Hi,
This series tightens the alignment requirements for buffers that are shared between confidential-computing guests and the host, and adds a common allocator for host-shared memory.
When a guest runs with private memory, buffers shared with the hypervisor are not only accessed by the guest. They are also accessed by the host kernel, and the host may manage the corresponding shared/private state at a granularity larger than the guest page size.
This matters for CCA systems where the Realm stage-2 mappings managed by the RMM can still operate at 4K granularity, while the non-secure host may manage the IPA state change at a larger page size, for example 64K. In that case, allowing a guest to convert and share only a 4K subrange of a host-managed granule is unsafe.
Architectures such as Arm can detect incorrect accesses to Realm physical address space PFNs through GPC faults. However, relying on that as the only line of defence is fragile and can still lead to kernel crashes. The risk is especially visible for shared buffers that are later mmapped into userspace, such as guest_memfd or dma-buf backed allocations. Once userspace can access the mapping, the kernel cannot guarantee that applications will only touch the intended 4K region rather than the whole host page mapped into their address space. Those userspace addresses may also be passed back into the kernel and accessed through the linear map, resulting in a GPC fault.
To avoid this, host-shared buffers must satisfy two constraints:
- the address must be aligned to the CoCo shared-granule size - the size must be a multiple of that granule size
The series adds a common CoCo shared-memory layer for enforcing these constraints. It provides shared-granule geometry and range-validation helpers, byte-oriented private/shared transition helpers, and alloc_cc_shared_pages() with a node-aware variant. The allocator rounds a request to the architecture shared granule, allocates suitably aligned contiguous pages, transitions the complete allocation to shared state, and returns the transitioned size alongside the page.
The corresponding free helper restores the complete allocation to private state before returning it to the buddy allocator. If private state cannot be restored safely, the allocation is deliberately leaked rather than returning potentially shared memory for unrelated use. Since a private-to-shared transition may modify memory contents, __GFP_ZERO is applied after the transition.
The generic shared-granule size defaults to PAGE_SIZE. For arm64 CCA, the series queries the host IPA state change alignment through the Realm Host Interface, caches it during Realm initialization, and exposes it through the arm64 memory-encryption operations.
The common allocator is used for host-shared allocations whose backing is owned by an individual caller:
- GIC ITS command queues and tables - dma-direct allocations backed by CMA or the page allocator - backing allocations for the CoCo atomic DMA pools - dma-buf system_cc_shared heap allocations
Hyper-V users of set_memory_encrypted() and set_memory_decrypted() are not changed by this series. Those paths are not currently used by the arm64 CCA code path, and therefore are not part of the arm64 CCA IPA state change alignment problem addressed here.
NOTE: I have not added explicit MAINTAINERS entries for mm/cc_shared.c and include/linux/cc_shared.h, as I am unsure whether we need a separate section for common CoCo-related files. I will add the entries based on feedback.
The series is based on: - https://lore.kernel.org/all/20260921053807.354802-1-aneesh.kumar@kernel.org
Changes from v6: https://lore.kernel.org/all/20260904103452.1197239-1-aneesh.kumar@kernel.org * Add a common allocator and geometry/transition helpers for CoCo host-shared memory. * Convert GIC ITS, dma-direct, atomic DMA pools, and the dma-buf system_cc_shared heap to the common allocator. * Limit dma-buf scatterlist entries to the requested buffer size so rounded backing is not exposed to importers.
Changes from v5: https://lore.kernel.org/all/20260706060432.1375570-1-aneesh.kumar@kernel.org * Rebased to latest kernel * Drop patch arm64: realm: Move Realm memory encryption ops to RSI code
Changes from v4: https://lore.kernel.org/all/20260427063108.909019-1-aneesh.kumar@kernel.org * Rename the helpers to use CoCo terminology (mem_cc_shared_granule_size() / mem_cc_align_to_shared_granule() instead of mem_decrypt_granule_size() / mem_decrypt_align()). * Use __DMA_ATTR_ALLOC_CC_SHARED to pass CoCo shared allocation requirements down to CMA-based allocation helpers. * Add validation for restricted DMA pools to reject pools that are not aligned to the shared granule size. * Add dma-buf system heap handling for cc-shared buffers. * Split the previous combined DMA/SWIOTLB/ITS change into smaller subsystem patches covering ITS, DMA direct, SWIOTLB, restricted DMA pools, dma-buf system heap, and arm64 Realm support. * Rework arm64 Realm support by moving Realm memory encryption ops into RSI code and exposing the CCA shared granule size through arm64_mem_crypt_ops.
Changes from v3: https://lore.kernel.org/all/20260309102625.2315725-1-aneesh.kumar@kernel.org * Fix build error reported by kernel test robot lkp@intel.com
Changes from v2: https://lore.kernel.org/all/20251221160920.297689-1-aneesh.kumar@kernel.org * Rebase to latest kernel * Consider swiotlb always decrypted and don't align when allocating from swiotlb.
Changes from v1: * Rename the helper to mem_encrypt_align * Improve the commit message * Handle DMA allocations from contiguous memory * Handle DMA allocations from the pool * swiotlb is still considered unencrypted. Support for an encrypted swiotlb pool is left as TODO and is independent of this series.
Cc: Andrew Morton akpm@linux-foundation.org Cc: Catalin Marinas catalin.marinas@arm.com Cc: christian.koenig@amd.com Cc: Jason Gunthorpe jgg@ziepe.ca Cc: Joerg Roedel (AMD) joro@8bytes.org Cc: Marc Zyngier maz@kernel.org Cc: Marek Szyprowski m.szyprowski@samsung.com Cc: Robin Murphy robin.murphy@arm.com Cc: Steven Price steven.price@arm.com Cc: Sumit Semwal sumit.semwal@linaro.org Cc: Suzuki K Poulose suzuki.poulose@arm.com Cc: Thomas Gleixner tglx@kernel.org Cc: Will Deacon will@kernel.org Cc: dri-devel@lists.freedesktop.org Cc: iommu@lists.linux.dev Cc: linaro-mm-sig@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: linux-kernel@vger.kernel.org Cc: linux-media@vger.kernel.org Cc: linux-mm@kvack.org
Aneesh Kumar K.V (Arm) (13): arm64: realm: Add RHI helper to query IPA state change alignment mm: Add an allocator for CoCo shared memory arm64: realm: Expose the CCA shared granule size through mem_encrypt ops irqchip/gic-v3-its: Resolve the default NUMA node explicitly irqchip/gic-v3-its: Allocate shared tables using CoCo shared memory allocator dma-contiguous: Accept an explicit minimum alignment dma-pool: Allocate CoCo atomic pools using CoCo shared memory allocator dma-direct: Align CoCo shared DMA allocations to the shared granule size swiotlb: Align shared IO TLB pools to the shared granule size swiotlb: Reject misaligned restricted DMA pools for CoCo guests dma-buf: system_heap: Limit scatterlist entries to the buffer size dma-buf: system_heap: Allocate shared buffers using CoCo shared memory allocator swiotlb: Make rounded shared pool capacity allocatable
MAINTAINERS | 1 + arch/arm/mm/dma-mapping.c | 5 +- arch/arm64/include/asm/mem_encrypt.h | 1 + arch/arm64/mm/mem_encrypt.c | 13 +- drivers/dma-buf/heaps/system_heap.c | 126 +++++++++----------- drivers/firmware/arm_rmm/rsi.c | 58 +++++++++ drivers/iommu/dma-iommu.c | 2 +- drivers/irqchip/irq-gic-v3-its.c | 43 +++---- include/linux/arm-rsi-cmds.h | 10 ++ include/linux/arm-smccc-rhi.h | 25 ++++ include/linux/arm-smccc-rsi.h | 7 ++ include/linux/cc_shared.h | 39 ++++++ include/linux/dma-map-ops.h | 10 +- kernel/dma/contiguous.c | 33 +++-- kernel/dma/direct.c | 55 +++++++-- kernel/dma/ops_helpers.c | 2 +- kernel/dma/pool.c | 23 +++- kernel/dma/swiotlb.c | 81 +++++++++---- kernel/kexec_file.c | 3 +- mm/Makefile | 1 + mm/cc_shared.c | 172 +++++++++++++++++++++++++++ 21 files changed, 552 insertions(+), 158 deletions(-) create mode 100644 include/linux/arm-smccc-rhi.h create mode 100644 include/linux/cc_shared.h create mode 100644 mm/cc_shared.c
Arm CCA guests can run with a PAGE_SIZE that is smaller than the granule size used by the host to track Realm IPA state. In that configuration, changing a guest PAGE_SIZE range between protected and shared state can cover only part of the host-owned state-change unit.
The Realm Host Interface specification (Arm DEN0148) [1] defines the Host Configuration interface for querying host capabilities, including the IPA state change alignment.
Add the RHI HostConf definitions and a get_ipa_state_change_alignment() helper. The helper uses RSI_HOST_CALL to query the supported HostConf version and features, and reads the IPA state change alignment when available. It falls back to PAGE_SIZE if the interface is unavailable or returns an invalid value.
[1] https://support.arm.com/documentation/den0148/latest/
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- MAINTAINERS | 1 + drivers/firmware/arm_rmm/rsi.c | 47 ++++++++++++++++++++++++++++++++++ include/linux/arm-rsi-cmds.h | 10 ++++++++ include/linux/arm-smccc-rhi.h | 25 ++++++++++++++++++ include/linux/arm-smccc-rsi.h | 7 +++++ 5 files changed, 90 insertions(+) create mode 100644 include/linux/arm-smccc-rhi.h
diff --git a/MAINTAINERS b/MAINTAINERS index 1e543163a8cc..87ec876868eb 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -3956,6 +3956,7 @@ F: drivers/firmware/arm_rmm/ F: drivers/virt/coco/arm-cca-guest/ F: drivers/virt/coco/pkvm-guest/ F: include/linux/arm-rsi-cmds.h +F: include/linux/arm-smccc-rhi.h F: include/linux/arm-smccc-rsi.h F: tools/testing/selftests/arm64/ X: arch/arm64/boot/dts/ diff --git a/drivers/firmware/arm_rmm/rsi.c b/drivers/firmware/arm_rmm/rsi.c index 0fcc4e33bd92..883f6091ff95 100644 --- a/drivers/firmware/arm_rmm/rsi.c +++ b/drivers/firmware/arm_rmm/rsi.c @@ -8,7 +8,9 @@ #include <linux/psci.h> #include <linux/swiotlb.h> #include <linux/arm-rsi-cmds.h> +#include <linux/arm-smccc-rhi.h> #include <linux/kobject.h> +#include <linux/string.h> #include <linux/sysfs.h>
#include <asm/io.h> @@ -162,6 +164,51 @@ static int realm_register_memory_enc_ops(void) return arm64_mem_crypt_ops_register(&realm_crypt_ops); }
+/* we need an aligned struct for rsi_host_call. slab is not yet ready */ +static struct rsi_host_call hostconf_call __initdata; +static unsigned long __maybe_unused __init get_ipa_state_change_alignment(void) +{ + long ret; + unsigned long shared_granule_size; + + memset(&hostconf_call, 0, sizeof(hostconf_call)); + hostconf_call.gprs[0] = RHI_HOSTCONF_VERSION; + ret = rsi_host_call(lm_alias(&hostconf_call)); + if (ret != RSI_SUCCESS) + goto err_out; + + if (hostconf_call.gprs[0] != RHI_HOSTCONF_VER_1_0) + goto err_out; + + memset(&hostconf_call, 0, sizeof(hostconf_call)); + hostconf_call.gprs[0] = RHI_HOSTCONF_FEATURES; + ret = rsi_host_call(lm_alias(&hostconf_call)); + if (ret != RSI_SUCCESS) + goto err_out; + + if (!(hostconf_call.gprs[0] & RHI_HOSTCONF_FEATURE_GET_IPA_CHANGE_ALIGNMENT)) + goto err_out; + + memset(&hostconf_call, 0, sizeof(hostconf_call)); + hostconf_call.gprs[0] = RHI_HOSTCONF_GET_IPA_CHANGE_ALIGNMENT; + ret = rsi_host_call(lm_alias(&hostconf_call)); + if (ret != RSI_SUCCESS) + goto err_out; + + shared_granule_size = hostconf_call.gprs[0]; + if (shared_granule_size & (SZ_4K - 1) || + !is_power_of_2(shared_granule_size)) + goto err_out; + + return max(PAGE_SIZE, shared_granule_size); +err_out: + /* + * For failure condition assume host is built with 4K page size + * and hence IPA state change alignment can be guest PAGE_SIZE. + */ + return PAGE_SIZE; +} + void __init arm64_rsi_init(void) { if (arm_smccc_1_1_get_conduit() != SMCCC_CONDUIT_SMC) diff --git a/include/linux/arm-rsi-cmds.h b/include/linux/arm-rsi-cmds.h index 3f7a6a833993..996f1621b996 100644 --- a/include/linux/arm-rsi-cmds.h +++ b/include/linux/arm-rsi-cmds.h @@ -236,4 +236,14 @@ static inline unsigned long rsi_attestation_token_continue(phys_addr_t granule, return res.a0; }
+static inline unsigned long rsi_host_call(struct rsi_host_call *rhi_call) +{ + phys_addr_t addr = virt_to_phys(rhi_call); + struct arm_smccc_res res; + + arm_smccc_1_1_invoke(SMC_RSI_HOST_CALL, addr, &res); + + return res.a0; +} + #endif /* __LINUX_ARM_RSI_CMDS_H_ */ diff --git a/include/linux/arm-smccc-rhi.h b/include/linux/arm-smccc-rhi.h new file mode 100644 index 000000000000..be89185cddae --- /dev/null +++ b/include/linux/arm-smccc-rhi.h @@ -0,0 +1,25 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (C) 2026 ARM Ltd. + */ + +#ifndef __LINUX_ARM_SMCCC_RHI_H_ +#define __LINUX_ARM_SMCCC_RHI_H_ + +#include <linux/arm-smccc.h> + +#define SMC_RHI_CALL(func) \ + ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, \ + ARM_SMCCC_SMC_64, \ + ARM_SMCCC_OWNER_STANDARD_HYP,\ + (func)) + +#define RHI_HOSTCONF_VER_1_0 0x10000 +#define RHI_HOSTCONF_VERSION SMC_RHI_CALL(0x004E) + +#define RHI_HOSTCONF_FEATURE_GET_IPA_CHANGE_ALIGNMENT BIT(0) +#define RHI_HOSTCONF_FEATURES SMC_RHI_CALL(0x004F) + +#define RHI_HOSTCONF_GET_IPA_CHANGE_ALIGNMENT SMC_RHI_CALL(0x0050) + +#endif /* __LINUX_ARM_SMCCC_RHI_H_ */ diff --git a/include/linux/arm-smccc-rsi.h b/include/linux/arm-smccc-rsi.h index fddb77986f70..3532a3f08f4e 100644 --- a/include/linux/arm-smccc-rsi.h +++ b/include/linux/arm-smccc-rsi.h @@ -182,6 +182,13 @@ struct realm_config { */ #define SMC_RSI_IPA_STATE_GET SMC_RSI_FID(0x198)
+struct rsi_host_call { + union { + u16 imm; + u64 padding0; + }; + u64 gprs[31]; +} __aligned(0x100); /* * Make a Host call. *
Confidential-computing guests may require memory shared with the host to be aligned and transitioned in units larger than PAGE_SIZE. Several DMA users need struct page-backed allocations satisfying these requirements. Provide a common allocator instead of requiring each user to open-code this sequence.
Add alloc_cc_shared_pages() and its node-aware variant. The allocator rounds the requested size to the architecture's shared granule, allocates suitably aligned contiguous pages and transitions the complete range to shared state. It also preserves the caller's GFP policy.
A private-to-shared transition may alter memory contents. Mask __GFP_ZERO from the underlying allocation and when requested, clear the complete transitioned range after cc_make_shared() succeeds.
Return the page and transitioned size so free_cc_shared_pages() can restore the complete range to private state before freeing it. If private state cannot be established, retain the allocation instead of returning a possibly shared page to the buddy allocator.
Also provide the shared-granule geometry and byte-oriented transition helpers used by the allocator and by callers managing their own backing memory.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- include/linux/cc_shared.h | 39 +++++++++ mm/Makefile | 1 + mm/cc_shared.c | 172 ++++++++++++++++++++++++++++++++++++++ 3 files changed, 212 insertions(+) create mode 100644 include/linux/cc_shared.h create mode 100644 mm/cc_shared.c
diff --git a/include/linux/cc_shared.h b/include/linux/cc_shared.h new file mode 100644 index 000000000000..5f8db7c468c5 --- /dev/null +++ b/include/linux/cc_shared.h @@ -0,0 +1,39 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _LINUX_CC_SHARED_H +#define _LINUX_CC_SHARED_H + +#include <linux/gfp_types.h> +#include <linux/types.h> + +struct page; + +struct cc_shared_pages { + struct page *page; + size_t shared_size; +}; + +struct cc_shared_layout { + size_t requested_size; + size_t shared_size; + size_t alignment; +}; + +/* + * Architectures may override this to return the granule used for transitions + * between private and shared memory. The value must be a power of two and no + * smaller than PAGE_SIZE. + */ +size_t arch_cc_shared_granule_size(void); + +size_t cc_shared_granule_size(void); +int cc_shared_calc_layout(size_t requested, struct cc_shared_layout *layout); +bool cc_shared_range_valid(phys_addr_t base, size_t size); +int cc_make_shared(void *addr, size_t size); +int cc_make_private(void *addr, size_t size); +int alloc_cc_shared_pages_node(int nid, gfp_t gfp, + size_t requested, struct cc_shared_pages *mem); +int alloc_cc_shared_pages(gfp_t gfp, + size_t requested, struct cc_shared_pages *mem); +void free_cc_shared_pages(struct cc_shared_pages *mem); + +#endif /* _LINUX_CC_SHARED_H */ diff --git a/mm/Makefile b/mm/Makefile index e7245cb88c66..6e6544428422 100644 --- a/mm/Makefile +++ b/mm/Makefile @@ -56,6 +56,7 @@ obj-y := filemap.o mempool.o oom_kill.o fadvise.o \ compaction.o show_mem.o \ interval_tree.o list_lru.o workingset.o \ debug.o gup.o mmap_lock.o vma_init.o $(mmu-y) +obj-y += cc_shared.o
# Give 'page_alloc' its own module-parameter namespace page-alloc-y := page_alloc.o diff --git a/mm/cc_shared.c b/mm/cc_shared.c new file mode 100644 index 000000000000..3e33681218f1 --- /dev/null +++ b/mm/cc_shared.c @@ -0,0 +1,172 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright (C) 2026 ARM Ltd. + */ +#include <linux/align.h> +#include <linux/cc_shared.h> +#include <linux/errno.h> +#include <linux/export.h> +#include <linux/gfp.h> +#include <linux/log2.h> +#include <linux/mm.h> +#include <linux/mem_encrypt.h> +#include <linux/numa.h> +#include <linux/overflow.h> +#include <linux/set_memory.h> + +size_t __weak arch_cc_shared_granule_size(void) +{ + return PAGE_SIZE; +} + +size_t cc_shared_granule_size(void) +{ + size_t granule = arch_cc_shared_granule_size(); + + if (WARN_ON_ONCE(granule < PAGE_SIZE || !is_power_of_2(granule))) + return PAGE_SIZE; + + return granule; +} +EXPORT_SYMBOL_GPL(cc_shared_granule_size); + +int cc_shared_calc_layout(size_t requested, struct cc_shared_layout *layout) +{ + size_t granule, rounded; + + if (!requested || !layout) + return -EINVAL; + + granule = cc_shared_granule_size(); + if (check_add_overflow(requested, granule - 1, &rounded)) + return -EOVERFLOW; + + rounded = ALIGN_DOWN(rounded, granule); + layout->requested_size = requested; + layout->shared_size = rounded; + layout->alignment = granule; + + return 0; +} +EXPORT_SYMBOL_GPL(cc_shared_calc_layout); + +bool cc_shared_range_valid(phys_addr_t base, size_t size) +{ + size_t granule = cc_shared_granule_size(); + + if (!size) + return false; + + return IS_ALIGNED(base, granule) && IS_ALIGNED(size, granule); +} +EXPORT_SYMBOL_GPL(cc_shared_range_valid); + +static int cc_validate_transition(void *addr, size_t size) +{ + phys_addr_t phys; + + if (!addr || !size || !PAGE_ALIGNED(addr) || + !virt_addr_valid(addr)) + return -EINVAL; + + phys = page_to_phys(virt_to_page(addr)); + if (!cc_shared_range_valid(phys, size)) + return -EINVAL; + + return 0; +} + +int cc_make_shared(void *addr, size_t size) +{ + int ret = cc_validate_transition(addr, size); + + if (ret) + return ret; + + return set_memory_decrypted((unsigned long)addr, size >> PAGE_SHIFT); +} + +int cc_make_private(void *addr, size_t size) +{ + int ret = cc_validate_transition(addr, size); + + if (ret) + return ret; + + return set_memory_encrypted((unsigned long)addr, size >> PAGE_SHIFT); +} + +int alloc_cc_shared_pages_node(int nid, gfp_t gfp, + size_t requested, struct cc_shared_pages *mem) +{ + struct cc_shared_layout layout; + struct page *page; + unsigned int order; + bool zero = gfp & __GFP_ZERO; + int ret; + + if (!mem) + return -EINVAL; + + ret = cc_shared_calc_layout(requested, &layout); + if (ret) + return ret; + + order = get_order(layout.shared_size); + if (order > MAX_PAGE_ORDER) + return -EINVAL; + + /* + * State transitions require a linear-map address and may modify memory. + * Allocate from low memory and defer requested zeroing until afterwards. + */ + gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO); + if (nid == NUMA_NO_NODE) + page = alloc_pages(gfp, order); + else + page = alloc_pages_node(nid, gfp, order); + if (!page) + return -ENOMEM; + + ret = cc_make_shared(page_address(page), layout.shared_size); + if (ret) { + if (!cc_make_private(page_address(page), layout.shared_size)) + __free_pages(page, order); + else + pr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n", + layout.shared_size); + return ret; + } + + if (zero) + memset(page_address(page), 0, layout.shared_size); + + mem->page = page; + mem->shared_size = layout.shared_size; + return 0; +} +EXPORT_SYMBOL_GPL(alloc_cc_shared_pages_node); + +int alloc_cc_shared_pages(gfp_t gfp, + size_t requested, struct cc_shared_pages *mem) +{ + return alloc_cc_shared_pages_node(NUMA_NO_NODE, gfp, requested, mem); +} +EXPORT_SYMBOL_GPL(alloc_cc_shared_pages); + +void free_cc_shared_pages(struct cc_shared_pages *mem) +{ + if (!mem || !mem->page) + return; + + if (cc_make_private(page_address(mem->page), mem->shared_size)) { + pr_warn_ratelimited("leaking %zu bytes that cannot be made private\n", + mem->shared_size); + return; + } + + __free_pages(mem->page, get_order(mem->shared_size)); + mem->page = NULL; + mem->shared_size = 0; +} +EXPORT_SYMBOL_GPL(free_cc_shared_pages);
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);
Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
On Tue, Sep 22, 2026 at 05:25:58PM +0100, Catalin Marinas wrote:
Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Same question for RMM? Or did they define that the RMM zeros on these transitions?
Jason
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
I don't think every architecture or platform can unconditionally zero memory in set_memory_decrypted(). Some callers may need to share valid contents with the host.
Also, if zeroing is added only to the CCA implementation, the allocator must retain __GFP_ZERO for platforms such as pKVM. This would cause the memory to be zeroed twice on CCA.
How about extending cc_make_shared() with a flag indicating that the memory must be zeroed, and passing that requirement down to the architecture-specific implementation? The implementation could then zero the memory at the appropriate point: before sharing for pKVM and after the destructive transition for CCA.
The allocator could derive this flag from __GFP_ZERO, remove __GFP_ZERO before calling alloc_pages(), and let the sharing operation perform the requested zeroing with the correct ordering.
-aneesh
Aneesh Kumar K.V aneesh.kumar@kernel.org writes:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
I don't think every architecture or platform can unconditionally zero memory in set_memory_decrypted(). Some callers may need to share valid contents with the host.
Also, if zeroing is added only to the CCA implementation, the allocator must retain __GFP_ZERO for platforms such as pKVM. This would cause the memory to be zeroed twice on CCA.
How about extending cc_make_shared() with a flag indicating that the memory must be zeroed, and passing that requirement down to the architecture-specific implementation? The implementation could then zero the memory at the appropriate point: before sharing for pKVM and after the destructive transition for CCA.
The allocator could derive this flag from __GFP_ZERO, remove __GFP_ZERO before calling alloc_pages(), and let the sharing operation perform the requested zeroing with the correct ordering.
I was pointed to this email thread:
https://lore.kernel.org/all/c25502d3-35c6-4281-a9ec-856f789fb1b4@arm.com
This makes a stronger case for having a CoCo shared memory allocator that captures all these restrictions. It also means that alloc_cc_shared_pages_node() needs:
if (WARN_ON_ONCE(!gfpflags_allow_blocking(gfp))) return -EINVAL;
might_sleep();
I guess this also requires the VPE L1 tables to be preallocated from a sleepable context.
-aneesh
On Wed, Sep 23, 2026 at 02:01:08PM +0530, Aneesh Kumar K.V wrote:
I was pointed to this email thread:
https://lore.kernel.org/all/c25502d3-35c6-4281-a9ec-856f789fb1b4@arm.com
This makes a stronger case for having a CoCo shared memory allocator that captures all these restrictions. It also means that alloc_cc_shared_pages_node() needs:
if (WARN_ON_ONCE(!gfpflags_allow_blocking(gfp))) return -EINVAL;
might_sleep();
I think gfpflags_allow_blocking() is the wrong predicate. It checks for __GFP_DIRECT_RECLAIM but we have callers like dma-buf which clear the flag (see HIGH_ORDER_GFP). ITS allocations also use GFP_NOWAIT, so they'll fail here.
I'd just keep a might_sleep() in cc_make_*() and ignore gfp flags checking. Passing a no-blocking GFP flag doesn't necessarily mean the caller cannot sleep. It might also happen early enough (like its_init()) not to care - __might_resched() skips the warning if SYSTEM_BOOTING.
On Wed, Sep 23, 2026 at 11:23:27AM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
I don't think every architecture or platform can unconditionally zero memory in set_memory_decrypted(). Some callers may need to share valid contents with the host.
Is there any? That would be a bad assumptions in the caller. Most set_memory_* backends don't preserve the content as they change the encryption key. So properly written code shouldn't rely on this unless it knows specifically it's only running on pKVM for example. The only use-case I see to avoid explicit zeroing is when the caller doesn't care about the page initialisation and wants to save some cycles. The encryption key change would take care of the security aspect.
Also, if zeroing is added only to the CCA implementation, the allocator must retain __GFP_ZERO for platforms such as pKVM. This would cause the memory to be zeroed twice on CCA.
What I meant is that we change the set_memory_decrypted() contract to always zero, assuming that all callers need to zero the pages anyway. If we do have cases where zeroing is not needed, we could make it explicit via a flag.
How about extending cc_make_shared() with a flag indicating that the memory must be zeroed, and passing that requirement down to the architecture-specific implementation? The implementation could then zero the memory at the appropriate point: before sharing for pKVM and after the destructive transition for CCA.
On pKVM, we want set_memory_decrypted() to zero the buffer before the host can access it (I guess currently relying on __GFP_ZERO allocations). Since no cryptographic encryption takes place, there's not much point in memset'ing again after the operation as the content was already zeroed.
I don't think cc_make_shared() has the right information on how to safely and efficiently do the zeroing. That's only known to the set_memory_* backend. So you'd have to propagate the flag down.
Catalin Marinas catalin.marinas@arm.com writes:
On Wed, Sep 23, 2026 at 11:23:27AM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
[ ... 79 lines skipped ... ]
On pKVM, we want set_memory_decrypted() to zero the buffer before the host can access it (I guess currently relying on __GFP_ZERO allocations). Since no cryptographic encryption takes place, there's not much point in memset'ing again after the operation as the content was already zeroed.
I don't think cc_make_shared() has the right information on how to safely and efficiently do the zeroing. That's only known to the set_memory_* backend. So you'd have to propagate the flag down.
This is my attempt to do that using Codex. Quite a few paths already call memset() outside set_memory_decrypted(), and there is a fixup series for the ITS and other paths here:
https://lore.kernel.org/all/c25502d3-35c6-4281-a9ec-856f789fb1b4@arm.com
commit 0317b02d6759a8b55e9ec854e15b5b94025800e3 Author: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org Date: Wed Sep 23 11:09:40 2026 +0530
mm: Add zeroing support to shared memory transitions
Architectures need to zero memory at different points in a private-to-shared transition. For example, pKVM needs to clear the memory before sharing it, while Arm CCA needs to clear it after the RSI transition has completed.
Add CC_SHARED_ZERO to cc_make_shared() and pass it through set_memory_decrypted() so each architecture or platform can select the safe ordering. Thread the flag through the arm64 memory-encryption operations and the x86 encryption-status hooks. Clear memory immediately before sharing in the other implementations, while keeping CCA zeroing after a successful RSI transition.
Keep allocations on platforms without memory encryption on the ordinary page allocator path so the original GFP constraints, including __GFP_ZERO, remain intact. Callers that need zero-filled memory request zeroing as part of an actual transition and explicitly clear the memory when no transition is needed. This also removes redundant post-transition memset() calls where the transition now provides that guarantee.
Assisted-by: Codex:gpt-5
diff --git a/arch/arm64/include/asm/mem_encrypt.h b/arch/arm64/include/asm/mem_encrypt.h index 636f45b4d8af..cf8dd5e84c86 100644 --- a/arch/arm64/include/asm/mem_encrypt.h +++ b/arch/arm64/include/asm/mem_encrypt.h @@ -9,14 +9,13 @@ struct device;
struct arm64_mem_crypt_ops { int (*encrypt)(unsigned long addr, int numpages); - int (*decrypt)(unsigned long addr, int numpages); + int (*decrypt)(unsigned long addr, int numpages, unsigned int flags); };
int arm64_mem_crypt_ops_register(const struct arm64_mem_crypt_ops *ops);
int set_memory_encrypted(unsigned long addr, int numpages); -int set_memory_decrypted(unsigned long addr, int numpages); - +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags); int realm_register_memory_enc_ops(void);
static inline bool force_dma_unencrypted(struct device *dev) diff --git a/arch/arm64/include/asm/set_memory.h b/arch/arm64/include/asm/set_memory.h index 90f61b17275e..10278a7ba5a9 100644 --- a/arch/arm64/include/asm/set_memory.h +++ b/arch/arm64/include/asm/set_memory.h @@ -17,6 +17,6 @@ int set_direct_map_valid_noflush(struct page *page, unsigned nr, bool valid); bool kernel_page_present(struct page *page);
int set_memory_encrypted(unsigned long addr, int numpages); -int set_memory_decrypted(unsigned long addr, int numpages); +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags);
#endif /* _ASM_ARM64_SET_MEMORY_H */ diff --git a/arch/arm64/mm/mem_encrypt.c b/arch/arm64/mm/mem_encrypt.c index ee3c0ab04384..4da91f73a620 100644 --- a/arch/arm64/mm/mem_encrypt.c +++ b/arch/arm64/mm/mem_encrypt.c @@ -40,11 +40,11 @@ int set_memory_encrypted(unsigned long addr, int numpages) } EXPORT_SYMBOL_GPL(set_memory_encrypted);
-int set_memory_decrypted(unsigned long addr, int numpages) +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags) { if (likely(!crypt_ops) || WARN_ON(!PAGE_ALIGNED(addr))) return 0;
- return crypt_ops->decrypt(addr, numpages); + return crypt_ops->decrypt(addr, numpages, flags); } EXPORT_SYMBOL_GPL(set_memory_decrypted); diff --git a/arch/arm64/mm/pageattr.c b/arch/arm64/mm/pageattr.c index bbe98ac9ad8c..f565996efbef 100644 --- a/arch/arm64/mm/pageattr.c +++ b/arch/arm64/mm/pageattr.c @@ -9,6 +9,7 @@ #include <linux/sched.h> #include <linux/vmalloc.h> #include <linux/pagewalk.h> +#include <linux/cc_shared.h>
#include <asm/cacheflush.h> #include <asm/pgtable-prot.h> @@ -335,10 +336,14 @@ static int realm_set_memory_encrypted(unsigned long addr, int numpages) return ret; }
-static int realm_set_memory_decrypted(unsigned long addr, int numpages) +static int realm_set_memory_decrypted(unsigned long addr, int numpages, + unsigned int flags) { int ret = __set_memory_enc_dec(addr, numpages, false);
+ if (!ret && (flags & CC_SHARED_ZERO)) + memset((void *)addr, 0, (size_t)numpages << PAGE_SHIFT); + WARN(ret, "Failed to decrypt memory, %d pages will be leaked", numpages);
diff --git a/arch/powerpc/include/asm/mem_encrypt.h b/arch/powerpc/include/asm/mem_encrypt.h index e355ca46fad9..e03c90d70d3c 100644 --- a/arch/powerpc/include/asm/mem_encrypt.h +++ b/arch/powerpc/include/asm/mem_encrypt.h @@ -19,6 +19,6 @@ static inline bool force_dma_unencrypted(struct device *dev) }
int set_memory_encrypted(unsigned long addr, int numpages); -int set_memory_decrypted(unsigned long addr, int numpages); +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags);
#endif /* _ASM_POWERPC_MEM_ENCRYPT_H */ diff --git a/arch/powerpc/platforms/pseries/svm.c b/arch/powerpc/platforms/pseries/svm.c index 7a403dbd35ee..46e940b40752 100644 --- a/arch/powerpc/platforms/pseries/svm.c +++ b/arch/powerpc/platforms/pseries/svm.c @@ -9,7 +9,9 @@ #include <linux/mm.h> #include <linux/memblock.h> #include <linux/mem_encrypt.h> +#include <linux/string.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h> #include <asm/machdep.h> #include <asm/svm.h> #include <asm/swiotlb.h> @@ -51,7 +53,7 @@ int set_memory_encrypted(unsigned long addr, int numpages) return 0; }
-int set_memory_decrypted(unsigned long addr, int numpages) +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags) { if (!cc_platform_has(CC_ATTR_MEM_ENCRYPT)) return 0; @@ -59,6 +61,8 @@ int set_memory_decrypted(unsigned long addr, int numpages) if (!PAGE_ALIGNED(addr)) return -EINVAL;
+ if (flags & CC_SHARED_ZERO) + memset((void *)addr, 0, (size_t)numpages << PAGE_SHIFT); uv_share_page(PHYS_PFN(__pa(addr)), numpages);
return 0; diff --git a/arch/s390/include/asm/mem_encrypt.h b/arch/s390/include/asm/mem_encrypt.h index 28c83ec1f243..97813680093c 100644 --- a/arch/s390/include/asm/mem_encrypt.h +++ b/arch/s390/include/asm/mem_encrypt.h @@ -5,7 +5,7 @@ #ifndef __ASSEMBLER__
int set_memory_encrypted(unsigned long vaddr, int numpages); -int set_memory_decrypted(unsigned long vaddr, int numpages); +int set_memory_decrypted(unsigned long vaddr, int numpages, unsigned int flags);
#endif /* __ASSEMBLER__ */
diff --git a/arch/s390/mm/init.c b/arch/s390/mm/init.c index be7e009e7b59..b7aaba663889 100644 --- a/arch/s390/mm/init.c +++ b/arch/s390/mm/init.c @@ -51,6 +51,7 @@ #include <linux/virtio_config.h> #include <linux/execmem.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h>
pgd_t swapper_pg_dir[PTRS_PER_PGD] __section(".bss..swapper_pg_dir"); pgd_t invalid_pg_dir[PTRS_PER_PGD] __section(".bss..invalid_pg_dir"); @@ -126,9 +127,13 @@ int set_memory_encrypted(unsigned long vaddr, int numpages) return 0; }
-int set_memory_decrypted(unsigned long vaddr, int numpages) +int set_memory_decrypted(unsigned long vaddr, int numpages, unsigned int flags) { int i; + + if (flags & CC_SHARED_ZERO) + memset((void *)vaddr, 0, (size_t)numpages << PAGE_SHIFT); + /* make specified pages shared (swiotlb, dma_alloca) */ for (i = 0; i < numpages; ++i) { uv_set_shared(virt_to_phys((void *)vaddr)); diff --git a/arch/x86/coco/sev/core.c b/arch/x86/coco/sev/core.c index cc292d7c6fd1..249054d53915 100644 --- a/arch/x86/coco/sev/core.c +++ b/arch/x86/coco/sev/core.c @@ -1497,7 +1497,8 @@ static void *alloc_shared_pages(size_t sz) if (!page) return NULL;
- ret = set_memory_decrypted((unsigned long)page_address(page), npages); + ret = set_memory_decrypted((unsigned long)page_address(page), npages, + 0); if (ret) { pr_err("failed to mark page shared, ret=%d\n", ret); __free_pages(page, get_order(sz)); diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c index f904a636d449..748e4d19b15e 100644 --- a/arch/x86/coco/tdx/tdx.c +++ b/arch/x86/coco/tdx/tdx.c @@ -5,6 +5,7 @@ #define pr_fmt(fmt) "tdx: " fmt
#include <linux/cpufeature.h> +#include <linux/cc_shared.h> #include <linux/export.h> #include <linux/io.h> #include <linux/kexec.h> @@ -976,8 +977,11 @@ static bool tdx_enc_status_changed(unsigned long vaddr, int numpages, bool enc) }
static int tdx_enc_status_change_prepare(unsigned long vaddr, int numpages, - bool enc) + bool enc, unsigned int flags) { + if (!enc && (flags & CC_SHARED_ZERO)) + memset((void *)vaddr, 0, (size_t)numpages << PAGE_SHIFT); + /* * Only handle shared->private conversion here. * See the comment in tdx_early_init(). @@ -989,7 +993,7 @@ static int tdx_enc_status_change_prepare(unsigned long vaddr, int numpages, }
static int tdx_enc_status_change_finish(unsigned long vaddr, int numpages, - bool enc) + bool enc, unsigned int flags) { /* * Only handle private->shared conversion here. diff --git a/arch/x86/hyperv/hv_init.c b/arch/x86/hyperv/hv_init.c index 0b4a1c0b0b16..9f5113868c7a 100644 --- a/arch/x86/hyperv/hv_init.c +++ b/arch/x86/hyperv/hv_init.c @@ -12,6 +12,7 @@ #include <linux/efi.h> #include <linux/types.h> #include <linux/bitfield.h> +#include <linux/cc_shared.h> #include <linux/io.h> #include <asm/apic.h> #include <asm/desc.h> @@ -156,8 +157,11 @@ static int hv_cpu_init(unsigned int cpu) * page in non-root partition here. */ if (*hvp && !ms_hyperv.paravisor_present && hv_isolation_type_snp()) { - WARN_ON_ONCE(set_memory_decrypted((unsigned long)(*hvp), 1)); - memset(*hvp, 0, PAGE_SIZE); + int ret; + + ret = set_memory_decrypted((unsigned long)*hvp, 1, + CC_SHARED_ZERO); + WARN_ON_ONCE(ret); } }
diff --git a/arch/x86/hyperv/ivm.c b/arch/x86/hyperv/ivm.c index 2ce4dfe53472..104e45d4605d 100644 --- a/arch/x86/hyperv/ivm.c +++ b/arch/x86/hyperv/ivm.c @@ -7,6 +7,7 @@ */
#include <linux/bitfield.h> +#include <linux/cc_shared.h> #include <linux/types.h> #include <linux/slab.h> #include <linux/cpu.h> @@ -753,8 +754,13 @@ static int hv_mark_gpa_visibility(u16 count, const u64 pfn[], * transition is complete, hv_vtom_set_host_visibility() marks the pages * as "present" again. */ -static int hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc) +static int hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc, + unsigned int flags) { + if (!enc && (flags & CC_SHARED_ZERO)) + memset((void *)kbuffer, 0, + (size_t)pagecount << PAGE_SHIFT); + return set_memory_np(kbuffer, pagecount); }
@@ -766,7 +772,8 @@ static int hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc) * with host. This function works as wrap of hv_mark_gpa_visibility() * with memory base and size. */ -static int hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, bool enc) +static int hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, + bool enc, unsigned int flags) { enum hv_mem_host_visibility visibility = enc ? VMBUS_PAGE_NOT_VISIBLE : VMBUS_PAGE_VISIBLE_READ_WRITE; @@ -816,7 +823,6 @@ static int hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, boo err = set_memory_p(kbuffer, pagecount); if (err && !ret) ret = err; - return ret; }
diff --git a/arch/x86/include/asm/set_memory.h b/arch/x86/include/asm/set_memory.h index 4362c26aa992..117f8ae05fee 100644 --- a/arch/x86/include/asm/set_memory.h +++ b/arch/x86/include/asm/set_memory.h @@ -51,7 +51,7 @@ int set_memory_4k(unsigned long addr, int numpages);
bool set_memory_enc_stop_conversion(void); int set_memory_encrypted(unsigned long addr, int numpages); -int set_memory_decrypted(unsigned long addr, int numpages); +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags);
int set_memory_np_noalias(unsigned long addr, int numpages); int set_memory_nonglobal(unsigned long addr, int numpages); diff --git a/arch/x86/include/asm/vga.h b/arch/x86/include/asm/vga.h index 46f9b2deab4d..b71311270b21 100644 --- a/arch/x86/include/asm/vga.h +++ b/arch/x86/include/asm/vga.h @@ -22,7 +22,7 @@ unsigned long start = (unsigned long)phys_to_virt(x); \ \ if (IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT)) \ - set_memory_decrypted(start, (s) >> PAGE_SHIFT); \ + set_memory_decrypted(start, (s) >> PAGE_SHIFT, 0); \ \ start; \ }) diff --git a/arch/x86/include/asm/x86_init.h b/arch/x86/include/asm/x86_init.h index 953d3199408a..a10de48b27d4 100644 --- a/arch/x86/include/asm/x86_init.h +++ b/arch/x86/include/asm/x86_init.h @@ -162,8 +162,10 @@ struct x86_init_acpi { * and with interrupts disabled. */ struct x86_guest { - int (*enc_status_change_prepare)(unsigned long vaddr, int npages, bool enc); - int (*enc_status_change_finish)(unsigned long vaddr, int npages, bool enc); + int (*enc_status_change_prepare)(unsigned long vaddr, int npages, bool enc, + unsigned int flags); + int (*enc_status_change_finish)(unsigned long vaddr, int npages, bool enc, + unsigned int flags); bool (*enc_tlb_flush_required)(bool enc); bool (*enc_cache_flush_required)(void); void (*enc_kexec_begin)(void); diff --git a/arch/x86/kernel/kvmclock.c b/arch/x86/kernel/kvmclock.c index cb3d0ca1fa22..4e87c0657db9 100644 --- a/arch/x86/kernel/kvmclock.c +++ b/arch/x86/kernel/kvmclock.c @@ -248,17 +248,17 @@ static void __init kvmclock_init_mem(void) * be mapped decrypted. */ if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) { - r = set_memory_decrypted((unsigned long) hvclock_mem, - 1UL << order); + r = set_memory_decrypted((unsigned long)hvclock_mem, + 1UL << order, CC_SHARED_ZERO); if (r) { __free_pages(p, order); hvclock_mem = NULL; pr_warn("kvmclock: set_memory_decrypted() failed. Disabling\n"); return; } + } else { + memset(hvclock_mem, 0, PAGE_SIZE << order); } - - memset(hvclock_mem, 0, PAGE_SIZE << order); }
static int __init kvm_setup_vsyscall_timeinfo(void) diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c index c3f4a389992d..3fe4cdc265d1 100644 --- a/arch/x86/kernel/machine_kexec_64.c +++ b/arch/x86/kernel/machine_kexec_64.c @@ -18,6 +18,7 @@ #include <linux/vmalloc.h> #include <linux/efi.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h>
#include <asm/init.h> #include <asm/tlbflush.h> @@ -693,7 +694,8 @@ int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp) * pages are not encrypted because when we boot to the new kernel the * pages won't be accessed encrypted (initially). */ - return set_memory_decrypted((unsigned long)vaddr, pages); + return set_memory_decrypted((unsigned long)vaddr, pages, + gfp & __GFP_ZERO ? CC_SHARED_ZERO : 0); }
void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages) diff --git a/arch/x86/kernel/x86_init.c b/arch/x86/kernel/x86_init.c index 252c5827d063..187e2a888c32 100644 --- a/arch/x86/kernel/x86_init.c +++ b/arch/x86/kernel/x86_init.c @@ -138,8 +138,17 @@ struct x86_cpuinit_ops x86_cpuinit = {
static void default_nmi_init(void) { };
-static int enc_status_change_prepare_noop(unsigned long vaddr, int npages, bool enc) { return 0; } -static int enc_status_change_finish_noop(unsigned long vaddr, int npages, bool enc) { return 0; } +static int enc_status_change_prepare_noop(unsigned long vaddr, int npages, bool enc, + unsigned int flags) +{ + return 0; +} + +static int enc_status_change_finish_noop(unsigned long vaddr, int npages, bool enc, + unsigned int flags) +{ + return 0; +} static bool enc_tlb_flush_required_noop(bool enc) { return false; } static bool enc_cache_flush_required_noop(void) { return false; } static void enc_kexec_begin_noop(void) {} diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c index 064ecc33b926..8fdec6c8090d 100644 --- a/arch/x86/kvm/mmu/mmu.c +++ b/arch/x86/kvm/mmu/mmu.c @@ -6852,7 +6852,7 @@ static int __kvm_mmu_create(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, struct k * by 32-bit kernels (when KVM itself uses 32-bit NPT). */ if (!tdp_enabled) - set_memory_decrypted((unsigned long)mmu->pae_root, 1); + set_memory_decrypted((unsigned long)mmu->pae_root, 1, 0); else WARN_ON_ONCE(shadow_me_value);
diff --git a/arch/x86/mm/mem_encrypt_amd.c b/arch/x86/mm/mem_encrypt_amd.c index 2f8c32173972..ba3cfb89d155 100644 --- a/arch/x86/mm/mem_encrypt_amd.c +++ b/arch/x86/mm/mem_encrypt_amd.c @@ -13,11 +13,13 @@ #include <linux/dma-direct.h> #include <linux/swiotlb.h> #include <linux/mem_encrypt.h> +#include <linux/string.h> #include <linux/device.h> #include <linux/kernel.h> #include <linux/bitops.h> #include <linux/dma-mapping.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h>
#include <asm/tlbflush.h> #include <asm/fixmap.h> @@ -283,8 +285,12 @@ static void enc_dec_hypercall(unsigned long vaddr, unsigned long size, bool enc) #endif }
-static int amd_enc_status_change_prepare(unsigned long vaddr, int npages, bool enc) +static int amd_enc_status_change_prepare(unsigned long vaddr, int npages, + bool enc, unsigned int flags) { + if (!enc && (flags & CC_SHARED_ZERO)) + memset((void *)vaddr, 0, (size_t)npages << PAGE_SHIFT); + /* * To maintain the security guarantees of SEV-SNP guests, make sure * to invalidate the memory before encryption attribute is cleared. @@ -296,7 +302,8 @@ static int amd_enc_status_change_prepare(unsigned long vaddr, int npages, bool e }
/* Return true unconditionally: return value doesn't matter for the SEV side */ -static int amd_enc_status_change_finish(unsigned long vaddr, int npages, bool enc) +static int amd_enc_status_change_finish(unsigned long vaddr, int npages, bool enc, + unsigned int flags) { /* * After memory is mapped encrypted in the page table, validate it diff --git a/arch/x86/mm/pat/set_memory.c b/arch/x86/mm/pat/set_memory.c index 4652487b5572..81be379f43b1 100644 --- a/arch/x86/mm/pat/set_memory.c +++ b/arch/x86/mm/pat/set_memory.c @@ -2420,7 +2420,8 @@ int set_memory_global(unsigned long addr, int numpages) * __set_memory_enc_pgtable() is used for the hypervisors that get * informed about "encryption" status via page tables. */ -static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc) +static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc, + unsigned int flags) { pgprot_t empty = __pgprot(0); struct cpa_data cpa; @@ -2446,7 +2447,8 @@ static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc) cpa_flush(&cpa, x86_platform.guest.enc_cache_flush_required());
/* Notify hypervisor that we are about to set/clr encryption attribute. */ - ret = x86_platform.guest.enc_status_change_prepare(addr, numpages, enc); + ret = x86_platform.guest.enc_status_change_prepare(addr, numpages, enc, + flags); if (ret) goto vmm_fail;
@@ -2465,7 +2467,8 @@ static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc) return ret;
/* Notify hypervisor that we have successfully set/clr encryption attribute. */ - ret = x86_platform.guest.enc_status_change_finish(addr, numpages, enc); + ret = x86_platform.guest.enc_status_change_finish(addr, numpages, enc, + flags); if (ret) goto vmm_fail;
@@ -2506,7 +2509,8 @@ bool set_memory_enc_stop_conversion(void) return true; }
-static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc) +static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc, + unsigned int flags) { int ret = 0;
@@ -2514,7 +2518,7 @@ static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc) if (!down_read_trylock(&mem_enc_lock)) return -EBUSY;
- ret = __set_memory_enc_pgtable(addr, numpages, enc); + ret = __set_memory_enc_pgtable(addr, numpages, enc, flags);
up_read(&mem_enc_lock); } @@ -2524,13 +2528,13 @@ static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc)
int set_memory_encrypted(unsigned long addr, int numpages) { - return __set_memory_enc_dec(addr, numpages, true); + return __set_memory_enc_dec(addr, numpages, true, 0); } EXPORT_SYMBOL_GPL(set_memory_encrypted);
-int set_memory_decrypted(unsigned long addr, int numpages) +int set_memory_decrypted(unsigned long addr, int numpages, unsigned int flags) { - return __set_memory_enc_dec(addr, numpages, false); + return __set_memory_enc_dec(addr, numpages, false, flags); } EXPORT_SYMBOL_GPL(set_memory_decrypted);
diff --git a/arch/x86/realmode/init.c b/arch/x86/realmode/init.c index 694d80a5c68e..1e15fb863927 100644 --- a/arch/x86/realmode/init.c +++ b/arch/x86/realmode/init.c @@ -111,7 +111,8 @@ static void __init setup_real_mode(void) * successfully. This is not needed for SEV. */ if (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT)) - set_memory_decrypted((unsigned long)base, size >> PAGE_SHIFT); + set_memory_decrypted((unsigned long)base, size >> PAGE_SHIFT, + 0);
memcpy(base, real_mode_blob, size);
diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c index 7e4cc6f55237..09f7ac96e475 100644 --- a/drivers/hv/channel.c +++ b/drivers/hv/channel.c @@ -474,7 +474,7 @@ static int __vmbus_establish_gpadl(struct vmbus_channel *channel, * on the free list. */ ret = set_memory_decrypted((unsigned long)kbuffer, - PFN_UP(size)); + PFN_UP(size), 0); if (ret) { dev_warn(&channel->device_obj->device, "Failed to set host visibility for new GPADL %d.\n", @@ -727,7 +727,7 @@ void *vmbus_alloc_buffer(struct vmbus_channel *channel, }
ret = set_memory_decrypted((unsigned long)page_address(page), - 1U << order); + 1U << order, 0); if (ret) { /* * set_memory_decrypted() failed; the page state is diff --git a/drivers/hv/connection.c b/drivers/hv/connection.c index 1ab3581b096a..cc9f73903c2f 100644 --- a/drivers/hv/connection.c +++ b/drivers/hv/connection.c @@ -13,6 +13,7 @@ #include <linux/sched.h> #include <linux/wait.h> #include <linux/delay.h> +#include <linux/cc_platform.h> #include <linux/mm.h> #include <linux/module.h> #include <linux/slab.h> @@ -263,29 +264,27 @@ int vmbus_connect(void) goto cleanup; }
- ret = set_memory_decrypted((unsigned long) - vmbus_connection.monitor_pages[0], 1); - ret |= set_memory_decrypted((unsigned long) - vmbus_connection.monitor_pages[1], 1); - if (ret) { - /* - * If set_memory_decrypted() fails, the encryption state - * of the memory is unknown. So leak the memory instead - * of risking returning decrypted memory to the free list. - * For simplicity, always handle both pages the same. - */ - vmbus_connection.monitor_pages[0] = NULL; - vmbus_connection.monitor_pages[1] = NULL; - goto cleanup; + if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) { + ret = set_memory_decrypted((unsigned long)vmbus_connection.monitor_pages[0], + 1, CC_SHARED_ZERO); + ret |= set_memory_decrypted((unsigned long)vmbus_connection.monitor_pages[1], + 1, CC_SHARED_ZERO); + if (ret) { + /* + * If set_memory_decrypted() fails, the encryption state + * of the memory is unknown. So leak the memory instead + * of risking returning decrypted memory to the free list. + * For simplicity, always handle both pages the same. + */ + vmbus_connection.monitor_pages[0] = NULL; + vmbus_connection.monitor_pages[1] = NULL; + goto cleanup; + } + } else { + memset(vmbus_connection.monitor_pages[0], 0, HV_HYP_PAGE_SIZE); + memset(vmbus_connection.monitor_pages[1], 0, HV_HYP_PAGE_SIZE); }
- /* - * Set_memory_decrypted() will change the memory contents if - * decryption occurs, so zero monitor pages here. - */ - memset(vmbus_connection.monitor_pages[0], 0x00, HV_HYP_PAGE_SIZE); - memset(vmbus_connection.monitor_pages[1], 0x00, HV_HYP_PAGE_SIZE); - msginfo = kzalloc(sizeof(*msginfo) + sizeof(struct vmbus_channel_initiate_contact), GFP_KERNEL); diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c index fe50090dcc01..f675fe90b78d 100644 --- a/drivers/hv/hv.c +++ b/drivers/hv/hv.c @@ -123,12 +123,14 @@ static int hv_alloc_page(void **page, bool decrypt, const char *note) if (!*page) return -ENOMEM;
- if (decrypt) - ret = set_memory_decrypted((unsigned long)*page, 1); - if (ret) - goto failed; - - memset(*page, 0, PAGE_SIZE); + if (decrypt) { + ret = set_memory_decrypted((unsigned long)*page, 1, + CC_SHARED_ZERO); + if (ret) + goto failed; + } else { + memset(*page, 0, PAGE_SIZE); + } return 0;
failed: diff --git a/drivers/hv/hv_common.c b/drivers/hv/hv_common.c index 31256cb22b39..84c950bd82b0 100644 --- a/drivers/hv/hv_common.c +++ b/drivers/hv/hv_common.c @@ -500,13 +500,13 @@ int hv_common_cpu_init(unsigned int cpu)
if (!ms_hyperv.paravisor_present && (hv_isolation_type_snp() || hv_isolation_type_tdx())) { - ret = set_memory_decrypted((unsigned long)mem, pgcount); + ret = set_memory_decrypted((unsigned long)mem, + pgcount, + CC_SHARED_ZERO); if (ret) { /* It may be unsafe to free 'mem' */ return ret; } - - memset(mem, 0x00, pgcount * HV_HYP_PAGE_SIZE); }
/* diff --git a/drivers/ptp/ptp_kvm_x86.c b/drivers/ptp/ptp_kvm_x86.c index 6cea4fe39bcf..9b0558af9a8e 100644 --- a/drivers/ptp/ptp_kvm_x86.c +++ b/drivers/ptp/ptp_kvm_x86.c @@ -34,7 +34,8 @@ int kvm_arch_ptp_init(void) return -ENOMEM;
clock_pair = page_address(p); - ret = set_memory_decrypted((unsigned long)clock_pair, 1); + ret = set_memory_decrypted((unsigned long)clock_pair, 1, + CC_SHARED_ZERO); if (ret) { __free_page(p); clock_pair = NULL; diff --git a/drivers/virt/coco/pkvm-guest/arm-pkvm-guest.c b/drivers/virt/coco/pkvm-guest/arm-pkvm-guest.c index 26fe9c3f22e3..87b6dbb468de 100644 --- a/drivers/virt/coco/pkvm-guest/arm-pkvm-guest.c +++ b/drivers/virt/coco/pkvm-guest/arm-pkvm-guest.c @@ -9,10 +9,12 @@
#include <linux/arm-smccc.h> #include <linux/array_size.h> +#include <linux/cc_shared.h> #include <linux/io.h> #include <linux/mem_encrypt.h> #include <linux/mm.h> #include <linux/pgtable.h> +#include <linux/string.h>
#include <asm/hypervisor.h>
@@ -59,8 +61,12 @@ static int pkvm_set_memory_encrypted(unsigned long addr, int numpages) addr, numpages); }
-static int pkvm_set_memory_decrypted(unsigned long addr, int numpages) +static int pkvm_set_memory_decrypted(unsigned long addr, int numpages, + unsigned int flags) { + if (flags & CC_SHARED_ZERO) + memset((void *)addr, 0, (size_t)numpages << PAGE_SHIFT); + return __set_memory_range(ARM_SMCCC_VENDOR_HYP_KVM_MEM_SHARE_FUNC_ID, addr, numpages); } diff --git a/drivers/virt/coco/sev-guest/sev-guest.c b/drivers/virt/coco/sev-guest/sev-guest.c index 935537a41469..3943c163965d 100644 --- a/drivers/virt/coco/sev-guest/sev-guest.c +++ b/drivers/virt/coco/sev-guest/sev-guest.c @@ -216,7 +216,8 @@ static int get_ext_report(struct snp_guest_dev *snp_dev, struct snp_guest_reques return -ENOMEM;
pfn = PHYS_PFN(virt_to_phys(req.certs_data)); - ret = set_memory_decrypted((unsigned long)req.certs_data, npages); + ret = set_memory_decrypted((unsigned long)req.certs_data, npages, + CC_SHARED_ZERO); if (ret) { pr_err("failed to mark page shared, ret=%d\n", ret); snp_leak_pages(pfn, npages); diff --git a/drivers/virt/coco/tdx-guest/tdx-guest.c b/drivers/virt/coco/tdx-guest/tdx-guest.c index d0303e31e816..db898564cfcf 100644 --- a/drivers/virt/coco/tdx-guest/tdx-guest.c +++ b/drivers/virt/coco/tdx-guest/tdx-guest.c @@ -232,7 +232,7 @@ static void *alloc_quote_buf(void) if (!addr) return NULL;
- if (set_memory_decrypted((unsigned long)addr, count)) + if (set_memory_decrypted((unsigned long)addr, count, CC_SHARED_ZERO)) return NULL;
return addr; diff --git a/include/linux/cc_shared.h b/include/linux/cc_shared.h index 5f8db7c468c5..35be90246b88 100644 --- a/include/linux/cc_shared.h +++ b/include/linux/cc_shared.h @@ -2,11 +2,15 @@ #ifndef _LINUX_CC_SHARED_H #define _LINUX_CC_SHARED_H
+#include <linux/bits.h> #include <linux/gfp_types.h> #include <linux/types.h>
struct page;
+/* Zero the range at an architecture-appropriate point while sharing it. */ +#define CC_SHARED_ZERO BIT(0) + struct cc_shared_pages { struct page *page; size_t shared_size; @@ -28,7 +32,7 @@ size_t arch_cc_shared_granule_size(void); size_t cc_shared_granule_size(void); int cc_shared_calc_layout(size_t requested, struct cc_shared_layout *layout); bool cc_shared_range_valid(phys_addr_t base, size_t size); -int cc_make_shared(void *addr, size_t size); +int cc_make_shared(void *addr, size_t size, unsigned int flags); int cc_make_private(void *addr, size_t size); int alloc_cc_shared_pages_node(int nid, gfp_t gfp, size_t requested, struct cc_shared_pages *mem); diff --git a/include/linux/set_memory.h b/include/linux/set_memory.h index 3030d9245f5a..a52713f3510c 100644 --- a/include/linux/set_memory.h +++ b/include/linux/set_memory.h @@ -5,6 +5,8 @@ #ifndef _LINUX_SET_MEMORY_H_ #define _LINUX_SET_MEMORY_H_
+#include <linux/cc_shared.h> + #ifdef CONFIG_ARCH_HAS_SET_MEMORY #include <asm/set_memory.h> #else @@ -78,7 +80,8 @@ static inline int set_memory_encrypted(unsigned long addr, int numpages) return 0; }
-static inline int set_memory_decrypted(unsigned long addr, int numpages) +static inline int set_memory_decrypted(unsigned long addr, int numpages, + unsigned int flags) { return 0; } diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c index d293198384c3..5b557ee27fd4 100644 --- a/kernel/dma/direct.c +++ b/kernel/dma/direct.c @@ -81,11 +81,12 @@ bool dma_coherent_ok(struct device *dev, phys_addr_t phys, size_t size) min_not_zero(dev->coherent_dma_mask, dev->bus_dma_limit); }
-static int dma_set_decrypted(struct device *dev, void *vaddr, size_t size) +static int dma_set_decrypted(struct device *dev, void *vaddr, size_t size, + unsigned int flags) { int ret;
- ret = cc_make_shared(vaddr, size); + ret = cc_make_shared(vaddr, size, flags); if (ret) pr_warn_ratelimited("leaking DMA memory that can't be decrypted\n"); return ret; @@ -213,7 +214,8 @@ void *dma_direct_alloc(struct device *dev, size_t size, if (force_dma_unencrypted(dev)) attrs |= __DMA_ATTR_ALLOC_CC_SHARED;
- if (attrs & __DMA_ATTR_ALLOC_CC_SHARED) { + mark_mem_decrypt = attrs & __DMA_ATTR_ALLOC_CC_SHARED; + if (mark_mem_decrypt) { /* * Unencrypted/shared DMA requires a linear-mapped buffer * address to look up the PFN and set architecture-required PFN @@ -221,7 +223,6 @@ void *dma_direct_alloc(struct device *dev, size_t size, * allocation. */ allow_highmem = false; - mark_mem_decrypt = true; }
size = PAGE_ALIGN(size); @@ -315,7 +316,7 @@ void *dma_direct_alloc(struct device *dev, size_t size, void *lm_addr;
lm_addr = page_address(page); - if (dma_set_decrypted(dev, lm_addr, size)) + if (dma_set_decrypted(dev, lm_addr, size, CC_SHARED_ZERO)) goto out_leak_pages; }
@@ -334,7 +335,9 @@ void *dma_direct_alloc(struct device *dev, size_t size, cpu_addr = page_address(page); }
- memset(cpu_addr, 0, size); + /* Zero after remapping because the page may be in HighMem. */ + if (!mark_mem_decrypt) + memset(cpu_addr, 0, size);
if (set_uncached) { void *uncached_cpu_addr; @@ -452,10 +455,13 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size, unsigned int align_order = 0; struct page *page; void *cpu_addr; + bool mark_mem_decrypt;
if (force_dma_unencrypted(dev)) attrs |= __DMA_ATTR_ALLOC_CC_SHARED;
+ mark_mem_decrypt = attrs & __DMA_ATTR_ALLOC_CC_SHARED; + if ((attrs & __DMA_ATTR_ALLOC_CC_SHARED) && dma_direct_use_pool(dev, gfp)) return dma_direct_alloc_from_pool(dev, size, dma_handle, &cpu_addr, gfp, attrs); @@ -466,10 +472,11 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size, return NULL;
cpu_addr = page_address(page); + mark_mem_decrypt = false; goto setup_page; }
- if (attrs & __DMA_ATTR_ALLOC_CC_SHARED) { + if (mark_mem_decrypt) { if (cc_shared_calc_layout(size, &layout)) return NULL; size = layout.shared_size; @@ -481,11 +488,13 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size, return NULL;
cpu_addr = page_address(page); - if ((attrs & __DMA_ATTR_ALLOC_CC_SHARED) && - dma_set_decrypted(dev, cpu_addr, size)) - goto out_leak_pages; setup_page: - memset(cpu_addr, 0, size); + if (mark_mem_decrypt) { + if (dma_set_decrypted(dev, cpu_addr, size, CC_SHARED_ZERO)) + goto out_leak_pages; + } else { + memset(cpu_addr, 0, size); + } *dma_handle = phys_to_dma_direct(dev, page_to_phys(page), attrs & __DMA_ATTR_ALLOC_CC_SHARED); return page; diff --git a/kernel/dma/pool.c b/kernel/dma/pool.c index 651d3a99c574..4298d5fddf57 100644 --- a/kernel/dma/pool.c +++ b/kernel/dma/pool.c @@ -138,7 +138,7 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, * shrink so no re-encryption occurs in dma_direct_free(). */ if (dma_pool->cc_shared) { - ret = cc_make_shared(page_to_virt(page), pool_size); + ret = cc_make_shared(page_to_virt(page), pool_size, 0); if (ret) { leak_pages = true; goto remove_mapping; diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index 9577a8807b07..281873ee8fe6 100644 --- a/kernel/dma/swiotlb.c +++ b/kernel/dma/swiotlb.c @@ -383,12 +383,10 @@ void __init swiotlb_update_mem_attributes(void) if (io_tlb_default_mem.cc_shared) { int ret;
- ret = cc_make_shared(mem->vaddr, bytes); + ret = cc_make_shared(mem->vaddr, bytes, CC_SHARED_ZERO); if (ret) { pr_warn("Failed to decrypt default memory pool, disabling it\n"); swiotlb_mark_pool_used(mem); - } else { - memset(mem->vaddr, 0, bytes); } } } @@ -642,7 +640,7 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask, goto error_slots;
if (io_tlb_default_mem.cc_shared) { - rc = cc_make_shared(vstart, nslabs << IO_TLB_SHIFT); + rc = cc_make_shared(vstart, nslabs << IO_TLB_SHIFT, 0); if (rc) { leak_pages = true; goto error_decrypt; @@ -746,7 +744,7 @@ static struct page *alloc_dma_pages(gfp_t gfp, size_t bytes, }
vaddr = phys_to_virt(paddr); - if (cc_shared && cc_make_shared(vaddr, bytes)) + if (cc_shared && cc_make_shared(vaddr, bytes, 0)) goto error; return page;
@@ -2069,7 +2067,8 @@ static int rmem_swiotlb_device_init(struct reserved_mem *rmem, int ret;
mem->cc_shared = true; - ret = cc_make_shared(phys_to_virt(rmem->base), rmem->size); + ret = cc_make_shared(phys_to_virt(rmem->base), + rmem->size, 0); if (ret) { dev_err(dev, "Failed to decrypt restricted DMA pool\n"); kfree(pool->areas); diff --git a/mm/cc_shared.c b/mm/cc_shared.c index 3e33681218f1..586a82116ba6 100644 --- a/mm/cc_shared.c +++ b/mm/cc_shared.c @@ -3,6 +3,7 @@ * Copyright (C) 2026 ARM Ltd. */ #include <linux/align.h> +#include <linux/cc_platform.h> #include <linux/cc_shared.h> #include <linux/errno.h> #include <linux/export.h> @@ -76,14 +77,17 @@ static int cc_validate_transition(void *addr, size_t size) return 0; }
-int cc_make_shared(void *addr, size_t size) +int cc_make_shared(void *addr, size_t size, unsigned int flags) { int ret = cc_validate_transition(addr, size);
if (ret) return ret; + if (flags & ~CC_SHARED_ZERO) + return -EINVAL;
- return set_memory_decrypted((unsigned long)addr, size >> PAGE_SHIFT); + return set_memory_decrypted((unsigned long)addr, size >> PAGE_SHIFT, + flags); }
int cc_make_private(void *addr, size_t size) @@ -96,8 +100,9 @@ int cc_make_private(void *addr, size_t size) return set_memory_encrypted((unsigned long)addr, size >> PAGE_SHIFT); }
-int alloc_cc_shared_pages_node(int nid, gfp_t gfp, - size_t requested, struct cc_shared_pages *mem) +static int __alloc_cc_shared_pages_node(int nid, gfp_t gfp, + size_t requested, + struct cc_shared_pages *mem) { struct cc_shared_layout layout; struct page *page; @@ -105,9 +110,6 @@ int alloc_cc_shared_pages_node(int nid, gfp_t gfp, bool zero = gfp & __GFP_ZERO; int ret;
- if (!mem) - return -EINVAL; - ret = cc_shared_calc_layout(requested, &layout); if (ret) return ret; @@ -118,7 +120,8 @@ int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
/* * State transitions require a linear-map address and may modify memory. - * Allocate from low memory and defer requested zeroing until afterwards. + * Allocate from low memory and let the architecture place requested + * zeroing at the appropriate point in the transition. */ gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO); if (nid == NUMA_NO_NODE) @@ -128,7 +131,8 @@ int alloc_cc_shared_pages_node(int nid, gfp_t gfp, if (!page) return -ENOMEM;
- ret = cc_make_shared(page_address(page), layout.shared_size); + ret = cc_make_shared(page_address(page), layout.shared_size, + zero ? CC_SHARED_ZERO : 0); if (ret) { if (!cc_make_private(page_address(page), layout.shared_size)) __free_pages(page, order); @@ -138,13 +142,39 @@ int alloc_cc_shared_pages_node(int nid, gfp_t gfp, return ret; }
- if (zero) - memset(page_address(page), 0, layout.shared_size); - mem->page = page; mem->shared_size = layout.shared_size; return 0; } + +int alloc_cc_shared_pages_node(int nid, gfp_t gfp, + size_t requested, + struct cc_shared_pages *mem) +{ + struct page *page; + unsigned int order; + + if (!mem || !requested) + return -EINVAL; + + if (cc_platform_has(CC_ATTR_MEM_ENCRYPT)) + return __alloc_cc_shared_pages_node(nid, gfp, requested, mem); + + order = get_order(requested); + if (order > MAX_PAGE_ORDER) + return -EINVAL; + + if (nid == NUMA_NO_NODE) + page = alloc_pages(gfp, order); + else + page = alloc_pages_node(nid, gfp, order); + if (!page) + return -ENOMEM; + + mem->page = page; + mem->shared_size = requested; + return 0; +} EXPORT_SYMBOL_GPL(alloc_cc_shared_pages_node);
int alloc_cc_shared_pages(gfp_t gfp, @@ -159,7 +189,8 @@ void free_cc_shared_pages(struct cc_shared_pages *mem) if (!mem || !mem->page) return;
- if (cc_make_private(page_address(mem->page), mem->shared_size)) { + if (cc_platform_has(CC_ATTR_MEM_ENCRYPT) && + cc_make_private(page_address(mem->page), mem->shared_size)) { pr_warn_ratelimited("leaking %zu bytes that cannot be made private\n", mem->shared_size); return;
Catalin Marinas catalin.marinas@arm.com writes:
On Wed, Sep 23, 2026 at 11:23:27AM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
I don't think every architecture or platform can unconditionally zero memory in set_memory_decrypted(). Some callers may need to share valid contents with the host.
Is there any? That would be a bad assumptions in the caller. Most set_memory_* backends don't preserve the content as they change the encryption key. So properly written code shouldn't rely on this unless it knows specifically it's only running on pKVM for example. The only use-case I see to avoid explicit zeroing is when the caller doesn't care about the page initialisation and wants to save some cycles. The encryption key change would take care of the security aspect.
I checked this, and you are right. We cannot expect the contents to remain valid across sharing; set_memory_decrypted() is destructive in that sense. Since pKVM does not rely on memory encryption, it needs to zero the memory unconditionally in set_memory_decrypted() to avoid exposing existing guest data. Other CoCo implementations may omit the memset(0). We still need to zero the memory when the caller requests __GFP_ZERO, where the memory location is expected to be zero.
We could either zero the memory unconditionally or pass a flag to allow this micro-optimization. We would also need to audit all call paths to avoid redundant zeroing after set_memory_decrypted(). Let me know if you have a preference for either approach.
-aneesh
On Wed, Sep 23, 2026 at 03:58:25PM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Wed, Sep 23, 2026 at 11:23:27AM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:36PM +0530, Aneesh Kumar K.V (Arm) wrote:
+int alloc_cc_shared_pages_node(int nid, gfp_t gfp,
size_t requested, struct cc_shared_pages *mem)+{
- struct cc_shared_layout layout;
- struct page *page;
- unsigned int order;
- bool zero = gfp & __GFP_ZERO;
- int ret;
- if (!mem)
return -EINVAL;- ret = cc_shared_calc_layout(requested, &layout);
- if (ret)
return ret;- order = get_order(layout.shared_size);
- if (order > MAX_PAGE_ORDER)
return -EINVAL;- /*
* State transitions require a linear-map address and may modify memory.* Allocate from low memory and defer requested zeroing until afterwards.*/- gfp &= ~(__GFP_HIGHMEM | __GFP_ZERO);
- if (nid == NUMA_NO_NODE)
page = alloc_pages(gfp, order);- else
page = alloc_pages_node(nid, gfp, order);- if (!page)
return -ENOMEM;- ret = cc_make_shared(page_address(page), layout.shared_size);
- if (ret) {
if (!cc_make_private(page_address(page), layout.shared_size))__free_pages(page, order);elsepr_warn_ratelimited("leaking %zu bytes with uncertain shared state\n",layout.shared_size);return ret;- }
- if (zero)
memset(page_address(page), 0, layout.shared_size);Does the memset() post sharing logic work for pKVM as well? If nothing clears it, we have a small window where guest data is leaked to the host.
Is there a case where we *do not* need the memory cleared? If not, maybe we can move the logic in the arch set_memory_decrypted().
I don't think every architecture or platform can unconditionally zero memory in set_memory_decrypted(). Some callers may need to share valid contents with the host.
Is there any? That would be a bad assumptions in the caller. Most set_memory_* backends don't preserve the content as they change the encryption key. So properly written code shouldn't rely on this unless it knows specifically it's only running on pKVM for example. The only use-case I see to avoid explicit zeroing is when the caller doesn't care about the page initialisation and wants to save some cycles. The encryption key change would take care of the security aspect.
I checked this, and you are right. We cannot expect the contents to remain valid across sharing; set_memory_decrypted() is destructive in that sense. Since pKVM does not rely on memory encryption, it needs to zero the memory unconditionally in set_memory_decrypted() to avoid exposing existing guest data. Other CoCo implementations may omit the memset(0). We still need to zero the memory when the caller requests __GFP_ZERO, where the memory location is expected to be zero.
We could either zero the memory unconditionally or pass a flag to allow this micro-optimization. We would also need to audit all call paths to avoid redundant zeroing after set_memory_decrypted(). Let me know if you have a preference for either approach.
The simplest is probably to always zero in the backend and ignore __GFP_ZERO to the allocator. But it's probably only marginally smaller than passing a CC_SHARED_ZERO flag down. Get codex to try this as well and compare the diffstat.
There's an argument for the flag approach from a performance perspective (avoid zeroing unnecessarily) but not sure how much it matters in practice.
On Wed, Sep 23, 2026 at 11:40:36AM +0100, Catalin Marinas wrote:
The simplest is probably to always zero in the backend and ignore __GFP_ZERO to the allocator. But it's probably only marginally smaller than passing a CC_SHARED_ZERO flag down. Get codex to try this as well and compare the diffstat.
For patch ordering I would convert to use the allocator first
The semantics of the new API should be clear
If you pass GFP_ZERO then the resulting allocated memory is zero
Otherwise the allocator does Whatever The Arch Needs to not leak private data out.
Once places are converted to the allocator lets go see what is left and ask why it is left and what API it actually needs.
I think HCH was right that nobody should be calling this in driver code, lets aim to unexport set_memory_decrypted as an ideal goal
Jason
Jason Gunthorpe jgg@ziepe.ca writes:
On Wed, Sep 23, 2026 at 11:40:36AM +0100, Catalin Marinas wrote:
The simplest is probably to always zero in the backend and ignore __GFP_ZERO to the allocator. But it's probably only marginally smaller than passing a CC_SHARED_ZERO flag down. Get codex to try this as well and compare the diffstat.
For patch ordering I would convert to use the allocator first
The semantics of the new API should be clear
If you pass GFP_ZERO then the resulting allocated memory is zero
Otherwise the allocator does Whatever The Arch Needs to not leak private data out.
ok
Once places are converted to the allocator lets go see what is left and ask why it is left and what API it actually needs.
I'm also considering requiring the address passed to cc_make_shared() to be in the linear map. This is currently required by both TDX and CCA, while AMD SNP appears to support vmalloc addresses. The only user of that vmalloc support is Hyper-V VMBus GPADL setup (vmbus_establish_gpadl()). How should the generic CoCo shared-memory allocator handle this?
-aneesh
On Wed, Sep 23, 2026 at 08:28:57PM +0530, Aneesh Kumar K.V wrote:
I'm also considering requiring the address passed to cc_make_shared() to be in the linear map. This is currently required by both TDX and CCA, while AMD SNP appears to support vmalloc addresses. The only user of that vmalloc support is Hyper-V VMBus GPADL setup (vmbus_establish_gpadl()). How should the generic CoCo shared-memory allocator handle this?
vmbus_establish_gpadl() is the the same wrong abstraction as the arch code. Get rid of it and use vmbus_establish_gpadl_caller_decrypted():
pdata->recv_buf = vzalloc(RECV_BUFFER_SIZE); if (!pdata->recv_buf) { ret = -ENOMEM; goto fail_free_ring; }
ret = vmbus_establish_gpadl(channel, pdata->recv_buf, RECV_BUFFER_SIZE, &pdata->recv_gpadl);
So you made an allocator that returns folios, now you just need to use that allocator to implement a kvzalloc wrapper. ARM can call vmap on decrypted memory with pgprot_decrypted(), right?
Or maybe this can use vmbus_alloc_buffer(), it already does it.
Jason
On Wed, Sep 23, 2026 at 10:42:39AM +0100, Catalin Marinas wrote:
On pKVM, we want set_memory_decrypted() to zero the buffer before the host can access it (I guess currently relying on __GFP_ZERO allocations). Since no cryptographic encryption takes place, there's not much point in memset'ing again after the operation as the content was already zeroed.
It sounds like this a pkvm unique thing, since RMM always makes the memory hidden, pkvm flow should do it too. Either inside its guest set_memory_decrypted() or inside the hypervisor like RMM.
Hypervisor doing it is the right place, hypervisor always needs to ensure guest memory is cleansed whenever the guest gives it up. Either via a private to shared, unmap or just exiting.
I don't think cc_make_shared() has the right information on how to safely and efficiently do the zeroing. That's only known to the set_memory_* backend. So you'd have to propagate the flag down.
+1
Jason
CCA guests must align shared/private memory transitions to the IPA state change alignment reported by the host. This alignment can be larger than PAGE_SIZE, so transitioning only a PAGE_SIZE-sized subrange may leave part of the host-managed granule in the wrong state.
Cache the RHI-reported alignment during Realm initialization and expose it through the arm64 memory encryption operations. Implement arch_cc_shared_granule_size() so the common CoCo shared-memory allocator uses the same constraint. Use PAGE_SIZE when no suitable backend is registered.
The common cc_make_shared() and cc_make_private() wrappers validate both the base and size against the architecture's shared granule size. Remove the PAGE_SIZE-only address checks from the lower-level arm64 transition hooks so that range validation remains in the common wrappers.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- arch/arm64/include/asm/mem_encrypt.h | 1 + arch/arm64/mm/mem_encrypt.c | 13 +++++++++++-- drivers/firmware/arm_rmm/rsi.c | 13 ++++++++++++- 3 files changed, 24 insertions(+), 3 deletions(-)
diff --git a/arch/arm64/include/asm/mem_encrypt.h b/arch/arm64/include/asm/mem_encrypt.h index ef8b8463e52b..32064338c9b0 100644 --- a/arch/arm64/include/asm/mem_encrypt.h +++ b/arch/arm64/include/asm/mem_encrypt.h @@ -10,6 +10,7 @@ struct device; struct arm64_mem_crypt_ops { int (*encrypt)(unsigned long addr, int numpages); int (*decrypt)(unsigned long addr, int numpages); + size_t (*shared_granule_size)(void); };
int arm64_mem_crypt_ops_register(const struct arm64_mem_crypt_ops *ops); diff --git a/arch/arm64/mm/mem_encrypt.c b/arch/arm64/mm/mem_encrypt.c index ee3c0ab04384..e6c059d20595 100644 --- a/arch/arm64/mm/mem_encrypt.c +++ b/arch/arm64/mm/mem_encrypt.c @@ -16,6 +16,7 @@ #include <linux/bug.h> #include <linux/compiler.h> #include <linux/err.h> +#include <linux/cc_shared.h> #include <linux/mm.h>
#include <asm/mem_encrypt.h> @@ -33,7 +34,7 @@ int arm64_mem_crypt_ops_register(const struct arm64_mem_crypt_ops *ops)
int set_memory_encrypted(unsigned long addr, int numpages) { - if (likely(!crypt_ops) || WARN_ON(!PAGE_ALIGNED(addr))) + if (likely(!crypt_ops)) return 0;
return crypt_ops->encrypt(addr, numpages); @@ -42,9 +43,17 @@ EXPORT_SYMBOL_GPL(set_memory_encrypted);
int set_memory_decrypted(unsigned long addr, int numpages) { - if (likely(!crypt_ops) || WARN_ON(!PAGE_ALIGNED(addr))) + if (likely(!crypt_ops)) return 0;
return crypt_ops->decrypt(addr, numpages); } EXPORT_SYMBOL_GPL(set_memory_decrypted); + +size_t arch_cc_shared_granule_size(void) +{ + if (likely(!crypt_ops) || !crypt_ops->shared_granule_size) + return PAGE_SIZE; + + return crypt_ops->shared_granule_size(); +} diff --git a/drivers/firmware/arm_rmm/rsi.c b/drivers/firmware/arm_rmm/rsi.c index 883f6091ff95..b59099d1f318 100644 --- a/drivers/firmware/arm_rmm/rsi.c +++ b/drivers/firmware/arm_rmm/rsi.c @@ -18,6 +18,7 @@ #include <asm/pgtable.h>
static struct realm_config config; +static size_t ipa_state_change_alignment = PAGE_SIZE;
unsigned long prot_ns_shared; EXPORT_SYMBOL(prot_ns_shared); @@ -154,9 +155,17 @@ static int realm_set_memory_decrypted(unsigned long addr, int numpages) return ret; }
+static size_t realm_shared_granule_size(void) +{ + if (is_realm_world()) + return ipa_state_change_alignment; + return PAGE_SIZE; +} + static const struct arm64_mem_crypt_ops realm_crypt_ops = { .encrypt = realm_set_memory_encrypted, .decrypt = realm_set_memory_decrypted, + .shared_granule_size = realm_shared_granule_size, };
static int realm_register_memory_enc_ops(void) @@ -166,7 +175,7 @@ static int realm_register_memory_enc_ops(void)
/* we need an aligned struct for rsi_host_call. slab is not yet ready */ static struct rsi_host_call hostconf_call __initdata; -static unsigned long __maybe_unused __init get_ipa_state_change_alignment(void) +static unsigned long __init get_ipa_state_change_alignment(void) { long ret; unsigned long shared_granule_size; @@ -217,6 +226,8 @@ void __init arm64_rsi_init(void) return; if (WARN_ON(rsi_get_realm_config(lm_alias(&config)))) return; + + ipa_state_change_alignment = get_ipa_state_change_alignment(); prot_ns_shared = __phys_to_pte_val(BIT(config.ipa_bits - 1));
if (arm64_ioremap_prot_hook_register(realm_ioremap_hook))
alloc_pages_node() resolves NUMA_NO_NODE to numa_mem_id() internally. In preparation for switching ITS allocations to the CoCo shared memory allocator, resolve the default node explicitly in its_alloc_pages().
The CoCo shared memory allocator interprets NUMA_NO_NODE as a request to apply the current task's memory policy. Passing numa_mem_id() instead preserves the existing nearest-memory-node placement across the allocator switch. This patch does not change the current allocation behavior.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- drivers/irqchip/irq-gic-v3-its.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c index e9807af23537..ac5507acb1a0 100644 --- a/drivers/irqchip/irq-gic-v3-its.c +++ b/drivers/irqchip/irq-gic-v3-its.c @@ -32,6 +32,7 @@ #include <linux/set_memory.h> #include <linux/slab.h> #include <linux/syscore_ops.h> +#include <linux/topology.h>
#include <linux/irqchip.h> #include <linux/irqchip/arm-gic-v3.h> @@ -236,7 +237,7 @@ static struct page *its_alloc_pages_node(int node, gfp_t gfp,
static struct page *its_alloc_pages(gfp_t gfp, unsigned int order) { - return its_alloc_pages_node(NUMA_NO_NODE, gfp, order); + return its_alloc_pages_node(numa_mem_id(), gfp, order); }
static void its_free_pages(void *addr, unsigned int order)
ITS command queues and tables are shared with the host in a confidential-computing guest. The ITS allocator currently assumes that the guest PAGE_SIZE is sufficient for both allocation alignment and private/shared state changes.
Route ITS page allocations through alloc_cc_shared_pages_node(). Keep the tracked table size based on the ITS-requested order. On release, reconstruct the common allocation metadata from the original ITS order and the architecture granule size so the same transitioned range and buddy order are used.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- drivers/irqchip/irq-gic-v3-its.c | 40 ++++++++++++-------------------- 1 file changed, 15 insertions(+), 25 deletions(-)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c index ac5507acb1a0..4814161df02e 100644 --- a/drivers/irqchip/irq-gic-v3-its.c +++ b/drivers/irqchip/irq-gic-v3-its.c @@ -8,6 +8,7 @@ #include <linux/acpi_iort.h> #include <linux/bitfield.h> #include <linux/bitmap.h> +#include <linux/cc_shared.h> #include <linux/cpu.h> #include <linux/crash_dump.h> #include <linux/delay.h> @@ -19,7 +20,6 @@ #include <linux/irqdomain.h> #include <linux/list.h> #include <linux/log2.h> -#include <linux/mem_encrypt.h> #include <linux/memblock.h> #include <linux/mm.h> #include <linux/msi.h> @@ -29,7 +29,6 @@ #include <linux/of_pci.h> #include <linux/of_platform.h> #include <linux/percpu.h> -#include <linux/set_memory.h> #include <linux/slab.h> #include <linux/syscore_ops.h> #include <linux/topology.h> @@ -214,25 +213,13 @@ static gfp_t gfp_flags_quirk; static struct page *its_alloc_pages_node(int node, gfp_t gfp, unsigned int order) { - struct page *page; - int ret = 0; - - page = alloc_pages_node(node, gfp | gfp_flags_quirk, order); - - if (!page) - return NULL; + struct cc_shared_pages mem;
- ret = set_memory_decrypted((unsigned long)page_address(page), - 1 << order); - /* - * If set_memory_decrypted() fails then we don't know what state the - * page is in, so we can't free it. Instead we leak it. - * set_memory_decrypted() will already have WARNed. - */ - if (ret) + if (alloc_cc_shared_pages_node(node, gfp | gfp_flags_quirk, + PAGE_SIZE << order, &mem)) return NULL;
- return page; + return mem.page; }
static struct page *its_alloc_pages(gfp_t gfp, unsigned int order) @@ -242,13 +229,15 @@ static struct page *its_alloc_pages(gfp_t gfp, unsigned int order)
static void its_free_pages(void *addr, unsigned int order) { - /* - * If the memory cannot be encrypted again then we must leak the pages. - * set_memory_encrypted() will already have WARNed. - */ - if (set_memory_encrypted((unsigned long)addr, 1 << order)) + struct cc_shared_layout layout; + struct cc_shared_pages mem; + + if (WARN_ON(cc_shared_calc_layout(PAGE_SIZE << order, &layout))) return; - free_pages((unsigned long)addr, order); + + mem.page = virt_to_page(addr); + mem.shared_size = layout.shared_size; + free_cc_shared_pages(&mem); }
static struct gen_pool *itt_pool; @@ -273,7 +262,8 @@ static void *itt_alloc_pool(int node, int size) if (!page) break;
- gen_pool_add(itt_pool, (unsigned long)page_address(page), PAGE_SIZE, node); + gen_pool_add(itt_pool, (unsigned long)page_address(page), + cc_shared_granule_size(), node); } while (!addr);
return (void *)addr;
Confidential-computing shared allocations need CMA backing allocations to be aligned to the architecture private/shared state-change granule size. Passing a DMA attribute into CMA would make a generic physical allocator interpret DMA and confidential-computing policy that has already been resolved by its caller.
Add an explicit minimum alignment order to dma_alloc_contiguous(). Preserve the existing size-derived alignment for ordinary callers by passing zero, and reject a request that exceeds CONFIG_CMA_ALIGNMENT. For supported requests, use the larger of the size-derived and requested orders, capped by the configured CMA alignment as before.
Also distinguish preferred and required alignment for dma_alloc_from_contiguous(). Existing callers pass zero as the required alignment and retain the current clamping behavior. Callers that require a minimum alignment can request it explicitly and receive NULL when CMA cannot satisfy it.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- arch/arm/mm/dma-mapping.c | 5 +++-- drivers/iommu/dma-iommu.c | 2 +- include/linux/dma-map-ops.h | 10 ++++++---- kernel/dma/contiguous.c | 33 +++++++++++++++++++++++---------- kernel/dma/direct.c | 2 +- kernel/dma/ops_helpers.c | 2 +- kernel/dma/pool.c | 2 +- kernel/kexec_file.c | 3 ++- 8 files changed, 38 insertions(+), 21 deletions(-)
diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c index 7761099dde9e..9714fcd51941 100644 --- a/arch/arm/mm/dma-mapping.c +++ b/arch/arm/mm/dma-mapping.c @@ -398,7 +398,8 @@ static void *__alloc_from_contiguous(struct device *dev, size_t size, struct page *page; void *ptr = NULL;
- page = dma_alloc_from_contiguous(dev, count, order, gfp & __GFP_NOWARN); + page = dma_alloc_from_contiguous(dev, count, order, 0, + gfp & __GFP_NOWARN); if (!page) return NULL;
@@ -866,7 +867,7 @@ static struct page **__iommu_alloc_buffer(struct device *dev, size_t size, unsigned long order = get_order(size); struct page *page;
- page = dma_alloc_from_contiguous(dev, count, order, + page = dma_alloc_from_contiguous(dev, count, order, 0, gfp & __GFP_NOWARN); if (!page) goto error; diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c index 58c624513cd4..59baf2687612 100644 --- a/drivers/iommu/dma-iommu.c +++ b/drivers/iommu/dma-iommu.c @@ -1627,7 +1627,7 @@ static void *iommu_dma_alloc_pages(struct device *dev, size_t size, struct page *page = NULL; void *cpu_addr;
- page = dma_alloc_contiguous(dev, alloc_size, gfp); + page = dma_alloc_contiguous(dev, alloc_size, gfp, 0); if (!page) page = alloc_pages_node(node, gfp, get_order(alloc_size)); if (!page) diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h index 8fae2b7deb20..5fccda7e5c69 100644 --- a/include/linux/dma-map-ops.h +++ b/include/linux/dma-map-ops.h @@ -99,10 +99,11 @@ int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base, phys_addr_t limit, struct cma **res_cma, bool fixed);
struct page *dma_alloc_from_contiguous(struct device *dev, size_t count, - unsigned int order, bool no_warn); + unsigned int order, unsigned int required_order, bool no_warn); bool dma_release_from_contiguous(struct device *dev, struct page *pages, int count); -struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp); +struct page *dma_alloc_contiguous(struct device *dev, size_t size, + gfp_t gfp, unsigned int align_order); void dma_free_contiguous(struct device *dev, struct page *page, size_t size);
void dma_contiguous_early_fixup(phys_addr_t base, unsigned long size); @@ -125,7 +126,8 @@ static inline int dma_contiguous_reserve_area(phys_addr_t size, return -ENOSYS; } static inline struct page *dma_alloc_from_contiguous(struct device *dev, - size_t count, unsigned int order, bool no_warn) + size_t count, unsigned int order, unsigned int required_order, + bool no_warn) { return NULL; } @@ -136,7 +138,7 @@ static inline bool dma_release_from_contiguous(struct device *dev, } /* Use fallback alloc() and free() when CONFIG_DMA_CMA=n */ static inline struct page *dma_alloc_contiguous(struct device *dev, size_t size, - gfp_t gfp) + gfp_t gfp, unsigned int align_order) { return NULL; } diff --git a/kernel/dma/contiguous.c b/kernel/dma/contiguous.c index 66093460584e..a3eb3299b817 100644 --- a/kernel/dma/contiguous.c +++ b/kernel/dma/contiguous.c @@ -357,19 +357,25 @@ int __init dma_contiguous_reserve_area(phys_addr_t size, phys_addr_t base, * dma_alloc_from_contiguous() - allocate pages from contiguous area * @dev: Pointer to device for which the allocation is performed. * @count: Requested number of pages. - * @align: Requested alignment of pages (in PAGE_SIZE order). + * @align: Preferred alignment of pages (in PAGE_SIZE order). + * @required_align: Minimum required alignment (in PAGE_SIZE order). * @no_warn: Avoid printing message about failed allocation. * * This function allocates memory buffer for specified device. It uses * device specific contiguous memory area if available or the default * global one. Requires architecture specific dev_get_cma_area() helper * function. + * + * The preferred alignment is capped at CONFIG_CMA_ALIGNMENT. Return NULL + * if the required alignment exceeds this limit. A required alignment of + * zero preserves the preferred-alignment clamping behavior. */ struct page *dma_alloc_from_contiguous(struct device *dev, size_t count, - unsigned int align, bool no_warn) + unsigned int align, unsigned int required_align, bool no_warn) { - if (align > CONFIG_CMA_ALIGNMENT) - align = CONFIG_CMA_ALIGNMENT; + if (required_align > CONFIG_CMA_ALIGNMENT) + return NULL; + align = min(max(align, required_align), CONFIG_CMA_ALIGNMENT);
return cma_alloc(dev_get_cma_area(dev), count, align, no_warn); } @@ -390,9 +396,14 @@ bool dma_release_from_contiguous(struct device *dev, struct page *pages, return cma_release(dev_get_cma_area(dev), pages, count); }
-static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp) +static struct page *cma_alloc_aligned(struct cma *cma, size_t size, + gfp_t gfp, unsigned int align_order) { - unsigned int align = min(get_order(size), CONFIG_CMA_ALIGNMENT); + unsigned int align; + + if (align_order > CONFIG_CMA_ALIGNMENT) + return NULL; + align = min(max(get_order(size), align_order), CONFIG_CMA_ALIGNMENT);
return cma_alloc(cma, size >> PAGE_SHIFT, align, gfp & __GFP_NOWARN); } @@ -402,6 +413,7 @@ static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp) * @dev: Pointer to device for which the allocation is performed. * @size: Requested allocation size. * @gfp: Allocation flags. + * @align_order: Minimum alignment as a power-of-two page order. * * tries to use device specific contiguous memory area if available, or it * tries to use per-numa cma, if the allocation fails, it will fallback to @@ -412,7 +424,8 @@ static struct page *cma_alloc_aligned(struct cma *cma, size_t size, gfp_t gfp) * there is no need to waste CMA pages for that kind; it also helps reduce * fragmentations. */ -struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) +struct page *dma_alloc_contiguous(struct device *dev, size_t size, + gfp_t gfp, unsigned int align_order) { #ifdef CONFIG_DMA_NUMA_CMA int nid = dev_to_node(dev); @@ -422,7 +435,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) if (!gfpflags_allow_blocking(gfp)) return NULL; if (dev->cma_area) - return cma_alloc_aligned(dev->cma_area, size, gfp); + return cma_alloc_aligned(dev->cma_area, size, gfp, align_order); if (size <= PAGE_SIZE) return NULL;
@@ -431,7 +444,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) struct cma *cma = dma_contiguous_numa_area[nid]; struct page *page; if (cma) { - page = cma_alloc_aligned(cma, size, gfp); + page = cma_alloc_aligned(cma, size, gfp, align_order); if (page) return page; } @@ -440,7 +453,7 @@ struct page *dma_alloc_contiguous(struct device *dev, size_t size, gfp_t gfp) if (!dma_contiguous_default_area) return NULL;
- return cma_alloc_aligned(dma_contiguous_default_area, size, gfp); + return cma_alloc_aligned(dma_contiguous_default_area, size, gfp, align_order); }
/** diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c index da665ca22d5c..d968a0c81e73 100644 --- a/kernel/dma/direct.c +++ b/kernel/dma/direct.c @@ -124,7 +124,7 @@ static struct page *__dma_direct_alloc_pages(struct device *dev, size_t size, WARN_ON_ONCE(!PAGE_ALIGNED(size));
gfp |= dma_direct_optimal_gfp_mask(dev, &phys_limit); - page = dma_alloc_contiguous(dev, size, gfp); + page = dma_alloc_contiguous(dev, size, gfp, 0); if (page) { if (dma_coherent_ok(dev, page_to_phys(page), size) && (allow_highmem || !PageHighMem(page))) diff --git a/kernel/dma/ops_helpers.c b/kernel/dma/ops_helpers.c index 6b5f9208d31c..8320cc0fada5 100644 --- a/kernel/dma/ops_helpers.c +++ b/kernel/dma/ops_helpers.c @@ -66,7 +66,7 @@ struct page *dma_common_alloc_pages(struct device *dev, size_t size, struct page *page; phys_addr_t phys;
- page = dma_alloc_contiguous(dev, size, gfp); + page = dma_alloc_contiguous(dev, size, gfp, 0); if (!page) page = alloc_pages_node(dev_to_node(dev), gfp, get_order(size)); if (!page) diff --git a/kernel/dma/pool.c b/kernel/dma/pool.c index 00f422a1e896..70b7f64b17ab 100644 --- a/kernel/dma/pool.c +++ b/kernel/dma/pool.c @@ -99,7 +99,7 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, pool_size = 1 << (PAGE_SHIFT + order); if (cma_in_zone(gfp)) page = dma_alloc_from_contiguous(NULL, 1 << order, - order, false); + order, 0, false); if (!page) page = alloc_pages(gfp | __GFP_NOWARN, order); } while (!page && order-- > 0); diff --git a/kernel/kexec_file.c b/kernel/kexec_file.c index 59fb9d71e9d8..2a337ce7264e 100644 --- a/kernel/kexec_file.c +++ b/kernel/kexec_file.c @@ -682,7 +682,8 @@ static int kexec_alloc_contig(struct kexec_buf *kbuf) if (kbuf->image->type == KEXEC_TYPE_CRASH) return -EPERM;
- p = dma_alloc_from_contiguous(NULL, nr_pages, get_order(kbuf->buf_align), true); + p = dma_alloc_from_contiguous(NULL, nr_pages, + get_order(kbuf->buf_align), 0, true); if (!p) return -ENOMEM;
On Mon, Sep 21, 2026 at 08:18:40PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -125,7 +126,8 @@ static inline int dma_contiguous_reserve_area(phys_addr_t size, return -ENOSYS; } static inline struct page *dma_alloc_from_contiguous(struct device *dev,
size_t count, unsigned int order, bool no_warn)
size_t count, unsigned int order, unsigned int required_order,bool no_warn){ return NULL; } @@ -136,7 +138,7 @@ static inline bool dma_release_from_contiguous(struct device *dev, } /* Use fallback alloc() and free() when CONFIG_DMA_CMA=n */ static inline struct page *dma_alloc_contiguous(struct device *dev, size_t size,
gfp_t gfp)
gfp_t gfp, unsigned int align_order)
Can we not just bake the alignment further down in these functions rather than getting the callers to pass the {required,align}_order?
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:40PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -125,7 +126,8 @@ static inline int dma_contiguous_reserve_area(phys_addr_t size, return -ENOSYS; } static inline struct page *dma_alloc_from_contiguous(struct device *dev,
size_t count, unsigned int order, bool no_warn)
size_t count, unsigned int order, unsigned int required_order,bool no_warn){ return NULL; } @@ -136,7 +138,7 @@ static inline bool dma_release_from_contiguous(struct device *dev, } /* Use fallback alloc() and free() when CONFIG_DMA_CMA=n */ static inline struct page *dma_alloc_contiguous(struct device *dev, size_t size,
gfp_t gfp)
gfp_t gfp, unsigned int align_order)Can we not just bake the alignment further down in these functions rather than getting the callers to pass the {required,align}_order?
But, we need this to be conditional on CoCo shared allocations. We could derive that from attrs and pass attrs instead of align_order, i.e. something like:
struct page *dma_alloc_from_contiguous(struct device *dev, size_t count, - unsigned int align, unsigned int required_align, bool no_warn) + unsigned int align, unsigned long attrs, bool no_warn) { - if (required_align > CONFIG_CMA_ALIGNMENT) + if (!dma_contiguous_resolve_alignment(attrs, &align)) return NULL; - align = min(max(align, required_align), CONFIG_CMA_ALIGNMENT);
return cma_alloc(dev_get_cma_area(dev), count, align, no_warn); }
with
+static bool dma_contiguous_resolve_alignment(unsigned long attrs, + unsigned int *align) +{ + unsigned int required_align = 0; + + if (attrs & __DMA_ATTR_ALLOC_CC_SHARED) + required_align = get_order(cc_shared_granule_size()); + if (required_align > CONFIG_CMA_ALIGNMENT) + return false; + + *align = min(max(*align, required_align), CONFIG_CMA_ALIGNMENT); + return true; +} +
-aneesh
On Wed, Sep 23, 2026 at 05:19:52PM +0530, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:40PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -125,7 +126,8 @@ static inline int dma_contiguous_reserve_area(phys_addr_t size, return -ENOSYS; } static inline struct page *dma_alloc_from_contiguous(struct device *dev,
size_t count, unsigned int order, bool no_warn)
size_t count, unsigned int order, unsigned int required_order,bool no_warn){ return NULL; } @@ -136,7 +138,7 @@ static inline bool dma_release_from_contiguous(struct device *dev, } /* Use fallback alloc() and free() when CONFIG_DMA_CMA=n */ static inline struct page *dma_alloc_contiguous(struct device *dev, size_t size,
gfp_t gfp)
gfp_t gfp, unsigned int align_order)Can we not just bake the alignment further down in these functions rather than getting the callers to pass the {required,align}_order?
But, we need this to be conditional on CoCo shared allocations. We could derive that from attrs and pass attrs instead of align_order, i.e. something like:
struct page *dma_alloc_from_contiguous(struct device *dev, size_t count,
unsigned int align, unsigned int required_align, bool no_warn)
unsigned int align, unsigned long attrs, bool no_warn){
- if (required_align > CONFIG_CMA_ALIGNMENT)
- if (!dma_contiguous_resolve_alignment(attrs, &align)) return NULL;
- align = min(max(align, required_align), CONFIG_CMA_ALIGNMENT);
return cma_alloc(dev_get_cma_area(dev), count, align, no_warn); }
with
+static bool dma_contiguous_resolve_alignment(unsigned long attrs,
unsigned int *align)+{
- unsigned int required_align = 0;
- if (attrs & __DMA_ATTR_ALLOC_CC_SHARED)
required_align = get_order(cc_shared_granule_size());- if (required_align > CONFIG_CMA_ALIGNMENT)
return false;- *align = min(max(*align, required_align), CONFIG_CMA_ALIGNMENT);
- return true;
+}
Yes, I think this would work.
Atomic DMA allocations cannot perform a potentially sleeping private/shared state transition at allocation time. The atomic DMA pools avoid this by transitioning their backing allocations in a sleepable context and suballocating memory that is already shared.
Architectures may require these transitions to use a shared granule size larger than PAGE_SIZE. The existing fallback loop can reduce the backing allocation below the order required by that size, producing a range that cannot be safely transitioned.
For pools marked cc_shared, round the requested pool size up to a multiple of the shared granule size and prevent allocation fallback below the order required by that size. Pass this minimum order to dma_alloc_from_contiguous() as a required alignment so that CMA cannot silently clamp it. If CMA cannot satisfy the alignment, fall back to the buddy allocator. Non-shared pools pass zero and retain the existing CMA alignment policy.
Use the cc_make_shared() and cc_make_private() helpers to transition the backing allocation and validate its address and size.
Individual atomic allocations may remain smaller than the shared granule size because the backing allocation remains owned by the pool and in the shared state.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- kernel/dma/pool.c | 23 +++++++++++++++++------ 1 file changed, 17 insertions(+), 6 deletions(-)
diff --git a/kernel/dma/pool.c b/kernel/dma/pool.c index 70b7f64b17ab..651d3a99c574 100644 --- a/kernel/dma/pool.c +++ b/kernel/dma/pool.c @@ -4,12 +4,12 @@ * Copyright (C) 2020 Google LLC */ #include <linux/cma.h> +#include <linux/cc_shared.h> #include <linux/debugfs.h> #include <linux/dma-map-ops.h> #include <linux/dma-direct.h> #include <linux/init.h> #include <linux/genalloc.h> -#include <linux/set_memory.h> #include <linux/slab.h> #include <linux/workqueue.h> #include <linux/cc_platform.h> @@ -85,6 +85,8 @@ static bool cma_in_zone(gfp_t gfp) static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, gfp_t gfp) { + struct cc_shared_layout layout; + unsigned int min_order = 0; unsigned int order; struct page *page = NULL; bool leak_pages = false; @@ -92,6 +94,16 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, int ret = -ENOMEM; pgprot_t prot __maybe_unused;
+ if (dma_pool->cc_shared) { + ret = cc_shared_calc_layout(pool_size, &layout); + if (ret) + goto out; + pool_size = layout.shared_size; + min_order = get_order(layout.alignment); + if (min_order > MAX_PAGE_ORDER) + return -E2BIG; + } + /* Cannot allocate larger than MAX_PAGE_ORDER */ order = min(get_order(pool_size), MAX_PAGE_ORDER);
@@ -99,10 +111,10 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, pool_size = 1 << (PAGE_SHIFT + order); if (cma_in_zone(gfp)) page = dma_alloc_from_contiguous(NULL, 1 << order, - order, 0, false); + order, min_order, false); if (!page) page = alloc_pages(gfp | __GFP_NOWARN, order); - } while (!page && order-- > 0); + } while (!page && order-- > min_order); if (!page) goto out;
@@ -126,8 +138,7 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size, * shrink so no re-encryption occurs in dma_direct_free(). */ if (dma_pool->cc_shared) { - ret = set_memory_decrypted((unsigned long)page_to_virt(page), - 1 << order); + ret = cc_make_shared(page_to_virt(page), pool_size); if (ret) { leak_pages = true; goto remove_mapping; @@ -144,7 +155,7 @@ static int atomic_pool_expand(struct dma_gen_pool *dma_pool, size_t pool_size,
encrypt_mapping: if (dma_pool->cc_shared && - set_memory_encrypted((unsigned long)page_to_virt(page), 1 << order)) + cc_make_private(page_to_virt(page), pool_size)) leak_pages = true;
remove_mapping:
Use the common CoCo shared-memory geometry helpers for allocations backed directly by CMA or the page allocator. Round the backing allocation to a whole shared granule, pass the required alignment order through the DMA contiguous allocator, and transition the complete range through the common shared/private helpers.
Recompute the layout before freeing ordinary direct allocations so the transition back to private memory and dma_free_contiguous() cover exactly the range acquired by the allocation path. If restoring private state fails, retain the existing fail-safe behavior and leak the pages rather than returning potentially shared memory to the allocator.
This also applies the same rules to dma_direct_alloc_pages(), covering callers which require a struct page result rather than a CPU virtual address.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- kernel/dma/direct.c | 55 +++++++++++++++++++++++++++++++++++---------- 1 file changed, 43 insertions(+), 12 deletions(-)
diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c index d968a0c81e73..d293198384c3 100644 --- a/kernel/dma/direct.c +++ b/kernel/dma/direct.c @@ -11,10 +11,10 @@ #include <linux/scatterlist.h> #include <linux/pfn.h> #include <linux/vmalloc.h> -#include <linux/set_memory.h> #include <linux/slab.h> #include <linux/pci-p2pdma.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h>
#include "direct.h"
@@ -85,7 +85,7 @@ static int dma_set_decrypted(struct device *dev, void *vaddr, size_t size) { int ret;
- ret = set_memory_decrypted((unsigned long)vaddr, PFN_UP(size)); + ret = cc_make_shared(vaddr, size); if (ret) pr_warn_ratelimited("leaking DMA memory that can't be decrypted\n"); return ret; @@ -95,7 +95,7 @@ static int dma_set_encrypted(struct device *dev, void *vaddr, size_t size) { int ret;
- ret = set_memory_encrypted((unsigned long)vaddr, PFN_UP(size)); + ret = cc_make_private(vaddr, size); if (ret) pr_warn_ratelimited("leaking DMA memory that can't be re-encrypted\n"); return ret; @@ -115,7 +115,7 @@ static struct page *dma_direct_alloc_swiotlb(struct device *dev, size_t size, }
static struct page *__dma_direct_alloc_pages(struct device *dev, size_t size, - gfp_t gfp, bool allow_highmem) + gfp_t gfp, bool allow_highmem, unsigned int align_order) { int node = dev_to_node(dev); struct page *page; @@ -124,7 +124,7 @@ static struct page *__dma_direct_alloc_pages(struct device *dev, size_t size, WARN_ON_ONCE(!PAGE_ALIGNED(size));
gfp |= dma_direct_optimal_gfp_mask(dev, &phys_limit); - page = dma_alloc_contiguous(dev, size, gfp, 0); + page = dma_alloc_contiguous(dev, size, gfp, align_order); if (page) { if (dma_coherent_ok(dev, page_to_phys(page), size) && (allow_highmem || !PageHighMem(page))) @@ -184,7 +184,7 @@ static void *dma_direct_alloc_no_mapping(struct device *dev, size_t size, { struct page *page;
- page = __dma_direct_alloc_pages(dev, size, gfp & ~__GFP_ZERO, true); + page = __dma_direct_alloc_pages(dev, size, gfp & ~__GFP_ZERO, true, 0); if (!page) return NULL;
@@ -205,6 +205,8 @@ void *dma_direct_alloc(struct device *dev, size_t size, bool remap = false, set_uncached = false; bool mark_mem_decrypt = false; bool allow_highmem = true; + struct cc_shared_layout layout; + unsigned int align_order = 0; struct page *page; void *cpu_addr;
@@ -285,8 +287,16 @@ void *dma_direct_alloc(struct device *dev, size_t size, return NULL; }
+ if (mark_mem_decrypt) { + if (cc_shared_calc_layout(size, &layout)) + return NULL; + size = layout.shared_size; + align_order = get_order(layout.alignment); + } + /* we always manually zero the memory once we are done */ - page = __dma_direct_alloc_pages(dev, size, gfp & ~__GFP_ZERO, allow_highmem); + page = __dma_direct_alloc_pages(dev, size, gfp & ~__GFP_ZERO, + allow_highmem, align_order); if (!page) return NULL;
@@ -305,7 +315,7 @@ void *dma_direct_alloc(struct device *dev, size_t size, void *lm_addr;
lm_addr = page_address(page); - if (set_memory_decrypted((unsigned long)lm_addr, PFN_UP(size))) + if (dma_set_decrypted(dev, lm_addr, size)) goto out_leak_pages; }
@@ -362,6 +372,7 @@ void dma_direct_free(struct device *dev, size_t size, phys_addr_t phys; bool mark_mem_encrypted = false; struct io_tlb_pool *swiotlb_pool; + struct cc_shared_layout layout; unsigned int page_order = get_order(size);
/* @@ -406,6 +417,12 @@ void dma_direct_free(struct device *dev, size_t size, /* Swiotlb doesn't need a page attribute update on free */ mark_mem_encrypted = false;
+ if (mark_mem_encrypted) { + if (WARN_ON_ONCE(cc_shared_calc_layout(size, &layout))) + return; + size = layout.shared_size; + } + if (is_vmalloc_addr(cpu_addr)) { vunmap(cpu_addr); } else { @@ -417,10 +434,8 @@ void dma_direct_free(struct device *dev, size_t size, void *lm_addr;
lm_addr = phys_to_virt(phys); - if (set_memory_encrypted((unsigned long)lm_addr, PFN_UP(size))) { - pr_warn_ratelimited("leaking DMA memory that can't be re-encrypted\n"); + if (dma_set_encrypted(dev, lm_addr, size)) return; - } }
if (swiotlb_pool) @@ -433,6 +448,8 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size, dma_addr_t *dma_handle, enum dma_data_direction dir, gfp_t gfp) { unsigned long attrs = 0; + struct cc_shared_layout layout; + unsigned int align_order = 0; struct page *page; void *cpu_addr;
@@ -452,7 +469,14 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size, goto setup_page; }
- page = __dma_direct_alloc_pages(dev, size, gfp, false); + if (attrs & __DMA_ATTR_ALLOC_CC_SHARED) { + if (cc_shared_calc_layout(size, &layout)) + return NULL; + size = layout.shared_size; + align_order = get_order(layout.alignment); + } + + page = __dma_direct_alloc_pages(dev, size, gfp, false, align_order); if (!page) return NULL;
@@ -476,6 +500,7 @@ void dma_direct_free_pages(struct device *dev, size_t size, phys_addr_t phys; void *vaddr = page_address(page); struct io_tlb_pool *swiotlb_pool; + struct cc_shared_layout layout; /* * if the device had requested for an unencrypted buffer, * convert it to encrypted on free @@ -492,6 +517,12 @@ void dma_direct_free_pages(struct device *dev, size_t size, if (swiotlb_pool) mark_mem_encrypted = false;
+ if (mark_mem_encrypted) { + if (WARN_ON_ONCE(cc_shared_calc_layout(size, &layout))) + return; + size = layout.shared_size; + } + if (mark_mem_encrypted && dma_set_encrypted(dev, vaddr, size)) return;
Align and size the early memblock pool to the common CoCo shared granule. Use the same rounded extent when the pool is transitioned and when it is released. Re-zero the extent after a successful private-to-shared transition because the architecture operation may change memory contents. This deliberately leaves the slot count unchanged: any rounded tail belongs to the pool allocation but is not advertised as allocatable SWIOTLB space.
Replace the default and dynamic pool set_memory calls with the helpers so alignment validation and architecture dispatch are kept in one place. As before, pages are intentionally leaked if their private state cannot be restored safely.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- kernel/dma/swiotlb.c | 42 +++++++++++++++++++++++++----------------- 1 file changed, 25 insertions(+), 17 deletions(-)
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index ded7016a46a7..9928d75efc2d 100644 --- a/kernel/dma/swiotlb.c +++ b/kernel/dma/swiotlb.c @@ -22,6 +22,7 @@
#include <linux/cache.h> #include <linux/cc_platform.h> +#include <linux/cc_shared.h> #include <linux/ctype.h> #include <linux/debugfs.h> #include <linux/dma-direct.h> @@ -38,7 +39,6 @@ #include <linux/pfn.h> #include <linux/rculist.h> #include <linux/scatterlist.h> -#include <linux/set_memory.h> #include <linux/spinlock.h> #include <linux/string.h> #include <linux/swiotlb.h> @@ -369,16 +369,18 @@ void __init swiotlb_update_mem_attributes(void)
if (!mem->nslabs || mem->late_alloc) return; - bytes = PAGE_ALIGN(mem->nslabs << IO_TLB_SHIFT); + + bytes = ALIGN(mem->nslabs << IO_TLB_SHIFT, cc_shared_granule_size());
if (io_tlb_default_mem.cc_shared) { int ret;
- ret = set_memory_decrypted((unsigned long)mem->vaddr, - bytes >> PAGE_SHIFT); + ret = cc_make_shared(mem->vaddr, bytes); if (ret) { pr_warn("Failed to decrypt default memory pool, disabling it\n"); swiotlb_mark_pool_used(mem); + } else { + memset(mem->vaddr, 0, bytes); } } } @@ -436,8 +438,8 @@ static void __init *swiotlb_memblock_alloc(unsigned long nslabs, unsigned int flags, int (*remap)(void *tlb, unsigned long nslabs)) { - size_t bytes = PAGE_ALIGN(nslabs << IO_TLB_SHIFT); void *tlb; + size_t bytes = ALIGN(nslabs << IO_TLB_SHIFT, cc_shared_granule_size());
/* * By default allocate the bounce buffer memory from low memory, but @@ -445,9 +447,9 @@ static void __init *swiotlb_memblock_alloc(unsigned long nslabs, * memory encryption. */ if (flags & SWIOTLB_ANY) - tlb = memblock_alloc(bytes, PAGE_SIZE); + tlb = memblock_alloc(bytes, cc_shared_granule_size()); else - tlb = memblock_alloc_low(bytes, PAGE_SIZE); + tlb = memblock_alloc_low(bytes, cc_shared_granule_size());
if (!tlb) { pr_warn("%s: Failed to allocate %zu bytes tlb structure\n", @@ -456,7 +458,7 @@ static void __init *swiotlb_memblock_alloc(unsigned long nslabs, }
if (remap && remap(tlb, nslabs) < 0) { - memblock_free(tlb, PAGE_ALIGN(bytes)); + memblock_free(tlb, bytes); pr_warn("%s: Failed to remap %zu bytes\n", __func__, bytes); return NULL; } @@ -578,7 +580,7 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask, swiotlb_adjust_nareas(num_possible_cpus());
retry: - order = get_order(nslabs << IO_TLB_SHIFT); + order = get_order(ALIGN(nslabs << IO_TLB_SHIFT, cc_shared_granule_size())); nslabs = SLABS_PER_PAGE << order;
while ((SLABS_PER_PAGE << order) > IO_TLB_MIN_SLABS) { @@ -587,6 +589,8 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask, if (vstart) break; order--; + if (order < get_order(cc_shared_granule_size())) + break; nslabs = SLABS_PER_PAGE << order; retried = true; } @@ -626,8 +630,7 @@ int swiotlb_init_late(size_t size, gfp_t gfp_mask, goto error_slots;
if (io_tlb_default_mem.cc_shared) { - rc = set_memory_decrypted((unsigned long)vstart, - (nslabs << IO_TLB_SHIFT) >> PAGE_SHIFT); + rc = cc_make_shared(vstart, nslabs << IO_TLB_SHIFT); if (rc) { leak_pages = true; goto error_decrypt; @@ -667,11 +670,11 @@ void __init swiotlb_exit(void)
pr_info("tearing down default memory pool\n"); tbl_vaddr = (unsigned long)phys_to_virt(mem->start); - tbl_size = PAGE_ALIGN(mem->end - mem->start); + tbl_size = ALIGN(mem->end - mem->start, cc_shared_granule_size()); slots_size = PAGE_ALIGN(array_size(sizeof(*mem->slots), mem->nslabs));
if (io_tlb_default_mem.cc_shared) { - if (set_memory_encrypted(tbl_vaddr, tbl_size >> PAGE_SHIFT)) + if (cc_make_private((void *)tbl_vaddr, tbl_size)) leak_pages = true; }
@@ -711,12 +714,15 @@ void __init swiotlb_exit(void) static struct page *alloc_dma_pages(gfp_t gfp, size_t bytes, u64 phys_limit, unsigned long attrs) { - unsigned int order = get_order(bytes); bool cc_shared = attrs & __DMA_ATTR_ALLOC_CC_SHARED; + unsigned int order; struct page *page; phys_addr_t paddr; void *vaddr;
+ if (cc_shared) + bytes = ALIGN(bytes, cc_shared_granule_size()); + order = get_order(bytes); page = alloc_pages(gfp, order); if (!page) return NULL; @@ -728,13 +734,13 @@ static struct page *alloc_dma_pages(gfp_t gfp, size_t bytes, }
vaddr = phys_to_virt(paddr); - if (cc_shared && set_memory_decrypted((unsigned long)vaddr, PFN_UP(bytes))) + if (cc_shared && cc_make_shared(vaddr, bytes)) goto error; return page;
error: /* Intentional leak if pages cannot be encrypted again. */ - if (cc_shared && !set_memory_encrypted((unsigned long)vaddr, PFN_UP(bytes))) + if (cc_shared && !cc_make_private(vaddr, bytes)) __free_pages(page, order); return NULL; } @@ -807,9 +813,11 @@ static void swiotlb_free_tlb(void *vaddr, size_t bytes, bool cc_shared) dma_free_from_pool(NULL, vaddr, bytes)) return;
+ if (cc_shared) + bytes = ALIGN(bytes, cc_shared_granule_size()); /* Intentional leak if pages cannot be encrypted again. */ if (!cc_shared || - !set_memory_encrypted((unsigned long)vaddr, PFN_UP(bytes))) + !cc_make_private(vaddr, bytes)) __free_pages(virt_to_page(vaddr), get_order(bytes)); }
A restricted DMA pool is a firmware-described ownership boundary. Unlike a pool allocated by SWIOTLB itself, its base cannot be moved and its size cannot be rounded without claiming memory outside the reserved region.
Confidential-computing guests may require private/shared state changes to cover units larger than PAGE_SIZE. Passing a misaligned restricted region to set_memory_decrypted() can therefore fail.
Validate both the physical base and the complete reserved size with the common CoCo shared-range helper before allocating SWIOTLB metadata or performing a state transition. Perform the transition through cc_make_shared().
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- kernel/dma/swiotlb.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index 9928d75efc2d..cb67105b8812 100644 --- a/kernel/dma/swiotlb.c +++ b/kernel/dma/swiotlb.c @@ -2010,6 +2010,14 @@ static int rmem_swiotlb_device_init(struct reserved_mem *rmem, return -EINVAL; }
+ if (cc_platform_has(CC_ATTR_MEM_ENCRYPT) && + !cc_shared_range_valid(rmem->base, rmem->size)) { + dev_err(dev, + "Restricted DMA pool must be aligned to %#zx bytes for memory encryption\n", + cc_shared_granule_size()); + return -EINVAL; + } + /* * Since multiple devices can share the same pool, the private data, * io_tlb_mem struct, will be initialized by the first device attached @@ -2043,8 +2051,7 @@ static int rmem_swiotlb_device_init(struct reserved_mem *rmem, int ret;
mem->cc_shared = true; - ret = set_memory_decrypted((unsigned long)phys_to_virt(rmem->base), - rmem->size >> PAGE_SHIFT); + ret = cc_make_shared(phys_to_virt(rmem->base), rmem->size); if (ret) { dev_err(dev, "Failed to decrypt restricted DMA pool\n"); kfree(pool->areas);
The system heap currently allocates each backing page no larger than the remaining dma-buf length. It can therefore use the complete compound-page size for every scatterlist entry while keeping the total length equal to the buffer size.
Shared backing allocations may need to be rounded up to an architecture shared granule size. A backing allocation can then be larger than the remaining buffer length. Describing the complete allocation in the scatterlist would incorrectly expose the rounded tail to scatterlist consumers as part of the dma-buf.
Track the remaining buffer length while constructing the scatterlist and limit each entry to the smaller of the compound-page size and the remaining length. The complete backing allocation remains owned by the heap and is still released normally.
This does not change behavior with the current allocation policy, but prepares the heap for shared-granule-sized backing allocations.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- drivers/dma-buf/heaps/system_heap.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/dma-buf/heaps/system_heap.c b/drivers/dma-buf/heaps/system_heap.c index c8959eadc71d..b5b8cdf65f23 100644 --- a/drivers/dma-buf/heaps/system_heap.c +++ b/drivers/dma-buf/heaps/system_heap.c @@ -406,6 +406,7 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, struct system_heap_buffer *buffer; DEFINE_DMA_BUF_EXPORT_INFO(exp_info); unsigned long size_remaining = len; + unsigned long sg_remaining = len; unsigned int max_order = orders[0]; struct system_heap_priv *priv = dma_heap_get_drvdata(heap); bool cc_shared = priv->cc_shared; @@ -454,7 +455,11 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap,
sg = table->sgl; list_for_each_entry_safe(page, tmp_page, &pages, lru) { - sg_set_page(sg, page, page_size(page), 0); + unsigned long sg_len; + + sg_len = min_t(unsigned long, page_size(page), sg_remaining); + sg_set_page(sg, page, sg_len, 0); + sg_remaining -= sg_len; sg = sg_next(sg); list_del(&page->lru); }
The system_cc_shared heap currently allocates its backing pages using the normal system-heap order policy and changes each resulting compound page to shared state. That is unsafe when an architecture requires state changes in units larger than PAGE_SIZE: an order-0 tail is neither sufficiently aligned nor large enough to transition independently.
Use the common CoCo shared-page allocator for every backing allocation of the shared heap. Preserve the existing preferred-order search and its GFP policy: each candidate order is passed to the common allocator as a byte request, and that allocator rounds it up when the architecture shared granule is larger.
Add __GFP_COMP for shared allocations because an order-0 candidate can be rounded into a high-order allocation. The system heap uses compound_order() and page_size() for accounting and release, so the returned allocation must retain compound-page semantics.
Calculate a rounded internal backing length but retain the original length in dma_buf::size. The preceding scatterlist-length change ensures that the rounded tail is not included in DMA mappings or other operations. This permits a 4 KiB request on a 64 KiB shared-granule system without exposing the extra 60 KiB to an importer.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- drivers/dma-buf/heaps/system_heap.c | 119 ++++++++++++---------------- 1 file changed, 51 insertions(+), 68 deletions(-)
diff --git a/drivers/dma-buf/heaps/system_heap.c b/drivers/dma-buf/heaps/system_heap.c index b5b8cdf65f23..35ac029dc41b 100644 --- a/drivers/dma-buf/heaps/system_heap.c +++ b/drivers/dma-buf/heaps/system_heap.c @@ -11,14 +11,13 @@ */
#include <linux/cc_platform.h> +#include <linux/cc_shared.h> #include <linux/dma-buf.h> #include <linux/dma-mapping.h> #include <linux/dma-heap.h> #include <linux/err.h> #include <linux/highmem.h> -#include <linux/mem_encrypt.h> #include <linux/mm.h> -#include <linux/set_memory.h> #include <linux/module.h> #include <linux/pgtable.h> #include <linux/scatterlist.h> @@ -65,34 +64,6 @@ static gfp_t order_flags[] = {HIGH_ORDER_GFP, HIGH_ORDER_GFP, LOW_ORDER_GFP}; static const unsigned int orders[] = {8, 4, 0}; #define NUM_ORDERS ARRAY_SIZE(orders)
-static int system_heap_set_page_decrypted(struct page *page) -{ - unsigned long addr = (unsigned long)page_address(page); - unsigned int nr_pages = 1 << compound_order(page); - int ret; - - ret = set_memory_decrypted(addr, nr_pages); - if (ret) - pr_warn_ratelimited("dma-buf system heap: failed to decrypt page at %p\n", - page_address(page)); - - return ret; -} - -static int system_heap_set_page_encrypted(struct page *page) -{ - unsigned long addr = (unsigned long)page_address(page); - unsigned int nr_pages = 1 << compound_order(page); - int ret; - - ret = set_memory_encrypted(addr, nr_pages); - if (ret) - pr_warn_ratelimited("dma-buf system heap: failed to re-encrypt page at %p, leaking memory\n", - page_address(page)); - - return ret; -} - static int dup_sg_table(struct sg_table *from, struct sg_table *to) { struct scatterlist *sg, *new_sg; @@ -337,6 +308,20 @@ static void system_heap_vunmap(struct dma_buf *dmabuf, struct iosys_map *map) iosys_map_clear(map); }
+static void system_heap_free_page(struct page *page, bool cc_shared) +{ + struct cc_shared_pages mem; + + if (!cc_shared) { + __free_pages(page, compound_order(page)); + return; + } + + mem.page = page; + mem.shared_size = page_size(page); + free_cc_shared_pages(&mem); +} + static void system_heap_dma_buf_release(struct dma_buf *dmabuf) { struct system_heap_buffer *buffer = dmabuf->priv; @@ -345,19 +330,8 @@ static void system_heap_dma_buf_release(struct dma_buf *dmabuf) int i;
table = &buffer->sg_table; - for_each_sgtable_sg(table, sg, i) { - struct page *page = sg_page(sg); - - /* - * Intentionally leak pages that cannot be re-encrypted - * to prevent shared memory from being reused. - */ - if (cc_shared_buffer(buffer) && - system_heap_set_page_encrypted(page)) - continue; - - __free_pages(page, compound_order(page)); - } + for_each_sgtable_sg(table, sg, i) + system_heap_free_page(sg_page(sg), cc_shared_buffer(buffer)); sg_free_table(table); kfree(buffer); } @@ -375,22 +349,39 @@ static const struct dma_buf_ops system_heap_buf_ops = { .release = system_heap_dma_buf_release, };
+static struct page *system_heap_alloc_order(unsigned int order, + gfp_t flags, bool cc_shared) +{ + struct cc_shared_pages mem; + + if (!cc_shared) + return alloc_pages(flags, order); + + /* The shared granule can raise the actual allocation order. */ + flags |= __GFP_COMP; + if (alloc_cc_shared_pages(flags, PAGE_SIZE << order, &mem)) + return NULL; + + return mem.page; +} + static struct page *alloc_largest_available(unsigned long size, - unsigned int max_order) + unsigned int max_order, bool cc_shared) { struct page *page; - int i; gfp_t flags; + int i;
for (i = 0; i < NUM_ORDERS; i++) { if (size < (PAGE_SIZE << orders[i])) continue; if (max_order < orders[i]) continue; + flags = order_flags[i]; if (mem_accounting) flags |= __GFP_ACCOUNT; - page = alloc_pages(flags, orders[i]); + page = system_heap_alloc_order(orders[i], flags, cc_shared); if (!page) continue; return page; @@ -405,6 +396,7 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, { struct system_heap_buffer *buffer; DEFINE_DMA_BUF_EXPORT_INFO(exp_info); + struct cc_shared_layout layout; unsigned long size_remaining = len; unsigned long sg_remaining = len; unsigned int max_order = orders[0]; @@ -417,6 +409,14 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, struct page *page, *tmp_page; int i, ret = -ENOMEM;
+ if (cc_shared) { + ret = cc_shared_calc_layout(len, &layout); + if (ret) + return ERR_PTR(ret); + + size_remaining = layout.shared_size; + } + buffer = kzalloc_obj(*buffer); if (!buffer) return ERR_PTR(-ENOMEM); @@ -439,7 +439,8 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, goto free_buffer; }
- page = alloc_largest_available(size_remaining, max_order); + page = alloc_largest_available(size_remaining, max_order, + cc_shared); if (!page) goto free_buffer;
@@ -464,14 +465,6 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, list_del(&page->lru); }
- if (cc_shared_buffer(buffer)) { - for_each_sgtable_sg(table, sg, i) { - ret = system_heap_set_page_decrypted(sg_page(sg)); - if (ret) - goto free_pages; - } - } - /* create the dmabuf */ exp_info.exp_name = dma_heap_get_name(heap); exp_info.ops = &system_heap_buf_ops; @@ -486,22 +479,12 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, return dmabuf;
free_pages: - for_each_sgtable_sg(table, sg, i) { - struct page *p = sg_page(sg); - - /* - * Intentionally leak pages that cannot be re-encrypted - * to prevent shared memory from being reused. - */ - if (cc_shared_buffer(buffer) && - system_heap_set_page_encrypted(p)) - continue; - __free_pages(p, compound_order(p)); - } + for_each_sgtable_sg(table, sg, i) + system_heap_free_page(sg_page(sg), cc_shared); sg_free_table(table); free_buffer: list_for_each_entry_safe(page, tmp_page, &pages, lru) - __free_pages(page, compound_order(page)); + system_heap_free_page(page, cc_shared); kfree(buffer);
return ERR_PTR(ret);
On Mon, Sep 21, 2026 at 08:18:46PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -405,6 +396,7 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, { struct system_heap_buffer *buffer; DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
- struct cc_shared_layout layout; unsigned long size_remaining = len; unsigned long sg_remaining = len; unsigned int max_order = orders[0];
@@ -417,6 +409,14 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, struct page *page, *tmp_page; int i, ret = -ENOMEM;
- if (cc_shared) {
ret = cc_shared_calc_layout(len, &layout);
You should not override 'ret' here, it was set to -ENOMEM for a reason. cc_shared_calc_layout() succeeds resetting 'ret' and some further down allocation failure, goto free will return ERR_PTR(0).
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:46PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -405,6 +396,7 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, { struct system_heap_buffer *buffer; DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
- struct cc_shared_layout layout; unsigned long size_remaining = len; unsigned long sg_remaining = len; unsigned int max_order = orders[0];
@@ -417,6 +409,14 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, struct page *page, *tmp_page; int i, ret = -ENOMEM;
- if (cc_shared) {
ret = cc_shared_calc_layout(len, &layout);You should not override 'ret' here, it was set to -ENOMEM for a reason. cc_shared_calc_layout() succeeds resetting 'ret' and some further down allocation failure, goto free will return ERR_PTR(0).
Agreed. I’ll fix this in the next revision.
-aneesh
On 9/23/26 10:32, Aneesh Kumar K.V wrote:
Catalin Marinas catalin.marinas@arm.com writes:
On Mon, Sep 21, 2026 at 08:18:46PM +0530, Aneesh Kumar K.V (Arm) wrote:
@@ -405,6 +396,7 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, { struct system_heap_buffer *buffer; DEFINE_DMA_BUF_EXPORT_INFO(exp_info);
- struct cc_shared_layout layout; unsigned long size_remaining = len; unsigned long sg_remaining = len; unsigned int max_order = orders[0];
@@ -417,6 +409,14 @@ static struct dma_buf *system_heap_allocate(struct dma_heap *heap, struct page *page, *tmp_page; int i, ret = -ENOMEM;
- if (cc_shared) {
ret = cc_shared_calc_layout(len, &layout);You should not override 'ret' here, it was set to -ENOMEM for a reason. cc_shared_calc_layout() succeeds resetting 'ret' and some further down allocation failure, goto free will return ERR_PTR(0).
Agreed. I’ll fix this in the next revision.
Please CC me on that as well, I will try to find time to take a look.
Regards, Christian.
-aneesh
CoCo shared memory may need to be allocated and transitioned in units larger than the requested object. Before this change, users handled the resulting capacity as follows:
User Rounded capacity reused dma-buf system heap no DMA-direct no regular GIC tables no small GIC ITTs yes, through a gen_pool early SWIOTLB pool no late SWIOTLB pool yes persistent dynamic SWIOTLB no transient dynamic SWIOTLB no, one mapping only atomic DMA pools yes, through a gen_pool restricted SWIOTLB pool no additional padding
Improve the early and persistent dynamic SWIOTLB pools. They already own and transition backing rounded to the shared granule size, and SWIOTLB is itself a suballocator. Advertise the rounded extent as slots, size the slot metadata to match. This makes the extra capacity available without reserving more backing memory.
Keep transient dynamic pools unchanged. A transient pool belongs to one DMA mapping and is destroyed when that mapping is unmapped, so its spare backing cannot satisfy a later request without changing the lifetime model.
Do not attempt the same optimization for dma-buf, DMA-direct or regular GIC objects. Those allocations have independent caller-visible sizes and lifetimes. Reusing their padding requires a shared-granule suballocator with reference counting, per-object mappings and accounting.
Note: For the current 64 KiB CCA shared granule size, SWIOTLB pool sizes are already multiples of the 256 KiB IO_TLB segment size. Consequently, the rounding does not change any runtime values on current CCA systems. It instead makes the code express the intended invariant that pool metadata describes the complete shared-granule-aligned backing allocation.
Signed-off-by: Aneesh Kumar K.V (Arm) aneesh.kumar@kernel.org --- kernel/dma/swiotlb.c | 30 ++++++++++++++++++++++++------ 1 file changed, 24 insertions(+), 6 deletions(-)
diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c index cb67105b8812..9577a8807b07 100644 --- a/kernel/dma/swiotlb.c +++ b/kernel/dma/swiotlb.c @@ -330,6 +330,14 @@ static inline unsigned long nr_slots(u64 val) return DIV_ROUND_UP(val, IO_TLB_SIZE); }
+static unsigned long swiotlb_align_nslabs(unsigned long nslabs) +{ + unsigned long granule_nslabs; + + granule_nslabs = cc_shared_granule_size() >> IO_TLB_SHIFT; + return ALIGN(nslabs, granule_nslabs); +} + static void swiotlb_mark_pool_used(struct io_tlb_pool *pool) { unsigned long i; @@ -435,11 +443,12 @@ static void add_mem_pool(struct io_tlb_mem *mem, struct io_tlb_pool *pool) }
static void __init *swiotlb_memblock_alloc(unsigned long nslabs, - unsigned int flags, + unsigned long *alloc_nslabs, unsigned int flags, int (*remap)(void *tlb, unsigned long nslabs)) { + unsigned long aligned_nslabs = swiotlb_align_nslabs(nslabs); + size_t bytes = aligned_nslabs << IO_TLB_SHIFT; void *tlb; - size_t bytes = ALIGN(nslabs << IO_TLB_SHIFT, cc_shared_granule_size());
/* * By default allocate the bounce buffer memory from low memory, but @@ -457,12 +466,13 @@ static void __init *swiotlb_memblock_alloc(unsigned long nslabs, return NULL; }
- if (remap && remap(tlb, nslabs) < 0) { + if (remap && remap(tlb, aligned_nslabs) < 0) { memblock_free(tlb, bytes); pr_warn("%s: Failed to remap %zu bytes\n", __func__, bytes); return NULL; }
+ *alloc_nslabs = aligned_nslabs; return tlb; }
@@ -475,6 +485,7 @@ void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags, { struct io_tlb_pool *mem = &io_tlb_default_mem.defpool; unsigned long nslabs; + unsigned long alloc_nslabs; unsigned int nareas; size_t alloc_size; void *tlb; @@ -499,13 +510,14 @@ void __init swiotlb_init_remap(bool addressing_limit, unsigned int flags, swiotlb_adjust_nareas(num_possible_cpus());
nslabs = default_nslabs; - nareas = limit_nareas(default_nareas, nslabs); - while ((tlb = swiotlb_memblock_alloc(nslabs, flags, remap)) == NULL) { + while ((tlb = swiotlb_memblock_alloc(nslabs, &alloc_nslabs, flags, + remap)) == NULL) { if (nslabs <= IO_TLB_MIN_SLABS) return; nslabs = ALIGN(nslabs >> 1, IO_TLB_SEGSIZE); - nareas = limit_nareas(nareas, nslabs); } + nslabs = alloc_nslabs; + nareas = limit_nareas(default_nareas, nslabs);
if (default_nslabs != nslabs) { pr_info("SWIOTLB bounce buffer size adjusted %lu -> %lu slabs", @@ -871,6 +883,12 @@ static struct io_tlb_pool *swiotlb_alloc_pool(struct device *dev, tlb_size = nslabs << IO_TLB_SHIFT; }
+ /* Transient pools are tied to one mapping and cannot reuse padding. */ + if (mem->cc_shared && !dev) { + nslabs = swiotlb_align_nslabs(nslabs); + tlb_size = nslabs << IO_TLB_SHIFT; + } + slot_order = get_order(array_size(sizeof(*pool->slots), nslabs)); pool->slots = (struct io_tlb_slot *) __get_free_pages(gfp, slot_order);
linaro-mm-sig@lists.linaro.org
-
Aneesh Kumar K.V -
Aneesh Kumar K.V (Arm)
-
Catalin Marinas -
Christian König -
Jason Gunthorpe