* Iterate over all standard PCI BARs and record only those that* correspond to MMIO regions. Skip non-memory resources (e.g. I/O* port BARs) since they cannot be used for peer-to-peer (P2P)* transactions.*/- for (i = 0; i < PCI_STD_NUM_BARS; i++) {
if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))continue;p2p->mem[i].owner = &pdev->dev;p2p->mem[i].bus_offset =pci_bus_address(pdev, i) - pci_resource_start(pdev, i);- }
- ret = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev);
- if (ret)
goto out_p2p;- rcu_assign_pointer(pdev->p2pdma, p2p);
- return 0;
+out_p2p:
- devm_kfree(&pdev->dev, p2p);
- return ret;
+} +EXPORT_SYMBOL_GPL(pcim_p2pdma_init);
+/**
- pcim_p2pdma_provider - Get peer-to-peer DMA provider
- @pdev: The PCI device to enable P2PDMA for
- @bar: BAR index to get provider
- This function gets peer-to-peer DMA provider for a PCI device. The lifetime
- of the provider (and of course the MMIO) is bound to the lifetime of the
- driver. A driver calling this function must ensure that all references to the
- provider, and any DMA mappings created for any MMIO, are all cleaned up
- before the driver remove() completes.
- Since P2P is almost always shared with a second driver this means some system
- to notify, invalidate and revoke the MMIO's DMA must be in place to use this
- function. For example a revoke can be built using DMABUF.
- */
+struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar) +{
- struct pci_p2pdma *p2p;
- if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM) ||
pdev->non_mappable_bars)return NULL;- p2p = rcu_dereference_protected(pdev->p2pdma, 1);
- if (WARN_ON(!p2p))
/* Someone forgot to call to pcim_p2pdma_init() before */return NULL;- return &p2p->mem[bar];
+} +EXPORT_SYMBOL_GPL(pcim_p2pdma_provider); diff --git a/include/linux/pci-p2pdma.h b/include/linux/pci-p2pdma.h index 873de20a2247..4c42a7b2ee85 100644 --- a/include/linux/pci-p2pdma.h +++ b/include/linux/pci-p2pdma.h @@ -67,9 +67,22 @@ enum pci_p2pdma_map_type { PCI_P2PDMA_MAP_THRU_HOST_BRIDGE, }; -#ifdef CONFIG_PCI_P2PDMA +#ifdef CONFIG_PCI_P2PDMA_CORE int pcim_p2pdma_init(struct pci_dev *pdev); struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev, int bar); +#else +static inline int pcim_p2pdma_init(struct pci_dev *pdev) +{
- return -EOPNOTSUPP;
+} +static inline struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev,
int bar)+{
- return NULL;
+} +#endif
+#ifdef CONFIG_PCI_P2PDMA int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, u64 offset); int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients, @@ -89,15 +102,6 @@ ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev, enum pci_p2pdma_map_type pci_p2pdma_map_type(struct p2pdma_provider *provider, struct device *dev); #else /* CONFIG_PCI_P2PDMA */ -static inline int pcim_p2pdma_init(struct pci_dev *pdev) -{
- return -EOPNOTSUPP;
-} -static inline struct p2pdma_provider *pcim_p2pdma_provider(struct pci_dev *pdev,
int bar)-{
- return NULL;
-} static inline int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, u64 offset) { diff --git a/include/linux/pci.h b/include/linux/pci.h index 64b308b6e61c..8adbabbec5bf 100644 --- a/include/linux/pci.h +++ b/include/linux/pci.h @@ -556,7 +556,7 @@ struct pci_dev { u16 pasid_cap; /* PASID Capability offset */ u16 pasid_features; #endif -#ifdef CONFIG_PCI_P2PDMA +#ifdef CONFIG_PCI_P2PDMA_CORE struct pci_p2pdma __rcu *p2pdma; #endif #ifdef CONFIG_PCI_DOE
Add vfio_pci_dma_buf_find_pfn(), which a VMA fault handler can use to find a PFN.
This supports multi-range DMABUFs, which typically would be used to represent scattered spans but might even represent overlapping or aliasing spans of PFNs.
Because this is intended to be used in vfio_pci_core.c, we also need to expose the struct vfio_pci_dma_buf in the vfio_pci_priv.h header.
Signed-off-by: Matt Evans matt@ozlabs.org --- drivers/vfio/pci/vfio_pci_dmabuf.c | 153 ++++++++++++++++++++++++++--- drivers/vfio/pci/vfio_pci_priv.h | 20 ++++ 2 files changed, 160 insertions(+), 13 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index c16f460c01d6..7c047400dfd1 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -9,19 +9,6 @@
MODULE_IMPORT_NS("DMA_BUF");
-struct vfio_pci_dma_buf { - struct dma_buf *dmabuf; - struct vfio_pci_core_device *vdev; - struct list_head dmabufs_elm; - size_t size; - struct phys_vec *phys_vec; - struct p2pdma_provider *provider; - u32 nr_ranges; - struct kref kref; - struct completion comp; - u8 revoked : 1; -}; - static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) { @@ -106,6 +93,146 @@ static const struct dma_buf_ops vfio_pci_dmabuf_ops = { .release = vfio_pci_dma_buf_release, };
+int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *priv, + struct vm_area_struct *vma, + unsigned long fault_addr, + unsigned int order, + unsigned long *out_pfn) +{ + /* + * Given a VMA (start, end, pgoffs) and a fault address, + * search the corresponding DMABUF's phys_vec[] to find the + * range representing the address's offset into the VMA, and + * its PFN. + * + * The phys_vec[] ranges represent contiguous spans of VAs + * upwards from the buffer offset 0; the actual PFNs might be + * in any order, overlap/alias, etc. Calculate an offset of + * the desired page given VMA start/pgoff and address, then + * search upwards from 0 to find which span contains it. + * + * On success, a valid PFN for a page sized by 'order' is + * returned into out_pfn. + * + * Failure occurs if: + * - A hugepage would cross the edge of the VMA, + * - A hugepage isn't entirely contained within a range + * (including where it straddles the boundary between + * ranges), + * - We find a range, but the final PFN isn't aligned to the + * requested order. + * + * Upon failure, -EAGAIN is returned and the caller is + * expected to try again with a smaller order, which will + * eventually succeed (order=0 will always work). + * + * It's suboptimal if DMABUFs are created with neighbouring + * ranges that are physically contiguous, since hugepages + * can't straddle range boundaries. (The construction of the + * ranges should merge them in this case.) + * + * Finally, vma_pgoff_adjust is used with a DMABUF created for + * a VFIO BAR mmap: a BAR mapped with vm_pgoff > 0 creates a + * DMABUF such that byte 0 of the VMA corresponds to byte 0 of + * the DMABUF and byte 'vm_pgoff << PAGE_SHIFT' into the BAR. + * To avoid double-offsetting in this scenario, subtracting + * vma_pgoff_adjust from this (non-zero) vm_pgoff generates + * the effective offset. + */ + + const unsigned long pagesize = PAGE_SIZE << order; + unsigned long vma_off = ((vma->vm_pgoff - priv->vma_pgoff_adjust) << + PAGE_SHIFT) & VFIO_PCI_OFFSET_MASK; + unsigned long rounded_page_addr = ALIGN_DOWN(fault_addr, pagesize); + unsigned long rounded_page_end = rounded_page_addr + pagesize; + unsigned long fault_offset; + unsigned long fault_offset_end; + unsigned long range_start_offset = 0; + unsigned int i; + int ret; + + if (rounded_page_addr < vma->vm_start || rounded_page_end > vma->vm_end) { + if (order > 0) + return -EAGAIN; + + /* A fault address outside of the VMA is absurd. */ + dev_warn_ratelimited( + &priv->vdev->pdev->dev, + "Fault addr 0x%lx outside VMA 0x%lx-0x%lx\n", + fault_addr, vma->vm_start, vma->vm_end); + return -EFAULT; + } + + /* + * fault_offset[_end] is the span within the DMABUF + * corresponding to the faulting page: + */ + if (unlikely(check_add_overflow(rounded_page_addr - vma->vm_start, + vma_off, &fault_offset) || + check_add_overflow(fault_offset, pagesize, + &fault_offset_end))) + return -EFAULT; + + /* + * Iterate over ranges in the buffer, summing their lengths: + * range_start_offset represents the current range's starting + * offset in the buffer (from 0 upwards). + * + * A failure for order == 0 is unexpected, and triggers a + * fault/warn. + */ + ret = (order == 0) ? -EFAULT : -EAGAIN; + + for (i = 0; i < priv->nr_ranges; i++) { + size_t range_len = priv->phys_vec[i].len; + + /* Early exit if range starts after the page end */ + if (fault_offset_end <= range_start_offset) + break; + + if (fault_offset >= range_start_offset && + fault_offset_end <= range_start_offset + range_len) { + /* + * The faulting page is wholly contained + * within the span represented by this range, + * so validate PFN alignment for the order. + * The if() condition ensures the pfn + * arithmetic won't overflow. + */ + unsigned long pfn = + ((fault_offset - range_start_offset) + + priv->phys_vec[i].paddr) >> PAGE_SHIFT; + + if (IS_ALIGNED(pfn, 1 << order)) { + *out_pfn = pfn; + ret = 0; + } + /* + * Else order > 0; EAGAIN retries with smaller + * order + */ + break; + } + range_start_offset += range_len; + } + + if (order == 0 && ret != 0) + /* + * The address fell outside of the span represented by + * the (concatenated) ranges. As setup of a mapping + * ensures that the VMA is <= the total size of the + * ranges this should never happen. If it does, warn + * and SIGBUS. + */ + dev_warn_ratelimited( + &priv->vdev->pdev->dev, + "No range for addr 0x%lx, order %d: VMA 0x%lx-0x%lx pgoff 0x%lx, %u ranges, size 0x%zx\n", + fault_addr, order, vma->vm_start, vma->vm_end, + vma->vm_pgoff, priv->nr_ranges, priv->size); + + return ret; +} + /* * This is a temporary "private interconnect" between VFIO DMABUF and iommufd. * It allows the two co-operating drivers to exchange the physical address of diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index fca9d0dfac90..c8f6f959056a 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -23,6 +23,20 @@ struct vfio_pci_ioeventfd { bool test_mem; };
+struct vfio_pci_dma_buf { + struct dma_buf *dmabuf; + struct vfio_pci_core_device *vdev; + struct list_head dmabufs_elm; + size_t size; + struct phys_vec *phys_vec; + struct p2pdma_provider *provider; + u32 nr_ranges; + struct kref kref; + struct completion comp; + unsigned long vma_pgoff_adjust; + u8 revoked : 1; +}; + bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev); void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev);
@@ -114,6 +128,12 @@ static inline bool vfio_pci_is_vga(struct pci_dev *pdev) return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA; }
+int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *vpdmabuf, + struct vm_area_struct *vma, + unsigned long address, + unsigned int order, + unsigned long *out_pfn); + #ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, struct vfio_device_feature_dma_buf __user *arg,
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:47 AM
Add vfio_pci_dma_buf_find_pfn(), which a VMA fault handler can use to find a PFN.
This supports multi-range DMABUFs, which typically would be used to represent scattered spans but might even represent overlapping or aliasing spans of PFNs.
Because this is intended to be used in vfio_pci_core.c, we also need to expose the struct vfio_pci_dma_buf in the vfio_pci_priv.h header.
Signed-off-by: Matt Evans matt@ozlabs.org
Reviewed-by: Kevin Tian kevin.tian@intel.com
On Wed, 15 Jul 2026 18:47:26 +0100 Matt Evans matt@ozlabs.org wrote:
Add vfio_pci_dma_buf_find_pfn(), which a VMA fault handler can use to find a PFN.
This supports multi-range DMABUFs, which typically would be used to represent scattered spans but might even represent overlapping or aliasing spans of PFNs.
Because this is intended to be used in vfio_pci_core.c, we also need to expose the struct vfio_pci_dma_buf in the vfio_pci_priv.h header.
Nit, it would be useful to mention the addition of vma_pgoff_adjust here and justification for it added here as a static zero versus later in the series where its value is updated.
Signed-off-by: Matt Evans matt@ozlabs.org
drivers/vfio/pci/vfio_pci_dmabuf.c | 153 ++++++++++++++++++++++++++--- drivers/vfio/pci/vfio_pci_priv.h | 20 ++++ 2 files changed, 160 insertions(+), 13 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index c16f460c01d6..7c047400dfd1 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -9,19 +9,6 @@ MODULE_IMPORT_NS("DMA_BUF"); -struct vfio_pci_dma_buf {
- struct dma_buf *dmabuf;
- struct vfio_pci_core_device *vdev;
- struct list_head dmabufs_elm;
- size_t size;
- struct phys_vec *phys_vec;
- struct p2pdma_provider *provider;
- u32 nr_ranges;
- struct kref kref;
- struct completion comp;
- u8 revoked : 1;
-};
static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) { @@ -106,6 +93,146 @@ static const struct dma_buf_ops vfio_pci_dmabuf_ops = { .release = vfio_pci_dma_buf_release, }; +int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *priv,
struct vm_area_struct *vma,unsigned long fault_addr,unsigned int order,unsigned long *out_pfn)+{
- /*
* Given a VMA (start, end, pgoffs) and a fault address,* search the corresponding DMABUF's phys_vec[] to find the* range representing the address's offset into the VMA, and* its PFN.** The phys_vec[] ranges represent contiguous spans of VAs* upwards from the buffer offset 0; the actual PFNs might be* in any order, overlap/alias, etc. Calculate an offset of* the desired page given VMA start/pgoff and address, then* search upwards from 0 to find which span contains it.** On success, a valid PFN for a page sized by 'order' is* returned into out_pfn.** Failure occurs if:* - A hugepage would cross the edge of the VMA,* - A hugepage isn't entirely contained within a range* (including where it straddles the boundary between* ranges),* - We find a range, but the final PFN isn't aligned to the* requested order.** Upon failure, -EAGAIN is returned and the caller is* expected to try again with a smaller order, which will* eventually succeed (order=0 will always work).
We typically use -EAGAIN to indicate a transient error where the caller should try again with the same args. We're documenting here that -EAGAIN means to try again with different args. Why wouldn't we use something like -ERANGE to indicate that?
** It's suboptimal if DMABUFs are created with neighbouring* ranges that are physically contiguous, since hugepages* can't straddle range boundaries. (The construction of the* ranges should merge them in this case.)** Finally, vma_pgoff_adjust is used with a DMABUF created for* a VFIO BAR mmap: a BAR mapped with vm_pgoff > 0 creates a* DMABUF such that byte 0 of the VMA corresponds to byte 0 of* the DMABUF and byte 'vm_pgoff << PAGE_SHIFT' into the BAR.* To avoid double-offsetting in this scenario, subtracting* vma_pgoff_adjust from this (non-zero) vm_pgoff generates* the effective offset.*/- const unsigned long pagesize = PAGE_SIZE << order;
- unsigned long vma_off = ((vma->vm_pgoff - priv->vma_pgoff_adjust) <<
PAGE_SHIFT) & VFIO_PCI_OFFSET_MASK;- unsigned long rounded_page_addr = ALIGN_DOWN(fault_addr, pagesize);
- unsigned long rounded_page_end = rounded_page_addr + pagesize;
- unsigned long fault_offset;
- unsigned long fault_offset_end;
- unsigned long range_start_offset = 0;
- unsigned int i;
- int ret;
AIUI, find_pfn should only act on a live dmabuf, but we're relying on the caller to guarantee that it's live and not racing a revoke. The dmabuf is live if !revoked under memory_lock. Should this annotate and validate that requirement by adding a vdev arg so we can do:
lockdep_assert_held(&vdev->memory_lock); if (priv->revoked) // later updated to status return -ENODEV;
This also removes one level of dereferencing in the below dev_warns, which is otherwise a bit suspect without the validation. I realize the mmap path does the validation, but this is an exposed internal API, not a static function there. Thanks,
Alex
- if (rounded_page_addr < vma->vm_start || rounded_page_end > vma->vm_end) {
if (order > 0)return -EAGAIN;/* A fault address outside of the VMA is absurd. */dev_warn_ratelimited(&priv->vdev->pdev->dev,"Fault addr 0x%lx outside VMA 0x%lx-0x%lx\n",fault_addr, vma->vm_start, vma->vm_end);return -EFAULT;- }
- /*
* fault_offset[_end] is the span within the DMABUF* corresponding to the faulting page:*/- if (unlikely(check_add_overflow(rounded_page_addr - vma->vm_start,
vma_off, &fault_offset) ||check_add_overflow(fault_offset, pagesize,&fault_offset_end)))return -EFAULT;- /*
* Iterate over ranges in the buffer, summing their lengths:* range_start_offset represents the current range's starting* offset in the buffer (from 0 upwards).** A failure for order == 0 is unexpected, and triggers a* fault/warn.*/- ret = (order == 0) ? -EFAULT : -EAGAIN;
- for (i = 0; i < priv->nr_ranges; i++) {
size_t range_len = priv->phys_vec[i].len;/* Early exit if range starts after the page end */if (fault_offset_end <= range_start_offset)break;if (fault_offset >= range_start_offset &&fault_offset_end <= range_start_offset + range_len) {/** The faulting page is wholly contained* within the span represented by this range,* so validate PFN alignment for the order.* The if() condition ensures the pfn* arithmetic won't overflow.*/unsigned long pfn =((fault_offset - range_start_offset) +priv->phys_vec[i].paddr) >> PAGE_SHIFT;if (IS_ALIGNED(pfn, 1 << order)) {*out_pfn = pfn;ret = 0;}/** Else order > 0; EAGAIN retries with smaller* order*/break;}range_start_offset += range_len;- }
- if (order == 0 && ret != 0)
/** The address fell outside of the span represented by* the (concatenated) ranges. As setup of a mapping* ensures that the VMA is <= the total size of the* ranges this should never happen. If it does, warn* and SIGBUS.*/dev_warn_ratelimited(&priv->vdev->pdev->dev,"No range for addr 0x%lx, order %d: VMA 0x%lx-0x%lx pgoff 0x%lx, %u ranges, size 0x%zx\n",fault_addr, order, vma->vm_start, vma->vm_end,vma->vm_pgoff, priv->nr_ranges, priv->size);- return ret;
+}
/*
- This is a temporary "private interconnect" between VFIO DMABUF and iommufd.
- It allows the two co-operating drivers to exchange the physical address of
diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index fca9d0dfac90..c8f6f959056a 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -23,6 +23,20 @@ struct vfio_pci_ioeventfd { bool test_mem; }; +struct vfio_pci_dma_buf {
- struct dma_buf *dmabuf;
- struct vfio_pci_core_device *vdev;
- struct list_head dmabufs_elm;
- size_t size;
- struct phys_vec *phys_vec;
- struct p2pdma_provider *provider;
- u32 nr_ranges;
- struct kref kref;
- struct completion comp;
- unsigned long vma_pgoff_adjust;
- u8 revoked : 1;
+};
bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev); void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev); @@ -114,6 +128,12 @@ static inline bool vfio_pci_is_vga(struct pci_dev *pdev) return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA; } +int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *vpdmabuf,
struct vm_area_struct *vma,unsigned long address,unsigned int order,unsigned long *out_pfn);#ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, struct vfio_device_feature_dma_buf __user *arg,
Hi Alex,
On 29/07/2026 18:52, Alex Williamson wrote:
On Wed, 15 Jul 2026 18:47:26 +0100 Matt Evans matt@ozlabs.org wrote:
Add vfio_pci_dma_buf_find_pfn(), which a VMA fault handler can use to find a PFN.
This supports multi-range DMABUFs, which typically would be used to represent scattered spans but might even represent overlapping or aliasing spans of PFNs.
Because this is intended to be used in vfio_pci_core.c, we also need to expose the struct vfio_pci_dma_buf in the vfio_pci_priv.h header.
Nit, it would be useful to mention the addition of vma_pgoff_adjust here and justification for it added here as a static zero versus later in the series where its value is updated.
Sure, good point.
Signed-off-by: Matt Evans matt@ozlabs.org
drivers/vfio/pci/vfio_pci_dmabuf.c | 153 ++++++++++++++++++++++++++--- drivers/vfio/pci/vfio_pci_priv.h | 20 ++++ 2 files changed, 160 insertions(+), 13 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index c16f460c01d6..7c047400dfd1 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -9,19 +9,6 @@ MODULE_IMPORT_NS("DMA_BUF"); -struct vfio_pci_dma_buf {
- struct dma_buf *dmabuf;
- struct vfio_pci_core_device *vdev;
- struct list_head dmabufs_elm;
- size_t size;
- struct phys_vec *phys_vec;
- struct p2pdma_provider *provider;
- u32 nr_ranges;
- struct kref kref;
- struct completion comp;
- u8 revoked : 1;
-};
static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) { @@ -106,6 +93,146 @@ static const struct dma_buf_ops vfio_pci_dmabuf_ops = { .release = vfio_pci_dma_buf_release, }; +int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *priv,
struct vm_area_struct *vma,unsigned long fault_addr,unsigned int order,unsigned long *out_pfn)+{
- /*
* Given a VMA (start, end, pgoffs) and a fault address,* search the corresponding DMABUF's phys_vec[] to find the* range representing the address's offset into the VMA, and* its PFN.** The phys_vec[] ranges represent contiguous spans of VAs* upwards from the buffer offset 0; the actual PFNs might be* in any order, overlap/alias, etc. Calculate an offset of* the desired page given VMA start/pgoff and address, then* search upwards from 0 to find which span contains it.** On success, a valid PFN for a page sized by 'order' is* returned into out_pfn.** Failure occurs if:* - A hugepage would cross the edge of the VMA,* - A hugepage isn't entirely contained within a range* (including where it straddles the boundary between* ranges),* - We find a range, but the final PFN isn't aligned to the* requested order.** Upon failure, -EAGAIN is returned and the caller is* expected to try again with a smaller order, which will* eventually succeed (order=0 will always work).We typically use -EAGAIN to indicate a transient error where the caller should try again with the same args. We're documenting here that -EAGAIN means to try again with different args. Why wouldn't we use something like -ERANGE to indicate that?
-EAGAIN didn't seem all that inappropriate, but you're correct that it isn't being used as a transient error here. I'll change it to -ERANGE as suggested.
** It's suboptimal if DMABUFs are created with neighbouring* ranges that are physically contiguous, since hugepages* can't straddle range boundaries. (The construction of the* ranges should merge them in this case.)** Finally, vma_pgoff_adjust is used with a DMABUF created for* a VFIO BAR mmap: a BAR mapped with vm_pgoff > 0 creates a* DMABUF such that byte 0 of the VMA corresponds to byte 0 of* the DMABUF and byte 'vm_pgoff << PAGE_SHIFT' into the BAR.* To avoid double-offsetting in this scenario, subtracting* vma_pgoff_adjust from this (non-zero) vm_pgoff generates* the effective offset.*/- const unsigned long pagesize = PAGE_SIZE << order;
- unsigned long vma_off = ((vma->vm_pgoff - priv->vma_pgoff_adjust) <<
PAGE_SHIFT) & VFIO_PCI_OFFSET_MASK;- unsigned long rounded_page_addr = ALIGN_DOWN(fault_addr, pagesize);
- unsigned long rounded_page_end = rounded_page_addr + pagesize;
- unsigned long fault_offset;
- unsigned long fault_offset_end;
- unsigned long range_start_offset = 0;
- unsigned int i;
- int ret;
AIUI, find_pfn should only act on a live dmabuf, but we're relying on the caller to guarantee that it's live and not racing a revoke. The dmabuf is live if !revoked under memory_lock. Should this annotate and validate that requirement by adding a vdev arg so we can do:
lockdep_assert_held(&vdev->memory_lock); if (priv->revoked) // later updated to status return -ENODEV;
Yes, it makes no sense to call this when the buffer's revoked, and it's a good idea to both explicitly document/assert this.
I think your point regarding the extra vdev parameter -- as opposed to a local 'vdev = priv->vdev' dereference -- is that it should also assert that vdev is non-NULL, but one can't just read priv->vdev. Dealing with the races having only priv takes some careful steps (see vfio_pci_mmap_huge_fault), so make it the caller's responsibility to provide a vdev. This function can both check it's non-NULL, and verify it relates to priv, and then annotate vdev->memory_lock etc.
This also removes one level of dereferencing in the below dev_warns, which is otherwise a bit suspect without the validation. I realize the mmap path does the validation, but this is an exposed internal API, not a static function there. Thanks,
[s/mmap path/the fault path/] Yep, checks here make sense, thanks for the suggestions.
Matt
Alex
- if (rounded_page_addr < vma->vm_start || rounded_page_end > vma->vm_end) {
if (order > 0)return -EAGAIN;/* A fault address outside of the VMA is absurd. */dev_warn_ratelimited(&priv->vdev->pdev->dev,"Fault addr 0x%lx outside VMA 0x%lx-0x%lx\n",fault_addr, vma->vm_start, vma->vm_end);return -EFAULT;- }
- /*
* fault_offset[_end] is the span within the DMABUF* corresponding to the faulting page:*/- if (unlikely(check_add_overflow(rounded_page_addr - vma->vm_start,
vma_off, &fault_offset) ||check_add_overflow(fault_offset, pagesize,&fault_offset_end)))return -EFAULT;- /*
* Iterate over ranges in the buffer, summing their lengths:* range_start_offset represents the current range's starting* offset in the buffer (from 0 upwards).** A failure for order == 0 is unexpected, and triggers a* fault/warn.*/- ret = (order == 0) ? -EFAULT : -EAGAIN;
- for (i = 0; i < priv->nr_ranges; i++) {
size_t range_len = priv->phys_vec[i].len;/* Early exit if range starts after the page end */if (fault_offset_end <= range_start_offset)break;if (fault_offset >= range_start_offset &&fault_offset_end <= range_start_offset + range_len) {/** The faulting page is wholly contained* within the span represented by this range,* so validate PFN alignment for the order.* The if() condition ensures the pfn* arithmetic won't overflow.*/unsigned long pfn =((fault_offset - range_start_offset) +priv->phys_vec[i].paddr) >> PAGE_SHIFT;if (IS_ALIGNED(pfn, 1 << order)) {*out_pfn = pfn;ret = 0;}/** Else order > 0; EAGAIN retries with smaller* order*/break;}range_start_offset += range_len;- }
- if (order == 0 && ret != 0)
/** The address fell outside of the span represented by* the (concatenated) ranges. As setup of a mapping* ensures that the VMA is <= the total size of the* ranges this should never happen. If it does, warn* and SIGBUS.*/dev_warn_ratelimited(&priv->vdev->pdev->dev,"No range for addr 0x%lx, order %d: VMA 0x%lx-0x%lx pgoff 0x%lx, %u ranges, size 0x%zx\n",fault_addr, order, vma->vm_start, vma->vm_end,vma->vm_pgoff, priv->nr_ranges, priv->size);- return ret;
+}
/*
- This is a temporary "private interconnect" between VFIO DMABUF and iommufd.
- It allows the two co-operating drivers to exchange the physical address of
diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index fca9d0dfac90..c8f6f959056a 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -23,6 +23,20 @@ struct vfio_pci_ioeventfd { bool test_mem; }; +struct vfio_pci_dma_buf {
- struct dma_buf *dmabuf;
- struct vfio_pci_core_device *vdev;
- struct list_head dmabufs_elm;
- size_t size;
- struct phys_vec *phys_vec;
- struct p2pdma_provider *provider;
- u32 nr_ranges;
- struct kref kref;
- struct completion comp;
- unsigned long vma_pgoff_adjust;
- u8 revoked : 1;
+};
bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev); void vfio_pci_intx_unmask(struct vfio_pci_core_device *vdev); @@ -114,6 +128,12 @@ static inline bool vfio_pci_is_vga(struct pci_dev *pdev) return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA; } +int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *vpdmabuf,
struct vm_area_struct *vma,unsigned long address,unsigned int order,unsigned long *out_pfn);#ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, struct vfio_device_feature_dma_buf __user *arg,
This helper, vfio_pci_core_mmap_prep_dmabuf(), creates a single-range DMABUF for the purpose of mapping a PCI BAR. This is used in a future commit by VFIO's ordinary mmap() path.
This function transfers ownership of the VFIO device fd to the DMABUF, which fput()s when it's released.
Refactor the existing vfio_pci_core_feature_dma_buf() to split out export code common to the two paths, VFIO_DEVICE_FEATURE_DMA_BUF and this new VFIO_BAR mmap().
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Pranjal Shrivastava praan@google.com --- drivers/vfio/pci/vfio_pci_dmabuf.c | 142 +++++++++++++++++++++++------ drivers/vfio/pci/vfio_pci_priv.h | 5 + 2 files changed, 117 insertions(+), 30 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 7c047400dfd1..74c02794bfe2 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -82,6 +82,8 @@ static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf) up_write(&priv->vdev->memory_lock); vfio_device_put_registration(&priv->vdev->vdev); } + if (priv->vfile) + fput(priv->vfile); kfree(priv->phys_vec); kfree(priv); } @@ -233,6 +235,45 @@ int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *priv, return ret; }
+/* + * Create a DMABUF corresponding to priv, add it to vdev->dmabufs list + * for tracking (meaning cleanup or revocation will zap it), and take + * a vfio_device registration. + */ +static int vfio_pci_dmabuf_export(struct vfio_pci_core_device *vdev, + struct vfio_pci_dma_buf *priv, u32 flags) +{ + DEFINE_DMA_BUF_EXPORT_INFO(exp_info); + + if (!vfio_device_try_get_registration(&vdev->vdev)) + return -ENODEV; + + exp_info.ops = &vfio_pci_dmabuf_ops; + exp_info.size = priv->size; + exp_info.flags = flags; + exp_info.priv = priv; + + priv->dmabuf = dma_buf_export(&exp_info); + if (IS_ERR(priv->dmabuf)) { + vfio_device_put_registration(&vdev->vdev); + return PTR_ERR(priv->dmabuf); + } + + kref_init(&priv->kref); + init_completion(&priv->comp); + + /* dma_buf_put() now frees priv */ + INIT_LIST_HEAD(&priv->dmabufs_elm); + down_write(&vdev->memory_lock); + dma_resv_lock(priv->dmabuf->resv, NULL); + priv->revoked = !__vfio_pci_memory_enabled(vdev); + list_add_tail(&priv->dmabufs_elm, &vdev->dmabufs); + dma_resv_unlock(priv->dmabuf->resv); + up_write(&vdev->memory_lock); + + return 0; +} + /* * This is a temporary "private interconnect" between VFIO DMABUF and iommufd. * It allows the two co-operating drivers to exchange the physical address of @@ -351,7 +392,6 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, { struct vfio_device_feature_dma_buf get_dma_buf = {}; struct vfio_region_dma_range *dma_ranges; - DEFINE_DMA_BUF_EXPORT_INFO(exp_info); struct vfio_pci_dma_buf *priv; size_t length; int ret; @@ -411,34 +451,9 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, kfree(dma_ranges); dma_ranges = NULL;
- if (!vfio_device_try_get_registration(&vdev->vdev)) { - ret = -ENODEV; + ret = vfio_pci_dmabuf_export(vdev, priv, get_dma_buf.open_flags); + if (ret) goto err_free_phys; - } - - exp_info.ops = &vfio_pci_dmabuf_ops; - exp_info.size = priv->size; - exp_info.flags = get_dma_buf.open_flags; - exp_info.priv = priv; - - priv->dmabuf = dma_buf_export(&exp_info); - if (IS_ERR(priv->dmabuf)) { - ret = PTR_ERR(priv->dmabuf); - goto err_dev_put; - } - - kref_init(&priv->kref); - init_completion(&priv->comp); - - /* dma_buf_put() now frees priv */ - INIT_LIST_HEAD(&priv->dmabufs_elm); - down_write(&vdev->memory_lock); - dma_resv_lock(priv->dmabuf->resv, NULL); - priv->revoked = !__vfio_pci_memory_enabled(vdev); - list_add_tail(&priv->dmabufs_elm, &vdev->dmabufs); - dma_resv_unlock(priv->dmabuf->resv); - up_write(&vdev->memory_lock); - /* * dma_buf_fd() consumes the reference, when the file closes the dmabuf * will be released. @@ -449,8 +464,6 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
return ret;
-err_dev_put: - vfio_device_put_registration(&vdev->vdev); err_free_phys: kfree(priv->phys_vec); err_free_priv: @@ -460,6 +473,75 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, return ret; }
+int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, + struct vm_area_struct *vma, + u64 phys_start, u64 req_len, + unsigned int res_index) +{ + struct vfio_pci_dma_buf *priv; + unsigned long vma_pgoff = vma->vm_pgoff & (VFIO_PCI_OFFSET_MASK >> PAGE_SHIFT); + int ret; + + priv = kzalloc_obj(*priv); + if (!priv) + return -ENOMEM; + + priv->phys_vec = kzalloc_obj(*priv->phys_vec); + if (!priv->phys_vec) { + ret = -ENOMEM; + goto err_free_priv; + } + + /* + * The DMABUF begins from the mmap()'s BAR offset, i.e. the + * start of the VMA corresponds to byte 0 of the DMABUF and + * byte (vma_pgoff << PAGE_SHIFT) of the BAR. + * + * vfio_pci_dma_buf_find_pfn() reverses this offset using + * vma_pgoff_adjust, so that ultimately a fault's offset from + * the start of the _VMA_ has a consistent usage whether the + * VMA originates from an mmap() of the VFIO device here or a + * direct DMABUF mmap(). + */ + priv->vdev = vdev; + priv->size = req_len; + priv->nr_ranges = 1; + priv->vma_pgoff_adjust = vma_pgoff; + priv->provider = pcim_p2pdma_provider(vdev->pdev, res_index); + if (!priv->provider) { + ret = -EINVAL; + goto err_free_phys; + } + + priv->phys_vec[0].paddr = phys_start + ((u64)vma_pgoff << PAGE_SHIFT); + priv->phys_vec[0].len = priv->size; + + ret = vfio_pci_dmabuf_export(vdev, priv, O_CLOEXEC | O_RDWR); + if (ret) + goto err_free_phys; + + /* + * Ownership of the DMABUF file transfers to the VMA so that + * other users can locate the DMABUF via a VA. Ownership of + * the original VFIO device file being mmap()ed transfers to + * priv, and is put when the DMABUF is released. This + * intentionally does not use get_file()/vma_set_file() + * because the references are already held, and ownership + * moves. + */ + priv->vfile = vma->vm_file; + vma->vm_file = priv->dmabuf->file; + vma->vm_private_data = priv; + + return 0; + +err_free_phys: + kfree(priv->phys_vec); +err_free_priv: + kfree(priv); + return ret; +} + void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked) { struct vfio_pci_dma_buf *priv; diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index c8f6f959056a..06dc0fd3e230 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -30,6 +30,7 @@ struct vfio_pci_dma_buf { size_t size; struct phys_vec *phys_vec; struct p2pdma_provider *provider; + struct file *vfile; u32 nr_ranges; struct kref kref; struct completion comp; @@ -133,6 +134,10 @@ int vfio_pci_dma_buf_find_pfn(struct vfio_pci_dma_buf *vpdmabuf, unsigned long address, unsigned int order, unsigned long *out_pfn); +int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, + struct vm_area_struct *vma, + u64 phys_start, u64 req_len, + unsigned int res_index);
#ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
Convert the VFIO device fd fops->mmap to create a DMABUF representing the BAR mapping, and make the VMA fault handler look up PFNs from the corresponding DMABUF. This supports future code mmap()ing BAR DMABUFs, and iommufd work to support Type1 P2P.
First, vfio_pci_core_mmap() uses the new vfio_pci_core_mmap_prep_dmabuf() helper to export a DMABUF representing a single BAR range. Then, the vfio_pci_mmap_huge_fault() callback is updated to understand revoked buffers, and uses the new vfio_pci_dma_buf_find_pfn() helper to determine the PFN for a given fault address.
Now that the VFIO DMABUFs can be mmap()ed, vfio_pci_dma_buf_move() zaps PTEs (used on the revocation and cleanup paths).
CONFIG_VFIO_PCI_CORE now unconditionally depends on CONFIG_DMA_SHARED_BUFFER and CONFIG_PCI_P2PDMA_CORE. The CONFIG_VFIO_PCI_DMABUF feature conditionally includes support for VFIO_DEVICE_FEATURE_DMA_BUF, depending on the availability of CONFIG_PCI_P2PDMA.
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Kevin Tian kevin.tian@intel.com Reviewed-by: Pranjal Shrivastava praan@google.com --- drivers/vfio/pci/Kconfig | 5 ++- drivers/vfio/pci/Makefile | 3 +- drivers/vfio/pci/vfio_pci_core.c | 67 ++++++++++++++++++------------ drivers/vfio/pci/vfio_pci_dmabuf.c | 12 ++++++ drivers/vfio/pci/vfio_pci_priv.h | 11 +---- 5 files changed, 59 insertions(+), 39 deletions(-)
diff --git a/drivers/vfio/pci/Kconfig b/drivers/vfio/pci/Kconfig index 296bf01e185e..67a2ae1fbc04 100644 --- a/drivers/vfio/pci/Kconfig +++ b/drivers/vfio/pci/Kconfig @@ -6,6 +6,8 @@ config VFIO_PCI_CORE tristate select VFIO_VIRQFD select IRQ_BYPASS_MANAGER + select PCI_P2PDMA_CORE + select DMA_SHARED_BUFFER
config VFIO_PCI_INTX def_bool y if !S390 @@ -56,7 +58,8 @@ config VFIO_PCI_ZDEV_KVM To enable s390x KVM vfio-pci extensions, say Y.
config VFIO_PCI_DMABUF - def_bool y if VFIO_PCI_CORE && PCI_P2PDMA && DMA_SHARED_BUFFER + def_bool y if PCI_P2PDMA + depends on VFIO_PCI_CORE
source "drivers/vfio/pci/mlx5/Kconfig"
diff --git a/drivers/vfio/pci/Makefile b/drivers/vfio/pci/Makefile index 6138f1bf241d..881452ea89be 100644 --- a/drivers/vfio/pci/Makefile +++ b/drivers/vfio/pci/Makefile @@ -1,8 +1,7 @@ # SPDX-License-Identifier: GPL-2.0-only
-vfio-pci-core-y := vfio_pci_core.o vfio_pci_intrs.o vfio_pci_rdwr.o vfio_pci_config.o +vfio-pci-core-y := vfio_pci_core.o vfio_pci_intrs.o vfio_pci_rdwr.o vfio_pci_config.o vfio_pci_dmabuf.o vfio-pci-core-$(CONFIG_VFIO_PCI_ZDEV_KVM) += vfio_pci_zdev.o -vfio-pci-core-$(CONFIG_VFIO_PCI_DMABUF) += vfio_pci_dmabuf.o obj-$(CONFIG_VFIO_PCI_CORE) += vfio-pci-core.o
vfio-pci-y := vfio_pci.o diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c index 3f11a9624b9c..adce779c722c 100644 --- a/drivers/vfio/pci/vfio_pci_core.c +++ b/drivers/vfio/pci/vfio_pci_core.c @@ -1739,18 +1739,6 @@ void vfio_pci_memory_unlock_and_restore(struct vfio_pci_core_device *vdev, u16 c up_write(&vdev->memory_lock); }
-static unsigned long vma_to_pfn(struct vm_area_struct *vma) -{ - struct vfio_pci_core_device *vdev = vma->vm_private_data; - int index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); - u64 pgoff; - - pgoff = vma->vm_pgoff & - ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); - - return (pci_resource_start(vdev->pdev, index) >> PAGE_SHIFT) + pgoff; -} - vm_fault_t vfio_pci_vmf_insert_pfn(struct vfio_pci_core_device *vdev, struct vm_fault *vmf, unsigned long pfn, @@ -1778,23 +1766,42 @@ static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf, unsigned int order) { struct vm_area_struct *vma = vmf->vma; - struct vfio_pci_core_device *vdev = vma->vm_private_data; - unsigned long addr = vmf->address & ~((PAGE_SIZE << order) - 1); - unsigned long pgoff = (addr - vma->vm_start) >> PAGE_SHIFT; - unsigned long pfn = vma_to_pfn(vma) + pgoff; - vm_fault_t ret = VM_FAULT_FALLBACK; - - if (is_aligned_for_order(vma, addr, pfn, order)) { - scoped_guard(rwsem_read, &vdev->memory_lock) - ret = vfio_pci_vmf_insert_pfn(vdev, vmf, pfn, order); + struct vfio_pci_dma_buf *priv = vma->vm_private_data; + struct vfio_pci_core_device *vdev; + unsigned long pfn = 0; + vm_fault_t ret = VM_FAULT_SIGBUS; + + /* + * We can rely on the existence of both a DMABUF (priv) and + * the VFIO device it was exported from (vdev). This fault's + * VMA was established using vfio_pci_core_mmap_prep_dmabuf() + * which transfers ownership of the VFIO device fd to the + * DMABUF, and so the VFIO device is held open because the + * VMA's vm_file (DMABUF) is open. + * + * Since vfio_pci_dma_buf_cleanup() cannot have happened, + * vdev must be valid; we can take memory_lock. + */ + vdev = priv->vdev; + + scoped_guard(rwsem_read, &vdev->memory_lock) { + if (!priv->revoked) { + int pres = vfio_pci_dma_buf_find_pfn(priv, vma, + vmf->address, + order, &pfn); + + if (pres == 0) + ret = vfio_pci_vmf_insert_pfn(vdev, vmf, + pfn, order); + else if (pres == -EAGAIN) + ret = VM_FAULT_FALLBACK; + } }
dev_dbg_ratelimited(&vdev->pdev->dev, - "%s(,order = %d) BAR %ld page offset 0x%lx: 0x%x\n", - __func__, order, - vma->vm_pgoff >> - (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT), - pgoff, (unsigned int)ret); + "%s(order = %d) PFN 0x%lx, VA 0x%lx, pgoff 0x%lx: 0x%x\n", + __func__, order, pfn, vmf->address, + vma->vm_pgoff, (unsigned int)ret);
return ret; } @@ -1819,6 +1826,7 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma unsigned int index; u64 phys_len, req_len, pgoff, req_start; void __iomem *bar_io; + int ret;
index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
@@ -1858,7 +1866,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma if (IS_ERR(bar_io)) return PTR_ERR(bar_io);
- vma->vm_private_data = vdev; + ret = vfio_pci_core_mmap_prep_dmabuf(vdev, vma, + pci_resource_start(pdev, index), + req_len, index); + if (ret) + return ret; + vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot);
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 74c02794bfe2..46b3cf5dc94b 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -9,6 +9,7 @@
MODULE_IMPORT_NS("DMA_BUF");
+#ifdef CONFIG_VFIO_PCI_DMABUF static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) { @@ -25,6 +26,7 @@ static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
return 0; } +#endif /* CONFIG_VFIO_PCI_DMABUF */
static void vfio_pci_dma_buf_done(struct kref *kref) { @@ -89,7 +91,9 @@ static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf) }
static const struct dma_buf_ops vfio_pci_dmabuf_ops = { +#ifdef CONFIG_VFIO_PCI_DMABUF .attach = vfio_pci_dma_buf_attach, +#endif .map_dma_buf = vfio_pci_dma_buf_map, .unmap_dma_buf = vfio_pci_dma_buf_unmap, .release = vfio_pci_dma_buf_release, @@ -274,6 +278,7 @@ static int vfio_pci_dmabuf_export(struct vfio_pci_core_device *vdev, return 0; }
+#ifdef CONFIG_VFIO_PCI_DMABUF /* * This is a temporary "private interconnect" between VFIO DMABUF and iommufd. * It allows the two co-operating drivers to exchange the physical address of @@ -472,6 +477,7 @@ int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, kfree(dma_ranges); return ret; } +#endif /* CONFIG_VFIO_PCI_DMABUF */
int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, struct vm_area_struct *vma, @@ -547,6 +553,10 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked) struct vfio_pci_dma_buf *priv; struct vfio_pci_dma_buf *tmp;
+ /* + * Holding memory_lock ensures a racing VMA fault observes + * priv->revoked properly. + */ lockdep_assert_held_write(&vdev->memory_lock);
list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) { @@ -565,6 +575,8 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked) if (revoked) { kref_put(&priv->kref, vfio_pci_dma_buf_done); wait_for_completion(&priv->comp); + unmap_mapping_range(priv->dmabuf->file->f_mapping, + 0, 0, true); /* * Re-arm the registered kref reference and the * completion so the post-revoke state matches the diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index 06dc0fd3e230..d38e1b98b2e9 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -138,13 +138,13 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, struct vm_area_struct *vma, u64 phys_start, u64 req_len, unsigned int res_index); +void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev); +void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked);
#ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, struct vfio_device_feature_dma_buf __user *arg, size_t argsz); -void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev); -void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked); #else static inline int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, @@ -153,13 +153,6 @@ vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, { return -ENOTTY; } -static inline void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev) -{ -} -static inline void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, - bool revoked) -{ -} #endif
#endif
Since converting BAR mmap()s to using DMABUFs, we lose the original device path in /proc/<pid>/maps, lsof, etc. Generate a debug-oriented synthetic 'filename' based on the cdev, plus BDF, plus resource index.
This applies only to BAR mappings via the VFIO device fd, as explicitly-exported DMABUFs are named by userspace via the DMA_BUF_SET_NAME ioctl.
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Pranjal Shrivastava praan@google.com --- drivers/vfio/pci/vfio_pci_dmabuf.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 46b3cf5dc94b..51606ff3cdc6 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -4,6 +4,7 @@ #include <linux/dma-buf-mapping.h> #include <linux/pci-p2pdma.h> #include <linux/dma-resv.h> +#include <uapi/linux/dma-buf.h>
#include "vfio_pci_priv.h"
@@ -486,6 +487,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, { struct vfio_pci_dma_buf *priv; unsigned long vma_pgoff = vma->vm_pgoff & (VFIO_PCI_OFFSET_MASK >> PAGE_SHIFT); + char *bufname; int ret;
priv = kzalloc_obj(*priv); @@ -498,6 +500,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, goto err_free_priv; }
+ bufname = kasprintf(GFP_KERNEL, "%s:%s/%x", + dev_name(&vdev->vdev.device), pci_name(vdev->pdev), + res_index); + + if (!bufname) { + ret = -ENOMEM; + goto err_free_phys; + } + /* * The DMABUF begins from the mmap()'s BAR offset, i.e. the * start of the VMA corresponds to byte 0 of the DMABUF and @@ -516,7 +527,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, priv->provider = pcim_p2pdma_provider(vdev->pdev, res_index); if (!priv->provider) { ret = -EINVAL; - goto err_free_phys; + goto err_free_name; }
priv->phys_vec[0].paddr = phys_start + ((u64)vma_pgoff << PAGE_SHIFT); @@ -524,7 +535,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev,
ret = vfio_pci_dmabuf_export(vdev, priv, O_CLOEXEC | O_RDWR); if (ret) - goto err_free_phys; + goto err_free_name;
/* * Ownership of the DMABUF file transfers to the VMA so that @@ -539,8 +550,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, vma->vm_file = priv->dmabuf->file; vma->vm_private_data = priv;
+ spin_lock(&priv->dmabuf->name_lock); + kfree(priv->dmabuf->name); + priv->dmabuf->name = bufname; + spin_unlock(&priv->dmabuf->name_lock); + return 0;
+err_free_name: + kfree(bufname); err_free_phys: kfree(priv->phys_vec); err_free_priv:
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:47 AM
Since converting BAR mmap()s to using DMABUFs, we lose the original device path in /proc/<pid>/maps, lsof, etc. Generate a debug-oriented synthetic 'filename' based on the cdev, plus BDF, plus resource index.
This applies only to BAR mappings via the VFIO device fd, as explicitly-exported DMABUFs are named by userspace via the DMA_BUF_SET_NAME ioctl.
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Pranjal Shrivastava praan@google.com
Reviewed-by: Kevin Tian kevin.tian@intel.com
On Wed, 15 Jul 2026 18:47:29 +0100 Matt Evans matt@ozlabs.org wrote:
Since converting BAR mmap()s to using DMABUFs, we lose the original device path in /proc/<pid>/maps, lsof, etc. Generate a debug-oriented synthetic 'filename' based on the cdev, plus BDF, plus resource index.
This applies only to BAR mappings via the VFIO device fd, as explicitly-exported DMABUFs are named by userspace via the DMA_BUF_SET_NAME ioctl.
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Pranjal Shrivastava praan@google.com
drivers/vfio/pci/vfio_pci_dmabuf.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 46b3cf5dc94b..51606ff3cdc6 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -4,6 +4,7 @@ #include <linux/dma-buf-mapping.h> #include <linux/pci-p2pdma.h> #include <linux/dma-resv.h> +#include <uapi/linux/dma-buf.h>
Is this required? It seems unused.
#include "vfio_pci_priv.h" @@ -486,6 +487,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, { struct vfio_pci_dma_buf *priv; unsigned long vma_pgoff = vma->vm_pgoff & (VFIO_PCI_OFFSET_MASK >> PAGE_SHIFT);
- char *bufname; int ret;
priv = kzalloc_obj(*priv); @@ -498,6 +500,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, goto err_free_priv; }
- bufname = kasprintf(GFP_KERNEL, "%s:%s/%x",
dev_name(&vdev->vdev.device), pci_name(vdev->pdev),res_index);- if (!bufname) {
ret = -ENOMEM;goto err_free_phys;- }
- /*
- The DMABUF begins from the mmap()'s BAR offset, i.e. the
- start of the VMA corresponds to byte 0 of the DMABUF and
@@ -516,7 +527,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, priv->provider = pcim_p2pdma_provider(vdev->pdev, res_index); if (!priv->provider) { ret = -EINVAL;
goto err_free_phys;
goto err_free_name;priv->phys_vec[0].paddr = phys_start + ((u64)vma_pgoff << PAGE_SHIFT); @@ -524,7 +535,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, ret = vfio_pci_dmabuf_export(vdev, priv, O_CLOEXEC | O_RDWR); if (ret)
goto err_free_phys;
goto err_free_name;/* * Ownership of the DMABUF file transfers to the VMA so that @@ -539,8 +550,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, vma->vm_file = priv->dmabuf->file; vma->vm_private_data = priv;
- spin_lock(&priv->dmabuf->name_lock);
- kfree(priv->dmabuf->name);
- priv->dmabuf->name = bufname;
- spin_unlock(&priv->dmabuf->name_lock);
This is simple, but shouldn't it really have a sanctioned, exported dmabuf helper rather than open coding this handling? Thanks,
Alex
- return 0;
+err_free_name:
- kfree(bufname);
err_free_phys: kfree(priv->phys_vec); err_free_priv:
Hi Alex,
On 29/07/2026 18:52, Alex Williamson wrote:
On Wed, 15 Jul 2026 18:47:29 +0100 Matt Evans matt@ozlabs.org wrote:
Since converting BAR mmap()s to using DMABUFs, we lose the original device path in /proc/<pid>/maps, lsof, etc. Generate a debug-oriented synthetic 'filename' based on the cdev, plus BDF, plus resource index.
This applies only to BAR mappings via the VFIO device fd, as explicitly-exported DMABUFs are named by userspace via the DMA_BUF_SET_NAME ioctl.
Signed-off-by: Matt Evans matt@ozlabs.org Reviewed-by: Pranjal Shrivastava praan@google.com
drivers/vfio/pci/vfio_pci_dmabuf.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 46b3cf5dc94b..51606ff3cdc6 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -4,6 +4,7 @@ #include <linux/dma-buf-mapping.h> #include <linux/pci-p2pdma.h> #include <linux/dma-resv.h> +#include <uapi/linux/dma-buf.h>
Is this required? It seems unused.
No, good catch. That originally pulled in DMA_BUF_NAME_LEN in the previous version snprintf'ing here. Removed.
#include "vfio_pci_priv.h" @@ -486,6 +487,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, { struct vfio_pci_dma_buf *priv; unsigned long vma_pgoff = vma->vm_pgoff & (VFIO_PCI_OFFSET_MASK >> PAGE_SHIFT);
- char *bufname; int ret;
priv = kzalloc_obj(*priv); @@ -498,6 +500,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, goto err_free_priv; }
- bufname = kasprintf(GFP_KERNEL, "%s:%s/%x",
dev_name(&vdev->vdev.device), pci_name(vdev->pdev),res_index);- if (!bufname) {
ret = -ENOMEM;goto err_free_phys;- }
- /*
- The DMABUF begins from the mmap()'s BAR offset, i.e. the
- start of the VMA corresponds to byte 0 of the DMABUF and
@@ -516,7 +527,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, priv->provider = pcim_p2pdma_provider(vdev->pdev, res_index); if (!priv->provider) { ret = -EINVAL;
goto err_free_phys;
goto err_free_name;priv->phys_vec[0].paddr = phys_start + ((u64)vma_pgoff << PAGE_SHIFT); @@ -524,7 +535,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, ret = vfio_pci_dmabuf_export(vdev, priv, O_CLOEXEC | O_RDWR); if (ret)
goto err_free_phys;
goto err_free_name;/* * Ownership of the DMABUF file transfers to the VMA so that @@ -539,8 +550,15 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, vma->vm_file = priv->dmabuf->file; vma->vm_private_data = priv;
- spin_lock(&priv->dmabuf->name_lock);
- kfree(priv->dmabuf->name);
- priv->dmabuf->name = bufname;
- spin_unlock(&priv->dmabuf->name_lock);
This is simple, but shouldn't it really have a sanctioned, exported dmabuf helper rather than open coding this handling? Thanks,
Hm, okay that's reasonable. I've refactored dma_buf_set_name() and used it from here (separating out a userspace-string-copying version for the original ioctl usage).
Thanks for looking at the rest of v5 also.
Matt
Alex
- return 0;
+err_free_name:
- kfree(bufname);
err_free_phys: kfree(priv->phys_vec); err_free_priv:
Previously, vfio_pci_zap_bars() (and the wrapper vfio_pci_zap_and_down_write_memory_lock()) calls were paired with calls to vfio_pci_dma_buf_move().
This commit replaces them with a unified new function, vfio_pci_revoke_bars() containing both the vfio_pci_dma_buf_move() and the unmap_mapping_range(), making it harder for callers to omit one. It adds a wrapper, vfio_pci_lock_revoke_bars(), which takes the write memory_lock before zapping, and adds a new vfio_pci_unrevoke_bars() for the re-enable path.
As of "vfio/pci: Convert BAR mmap() to use a DMABUF", the zap via unmap_mapping_range() is no longer performed for vfio-pci since the DMABUFs used for BAR mappings already zap PTEs when the vfio_pci_dma_buf_move() occurs.
However, it must be assumed that VFIO drivers which override the .mmap op could create mappings _not_ backed by DMABUFs. So, the zap is still performed on revoke if .mmap is overridden, using a new zap_bars_on_revoke flag. A driver can explicitly opt out; the flag is cleared by the hisi_acc_vfio_pci driver, since its .mmap just wraps vfio_pci_core_mmap() and so still uses DMABUFs.
Signed-off-by: Matt Evans matt@ozlabs.org --- .../vfio/pci/hisilicon/hisi_acc_vfio_pci.c | 8 ++ drivers/vfio/pci/vfio_pci_config.c | 30 +++----- drivers/vfio/pci/vfio_pci_core.c | 74 ++++++++++++++----- drivers/vfio/pci/vfio_pci_priv.h | 3 +- include/linux/vfio_pci_core.h | 1 + 5 files changed, 77 insertions(+), 39 deletions(-)
diff --git a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c index 86362ec424a5..51990f6d66d5 100644 --- a/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c +++ b/drivers/vfio/pci/hisilicon/hisi_acc_vfio_pci.c @@ -1692,6 +1692,14 @@ static int hisi_acc_vfio_pci_probe(struct pci_dev *pdev, const struct pci_device if (ret) goto out_put_vdev;
+ /* + * hisi_acc_vfio_pci_mmap() calls down to + * vfio_pci_core_mmap(), so BAR mappings are still + * DMABUF-backed. They don't require a zap on revoke, so opt + * out: + */ + hisi_acc_vdev->core_device.zap_bars_on_revoke = false; + hisi_acc_vfio_debug_init(hisi_acc_vdev); return 0;
diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c index a10ed733f0e3..bc6ccb2e135c 100644 --- a/drivers/vfio/pci/vfio_pci_config.c +++ b/drivers/vfio/pci/vfio_pci_config.c @@ -590,12 +590,10 @@ static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos, virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY); new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
- if (!new_mem) { - vfio_pci_zap_and_down_write_memory_lock(vdev); - vfio_pci_dma_buf_move(vdev, true); - } else { + if (!new_mem) + vfio_pci_lock_revoke_bars(vdev); + else down_write(&vdev->memory_lock); - }
/* * If the user is writing mem/io enable (new_mem/io) and we @@ -631,7 +629,7 @@ static int vfio_basic_config_write(struct vfio_pci_core_device *vdev, int pos, *virt_cmd |= cpu_to_le16(new_cmd & mask);
if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); }
@@ -712,16 +710,14 @@ static int __init init_pci_cap_basic_perm(struct perm_bits *perm) static void vfio_lock_and_set_power_state(struct vfio_pci_core_device *vdev, pci_power_t state) { - if (state >= PCI_D3hot) { - vfio_pci_zap_and_down_write_memory_lock(vdev); - vfio_pci_dma_buf_move(vdev, true); - } else { + if (state >= PCI_D3hot) + vfio_pci_lock_revoke_bars(vdev); + else down_write(&vdev->memory_lock); - }
vfio_pci_set_power_state(vdev, state); if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); }
@@ -908,11 +904,10 @@ static int vfio_exp_config_write(struct vfio_pci_core_device *vdev, int pos, &cap);
if (!ret && (cap & PCI_EXP_DEVCAP_FLR)) { - vfio_pci_zap_and_down_write_memory_lock(vdev); - vfio_pci_dma_buf_move(vdev, true); + vfio_pci_lock_revoke_bars(vdev); pci_try_reset_function(vdev->pdev); if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); } } @@ -993,11 +988,10 @@ static int vfio_af_config_write(struct vfio_pci_core_device *vdev, int pos, &cap);
if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP)) { - vfio_pci_zap_and_down_write_memory_lock(vdev); - vfio_pci_dma_buf_move(vdev, true); + vfio_pci_lock_revoke_bars(vdev); pci_try_reset_function(vdev->pdev); if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); } } diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c index adce779c722c..74ba52cf647c 100644 --- a/drivers/vfio/pci/vfio_pci_core.c +++ b/drivers/vfio/pci/vfio_pci_core.c @@ -375,8 +375,7 @@ static int vfio_pci_runtime_pm_entry(struct vfio_pci_core_device *vdev, * The vdev power related flags are protected with 'memory_lock' * semaphore. */ - vfio_pci_zap_and_down_write_memory_lock(vdev); - vfio_pci_dma_buf_move(vdev, true); + vfio_pci_lock_revoke_bars(vdev);
if (vdev->pm_runtime_engaged) { up_write(&vdev->memory_lock); @@ -462,7 +461,7 @@ static void vfio_pci_runtime_pm_exit(struct vfio_pci_core_device *vdev) down_write(&vdev->memory_lock); __vfio_pci_runtime_pm_exit(vdev); if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); }
@@ -1312,6 +1311,8 @@ static int vfio_pci_ioctl_set_irqs(struct vfio_pci_core_device *vdev, return ret; }
+static void vfio_pci_revoke_bars(struct vfio_pci_core_device *vdev); + static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev, void __user *arg) { @@ -1320,7 +1321,7 @@ static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev, if (!vdev->reset_works) return -EINVAL;
- vfio_pci_zap_and_down_write_memory_lock(vdev); + down_write(&vdev->memory_lock);
/* * This function can be invoked while the power state is non-D0. If @@ -1330,13 +1331,18 @@ static int vfio_pci_ioctl_reset(struct vfio_pci_core_device *vdev, * have NoSoftRst-, the reset function can cause the PCI config space * reset without restoring the original state (saved locally in * 'vdev->pm_save'). + * + * The zap is done after making the device accessible in D0, + * because a DMABUF importer could access the device as part + * of its revocation cleanup. */ vfio_pci_set_power_state(vdev, PCI_D0);
- vfio_pci_dma_buf_move(vdev, true); + vfio_pci_revoke_bars(vdev); + ret = pci_try_reset_function(vdev->pdev); if (__vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock);
return ret; @@ -1704,20 +1710,37 @@ ssize_t vfio_pci_core_write(struct vfio_device *core_vdev, const char __user *bu } EXPORT_SYMBOL_GPL(vfio_pci_core_write);
-static void vfio_pci_zap_bars(struct vfio_pci_core_device *vdev) +static void vfio_pci_revoke_bars(struct vfio_pci_core_device *vdev) { - struct vfio_device *core_vdev = &vdev->vdev; - loff_t start = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_BAR0_REGION_INDEX); - loff_t end = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_ROM_REGION_INDEX); - loff_t len = end - start; + lockdep_assert_held_write(&vdev->memory_lock); + vfio_pci_dma_buf_move(vdev, true);
- unmap_mapping_range(core_vdev->inode->i_mapping, start, len, true); + /* + * If a driver could possibly create BAR mappings in the + * vdev's address_space, do an additional zap on revoke. See + * vfio_pci_core_register_device(). + */ + if (vdev->zap_bars_on_revoke) { + struct vfio_device *core_vdev = &vdev->vdev; + loff_t start = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_BAR0_REGION_INDEX); + loff_t end = VFIO_PCI_INDEX_TO_OFFSET(VFIO_PCI_ROM_REGION_INDEX); + loff_t len = end - start; + + unmap_mapping_range(core_vdev->inode->i_mapping, + start, len, true); + } }
-void vfio_pci_zap_and_down_write_memory_lock(struct vfio_pci_core_device *vdev) +void vfio_pci_lock_revoke_bars(struct vfio_pci_core_device *vdev) { down_write(&vdev->memory_lock); - vfio_pci_zap_bars(vdev); + vfio_pci_revoke_bars(vdev); +} + +void vfio_pci_unrevoke_bars(struct vfio_pci_core_device *vdev) +{ + lockdep_assert_held_write(&vdev->memory_lock); + vfio_pci_dma_buf_move(vdev, false); }
u16 vfio_pci_memory_lock_and_enable(struct vfio_pci_core_device *vdev) @@ -2290,6 +2313,17 @@ int vfio_pci_core_register_device(struct vfio_pci_core_device *vdev) if (ret) goto out_vf;
+ /* + * If a driver overrides .mmap, it has to be assumed that it + * might not use the DMABUF-backed core mmap; this flag + * enables a zap at revoke time. Drivers can opt out by + * clearing this flag after registration, e.g. if their .mmap + * override calls down to vfio_pci_core_mmap() so that maps + * are DMABUF-backed. + */ + if (vdev->vdev.ops->mmap != vfio_pci_core_mmap) + vdev->zap_bars_on_revoke = true; + vfio_pci_probe_power_state(vdev);
/* @@ -2577,9 +2611,10 @@ static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set, }
/* - * Take the memory write lock for each device and zap BAR - * mappings to prevent the user accessing the device while in - * reset. Locking multiple devices is prone to deadlock, + * Take the memory write lock for each device and + * zap/revoke BAR mappings to prevent the user (or + * peers) accessing the device while in reset. + * Locking multiple devices is prone to deadlock, * runaway and unwind if we hit contention. */ if (!down_write_trylock(&vdev->memory_lock)) { @@ -2587,8 +2622,7 @@ static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set, break; }
- vfio_pci_dma_buf_move(vdev, true); - vfio_pci_zap_bars(vdev); + vfio_pci_revoke_bars(vdev); }
if (!list_entry_is_head(vdev, @@ -2618,7 +2652,7 @@ static int vfio_pci_dev_set_hot_reset(struct vfio_device_set *dev_set, list_for_each_entry_from_reverse(vdev, &dev_set->device_list, vdev.dev_set_list) { if (vdev->vdev.open_count && __vfio_pci_memory_enabled(vdev)) - vfio_pci_dma_buf_move(vdev, false); + vfio_pci_unrevoke_bars(vdev); up_write(&vdev->memory_lock); }
diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index d38e1b98b2e9..a8671d63b9ca 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -83,7 +83,8 @@ void vfio_config_free(struct vfio_pci_core_device *vdev); int vfio_pci_set_power_state(struct vfio_pci_core_device *vdev, pci_power_t state);
-void vfio_pci_zap_and_down_write_memory_lock(struct vfio_pci_core_device *vdev); +void vfio_pci_lock_revoke_bars(struct vfio_pci_core_device *vdev); +void vfio_pci_unrevoke_bars(struct vfio_pci_core_device *vdev); u16 vfio_pci_memory_lock_and_enable(struct vfio_pci_core_device *vdev); void vfio_pci_memory_unlock_and_restore(struct vfio_pci_core_device *vdev, u16 cmd); diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h index 9a1674c152aa..e2b4252e7c3f 100644 --- a/include/linux/vfio_pci_core.h +++ b/include/linux/vfio_pci_core.h @@ -134,6 +134,7 @@ struct vfio_pci_core_device { bool pm_intx_masked; bool pm_runtime_engaged; bool sriov_active; + bool zap_bars_on_revoke; struct pci_saved_state *pci_saved_state; struct pci_saved_state *pm_save; int ioeventfds_nr;
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:48 AM
Previously, vfio_pci_zap_bars() (and the wrapper vfio_pci_zap_and_down_write_memory_lock()) calls were paired with calls to vfio_pci_dma_buf_move().
This commit replaces them with a unified new function, vfio_pci_revoke_bars() containing both the vfio_pci_dma_buf_move() and the unmap_mapping_range(), making it harder for callers to omit one. It adds a wrapper, vfio_pci_lock_revoke_bars(), which takes the write memory_lock before zapping, and adds a new vfio_pci_unrevoke_bars() for the re-enable path.
As of "vfio/pci: Convert BAR mmap() to use a DMABUF", the zap via unmap_mapping_range() is no longer performed for vfio-pci since the DMABUFs used for BAR mappings already zap PTEs when the vfio_pci_dma_buf_move() occurs.
However, it must be assumed that VFIO drivers which override the .mmap op could create mappings _not_ backed by DMABUFs. So, the zap is still performed on revoke if .mmap is overridden, using a new zap_bars_on_revoke flag. A driver can explicitly opt out; the flag is cleared by the hisi_acc_vfio_pci driver, since its .mmap just wraps vfio_pci_core_mmap() and so still uses DMABUFs.
Signed-off-by: Matt Evans matt@ozlabs.org
Reviewed-by: Kevin Tian kevin.tian@intel.com
On Wed, 15 Jul 2026 18:47:30 +0100 Matt Evans matt@ozlabs.org wrote:
diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h index 9a1674c152aa..e2b4252e7c3f 100644 --- a/include/linux/vfio_pci_core.h +++ b/include/linux/vfio_pci_core.h @@ -134,6 +134,7 @@ struct vfio_pci_core_device { bool pm_intx_masked; bool pm_runtime_engaged; bool sriov_active;
- bool zap_bars_on_revoke; struct pci_saved_state *pci_saved_state; struct pci_saved_state *pm_save; int ioeventfds_nr;
This should be in the bitfield usage group since it's only modified at init time. Thanks,
Alex
Hi Alex,
On 29/07/2026 18:52, Alex Williamson wrote:
On Wed, 15 Jul 2026 18:47:30 +0100 Matt Evans matt@ozlabs.org wrote:
diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h index 9a1674c152aa..e2b4252e7c3f 100644 --- a/include/linux/vfio_pci_core.h +++ b/include/linux/vfio_pci_core.h @@ -134,6 +134,7 @@ struct vfio_pci_core_device { bool pm_intx_masked; bool pm_runtime_engaged; bool sriov_active;
- bool zap_bars_on_revoke; struct pci_saved_state *pci_saved_state; struct pci_saved_state *pm_save; int ioeventfds_nr;
This should be in the bitfield usage group since it's only modified at init time.
This was intentional, but happy to change it if you're certain ofc. Is it inconceivable that a sub-driver could set it after init? I'd say they _shouldn't_, but only review will stop them and this placement intended to be cautious. It seemed a low cost way to avoid issues around synchronisation on the bitfield.
Thanks,
Matt
On Thu, 30 Jul 2026 15:47:13 +0100 Matt Evans matt@ozlabs.org wrote:
Hi Alex,
On 29/07/2026 18:52, Alex Williamson wrote:
On Wed, 15 Jul 2026 18:47:30 +0100 Matt Evans matt@ozlabs.org wrote:
diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h index 9a1674c152aa..e2b4252e7c3f 100644 --- a/include/linux/vfio_pci_core.h +++ b/include/linux/vfio_pci_core.h @@ -134,6 +134,7 @@ struct vfio_pci_core_device { bool pm_intx_masked; bool pm_runtime_engaged; bool sriov_active;
- bool zap_bars_on_revoke; struct pci_saved_state *pci_saved_state; struct pci_saved_state *pm_save; int ioeventfds_nr;
This should be in the bitfield usage group since it's only modified at init time.
This was intentional, but happy to change it if you're certain ofc. Is it inconceivable that a sub-driver could set it after init? I'd say they _shouldn't_, but only review will stop them and this placement intended to be cautious. It seemed a low cost way to avoid issues around synchronisation on the bitfield.
I'd agree with the statement that they shouldn't, it would be difficult to synchronize setting the flag once there are any active mappings of the BARs. Also, if we put it in the bitfield category under the comment that the value is only modified at setup/release, it documents the intentions, hopefully to the extent the author or reviewers notice. An argument can always be made to change it if there's a worthwhile use case. Thanks,
Alex
A VFIO DMABUF can export a subset of a BAR to userspace by fd; add support for mmap() of this fd. This provides another route for a process to map BARs, in which the process can only map a specific subset of a BAR represented by the exported DMABUF.
mmap() support enables userspace driver designs that safely delegate access to BAR sub-ranges to other client processes by sharing a DMABUF fd, without having to share the (omnipotent) VFIO device fd with them.
Since the main VFIO BAR mmap() is now DMABUF-aware, the new mmap() reuses the existing vm_ops.
The lifecycle of an exported DMABUF remains decoupled from that of the device fd it came from, i.e. the device fd could be closed with DMABUF VMAs present, meaning a fault on a VMA could happen concurrently with vfio_pci_dma_buf_cleanup().
To deal with this scenario, the fault handler now temporarily takes a VFIO device registration to ensure the vdev remains valid, and then vdev->memory_lock can be taken on it.
Signed-off-by: Matt Evans matt@ozlabs.org --- drivers/vfio/pci/vfio_pci_core.c | 83 +++++++++++++++++++++++++++--- drivers/vfio/pci/vfio_pci_dmabuf.c | 34 ++++++++++++ drivers/vfio/pci/vfio_pci_priv.h | 1 + 3 files changed, 110 insertions(+), 8 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c index 74ba52cf647c..ef93fbd5f1b6 100644 --- a/drivers/vfio/pci/vfio_pci_core.c +++ b/drivers/vfio/pci/vfio_pci_core.c @@ -13,6 +13,8 @@ #include <linux/aperture.h> #include <linux/debugfs.h> #include <linux/device.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h> #include <linux/eventfd.h> #include <linux/file.h> #include <linux/interrupt.h> @@ -1795,19 +1797,78 @@ static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf, vm_fault_t ret = VM_FAULT_SIGBUS;
/* - * We can rely on the existence of both a DMABUF (priv) and - * the VFIO device it was exported from (vdev). This fault's - * VMA was established using vfio_pci_core_mmap_prep_dmabuf() - * which transfers ownership of the VFIO device fd to the - * DMABUF, and so the VFIO device is held open because the - * VMA's vm_file (DMABUF) is open. + * The only thing this can rely on is that the DMABUF relating + * to the VMA's vm_file exists (priv). * - * Since vfio_pci_dma_buf_cleanup() cannot have happened, - * vdev must be valid; we can take memory_lock. + * A DMABUF for a VFIO device fd mmap() holds a reference to + * the original VFIO device fd, but an explicitly-exported + * DMABUF does not. The original fd might have closed, + * meaning this fault can race with + * vfio_pci_dma_buf_cleanup(), meaning the buffer could have + * been revoked (in which case priv->vdev might be NULL), and + * the VFIO device registration might have been dropped. + * + * With the goal of taking vdev->memory_lock in a world where + * vdev might not still exist: + * + * 1. Take the resv lock on the DMABUF: + * - If racing cleanup got in first, the buffer is revoked; + * stop/exit if so. + * - If we got in first, the buffer is not revoked so vdev is + * non-NULL, accessible, and cleanup _has not yet put the + * VFIO device registration_. So, the device refcount must + * be >0. + * + * 2. Take vfio_device registration (refcount guaranteed >0 + * hereafter). + * + * 3. Unlock the DMABUF's resv lock: + * - A racing cleanup can now complete. + * - But, the device refcount >0, meaning the vfio_device + * (and vfio_pcie_core device vdev) have not yet been + * freed. vdev is accessible, even if the DMABUF has been + * revoked or cleanup has happened, because + * vfio_unregister_group_dev() can't complete. + * + * 4. Take the vdev->memory_lock + * - Either the DMABUF is usable, or has been cleaned up. + * Whichever, it can no longer change under us. + * - Test the DMABUF revocation status again: if it was + * revoked between 1 and 4 return a SIGBUS. Otherwise, + * return a PFN. + * - It's not necessary to also take the resv lock, because + * the status/vdev can't change while memory_lock is held. + * + * 5. Unlock, done. */ + + dma_resv_lock(priv->dmabuf->resv, NULL); + + if (priv->revoked) { + pr_debug_ratelimited("%s VA 0x%lx, pgoff 0x%lx: DMABUF revoked/cleaned up\n", + __func__, vmf->address, vma->vm_pgoff); + dma_resv_unlock(priv->dmabuf->resv); + return VM_FAULT_SIGBUS; + } + + /* If the buffer isn't revoked, vdev is valid */ vdev = priv->vdev;
+ if (!vfio_device_try_get_registration(&vdev->vdev)) { + /* + * If vdev != NULL (above), the registration should + * already be >0 and so this try_get should never + * fail. + */ + dev_warn(&vdev->pdev->dev, "%s: Unexpected registration failure\n", + __func__); + dma_resv_unlock(priv->dmabuf->resv); + return VM_FAULT_SIGBUS; + } + dma_resv_unlock(priv->dmabuf->resv); + scoped_guard(rwsem_read, &vdev->memory_lock) { + /* Revocation status must be re-read, under memory_lock */ if (!priv->revoked) { int pres = vfio_pci_dma_buf_find_pfn(priv, vma, vmf->address, @@ -1826,6 +1887,7 @@ static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf, __func__, order, pfn, vmf->address, vma->vm_pgoff, (unsigned int)ret);
+ vfio_device_put_registration(&vdev->vdev); return ret; }
@@ -1841,6 +1903,11 @@ static const struct vm_operations_struct vfio_pci_mmap_ops = { #endif };
+void vfio_pci_set_vma_ops(struct vm_area_struct *vma) +{ + vma->vm_ops = &vfio_pci_mmap_ops; +} + int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma) { struct vfio_pci_core_device *vdev = diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index 51606ff3cdc6..ccfe861f88b6 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -27,6 +27,39 @@ static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
return 0; } + +static int vfio_pci_dma_buf_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma) +{ + struct vfio_pci_dma_buf *priv = dmabuf->priv; + + /* + * dma_buf_mmap_internal() has asserted that the VMA is + * contained within the DMABUF size before calling this. + * + * Also, if we observe that the buffer is revoked now then + * refuse the mmap(). This is a belt-and-braces early failure + * to ease debugging a revoked buffer being used. Userspace + * might also race an mmap() against an explicit revocation, + * or an action doing a temporary revoke; race scenarios are + * still safe because the fault handler ultimately prevents + * access to a revoked buffer if it isn't caught here. + */ + if (READ_ONCE(priv->revoked)) + return -ENODEV; + if ((vma->vm_flags & VM_SHARED) == 0) + return -EINVAL; + + vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); + vma->vm_page_prot = pgprot_decrypted(vma->vm_page_prot); + + /* See comments in vfio_pci_core_mmap() re VM_ALLOW_ANY_UNCACHED. */ + vm_flags_set(vma, VM_ALLOW_ANY_UNCACHED | VM_IO | VM_PFNMAP | + VM_DONTEXPAND | VM_DONTDUMP); + vma->vm_private_data = priv; + vfio_pci_set_vma_ops(vma); + + return 0; +} #endif /* CONFIG_VFIO_PCI_DMABUF */
static void vfio_pci_dma_buf_done(struct kref *kref) @@ -94,6 +127,7 @@ static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf) static const struct dma_buf_ops vfio_pci_dmabuf_ops = { #ifdef CONFIG_VFIO_PCI_DMABUF .attach = vfio_pci_dma_buf_attach, + .mmap = vfio_pci_dma_buf_mmap, #endif .map_dma_buf = vfio_pci_dma_buf_map, .unmap_dma_buf = vfio_pci_dma_buf_unmap, diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index a8671d63b9ca..c3fa35381679 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -141,6 +141,7 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, unsigned int res_index); void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev); void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked); +void vfio_pci_set_vma_ops(struct vm_area_struct *vma);
#ifdef CONFIG_VFIO_PCI_DMABUF int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags,
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:48 AM
A VFIO DMABUF can export a subset of a BAR to userspace by fd; add support for mmap() of this fd. This provides another route for a process to map BARs, in which the process can only map a specific subset of a BAR represented by the exported DMABUF.
mmap() support enables userspace driver designs that safely delegate access to BAR sub-ranges to other client processes by sharing a DMABUF fd, without having to share the (omnipotent) VFIO device fd with them.
Since the main VFIO BAR mmap() is now DMABUF-aware, the new mmap() reuses the existing vm_ops.
The lifecycle of an exported DMABUF remains decoupled from that of the device fd it came from, i.e. the device fd could be closed with DMABUF VMAs present, meaning a fault on a VMA could happen concurrently with vfio_pci_dma_buf_cleanup().
To deal with this scenario, the fault handler now temporarily takes a VFIO device registration to ensure the vdev remains valid, and then vdev->memory_lock can be taken on it.
Signed-off-by: Matt Evans matt@ozlabs.org
Reviewed-by: Kevin Tian kevin.tian@intel.com
Expand the VFIO DMABUF revocation state to three states: Not revoked, temporarily revoked, and permanently revoked.
The first two are for existing transient revocation, e.g. across a function reset, and the DMABUF is put into the last in response to a new VFIO feature VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE.
VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE passes a DMABUF by fd and requests that the DMABUF is permanently revoked. On success, it's guaranteed that the buffer can never be imported/attached/mmap()ed in future, that dynamic imports have been cleanly detached, and that all mappings have been made inaccessible/PTEs zapped.
This is useful for lifecycle management, to reclaim VFIO PCI BAR ranges previously delegated to a subordinate client process: by revoking, the driver process can ensure that the loaned resources are made inaccessible when the client is deemed "done". The original DMABUF is defunct, and BAR resources can then be safely re-exported for use by new clients.
Refactor the revocation code out of vfio_pci_dma_buf_move() to a function common to move and the new feature request path. Note: this now only calls dma_buf_invalidate_mappings()/dma_resv_wait_timeout() on the revoke path, whereas vfio_pci_dma_buf_move() originally called them for both revoke and (unnecessarily) un-revoke.
Signed-off-by: Matt Evans matt@ozlabs.org --- drivers/vfio/pci/vfio_pci_core.c | 6 +- drivers/vfio/pci/vfio_pci_dmabuf.c | 166 ++++++++++++++++++++++------- drivers/vfio/pci/vfio_pci_priv.h | 19 +++- include/uapi/linux/vfio.h | 25 +++++ 4 files changed, 174 insertions(+), 42 deletions(-)
diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c index ef93fbd5f1b6..0fa855ba537e 100644 --- a/drivers/vfio/pci/vfio_pci_core.c +++ b/drivers/vfio/pci/vfio_pci_core.c @@ -1633,6 +1633,8 @@ int vfio_pci_core_ioctl_feature(struct vfio_device *device, u32 flags, return vfio_pci_core_feature_token(vdev, flags, arg, argsz); case VFIO_DEVICE_FEATURE_DMA_BUF: return vfio_pci_core_feature_dma_buf(vdev, flags, arg, argsz); + case VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE: + return vfio_pci_core_feature_dma_buf_revoke(vdev, flags, arg, argsz); default: return -ENOTTY; } @@ -1844,7 +1846,7 @@ static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf,
dma_resv_lock(priv->dmabuf->resv, NULL);
- if (priv->revoked) { + if (priv->status != VFIO_PCI_DMABUF_OK) { pr_debug_ratelimited("%s VA 0x%lx, pgoff 0x%lx: DMABUF revoked/cleaned up\n", __func__, vmf->address, vma->vm_pgoff); dma_resv_unlock(priv->dmabuf->resv); @@ -1869,7 +1871,7 @@ static vm_fault_t vfio_pci_mmap_huge_fault(struct vm_fault *vmf,
scoped_guard(rwsem_read, &vdev->memory_lock) { /* Revocation status must be re-read, under memory_lock */ - if (!priv->revoked) { + if (priv->status == VFIO_PCI_DMABUF_OK) { int pres = vfio_pci_dma_buf_find_pfn(priv, vma, vmf->address, order, &pfn); diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c index ccfe861f88b6..36bf07530840 100644 --- a/drivers/vfio/pci/vfio_pci_dmabuf.c +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c @@ -19,7 +19,7 @@ static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf, if (!attachment->peer2peer) return -EOPNOTSUPP;
- if (priv->revoked) + if (priv->status != VFIO_PCI_DMABUF_OK) return -ENODEV;
if (!dma_buf_attach_revocable(attachment)) @@ -44,7 +44,7 @@ static int vfio_pci_dma_buf_mmap(struct dma_buf *dmabuf, struct vm_area_struct * * still safe because the fault handler ultimately prevents * access to a revoked buffer if it isn't caught here. */ - if (READ_ONCE(priv->revoked)) + if (READ_ONCE(priv->status) != VFIO_PCI_DMABUF_OK) return -ENODEV; if ((vma->vm_flags & VM_SHARED) == 0) return -EINVAL; @@ -79,7 +79,7 @@ vfio_pci_dma_buf_map(struct dma_buf_attachment *attachment,
dma_resv_assert_held(priv->dmabuf->resv);
- if (priv->revoked) + if (priv->status != VFIO_PCI_DMABUF_OK) return ERR_PTR(-ENODEV);
ret = dma_buf_phys_vec_to_sgt(attachment, priv->provider, @@ -305,7 +305,8 @@ static int vfio_pci_dmabuf_export(struct vfio_pci_core_device *vdev, INIT_LIST_HEAD(&priv->dmabufs_elm); down_write(&vdev->memory_lock); dma_resv_lock(priv->dmabuf->resv, NULL); - priv->revoked = !__vfio_pci_memory_enabled(vdev); + priv->status = __vfio_pci_memory_enabled(vdev) ? VFIO_PCI_DMABUF_OK : + VFIO_PCI_DMABUF_TEMP_REVOKED; list_add_tail(&priv->dmabufs_elm, &vdev->dmabufs); dma_resv_unlock(priv->dmabuf->resv); up_write(&vdev->memory_lock); @@ -336,7 +337,7 @@ int vfio_pci_dma_buf_iommufd_map(struct dma_buf_attachment *attachment, return -EOPNOTSUPP;
priv = attachment->dmabuf->priv; - if (priv->revoked) + if (priv->status != VFIO_PCI_DMABUF_OK) return -ENODEV;
/* More than one range to iommufd will require proper DMABUF support */ @@ -600,6 +601,59 @@ int vfio_pci_core_mmap_prep_dmabuf(struct vfio_pci_core_device *vdev, return ret; }
+/* Set the DMABUF's revocation status (OK or temporarily/permanently revoked) */ +static void vfio_pci_dma_buf_set_status(struct vfio_pci_dma_buf *priv, + enum vfio_pci_dma_buf_status new_status) +{ + bool was_revoked; + + lockdep_assert_held_write(&priv->vdev->memory_lock); + + if (priv->status == VFIO_PCI_DMABUF_PERM_REVOKED || + priv->status == new_status) + return; + + dma_resv_lock(priv->dmabuf->resv, NULL); + was_revoked = (priv->status == VFIO_PCI_DMABUF_TEMP_REVOKED); + + if (new_status != VFIO_PCI_DMABUF_OK) { + priv->status = new_status; /* Temp or permanently revoked */ + + if (was_revoked) { + /* + * TEMP_REVOKED is being upgraded to + * PERM_REVOKED. The buffer is already gone, + * don't wait on it again. + */ + dma_resv_unlock(priv->dmabuf->resv); + return; + } + dma_buf_invalidate_mappings(priv->dmabuf); + dma_resv_wait_timeout(priv->dmabuf->resv, + DMA_RESV_USAGE_BOOKKEEP, false, + MAX_SCHEDULE_TIMEOUT); + dma_resv_unlock(priv->dmabuf->resv); + kref_put(&priv->kref, vfio_pci_dma_buf_done); + wait_for_completion(&priv->comp); + unmap_mapping_range(priv->dmabuf->file->f_mapping, + 0, 0, true); + /* + * Re-arm the registered kref reference and the + * completion so the post-revoke state matches the + * post-creation state. An un-revoke followed by a + * new mapping needs the kref to be non-zero before + * kref_get(), and vfio_pci_dma_buf_cleanup() + * delegates its drain back through this revoke + * path on a possibly-already-revoked dma-buf. + */ + kref_init(&priv->kref); + reinit_completion(&priv->comp); + } else { + priv->status = VFIO_PCI_DMABUF_OK; + dma_resv_unlock(priv->dmabuf->resv); + } +} + void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked) { struct vfio_pci_dma_buf *priv; @@ -607,45 +661,16 @@ void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
/* * Holding memory_lock ensures a racing VMA fault observes - * priv->revoked properly. + * priv->status properly. */ lockdep_assert_held_write(&vdev->memory_lock);
list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) { if (!get_file_active(&priv->dmabuf->file)) continue; - - if (priv->revoked != revoked) { - dma_resv_lock(priv->dmabuf->resv, NULL); - if (revoked) - priv->revoked = true; - dma_buf_invalidate_mappings(priv->dmabuf); - dma_resv_wait_timeout(priv->dmabuf->resv, - DMA_RESV_USAGE_BOOKKEEP, false, - MAX_SCHEDULE_TIMEOUT); - dma_resv_unlock(priv->dmabuf->resv); - if (revoked) { - kref_put(&priv->kref, vfio_pci_dma_buf_done); - wait_for_completion(&priv->comp); - unmap_mapping_range(priv->dmabuf->file->f_mapping, - 0, 0, true); - /* - * Re-arm the registered kref reference and the - * completion so the post-revoke state matches the - * post-creation state. An un-revoke followed by a - * new mapping needs the kref to be non-zero before - * kref_get(), and vfio_pci_dma_buf_cleanup() - * delegates its drain back through this revoke - * path on a possibly-already-revoked dma-buf. - */ - kref_init(&priv->kref); - reinit_completion(&priv->comp); - } else { - dma_resv_lock(priv->dmabuf->resv, NULL); - priv->revoked = false; - dma_resv_unlock(priv->dmabuf->resv); - } - } + vfio_pci_dma_buf_set_status(priv, revoked ? + VFIO_PCI_DMABUF_TEMP_REVOKED : + VFIO_PCI_DMABUF_OK); fput(priv->dmabuf->file); } } @@ -671,9 +696,72 @@ void vfio_pci_dma_buf_cleanup(struct vfio_pci_core_device *vdev) continue;
list_del_init(&priv->dmabufs_elm); - priv->vdev = NULL; + WRITE_ONCE(priv->vdev, NULL); vfio_device_put_registration(&vdev->vdev); fput(priv->dmabuf->file); } up_write(&vdev->memory_lock); } + +#ifdef CONFIG_VFIO_PCI_DMABUF +int vfio_pci_core_feature_dma_buf_revoke( + struct vfio_pci_core_device *vdev, u32 flags, + struct vfio_device_feature_dma_buf_revoke __user *arg, + size_t argsz) +{ + struct vfio_device_feature_dma_buf_revoke db_revoke; + struct vfio_pci_dma_buf *priv; + struct dma_buf *dmabuf; + int ret; + + if (!vdev->pci_ops || !vdev->pci_ops->get_dmabuf_phys) + return -EOPNOTSUPP; + + ret = vfio_check_feature(flags, argsz, + VFIO_DEVICE_FEATURE_SET, + sizeof(db_revoke)); + if (ret != 1) + return ret; + + if (copy_from_user(&db_revoke, arg, sizeof(db_revoke))) + return -EFAULT; + + dmabuf = dma_buf_get(db_revoke.dmabuf_fd); + if (IS_ERR(dmabuf)) + return PTR_ERR(dmabuf); + + priv = dmabuf->priv; + /* + * Sanity-check the DMABUF is really a vfio_pci_dma_buf _and_ + * relates to the VFIO device it was provided with. + * + * If the DMABUF relates to this vdev then priv->vdev is + * stable because this open fd prevents cleanup. + * + * If it relates to a different vdev, reading priv->vdev might + * race with a concurrent cleanup on that device. But if so, + * it points to a non-matching vdev or NULL and is unusable + * either way. + */ + if (dmabuf->ops != &vfio_pci_dmabuf_ops || + READ_ONCE(priv->vdev) != vdev) { + ret = -ENODEV; + goto out_put_buf; + } + + scoped_guard(rwsem_write, &vdev->memory_lock) { + if (priv->status == VFIO_PCI_DMABUF_PERM_REVOKED) { + ret = -EBADFD; + } else { + vfio_pci_dma_buf_set_status(priv, + VFIO_PCI_DMABUF_PERM_REVOKED); + ret = 0; + } + } + +out_put_buf: + dma_buf_put(dmabuf); + + return ret; +} +#endif /* CONFIG_VFIO_PCI_DMABUF */ diff --git a/drivers/vfio/pci/vfio_pci_priv.h b/drivers/vfio/pci/vfio_pci_priv.h index c3fa35381679..8741abd04461 100644 --- a/drivers/vfio/pci/vfio_pci_priv.h +++ b/drivers/vfio/pci/vfio_pci_priv.h @@ -23,6 +23,12 @@ struct vfio_pci_ioeventfd { bool test_mem; };
+enum vfio_pci_dma_buf_status { + VFIO_PCI_DMABUF_OK = 0, + VFIO_PCI_DMABUF_TEMP_REVOKED = 1, + VFIO_PCI_DMABUF_PERM_REVOKED = 2, +}; + struct vfio_pci_dma_buf { struct dma_buf *dmabuf; struct vfio_pci_core_device *vdev; @@ -35,7 +41,7 @@ struct vfio_pci_dma_buf { struct kref kref; struct completion comp; unsigned long vma_pgoff_adjust; - u8 revoked : 1; + enum vfio_pci_dma_buf_status status; };
bool vfio_pci_intx_mask(struct vfio_pci_core_device *vdev); @@ -147,6 +153,10 @@ void vfio_pci_set_vma_ops(struct vm_area_struct *vma); int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, struct vfio_device_feature_dma_buf __user *arg, size_t argsz); +int vfio_pci_core_feature_dma_buf_revoke( + struct vfio_pci_core_device *vdev, u32 flags, + struct vfio_device_feature_dma_buf_revoke __user *arg, + size_t argsz); #else static inline int vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, @@ -155,6 +165,13 @@ vfio_pci_core_feature_dma_buf(struct vfio_pci_core_device *vdev, u32 flags, { return -ENOTTY; } +static inline int vfio_pci_core_feature_dma_buf_revoke( + struct vfio_pci_core_device *vdev, u32 flags, + struct vfio_device_feature_dma_buf_revoke __user *arg, + size_t argsz) +{ + return -ENOTTY; +} #endif
#endif diff --git a/include/uapi/linux/vfio.h b/include/uapi/linux/vfio.h index 5de618a3a5ee..cbe03e5d1009 100644 --- a/include/uapi/linux/vfio.h +++ b/include/uapi/linux/vfio.h @@ -1534,6 +1534,31 @@ struct vfio_device_feature_dma_buf { */ #define VFIO_DEVICE_FEATURE_MIG_PRECOPY_INFOv2 12
+/** + * Given a dma_buf fd previously exported from the same device by + * VFIO_DEVICE_FEATURE_DMA_BUF, a SET of this feature requests that + * access to the corresponding DMABUF is immediately and permanently + * revoked. On successful return, the buffer is not accessible + * through any mmap() or dma-buf import. The buffer is permanently + * disabled, and VFIO refuses all map, mmap, attach, etc. requests. + * + * Return: 0 on success, -1 and errno is set on failure: + * + * EBADF, EINVAL: dmabuf_fd is not a DMABUF fd. + * EOPNOTSUPP: The VFIO device does not support DMABUF export. + * ENODEV: The device supports export but dmabuf_fd was exported + * from a different device. + * EBADFD: The DMABUF is already revoked. + * + * Additionally, common errors can occur: EFAULT accessing the struct, + * or EINVAL requesting an unsupported feature op. + */ +#define VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE 13 + +struct vfio_device_feature_dma_buf_revoke { + __s32 dmabuf_fd; +}; + /* -------- API for Type1 VFIO IOMMU -------- */
/**
From: Matt Evans matt@ozlabs.org Sent: Thursday, July 16, 2026 1:48 AM
Expand the VFIO DMABUF revocation state to three states: Not revoked, temporarily revoked, and permanently revoked.
The first two are for existing transient revocation, e.g. across a function reset, and the DMABUF is put into the last in response to a new VFIO feature VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE.
VFIO_DEVICE_FEATURE_DMA_BUF_REVOKE passes a DMABUF by fd and requests that the DMABUF is permanently revoked. On success, it's guaranteed that the buffer can never be imported/attached/mmap()ed in future, that dynamic imports have been cleanly detached, and that all mappings have been made inaccessible/PTEs zapped.
This is useful for lifecycle management, to reclaim VFIO PCI BAR ranges previously delegated to a subordinate client process: by revoking, the driver process can ensure that the loaned resources are made inaccessible when the client is deemed "done". The original DMABUF is defunct, and BAR resources can then be safely re-exported for use by new clients.
Refactor the revocation code out of vfio_pci_dma_buf_move() to a function common to move and the new feature request path. Note: this now only calls dma_buf_invalidate_mappings()/dma_resv_wait_timeout() on the revoke path, whereas vfio_pci_dma_buf_move() originally called them for both revoke and (unnecessarily) un-revoke.
Signed-off-by: Matt Evans matt@ozlabs.org
Reviewed-by: Kevin Tian kevin.tian@intel.com
On Wed, Jul 15, 2026 at 10:47 AM Matt Evans matt@ozlabs.org wrote:
This is based on v7.2-rc3.
These commits are on GitHub for easier browsing, along with "[RFC ONLY] selftests: vfio: Add standalone vfio_dmabuf_mmap_test":
https://github.com/metamev/linux/compare/v7.2-rc3...dev/mev/vfio-dmabuf-mmap...
It'd be great to have this test upstream. I'm happy to review it when you're ready. Looks like it just needs to be redone to use the VFIO selftests library and kselftests harness. AI could probably do the conversion pretty quick :)
Hi David,
On 15/07/2026 19:12, David Matlack wrote:
On Wed, Jul 15, 2026 at 10:47 AM Matt Evans matt@ozlabs.org wrote:
This is based on v7.2-rc3.
These commits are on GitHub for easier browsing, along with "[RFC ONLY] selftests: vfio: Add standalone vfio_dmabuf_mmap_test":
https://github.com/metamev/linux/compare/v7.2-rc3...dev/mev/vfio-dmabuf-mmap...
It'd be great to have this test upstream. I'm happy to review it when you're ready. Looks like it just needs to be redone to use the VFIO selftests library and kselftests harness. AI could probably do the conversion pretty quick :)
For sure, I'd intended to catch up with you on best approach here. :)
Aside from the organic structure of the test (the open-coded VFIO device/group setup/init needs to go), the main issue is that it relies on a hacked/out of tree QEMU "EDU++" device with a second larger BAR (containing freely read-writable memory). A subset of tests run with the in-tree EDU device, but coverage is too low.
The desirable properties are:
- Having a BAR that is pure memory (all locations present, writable without disruptive side-effects) so that mapping aliases can be constructed and detected. This is good to test things like non-zero vm_pgoffs and VA space presentation of physically-discontiguous DMABUFs.
- BAR >> hugepage size so we can eyeball huge mappings work (or better, mechanically test for them). At least 32MB would tick this box for 4K, 16K page systems.
- Something QEMU supports*, so one can run the test in a VM/TCG system.
There were some real device models in QEMU that could be used this way, but needed a fair bit of setup; I didn't want to rathole vfio_dmabuf_mmap_test on including a ton of device-specific code for some video card or similar.
I'll dig more for a simple target that provides these properties -- obviously it would be better to point this test at an off-the-shelf device (including silicon!). And, proposing EDU extensions to the QEMU folks may be useful (there're uses for a better EDU in other contexts too).
Since this test uses MMIO for a specific [class of] function, my first thought is it should be another VFIO driver-type test sibling of vfio_pci_driver_test. For example, we could extend the driver-type tests' backend struct vfio_pci_driver_ops for functions capable of providing a Big Memory BAR, like QEMU EDU++. EDU can also memcpy, so could also support vfio_pci_driver_test.
The spirit of the device backends hiding setup of a complex device is handy, and it's plausible that several backends could provide this "big memory BAR" service. What do you think, any concerns with extending vfio_pci_driver_ops like that?
Thanks,
Matt
linaro-mm-sig@lists.linaro.org
-
Alex Williamson -
David Matlack -
Jason Gunthorpe -
Leon Romanovsky -
Matt Evans -
Tian, Kevin