* Re: [PATCH v9 14/23] x86/virt/seamldr: Shut down the current TDX module
From: Edgecombe, Rick P @ 2026-05-19 3:00 UTC (permalink / raw)
To: kvm@vger.kernel.org, linux-coco@lists.linux.dev,
linux-kernel@vger.kernel.org, Gao, Chao
Cc: Li, Xiaoyao, Huang, Kai, Zhao, Yan Y, dave.hansen@linux.intel.com,
kas@kernel.org, Chatre, Reinette, seanjc@google.com,
pbonzini@redhat.com, binbin.wu@linux.intel.com, Verma, Vishal L,
nik.borisov@suse.com, mingo@redhat.com, Weiny, Ira,
tony.lindgren@linux.intel.com, Annapurve, Vishal, Shahar, Sagi,
djbw@kernel.org, tglx@kernel.org, paulmck@kernel.org,
hpa@zytor.com, bp@alien8.de, yilun.xu@linux.intel.com,
x86@kernel.org
In-Reply-To: <20260513151045.1420990-15-chao.gao@intel.com>
On Wed, 2026-05-13 at 08:09 -0700, Chao Gao wrote:
> The first step of TDX module updates is shutting down the current TDX
> module. This step also packs state information that needs to be
> preserved across updates as handoff data,
>
kinda reads like handoff data is an existing term, but its the first reference
in this series.
Maybe packs state information that needs to be preserved across updates, called
"handoff data". This handoff data is consumed...
> which will be consumed by the
> updated module. The handoff data is stored internally in the SEAM range
> and is hidden from the kernel.
>
> To ensure a successful update, the new module must be able to consume
> the handoff data generated by the old module.
>
Is it too obvious thing to state? Above you already say it's needed.
> Since handoff data layout
> may change between modules, the handoff data is versioned. Each module
> has a native handoff version and provides backward support for several
> older versions.
>
> The complete handoff versioning protocol is complex as it supports both
> module upgrades and downgrades. See details in Intel® Trust Domain
> Extensions (Intel® TDX) Module Base Architecture Specification, Chapter
> "Handoff Versioning".
>
> Ideally, the kernel needs to retrieve the handoff versions supported by
> the current module and the new module and select a version supported by
> both. But, since this implementation chooses to only support module
> upgrades, simply request the current module to generate handoff data
> using its highest supported version, expecting that the new module will
> likely support it.
Hmm, "likely"? Is this trying to justify the kernel's policy? Dunno, stands out
as weird to me. Like "this will mostly work". Sounds incomplete, rather than a
reason of "this policy is the optimal initial implementation" or something like
that.
>
> Retrieve the module's handoff version from TDX global metadata and add an
> update step to shut down the module.
>
This is small patch with both things, but it's almost two changes.
> Module shutdown has global effect, so
> it only needs to run on one CPU.
I wouldn't think having some global effect would necessarily exclude having to
run on multiple CPUs. Or at least I don't follow. Is it a TDX arch thing? I
guess it's ok.
>
> Note that the handoff information isn't cached in tdx_sysinfo. It is used
> only for module shutdown, and is present only when the TDX module supports
> updates. Caching it in get_tdx_sys_info() would require extra update-support
> guards and refreshing the cached value across module updates.
Instead of being a "note", could this be just an imperative: Don't cache the
handoff information in tdx_sysinfo...
>
> Signed-off-by: Chao Gao <chao.gao@intel.com>
> Reviewed-by: Tony Lindgren <tony.lindgren@linux.intel.com>
> Reviewed-by: Xu Yilun <yilun.xu@linux.intel.com>
> Reviewed-by: Kai Huang <kai.huang@intel.com>
> Reviewed-by: Kiryl Shutsemau (Meta) <kas@kernel.org>
> ---
> v9:
> - Use CPU0 as the primary CPU
> ---
> arch/x86/include/asm/tdx_global_metadata.h | 4 ++++
> arch/x86/virt/vmx/tdx/seamldr.c | 15 ++++++++++++++-
> arch/x86/virt/vmx/tdx/tdx.c | 19 ++++++++++++++++++-
> arch/x86/virt/vmx/tdx/tdx.h | 3 +++
> arch/x86/virt/vmx/tdx/tdx_global_metadata.c | 13 +++++++++++++
> 5 files changed, 52 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/include/asm/tdx_global_metadata.h b/arch/x86/include/asm/tdx_global_metadata.h
> index 40689c8dc67e..41150d546589 100644
> --- a/arch/x86/include/asm/tdx_global_metadata.h
> +++ b/arch/x86/include/asm/tdx_global_metadata.h
> @@ -40,6 +40,10 @@ struct tdx_sys_info_td_conf {
> u64 cpuid_config_values[128][2];
> };
>
> +struct tdx_sys_info_handoff {
> + u16 module_hv;
> +};
> +
> struct tdx_sys_info {
> struct tdx_sys_info_version version;
> struct tdx_sys_info_features features;
> diff --git a/arch/x86/virt/vmx/tdx/seamldr.c b/arch/x86/virt/vmx/tdx/seamldr.c
> index 48fe71319fea..6114cab46196 100644
> --- a/arch/x86/virt/vmx/tdx/seamldr.c
> +++ b/arch/x86/virt/vmx/tdx/seamldr.c
> @@ -15,6 +15,7 @@
> #include <asm/seamldr.h>
>
> #include "seamcall_internal.h"
> +#include "tdx.h"
>
> /* P-SEAMLDR SEAMCALL leaf function */
> #define P_SEAMLDR_INFO 0x8000000000000000
> @@ -164,6 +165,7 @@ static int init_seamldr_params(struct seamldr_params *params, const u8 *data, u3
> */
> enum module_update_state {
> MODULE_UPDATE_START,
> + MODULE_UPDATE_SHUTDOWN,
> MODULE_UPDATE_DONE,
> };
>
> @@ -214,8 +216,16 @@ static void init_state(struct update_ctrl *ctrl)
> static int do_seamldr_install_module(void *seamldr_params)
> {
> enum module_update_state newstate, curstate = MODULE_UPDATE_START;
> + int cpu = smp_processor_id();
> + bool primary;
> int ret = 0;
>
> + /*
> + * Use CPU 0 to execute update steps that must run exactly once.
> + * Note CPU 0 is always online.
> + */
> + primary = cpu == 0;
> +
Where does the term 'primary' come from? I'm guessing that the global steps must
each be run on the same CPU? Is that right? And we just pick the cpu that we
know much be online? Or can the global steps be run on different CPUs? Or they
*have* to be run on cpu 0? It might be worth some comments explaining, depending
on the answers to those questions.
> do {
> newstate = READ_ONCE(update_ctrl.state);
>
> @@ -226,7 +236,10 @@ static int do_seamldr_install_module(void *seamldr_params)
>
> curstate = newstate;
> switch (curstate) {
> - /* TODO: add the update steps. */
> + case MODULE_UPDATE_SHUTDOWN:
> + if (primary)
> + ret = tdx_module_shutdown();
> + break;
> default:
> break;
> }
> diff --git a/arch/x86/virt/vmx/tdx/tdx.c b/arch/x86/virt/vmx/tdx/tdx.c
> index 1621695d7561..da3c1e857b26 100644
> --- a/arch/x86/virt/vmx/tdx/tdx.c
> +++ b/arch/x86/virt/vmx/tdx/tdx.c
> @@ -321,7 +321,7 @@ static __init int build_tdx_memlist(struct list_head *tmb_list)
> return ret;
> }
>
> -static __init int read_sys_metadata_field(u64 field_id, u64 *data)
> +static int read_sys_metadata_field(u64 field_id, u64 *data)
> {
> struct tdx_module_args args = {};
> int ret;
> @@ -1267,6 +1267,23 @@ static __init int tdx_enable(void)
> }
> subsys_initcall(tdx_enable);
>
> +int tdx_module_shutdown(void)
> +{
> + struct tdx_sys_info_handoff handoff = {};
> + struct tdx_module_args args = {};
> + int ret;
> +
> + ret = get_tdx_sys_info_handoff(&handoff);
> + WARN_ON_ONCE(ret);
Take or leave it:
Why not just WARN_ON_ONCE(get_tdx_sys_info_handoff(&handoff));
And we can drop the ret var. Save 2 LOC.
> +
> + /*
> + * Use the module's handoff version as it is the highest the
> + * module can produce and most likely supported by newer modules.
> + */
> + args.rcx = handoff.module_hv;
> + return seamcall_prerr(TDH_SYS_SHUTDOWN, &args);
> +}
> +
> static bool is_pamt_page(unsigned long phys)
> {
> struct tdmr_info_list *tdmr_list = &tdx_tdmr_list;
> diff --git a/arch/x86/virt/vmx/tdx/tdx.h b/arch/x86/virt/vmx/tdx/tdx.h
> index 76c5fb1e1ffe..f0c20dea0388 100644
> --- a/arch/x86/virt/vmx/tdx/tdx.h
> +++ b/arch/x86/virt/vmx/tdx/tdx.h
> @@ -46,6 +46,7 @@
> #define TDH_PHYMEM_PAGE_WBINVD 41
> #define TDH_VP_WR 43
> #define TDH_SYS_CONFIG 45
> +#define TDH_SYS_SHUTDOWN 52
> #define TDH_SYS_DISABLE 69
>
> /*
> @@ -108,4 +109,6 @@ struct tdmr_info_list {
> int max_tdmrs; /* How many 'tdmr_info's are allocated */
> };
>
> +int tdx_module_shutdown(void);
> +
> #endif
> diff --git a/arch/x86/virt/vmx/tdx/tdx_global_metadata.c b/arch/x86/virt/vmx/tdx/tdx_global_metadata.c
> index d54d4227990c..e793dec688ab 100644
> --- a/arch/x86/virt/vmx/tdx/tdx_global_metadata.c
> +++ b/arch/x86/virt/vmx/tdx/tdx_global_metadata.c
> @@ -100,6 +100,19 @@ static __init int get_tdx_sys_info_td_conf(struct tdx_sys_info_td_conf *sysinfo_
> return ret;
> }
>
> +static int get_tdx_sys_info_handoff(struct tdx_sys_info_handoff *sysinfo_handoff)
> +{
> + int ret;
> + u64 val;
> +
> + ret = read_sys_metadata_field(0x8900000100000000, &val);
> + if (ret)
> + return ret;
> +
> + sysinfo_handoff->module_hv = val;
> + return 0;
> +}
> +
> static __init int get_tdx_sys_info(struct tdx_sys_info *sysinfo)
> {
> int ret = 0;
^ permalink raw reply
* Re: [PATCH v9 19/23] x86/virt/tdx: Refresh TDX module version after update
From: Edgecombe, Rick P @ 2026-05-19 3:16 UTC (permalink / raw)
To: kvm@vger.kernel.org, linux-coco@lists.linux.dev,
linux-kernel@vger.kernel.org, Gao, Chao
Cc: Li, Xiaoyao, Huang, Kai, Zhao, Yan Y, dave.hansen@linux.intel.com,
kas@kernel.org, Chatre, Reinette, seanjc@google.com,
pbonzini@redhat.com, binbin.wu@linux.intel.com, Verma, Vishal L,
nik.borisov@suse.com, mingo@redhat.com, Weiny, Ira,
tony.lindgren@linux.intel.com, Annapurve, Vishal, Shahar, Sagi,
djbw@kernel.org, tglx@kernel.org, paulmck@kernel.org,
hpa@zytor.com, bp@alien8.de, yilun.xu@linux.intel.com,
x86@kernel.org
In-Reply-To: <20260513151045.1420990-20-chao.gao@intel.com>
On Wed, 2026-05-13 at 08:10 -0700, Chao Gao wrote:
> The kernel exposes the TDX module version through sysfs so userspace can
> check update compatibility. That information needs to remain accurate
> across runtime updates.
>
> A runtime update may change the module's update_version, so refresh the
> cached version right after a successful update.
>
> Drop __ro_after_init from tdx_sysinfo because it is now updated at runtime.
>
> Do not refresh the rest of tdx_sysinfo, even if some values change across
> updates. TDX module updates are backward compatible, so existing
> tdx_sysinfo consumers, e.g. KVM, can continue to operate without seeing the
> new values.
>
> Refreshing the full structure would be risky. A tdx_sysinfo consumer may
> initialize its TDX support based on the features originally reported in
> tdx_sysinfo. If a runtime update adds new features and the full structure
> is refreshed, that consumer could observe and use the newly reported
> features without having performed the setup required to use them safely.
>
> Signed-off-by: Chao Gao <chao.gao@intel.com>
> ---
Reviewed-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
The only thing I saw missing from Dave's last comments was:
---
> Note that major and minor versions are not refreshed because runtime updates
> are supported only between releases with identical major and minor versions.
I'd rather have this in code than a changelog comment.
If they can't change then warn if they do.
---
But I think we discussed offline to not do this, is it right?
^ permalink raw reply
* Re: [PATCH v14 05/44] arm64: RMI: Add wrappers for RMI calls
From: Aneesh Kumar K.V @ 2026-05-19 5:35 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-6-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> The wrappers make the call sites easier to read and deal with the
> boiler plate of handling the error codes from the RMM.
>
> Signed-off-by: Steven Price <steven.price@arm.com>
> +#define rmi_smccc(...) do { \
> + arm_smccc_1_1_invoke(__VA_ARGS__); \
> +} while (RMI_RETURN_STATUS(res.a0) == RMI_BUSY || \
> + RMI_RETURN_STATUS(res.a0) == RMI_BLOCKED)
> +
I guess this is not used. Also, that would require the call site to have a struct arm_smccc_res res.
-aneesh
^ permalink raw reply
* Re: [PATCH v14 08/44] arm64: RMI: Ensure that the RMM has GPT entries for memory
From: Aneesh Kumar K.V @ 2026-05-19 5:55 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-9-steven.price@arm.com>
> +
> +bool rmi_is_available(void)
> +{
> + return arm64_rmi_is_available;
> +}
> +
Can we rename to is_rmi_available(void) ?
-aneesh
^ permalink raw reply
* Re: [PATCH v14 10/44] arm64: RMI: Add support for SRO
From: Aneesh Kumar K.V @ 2026-05-19 6:02 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-11-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> +static unsigned long donate_req_to_size(unsigned long donatereq)
> +{
> + unsigned long unit_size = RMI_DONATE_SIZE(donatereq);
> +
> + switch (unit_size) {
> + case 0:
> + return PAGE_SIZE;
> + case 1:
> + return PMD_SIZE;
> + case 2:
> + return PUD_SIZE;
> + case 3:
> + return P4D_SIZE;
> + }
> + unreachable();
> +}
>
> +
> +static void rmi_smccc_invoke(struct arm_smccc_1_2_regs *regs_in,
> + struct arm_smccc_1_2_regs *regs_out)
> +{
> + struct arm_smccc_1_2_regs regs = *regs_in;
> + unsigned long status;
> +
> + do {
> + arm_smccc_1_2_invoke(®s, regs_out);
> + status = RMI_RETURN_STATUS(regs_out->a0);
> + } while (status == RMI_BUSY || status == RMI_BLOCKED);
> +}
> +
> +int free_delegated_page(phys_addr_t phys)
> +{
> + if (WARN_ON(rmi_undelegate_page(phys))) {
> + /* Undelegate failed: leak the page */
> + return -EBUSY;
> + }
> +
> + free_page((unsigned long)phys_to_virt(phys));
> +
> + return 0;
> +}
> +
> +static int rmi_sro_ensure_capacity(struct rmi_sro_state *sro,
> + unsigned long count)
> +{
> + if (WARN_ON_ONCE(sro->addr_count > RMI_MAX_ADDR_LIST))
> + return -EOVERFLOW;
> +
> + if (count > RMI_MAX_ADDR_LIST - sro->addr_count)
> + return -ENOSPC;
> +
> + return 0;
> +}
> +
> +static int rmi_sro_donate_contig(struct rmi_sro_state *sro,
> + unsigned long sro_handle,
> + unsigned long donatereq,
> + struct arm_smccc_1_2_regs *out_regs,
> + gfp_t gfp)
> +{
> + unsigned long unit_size = RMI_DONATE_SIZE(donatereq);
> + unsigned long unit_size_bytes = donate_req_to_size(donatereq);
> + unsigned long count = RMI_DONATE_COUNT(donatereq);
> + unsigned long state = RMI_DONATE_STATE(donatereq);
> + unsigned long size = unit_size_bytes * count;
> + unsigned long addr_range;
>
Looking at above and the related code, I am wondering whether we should
use u64 instead of unsigned long for everything that the specification
defines as 64-bit.
-aneesh
^ permalink raw reply
* Re: [PATCH v14 14/44] arm64: RMI: Basic infrastructure for creating a realm.
From: Aneesh Kumar K.V @ 2026-05-19 6:31 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-15-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> @@ -1114,7 +1119,10 @@ void kvm_free_stage2_pgd(struct kvm_s2_mmu *mmu)
> write_unlock(&kvm->mmu_lock);
>
> if (pgt) {
> - kvm_stage2_destroy(pgt);
> + if (!kvm_is_realm(kvm))
> + kvm_stage2_destroy(pgt);
> + else
> + kvm_pgtable_stage2_destroy_pgd(pgt);
> kfree(pgt);
> }
> }
Maybe add a comment here explaining the difference.
We now have:
kvm_arch_destroy_vm()
-> kvm_uninit_stage2_mmu()
-> kvm_realm_uninit_stage2()
-> unmap_range(0, max_ipa) // for Realm VMs
-> kvm_free_stage2_pgd()
-> unmap and free PGD // for non-Realm VMs
-> kvm_destroy_realm() // for Realm VMs
-> kvm_free_stage2_pgd()
-> free PGD // for Realm VMs
I wonder whether this can be simplified using different functions names?
(can we call kvm_pgtable_stage2_destroy_pgd() from kvm_destroy_realm()? )
-aneesh
^ permalink raw reply
* Re: [PATCH v14 17/44] arm64: RMI: RTT tear down
From: Aneesh Kumar K.V @ 2026-05-19 6:54 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-18-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> +static void kvm_realm_uninit_stage2(struct kvm_s2_mmu *mmu)
> +{
> + struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
> + struct realm *realm = &kvm->arch.realm;
> +
> + if (kvm_realm_state(kvm) != REALM_STATE_ACTIVE)
> + return;
> +
> + write_lock(&kvm->mmu_lock);
> + kvm_stage2_unmap_range(mmu, 0, BIT(realm->ia_bits - 1), true);
> + write_unlock(&kvm->mmu_lock);
> + kvm_realm_destroy_rtts(kvm);
> +}
> +
We also call kvm_stage2_unmap_range in kvm_destroy_realm()
void kvm_destroy_realm(struct kvm *kvm)
{
...
write_lock(&kvm->mmu_lock);
kvm_stage2_unmap_range(&kvm->arch.mmu, 0,
BIT(realm->ia_bits - 1), true);
write_unlock(&kvm->mmu_lock);
-aneesh
^ permalink raw reply
* Re: [RFC PATCH v4 00/14] coco/TSM: Host-side Arm CCA IDE setup via connect/disconnect callbacks
From: Suzuki K Poulose @ 2026-05-19 8:24 UTC (permalink / raw)
To: Will Deacon, Aneesh Kumar K.V (Arm)
Cc: linux-coco, kvmarm, linux-arm-kernel, linux-kernel,
Alexey Kardashevskiy, Catalin Marinas, Dan Williams,
Jason Gunthorpe, Jonathan Cameron, Marc Zyngier, Samuel Ortiz,
Steven Price, Xu Yilun
In-Reply-To: <agsNO9cc7H-b0H8L@willie-the-truck>
On 18/05/2026 13:59, Will Deacon wrote:
> On Mon, Apr 27, 2026 at 12:21:07PM +0530, Aneesh Kumar K.V (Arm) wrote:
>> arch/arm64/include/asm/rmi_cmds.h | 85 +++
>> arch/arm64/include/asm/rmi_smc.h | 168 +++++
>
> Curious, but why does this stuff have to live in the arch code? Wouldn't
> it be better off somewhere like drivers/firmware/ or
> include/linux/arm-rmi.h?
Good point. RMI interface is only available for arm64 (not in Arm32).
That said, it is indeed a firmware ! ;-) interface. The APIs are closely
integrated with the KVM Realm management. If the general consensus is
to move them under drivers/firmware (like PSCI), we could take that
approach.
Suzuki
>
> Will
^ permalink raw reply
* Re: [PATCH v14 23/44] arm64: RMI: Handle RMI_EXIT_RIPAS_CHANGE
From: Aneesh Kumar K.V @ 2026-05-19 9:40 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-24-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
...
> +void kvm_realm_unmap_range(struct kvm *kvm, unsigned long start,
> + unsigned long size, bool unmap_private,
> + bool may_block)
> +{
> + unsigned long end = start + size;
> + struct realm *realm = &kvm->arch.realm;
> +
> + if (!kvm_realm_is_created(kvm))
> + return;
> +
> + end = min(BIT(realm->ia_bits - 1), end);
> +
> + realm_unmap_shared_range(kvm, start, end, may_block);
> + if (unmap_private)
> + realm_unmap_private_range(kvm, start, end, may_block);
> +}
> +
kvm_gmem_invalidate_begin() indicates a private-only invalidation. How
is that supported?
-aneesh
^ permalink raw reply
* Re: [RFC PATCH v4 00/14] coco/TSM: Host-side Arm CCA IDE setup via connect/disconnect callbacks
From: Will Deacon @ 2026-05-19 9:46 UTC (permalink / raw)
To: Suzuki K Poulose
Cc: Aneesh Kumar K.V (Arm), linux-coco, kvmarm, linux-arm-kernel,
linux-kernel, Alexey Kardashevskiy, Catalin Marinas, Dan Williams,
Jason Gunthorpe, Jonathan Cameron, Marc Zyngier, Samuel Ortiz,
Steven Price, Xu Yilun
In-Reply-To: <afca6cdd-f9e3-4b5f-9f70-940c8dbb7a80@arm.com>
On Tue, May 19, 2026 at 09:24:07AM +0100, Suzuki K Poulose wrote:
> On 18/05/2026 13:59, Will Deacon wrote:
> > On Mon, Apr 27, 2026 at 12:21:07PM +0530, Aneesh Kumar K.V (Arm) wrote:
> > > arch/arm64/include/asm/rmi_cmds.h | 85 +++
> > > arch/arm64/include/asm/rmi_smc.h | 168 +++++
> >
> > Curious, but why does this stuff have to live in the arch code? Wouldn't
> > it be better off somewhere like drivers/firmware/ or
> > include/linux/arm-rmi.h?
>
> Good point. RMI interface is only available for arm64 (not in Arm32). That
> said, it is indeed a firmware ! ;-) interface. The APIs are closely
> integrated with the KVM Realm management. If the general consensus is
> to move them under drivers/firmware (like PSCI), we could take that
> approach.
I'd certainly prefer that as it means it's co-located with other firmware
interface code and also means that the arch maintainers don't need to
worry about changes to driver code :p
Will
^ permalink raw reply
* Re: [PATCH v14 27/44] arm64: RMI: Set RIPAS of initial memslots
From: Aneesh Kumar K.V @ 2026-05-19 10:02 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-28-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> The memory which the realm guest accesses must be set to RIPAS_RAM.
> Iterate over the memslots and set all gmem memslots to RIPAS_RAM.
>
> Signed-off-by: Steven Price <steven.price@arm.com>
> ---
...
> +static int set_ripas_of_protected_regions(struct kvm *kvm)
> +{
> + struct kvm_memslots *slots;
> + struct kvm_memory_slot *memslot;
> + int idx, bkt;
> + int ret = 0;
> +
> + idx = srcu_read_lock(&kvm->srcu);
> +
> + slots = kvm_memslots(kvm);
> + kvm_for_each_memslot(memslot, bkt, slots) {
> + if (!kvm_slot_has_gmem(memslot))
> + continue;
> +
> + ret = realm_init_ipa_state(kvm, memslot->base_gfn,
> + memslot->npages);
> + if (ret)
> + break;
> + }
> + srcu_read_unlock(&kvm->srcu, idx);
> +
> + return ret;
> +}
> +
> int kvm_arm_rmi_populate(struct kvm *kvm,
> struct kvm_arm_rmi_populate *args)
> {
> @@ -890,6 +922,10 @@ int kvm_activate_realm(struct kvm *kvm)
> return ret;
> }
>
> + ret = set_ripas_of_protected_regions(kvm);
> + if (ret)
> + return ret;
> +
> ret = rmi_realm_activate(virt_to_phys(realm->rd));
> if (ret)
> return -ENXIO;
relam guest already does.
for_each_mem_range(i, &start, &end) {
if (rsi_set_memory_range_protected_safe(start, end)) {
panic("Failed to set memory range to protected: %pa-%pa",
&start, &end);
}
}
if so why is host required to do this ?
-aneesh
^ permalink raw reply
* Re: [PATCH v14 27/44] arm64: RMI: Set RIPAS of initial memslots
From: Suzuki K Poulose @ 2026-05-19 10:13 UTC (permalink / raw)
To: Aneesh Kumar K.V, Steven Price, kvm, kvmarm
Cc: Catalin Marinas, Marc Zyngier, Will Deacon, James Morse,
Oliver Upton, Zenghui Yu, linux-arm-kernel, linux-kernel,
Joey Gouly, Alexandru Elisei, Christoffer Dall, Fuad Tabba,
linux-coco, Ganapatrao Kulkarni, Gavin Shan, Shanker Donthineni,
Alper Gun, Emi Kisanuki, Vishal Annapurve, WeiLin.Chang,
Lorenzo.Pieralisi2
In-Reply-To: <yq5av7cjsonr.fsf@kernel.org>
On 19/05/2026 11:02, Aneesh Kumar K.V wrote:
> Steven Price <steven.price@arm.com> writes:
>
>> The memory which the realm guest accesses must be set to RIPAS_RAM.
>> Iterate over the memslots and set all gmem memslots to RIPAS_RAM.
>>
>> Signed-off-by: Steven Price <steven.price@arm.com>
>> ---
>
> ...
>
>> +static int set_ripas_of_protected_regions(struct kvm *kvm)
>> +{
>> + struct kvm_memslots *slots;
>> + struct kvm_memory_slot *memslot;
>> + int idx, bkt;
>> + int ret = 0;
>> +
>> + idx = srcu_read_lock(&kvm->srcu);
>> +
>> + slots = kvm_memslots(kvm);
>> + kvm_for_each_memslot(memslot, bkt, slots) {
>> + if (!kvm_slot_has_gmem(memslot))
>> + continue;
>> +
>> + ret = realm_init_ipa_state(kvm, memslot->base_gfn,
>> + memslot->npages);
>> + if (ret)
>> + break;
>> + }
>> + srcu_read_unlock(&kvm->srcu, idx);
>> +
>> + return ret;
>> +}
>> +
>> int kvm_arm_rmi_populate(struct kvm *kvm,
>> struct kvm_arm_rmi_populate *args)
>> {
>> @@ -890,6 +922,10 @@ int kvm_activate_realm(struct kvm *kvm)
>> return ret;
>> }
>>
>> + ret = set_ripas_of_protected_regions(kvm);
>> + if (ret)
>> + return ret;
>> +
>> ret = rmi_realm_activate(virt_to_phys(realm->rd));
>> if (ret)
>> return -ENXIO;
>
> relam guest already does.
> for_each_mem_range(i, &start, &end) {
> if (rsi_set_memory_range_protected_safe(start, end)) {
> panic("Failed to set memory range to protected: %pa-%pa",
> &start, &end);
> }
> }
>
> if so why is host required to do this ?
Ideally this should be a call from the VMM (i.e., user). Irrespective of
what the guest does (which the host has no knowledge about), the VMM/
user is better aware of what to do for a given guest. We have done this
implicitly in the KVM as a start, to keep the initial implementation
simple. This could be moved out to the VMM as UABI, if there is
sufficient demand for it.
TL,DR: This should be a host/deployer decision, not the Guest. There
may other guest OS, which do not do RIPAS_RAM early enough.
Suzuki
^ permalink raw reply
* Re: [PATCH v9 13/23] x86/virt/seamldr: Abort updates after a failed step
From: Chao Gao @ 2026-05-19 10:20 UTC (permalink / raw)
To: Edgecombe, Rick P
Cc: kvm@vger.kernel.org, linux-coco@lists.linux.dev,
linux-kernel@vger.kernel.org, Li, Xiaoyao, Huang, Kai,
Zhao, Yan Y, dave.hansen@linux.intel.com, kas@kernel.org,
Chatre, Reinette, seanjc@google.com, pbonzini@redhat.com,
binbin.wu@linux.intel.com, Verma, Vishal L, nik.borisov@suse.com,
mingo@redhat.com, Weiny, Ira, tony.lindgren@linux.intel.com,
Annapurve, Vishal, Shahar, Sagi, djbw@kernel.org, tglx@kernel.org,
paulmck@kernel.org, hpa@zytor.com, bp@alien8.de,
yilun.xu@linux.intel.com, x86@kernel.org
In-Reply-To: <c0c5edb3cf4dfeee986a478aa3db886aa8d5db38.camel@intel.com>
On Tue, May 19, 2026 at 10:34:31AM +0800, Edgecombe, Rick P wrote:
>On Wed, 2026-05-13 at 08:09 -0700, Chao Gao wrote:
>> A TDX module update is a multi-step process, and any step can fail.
>>
>> The current update flow continues to later steps after an error.
>> Continuing after a failure can leave the TDX module in an unrecoverable
>> state.
>
>I get what you are saying here, but "continuing" vs "leaving" is a tiny bit
>confusing to me. Maybe: Continuing with subsequent update steps after a failure
>can cause the TDX module to enter an unrecoverable state?
Yes. it is better.
>
>>
>> One failure case must remain recoverable: update contention with an ongoing
>> TD build. The agreed kernel behavior for this case [1] is to fail the
>> update with -EBUSY so userspace can retry later.
>
>The link to the discussion is nice, but the explanation of just that there was
>an agreement is not saying much. But the reasoning around AVOID_COMPAT_SENSITIVE
>*is* handled in later patch. So can we say future changes will want to return
>errors to userspace for certain update failures? Then we can discuss the
>specifics when code is actual error is added?
yes. the main point is certain update failures should be recoverable so userspace
can retry.
>
>And why talk about EBUSY specifically? It is not in this patch. Stale log?
Sure. there is no need to single out EBUSY.
>
>>
>> Abort the update on any failure. This also makes the TD-build contention
>> case recoverable, because that failure occurs before any TDX module state
>> is changed.
>>
>
>Oh, maybe I didn't get what you meant above actually. The contention case is
>only recoverable because we detect it at the first step? Does "Continuing after
>a failure can leave the TDX module in an unrecoverable" really mean that any
>failure after the first step is unrecoverable?
You are right. Any failure after the initial module shutdown step is
unrecoverable.
> Or can we put it in some other
>more specific terms like that. Terms which are more specific but still not
>overly complex description of TDX module update flows?
>
>> Apply the same rule to all errors instead of special-casing
>> -EBUSY.
>
>It seems like actually it is not special cased...? The error returned is
>whatever is returned from the step.
>
>>
>> Track per-step failures, stop the update loop once a failure is observed,
>> and do not advance the state machine to the next step.
>
>Hmm, so this is actually a bunch of generic handling for each step, that really
>only works for the first one? Is the generic handling really needed?
We could special-case the first step, but that would add step-specific
error handling to the update loop. I think the simpler rule is to abort the
update on the first observed failure, regardless of which step reports it.
how about:
A TDX module update is a multi-step process, and any step can fail.
The current update flow continues to later steps after an error.
Continuing after a failure can cause the TDX module to enter an
unrecoverable state.
But certain failures during the initial module shutdown step should
simply return an error to userspace, so the update can be retried cleanly.
To preserve that recoverability, one option would be to abort the update
only for those failures, since they occur before any TDX module state is
changed. But special-casing specific failures in specific steps would
complicate the do-while() update loop for no benefit.
Simply abort update on any failure, at any step.
Track failures for each step, stop the update loop once a failure is
observed, and do not advance the state machine to the next step.
>
>>
>> Signed-off-by: Chao Gao <chao.gao@intel.com>
>> Reviewed-by: Xu Yilun <yilun.xu@linux.intel.com>
>> Reviewed-by: Tony Lindgren <tony.lindgren@linux.intel.com>
>> Reviewed-by: Kai Huang <kai.huang@intel.com>
>> Reviewed-by: Kiryl Shutsemau (Meta) <kas@kernel.org>
>> Link: https://lore.kernel.org/linux-coco/aQFmOZCdw64z14cJ@google.com/ # [1]
>> ---
>> v9:
>> - Avoid nested if/else by deferring failure accounting to ack_state().
>> - Reduce indentation of the main flow.
>> - Convert the failed flag into a counter. This avoids a conditional
>> update of the flag; the counter can simply accumulate failures.
>> ---
>> arch/x86/virt/vmx/tdx/seamldr.c | 11 +++++++----
>> 1 file changed, 7 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/x86/virt/vmx/tdx/seamldr.c b/arch/x86/virt/vmx/tdx/seamldr.c
>> index 7befe4a08f33..48fe71319fea 100644
>> --- a/arch/x86/virt/vmx/tdx/seamldr.c
>> +++ b/arch/x86/virt/vmx/tdx/seamldr.c
>> @@ -170,6 +170,7 @@ enum module_update_state {
>> static struct update_ctrl {
>> enum module_update_state state;
>> int num_ack;
>> + int num_failed;
>
>Was there past discussion on why it keeps a failed count? All we need to know is
>if anything failed right? So a bool is fine too?
Kiryl suggested a boolean, and I used that in earlier versions. In v9 I
moved the failure tracking next to the ack counting in ack_state(). A
boolean still works, but it needs an extra conditional to latch the failure
state.
static void ack_state(struct update_ctrl *ctrl, int result)
{
raw_spin_lock(&ctrl->lock);
- ctrl->num_failed += !!ret;
+ if (!ctrl->failed)
+ ctrl->failed = !!ret;
ctrl->num_ack++;
if (ctrl->num_ack == num_online_cpus())
if (ctrl->num_ack == num_online_cpus() && !ctrl->num_failed)
__set_target_state(ctrl, ctrl->state + 1);
Using an int mainly to keep the failure and ack tracking similar
and avoid the extra if. (I put a note under --- to explain this.)
If you prefer, I can switch it back to bool.
^ permalink raw reply
* Re: [PATCH v9 02/23] x86/virt/tdx: Move TDX_FEATURES0 bits to asm/tdx.h
From: Chao Gao @ 2026-05-19 10:24 UTC (permalink / raw)
To: Edgecombe, Rick P
Cc: Hansen, Dave, linux-kernel@vger.kernel.org,
linux-coco@lists.linux.dev, Huang, Kai, kvm@vger.kernel.org,
Li, Xiaoyao, Zhao, Yan Y, dave.hansen@linux.intel.com,
tony.lindgren@linux.intel.com, Chatre, Reinette,
seanjc@google.com, pbonzini@redhat.com, binbin.wu@linux.intel.com,
Weiny, Ira, nik.borisov@suse.com, mingo@redhat.com,
Verma, Vishal L, kas@kernel.org, Shahar, Sagi, Annapurve, Vishal,
djbw@kernel.org, tglx@kernel.org, paulmck@kernel.org,
hpa@zytor.com, bp@alien8.de, yilun.xu@linux.intel.com,
x86@kernel.org
In-Reply-To: <60dfac5273fa3bd5b5d31dbe6b8f32a60d329a78.camel@intel.com>
On Tue, May 19, 2026 at 09:59:02AM +0800, Edgecombe, Rick P wrote:
>On Mon, 2026-05-18 at 09:57 -0700, Rick Edgecombe wrote:
>> On Mon, 2026-05-18 at 15:52 +0800, Chao Gao wrote:
>> > On Fri, May 15, 2026 at 09:15:47AM -0700, Dave Hansen wrote:
>> > > On 5/13/26 08:09, Chao Gao wrote:
>> > > > This prepares for TDX module update [1] and Dynamic PAMT [2] support. Both
>> > > > add new TDX_FEATURES0 capability bits, and both need those capabilities to
>> > > > be queried from code outside arch/x86/virt. The corresponding feature-query
>> > > > helpers therefore need to live in the public asm/tdx.h header, so move the
>> > > > existing bit definitions there first.
>> > >
>> > > Please don't add unnecessary changelog cruft. If you need this move for
>> > > this series, that's enough.
>> >
>> > Sure. Will remove "Dynamic PAMT" stuff from the changelog.
>>
>> I think it should not link to old versions of this series to explain the
>> preparation. That is very confusing. We can just explain what will come in the
>> later patches of *this* series. I'll circle back and propose some verbiage.
>
>How about?
>
>Future changes will add support for new TDX features exposed as TDX_FEATURES0
>bits. The presence of these features will need to be checked outside of
>arch/x86/virt. So the feature query helpers, and the TDX_FEATURES0 defines they
>reference, will need to live in the widely accessible asm/tdx.h helper. Move the
>existing TDX_FEATURES0 to asm/tdx.h so that they can all be kept together.
Yes. This looks much clearer.
Thanks for helping me on this.
^ permalink raw reply
* Re: [PATCH v14 37/44] arm64: RMI: Prevent Device mappings for Realms
From: Aneesh Kumar K.V @ 2026-05-19 10:25 UTC (permalink / raw)
To: Steven Price, kvm, kvmarm
Cc: Steven Price, Catalin Marinas, Marc Zyngier, Will Deacon,
James Morse, Oliver Upton, Suzuki K Poulose, Zenghui Yu,
linux-arm-kernel, linux-kernel, Joey Gouly, Alexandru Elisei,
Christoffer Dall, Fuad Tabba, linux-coco, Ganapatrao Kulkarni,
Gavin Shan, Shanker Donthineni, Alper Gun, Emi Kisanuki,
Vishal Annapurve, WeiLin.Chang, Lorenzo.Pieralisi2
In-Reply-To: <20260513131757.116630-38-steven.price@arm.com>
Steven Price <steven.price@arm.com> writes:
> Physical device assignment is not yet supported. RMM v2.0 does add the
> relevant APIs, but device assignment is a big topic so will be handled
> in a future patch series. For now prevent device mappings when the guest
> is a realm.
>
> Signed-off-by: Steven Price <steven.price@arm.com>
> ---
> Changes from v6:
> * Fix the check in user_mem_abort() to prevent all pages that are not
> guest_memfd() from being mapped into the protected half of the IPA.
> Changes from v5:
> * Also prevent accesses in user_mem_abort()
> ---
> arch/arm64/kvm/mmu.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 776ffe56d17e..7678226ffd38 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -1230,6 +1230,10 @@ int kvm_phys_addr_ioremap(struct kvm *kvm, phys_addr_t guest_ipa,
> if (is_protected_kvm_enabled())
> return -EPERM;
>
> + /* We don't support mapping special pages into a Realm */
> + if (kvm_is_realm(kvm))
> + return -EPERM;
> +
> size += offset_in_page(guest_ipa);
> guest_ipa &= PAGE_MASK;
>
The commit message suggests that this will need to be updated to support
Device Assignment, but that is not true. IIUC, this is only used by
GICv2?. Can we update the commit message?
-aneesh
^ permalink raw reply
* Re: [PATCH v9 19/23] x86/virt/tdx: Refresh TDX module version after update
From: Chao Gao @ 2026-05-19 10:42 UTC (permalink / raw)
To: Edgecombe, Rick P
Cc: kvm@vger.kernel.org, linux-coco@lists.linux.dev,
linux-kernel@vger.kernel.org, Li, Xiaoyao, Huang, Kai,
Zhao, Yan Y, dave.hansen@linux.intel.com, kas@kernel.org,
Chatre, Reinette, seanjc@google.com, pbonzini@redhat.com,
binbin.wu@linux.intel.com, Verma, Vishal L, nik.borisov@suse.com,
mingo@redhat.com, Weiny, Ira, tony.lindgren@linux.intel.com,
Annapurve, Vishal, Shahar, Sagi, djbw@kernel.org, tglx@kernel.org,
paulmck@kernel.org, hpa@zytor.com, bp@alien8.de,
yilun.xu@linux.intel.com, x86@kernel.org
In-Reply-To: <dbfb8fa01ee6035576c6cb2f665372323117a63f.camel@intel.com>
On Tue, May 19, 2026 at 11:16:35AM +0800, Edgecombe, Rick P wrote:
>On Wed, 2026-05-13 at 08:10 -0700, Chao Gao wrote:
>> The kernel exposes the TDX module version through sysfs so userspace can
>> check update compatibility. That information needs to remain accurate
>> across runtime updates.
>>
>> A runtime update may change the module's update_version, so refresh the
>> cached version right after a successful update.
I will use version instead of update_version there. There is no need to
distinguish it from the major/minor version fields.
>>
>> Drop __ro_after_init from tdx_sysinfo because it is now updated at runtime.
>>
>> Do not refresh the rest of tdx_sysinfo, even if some values change across
>> updates. TDX module updates are backward compatible, so existing
>> tdx_sysinfo consumers, e.g. KVM, can continue to operate without seeing the
>> new values.
>>
>> Refreshing the full structure would be risky. A tdx_sysinfo consumer may
>> initialize its TDX support based on the features originally reported in
>> tdx_sysinfo. If a runtime update adds new features and the full structure
>> is refreshed, that consumer could observe and use the newly reported
>> features without having performed the setup required to use them safely.
>>
>> Signed-off-by: Chao Gao <chao.gao@intel.com>
>> ---
>
>Reviewed-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
>
>The only thing I saw missing from Dave's last comments was:
>---
>> Note that major and minor versions are not refreshed because runtime updates
>> are supported only between releases with identical major and minor versions.
>
>I'd rather have this in code than a changelog comment.
>
>If they can't change then warn if they do.
>---
>
>But I think we discussed offline to not do this, is it right?
We didn't reach a firm conclusion on that.
But I think there is good reason not to do that, as I explained in my v8
reply:
: Maybe I can just drop the note as I don't want to add code to preemptively
: catch theoretical module bugs.
:
: I added it because Sashiko pointed out that assigning the whole version struct
: outside stop_machine() could allow sysfs readers to observe a partially updated
: version. As we don't need to print new module version, I will move that
: assignment into stop_machine(), which addresses that issue. After that, there
: is no need to mention that major/minor versions are identical across updates.
^ permalink raw reply
* Re: [PATCH v4 04/13] dma: swiotlb: track pool encryption state and honor DMA_ATTR_CC_SHARED
From: Mostafa Saleh @ 2026-05-19 11:04 UTC (permalink / raw)
To: Aneesh Kumar K.V
Cc: iommu, linux-arm-kernel, linux-kernel, linux-coco, Robin Murphy,
Marek Szyprowski, Will Deacon, Marc Zyngier, Steven Price,
Suzuki K Poulose, Catalin Marinas, Jiri Pirko, Jason Gunthorpe,
Petr Tesarik, Alexey Kardashevskiy, Dan Williams, Xu Yilun,
linuxppc-dev, linux-s390, Madhavan Srinivasan, Michael Ellerman,
Nicholas Piggin, Christophe Leroy (CS GROUP), Alexander Gordeev,
Gerald Schaefer, Heiko Carstens, Vasily Gorbik,
Christian Borntraeger, Sven Schnelle, x86
In-Reply-To: <yq5ah5oa59wy.fsf@kernel.org>
On Thu, May 14, 2026 at 08:13:25PM +0530, Aneesh Kumar K.V wrote:
> >>
> >> What I meant was that we need a generic way to identify a pKVM guest, so
> >> that we can use it in the conditional above.
> >
> > I have this patch, with that I can boot with your series unmodified,
> > but I will need to do more testing.
> >
>
> Thanks, I can add this to the series once you complete the required testing.
>
I am still running more tests, but looking more into it. Setting
force_dma_unencrypted() to true for pKVM guests is wrong, as the
guest shouldn’t try to decrypt arbitrary memory as it can include
sensitive information (for example in case of virtio sub-page
allocation) and should strictly rely on the restricted-dma-pool
for that.
However, with my patch and setting force_dma_unencrypted() to false
on top of this series, it fails on pKVM due to a missing shared
attribute as Alexey mentioned, as now SWIOTLB rejects non shared
attrs, so, the DMA-API has to pass it. With that, I can boot again:
diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 5103a04df99f..b19aeec03f27 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -286,6 +286,8 @@ void *dma_direct_alloc(struct device *dev, size_t size,
}
if (is_swiotlb_for_alloc(dev)) {
+ attrs |= DMA_ATTR_CC_SHARED;
+
page = dma_direct_alloc_swiotlb(dev, size, attrs);
if (page) {
/*
@@ -449,6 +451,8 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
&cpu_addr, gfp, attrs);
if (is_swiotlb_for_alloc(dev)) {
+ attrs |= DMA_ATTR_CC_SHARED;
+
page = dma_direct_alloc_swiotlb(dev, size, attrs);
if (!page)
return NULL;
diff --git a/kernel/dma/direct.h b/kernel/dma/direct.h
index 4e35264ab6f8..8ee5bbf78cfb 100644
--- a/kernel/dma/direct.h
+++ b/kernel/dma/direct.h
@@ -92,6 +92,7 @@ static inline dma_addr_t dma_direct_map_phys(struct device *dev,
if (attrs & (DMA_ATTR_MMIO | DMA_ATTR_REQUIRE_COHERENT))
return DMA_MAPPING_ERROR;
+ attrs |= DMA_ATTR_CC_SHARED;
return swiotlb_map(dev, phys, size, dir, attrs);
}
--
I will keep testing and let you know how it goes. If there is nothing
else required to convert pKVM guests to CC, I can just post the patch
separately as it has no dependency on this series.
Re force_dma_unencrypted(), I am looking into a safe way to use it
for pKVM as I beleive it will be useful to eliminate some bouncing.
However, that’s not critical for this series and can be added later
as I am still investigating it, if I reach something I can post it
along the pKVM patch above.
Thanks,
Mostafa
>
>
> -aneesh
^ permalink raw reply related
* Re: [PATCH v4 04/13] dma: swiotlb: track pool encryption state and honor DMA_ATTR_CC_SHARED
From: Mostafa Saleh @ 2026-05-19 11:06 UTC (permalink / raw)
To: Jason Gunthorpe
Cc: Aneesh Kumar K.V (Arm), iommu, linux-arm-kernel, linux-kernel,
linux-coco, Robin Murphy, Marek Szyprowski, Will Deacon,
Marc Zyngier, Steven Price, Suzuki K Poulose, Catalin Marinas,
Jiri Pirko, Petr Tesarik, Alexey Kardashevskiy, Dan Williams,
Xu Yilun, linuxppc-dev, linux-s390, Madhavan Srinivasan,
Michael Ellerman, Nicholas Piggin, Christophe Leroy (CS GROUP),
Alexander Gordeev, Gerald Schaefer, Heiko Carstens, Vasily Gorbik,
Christian Borntraeger, Sven Schnelle, x86
In-Reply-To: <20260515225113.GN7702@ziepe.ca>
On Fri, May 15, 2026 at 07:51:13PM -0300, Jason Gunthorpe wrote:
> On Thu, May 14, 2026 at 02:43:39PM +0000, Mostafa Saleh wrote:
> > > That's a somewhat different problem, we have the dev->trusted stuff
> > > that is supposed to deal with this kind of security. We need it for
> > > IOMMU based systems too, eg hot plug thunderbolt should have it.
> >
> > I see that it is used only for dma-iommu and for PCI devices.
> > However, I think that should be a problem with other CCA solutions
> > with emulated devices as they are untrusted. As I'd expect they
> > would have virtio devices.
>
> Yes, any security solution with an out of TCB device should be using either
> memory encryption so the kernel already bounces or this trusted stuff
> and a force strict dma-iommu so the dma layer is careful.
>
> This is more policy from userspace what devices they want in or out of
> their TCB. Like you make accept the device into T=1 but then still
> want to keep it out of your TCB with the vIOMMU, I can see good
> arguments for something like that.
>
> > > > While we can debate the aesthetics of the setup , this is
> > > > the exisitng behaviour for Linux, which existed for years
> > > > and pKVM relies on and is used extensively.
> > > > And, this patch alters that long-standing logic and introduces
> > > > a functional regression.
> > >
> > > Yeah, Aneesh needs to do something here, I'm pointing out it is
> > > entirely seperate thing from the CC path we are working on which is
> > > decoupling CC from reylying on force swiotlb.
> >
> > I am looking into converting pKVM to use the CC stuff, I replied with
> > a patch to Aneesh in this thread. However, I need to do more testing
> > and make sure there are not any unwanted consequences.
>
> Yeah, it is a nice patch and I think it will help reduce the
> complexity if it aligns to CCA type stuff.
>
> > > In a pkvm world it should be the same, the S2 table for the SMMU will
> > > control what the device can access, and if the SMMU points to a
> > > "private" or "shared" page is not something the device needs to know
> > > or care about.
> >
> > I see that's because dma-iommu chooses the attrs for iommu_map().
>
> Long term the DMA API path through the dma-iommu will pass the
> ATTR_CC_SHARED through to iommu_map so when the arch requires a
> different IOPTE it can construct it.
>
> > In pKVM, dma_addr_t and IOPTE are the same for private and shared,
> > so nothing differs in that case.
>
> Yes, so you don't have to worry.
>
> > We don’t expect pass-through devices to interact with shared
> > memory (T=0) at the moment.
> > However, I can see use cases for that, where the host and the guest
> > collaborate with device passthrough and require zero copy.
>
> Once you add the CC patch it becomes immediately possible though
> because the user can allocate a CC shared DMA HEAP and feed that all
> over the place.
>
> > One other interesting case for device-passthrough is non-coherent
> > devices which then require private pools for bouncing.
>
> Why does shared/private matter for bouncing? Why do you need to bounce
> at all? Do cmo's not work in pkvm guests?
At the moment, in iommu_dma_map_phys(), if a non coherent device
tries to map an unaligned address or size it will be bounced.
In pKVM, dma-iommu is used for assigned devices which operate on
private memory, so bouncing that through the SWIOTLB would leak
information from the guest as the SWIOTLB is decrypted.
In that case, the device needs a pool which remains private.
Thanks,
Mostafa
>
> Jason
^ permalink raw reply
* Re: [PATCH v9 14/23] x86/virt/seamldr: Shut down the current TDX module
From: Chao Gao @ 2026-05-19 12:05 UTC (permalink / raw)
To: Edgecombe, Rick P
Cc: kvm@vger.kernel.org, linux-coco@lists.linux.dev,
linux-kernel@vger.kernel.org, Li, Xiaoyao, Huang, Kai,
Zhao, Yan Y, dave.hansen@linux.intel.com, kas@kernel.org,
Chatre, Reinette, seanjc@google.com, pbonzini@redhat.com,
binbin.wu@linux.intel.com, Verma, Vishal L, nik.borisov@suse.com,
mingo@redhat.com, Weiny, Ira, tony.lindgren@linux.intel.com,
Annapurve, Vishal, Shahar, Sagi, djbw@kernel.org, tglx@kernel.org,
paulmck@kernel.org, hpa@zytor.com, bp@alien8.de,
yilun.xu@linux.intel.com, x86@kernel.org
In-Reply-To: <329d3811e8acbfc2ecdb1c7ba443f23161329e2a.camel@intel.com>
On Tue, May 19, 2026 at 11:00:54AM +0800, Edgecombe, Rick P wrote:
>On Wed, 2026-05-13 at 08:09 -0700, Chao Gao wrote:
>> The first step of TDX module updates is shutting down the current TDX
>> module. This step also packs state information that needs to be
>> preserved across updates as handoff data,
>>
>
>kinda reads like handoff data is an existing term, but its the first reference
>in this series.
>
>Maybe packs state information that needs to be preserved across updates, called
>"handoff data". This handoff data is consumed...
Sure. Will do.
>
>> which will be consumed by the
>> updated module. The handoff data is stored internally in the SEAM range
>> and is hidden from the kernel.
>>
>> To ensure a successful update, the new module must be able to consume
>> the handoff data generated by the old module.
>>
>
>Is it too obvious thing to state? Above you already say it's needed.
Ok. Let me drop it.
>
>> Since handoff data layout
>> may change between modules, the handoff data is versioned. Each module
>> has a native handoff version and provides backward support for several
>> older versions.
>>
>> The complete handoff versioning protocol is complex as it supports both
>> module upgrades and downgrades. See details in Intel® Trust Domain
>> Extensions (Intel® TDX) Module Base Architecture Specification, Chapter
>> "Handoff Versioning".
>>
>> Ideally, the kernel needs to retrieve the handoff versions supported by
>> the current module and the new module and select a version supported by
>> both. But, since this implementation chooses to only support module
>> upgrades, simply request the current module to generate handoff data
>> using its highest supported version, expecting that the new module will
>> likely support it.
>
>Hmm, "likely"? Is this trying to justify the kernel's policy? Dunno, stands out
>as weird to me. Like "this will mostly work". Sounds incomplete, rather than a
>reason of "this policy is the optimal initial implementation" or something like
>that.
how about:
... But since this implementation only supports module upgrades, simply request
handoff data from the current module using its highest supported version.
That is sufficient for this upgrade-only implementation.
>
>>
>> Retrieve the module's handoff version from TDX global metadata and add an
>> update step to shut down the module.
>>
>
>This is small patch with both things, but it's almost two changes.
>
>> Module shutdown has global effect, so
>> it only needs to run on one CPU.
>
>I wouldn't think having some global effect would necessarily exclude having to
>run on multiple CPUs. Or at least I don't follow. Is it a TDX arch thing? I
>guess it's ok.
Yes. This comes from the TDX architecture. I will just say in the changelog
that module shutdown only needs to run on one CPU.
>
>>
>> Note that the handoff information isn't cached in tdx_sysinfo. It is used
>> only for module shutdown, and is present only when the TDX module supports
>> updates. Caching it in get_tdx_sys_info() would require extra update-support
>> guards and refreshing the cached value across module updates.
>
>Instead of being a "note", could this be just an imperative: Don't cache the
>handoff information in tdx_sysinfo...
Sure. Will do.
>
>> @@ -214,8 +216,16 @@ static void init_state(struct update_ctrl *ctrl)
>> static int do_seamldr_install_module(void *seamldr_params)
>> {
>> enum module_update_state newstate, curstate = MODULE_UPDATE_START;
>> + int cpu = smp_processor_id();
>> + bool primary;
>> int ret = 0;
>>
>> + /*
>> + * Use CPU 0 to execute update steps that must run exactly once.
>> + * Note CPU 0 is always online.
>> + */
>> + primary = cpu == 0;
>> +
>
>Where does the term 'primary' come from?
>I'm guessing that the global steps must
>each be run on the same CPU? Is that right? And we just pick the cpu that we
>know much be online? Or can the global steps be run on different CPUs? Or they
>*have* to be run on cpu 0? It might be worth some comments explaining, depending
>on the answers to those questions.
"primary" is just my name for the CPU that runs the global steps. There is
nothing special about CPU 0 beyond the fact that it is guaranteed to be
online, so it is a convenient choice.
I can rename it to something like 'is_primary_cpu' or 'is_global_step_cpu'
for clarity.
how about:
/*
* Some steps must be run on exactly one CPU. Pick CPU 0 to execute those
* steps because CPU 0 is always online.
*/
>
>> do {
>> newstate = READ_ONCE(update_ctrl.state);
>>
>> @@ -226,7 +236,10 @@ static int do_seamldr_install_module(void *seamldr_params)
>>
>> curstate = newstate;
>> switch (curstate) {
>> - /* TODO: add the update steps. */
>> + case MODULE_UPDATE_SHUTDOWN:
>> + if (primary)
>> + ret = tdx_module_shutdown();
>> + break;
>> default:
>> break;
>> }
>> diff --git a/arch/x86/virt/vmx/tdx/tdx.c b/arch/x86/virt/vmx/tdx/tdx.c
>> index 1621695d7561..da3c1e857b26 100644
>> --- a/arch/x86/virt/vmx/tdx/tdx.c
>> +++ b/arch/x86/virt/vmx/tdx/tdx.c
>> @@ -321,7 +321,7 @@ static __init int build_tdx_memlist(struct list_head *tmb_list)
>> return ret;
>> }
>>
>> -static __init int read_sys_metadata_field(u64 field_id, u64 *data)
>> +static int read_sys_metadata_field(u64 field_id, u64 *data)
>> {
>> struct tdx_module_args args = {};
>> int ret;
>> @@ -1267,6 +1267,23 @@ static __init int tdx_enable(void)
>> }
>> subsys_initcall(tdx_enable);
>>
>> +int tdx_module_shutdown(void)
>> +{
>> + struct tdx_sys_info_handoff handoff = {};
>> + struct tdx_module_args args = {};
>> + int ret;
>> +
>> + ret = get_tdx_sys_info_handoff(&handoff);
>> + WARN_ON_ONCE(ret);
>
>Take or leave it:
>
> Why not just WARN_ON_ONCE(get_tdx_sys_info_handoff(&handoff));
> And we can drop the ret var. Save 2 LOC.
Dave had a different preference here:
https://lore.kernel.org/kvm/8b9d7fa7-6534-48e7-a4fa-c21260b1c762@intel.com/
^ permalink raw reply
* Re: [PATCH v2 15/15] KVM: x86: Move the bulk of register specific code from x86.c to regs.c
From: Huang, Kai @ 2026-05-19 12:16 UTC (permalink / raw)
To: pbonzini@redhat.com, kas@kernel.org, seanjc@google.com,
vkuznets@redhat.com, dwmw2@infradead.org, paul@xen.org
Cc: Edgecombe, Rick P, x86@kernel.org, binbin.wu@linux.intel.com,
dave.hansen@linux.intel.com, linux-kernel@vger.kernel.org,
yosry@kernel.org, kvm@vger.kernel.org, linux-coco@lists.linux.dev
In-Reply-To: <20260514215355.1648463-16-seanjc@google.com>
> +void kvm_run_get_regs(struct kvm_vcpu *vcpu)
> +{
> + BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
> +
> + if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
> + __get_regs(vcpu, &vcpu->run->s.regs.regs);
> +
> + if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
> + __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
> +}
> +
> +int kvm_run_set_regs(struct kvm_vcpu *vcpu)
> +{
> + if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
> + __set_regs(vcpu, &vcpu->run->s.regs.regs);
> + vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
> + }
> +
> + if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
> + struct kvm_sregs sregs = vcpu->run->s.regs.sregs;
> +
> + if (__set_sregs(vcpu, &sregs))
> + return -EINVAL;
> +
> + vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
> + }
> +
> + return 0;
> +}
>
[...]
> @@ -12699,11 +11904,7 @@ static void store_regs(struct kvm_vcpu *vcpu)
> {
> BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
>
> - if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
> - __get_regs(vcpu, &vcpu->run->s.regs.regs);
> -
> - if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
> - __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
> + kvm_run_get_regs(vcpu);
>
> if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
> kvm_vcpu_ioctl_x86_get_vcpu_events(
> @@ -12712,19 +11913,8 @@ static void store_regs(struct kvm_vcpu *vcpu)
>
> static int sync_regs(struct kvm_vcpu *vcpu)
> {
> - if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
> - __set_regs(vcpu, &vcpu->run->s.regs.regs);
> - vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
> - }
> -
> - if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
> - struct kvm_sregs sregs = vcpu->run->s.regs.sregs;
> -
> - if (__set_sregs(vcpu, &sregs))
> - return -EINVAL;
> -
> - vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
> - }
> + if (kvm_run_set_regs(vcpu))
> + return -EINVAL;
Nit:
Do you think 'kvm_run_sync_regs()' is better than 'kvm_run_set_regs()'?
Because I think "sync" reflects better that vcpu->run->kvm_dirty_regs is cleared
after the "set" operation.
>
> if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
> struct kvm_vcpu_events events = vcpu->run->s.regs.events;
Also, I wonder whether it's better to add a helper for events so sync_regs() and
store_regs() can be simplified to:
static int sync_regs(struct kvm_vcpu *vcpu)
{
if (kvm_run_sync_regs(vcpu))
return -EINVAL;
return kvm_run_sync_events(vcpu);
}
static void store_regs(struct kvm_vcpu *vcpu)
{
kvm_run_get_regs(vcpu);
kvm_run_get_events(vcpu);
}
And maybe 'kvm_run_get_regs()' could be 'kvm_run_store_regs()' too , so that the
store_regs() could be:
static void store_regs(struct kvm_vcpu *vcpu)
{
kvm_run_store_regs(vcpu);
kvm_run_store_events(vcpu);
}
> diff --git a/arch/x86/kvm/x86.h b/arch/x86/kvm/x86.h
> index 185062a26924..fd55cd031b1c 100644
> --- a/arch/x86/kvm/x86.h
> +++ b/arch/x86/kvm/x86.h
> @@ -414,6 +414,7 @@ int handle_ud(struct kvm_vcpu *vcpu);
>
> void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
> struct kvm_queued_exception *ex);
> +void kvm_handle_exception_payload_quirk(struct kvm_vcpu *vcpu);
>
> int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
> int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
> @@ -604,6 +605,7 @@ static inline void kvm_machine_check(void)
> int kvm_spec_ctrl_test_value(u64 value);
> int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
> struct x86_exception *e);
> +void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid);
>
If I read correct, this is because "regs.c" calls kvm_invalidate_pcid() but you
want to keep it in x86.c. But it seems the "x86.h" isn't included by "regs.c"
directly but via other headers ("mmu.h" does include "x86.h").
Should the "regs.c" include "x86.h" directly?
Btw, I am a bit confused the relationship between "x86.h" and other headers like
"mmu.h" and the new "regs.h". That is, headers like "mmu.h" include "x86.h",
but headers like "regs.h" do not (instead, "x86.h" includes them).
^ permalink raw reply
* Re: [PATCH v4 04/13] dma: swiotlb: track pool encryption state and honor DMA_ATTR_CC_SHARED
From: Aneesh Kumar K.V @ 2026-05-19 12:27 UTC (permalink / raw)
To: Mostafa Saleh
Cc: iommu, linux-arm-kernel, linux-kernel, linux-coco, Robin Murphy,
Marek Szyprowski, Will Deacon, Marc Zyngier, Steven Price,
Suzuki K Poulose, Catalin Marinas, Jiri Pirko, Jason Gunthorpe,
Petr Tesarik, Alexey Kardashevskiy, Dan Williams, Xu Yilun,
linuxppc-dev, linux-s390, Madhavan Srinivasan, Michael Ellerman,
Nicholas Piggin, Christophe Leroy (CS GROUP), Alexander Gordeev,
Gerald Schaefer, Heiko Carstens, Vasily Gorbik,
Christian Borntraeger, Sven Schnelle, x86
In-Reply-To: <agxDxdxynp4KEovA@google.com>
Mostafa Saleh <smostafa@google.com> writes:
> On Thu, May 14, 2026 at 08:13:25PM +0530, Aneesh Kumar K.V wrote:
>> >>
>> >> What I meant was that we need a generic way to identify a pKVM guest, so
>> >> that we can use it in the conditional above.
>> >
>> > I have this patch, with that I can boot with your series unmodified,
>> > but I will need to do more testing.
>> >
>>
>> Thanks, I can add this to the series once you complete the required testing.
>>
>
> I am still running more tests, but looking more into it. Setting
> force_dma_unencrypted() to true for pKVM guests is wrong, as the
> guest shouldn’t try to decrypt arbitrary memory as it can include
> sensitive information (for example in case of virtio sub-page
> allocation) and should strictly rely on the restricted-dma-pool
> for that.
>
> However, with my patch and setting force_dma_unencrypted() to false
> on top of this series, it fails on pKVM due to a missing shared
> attribute as Alexey mentioned, as now SWIOTLB rejects non shared
> attrs, so, the DMA-API has to pass it. With that, I can boot again:
>
> diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
> index 5103a04df99f..b19aeec03f27 100644
> --- a/kernel/dma/direct.c
> +++ b/kernel/dma/direct.c
> @@ -286,6 +286,8 @@ void *dma_direct_alloc(struct device *dev, size_t size,
> }
>
> if (is_swiotlb_for_alloc(dev)) {
> + attrs |= DMA_ATTR_CC_SHARED;
> +
> page = dma_direct_alloc_swiotlb(dev, size, attrs);
> if (page) {
> /*
> @@ -449,6 +451,8 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
> &cpu_addr, gfp, attrs);
>
> if (is_swiotlb_for_alloc(dev)) {
> + attrs |= DMA_ATTR_CC_SHARED;
> +
> page = dma_direct_alloc_swiotlb(dev, size, attrs);
> if (!page)
> return NULL;
> diff --git a/kernel/dma/direct.h b/kernel/dma/direct.h
> index 4e35264ab6f8..8ee5bbf78cfb 100644
> --- a/kernel/dma/direct.h
> +++ b/kernel/dma/direct.h
> @@ -92,6 +92,7 @@ static inline dma_addr_t dma_direct_map_phys(struct device *dev,
> if (attrs & (DMA_ATTR_MMIO | DMA_ATTR_REQUIRE_COHERENT))
> return DMA_MAPPING_ERROR;
>
> + attrs |= DMA_ATTR_CC_SHARED;
> return swiotlb_map(dev, phys, size, dir, attrs);
> }
>
> --
>
How about the below?
modified kernel/dma/direct.c
@@ -278,6 +278,10 @@ void *dma_direct_alloc(struct device *dev, size_t size,
}
if (is_swiotlb_for_alloc(dev)) {
+
+ if (dev->dma_io_tlb_mem->unencrypted)
+ attrs |= DMA_ATTR_CC_SHARED;
+
page = dma_direct_alloc_swiotlb(dev, size, attrs);
if (page) {
/*
@@ -451,6 +455,10 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
&cpu_addr, gfp, attrs);
if (is_swiotlb_for_alloc(dev)) {
+
+ if (dev->dma_io_tlb_mem->unencrypted)
+ attrs |= DMA_ATTR_CC_SHARED;
+
page = dma_direct_alloc_swiotlb(dev, size, attrs);
if (!page)
return NULL;
modified kernel/dma/direct.h
@@ -92,6 +92,9 @@ static inline dma_addr_t dma_direct_map_phys(struct device *dev,
if (attrs & (DMA_ATTR_MMIO | DMA_ATTR_REQUIRE_COHERENT))
return DMA_MAPPING_ERROR;
+ if (dev->dma_io_tlb_mem->unencrypted)
+ attrs |= DMA_ATTR_CC_SHARED;
+
return swiotlb_map(dev, phys, size, dir, attrs);
}
>
>
> I will keep testing and let you know how it goes. If there is nothing
> else required to convert pKVM guests to CC, I can just post the patch
> separately as it has no dependency on this series.
>
That would be useful. I can then carry the patch as a dependent change,
which can also be merged separately
>
> Re force_dma_unencrypted(), I am looking into a safe way to use it
> for pKVM as I beleive it will be useful to eliminate some bouncing.
> However, that’s not critical for this series and can be added later
> as I am still investigating it, if I reach something I can post it
> along the pKVM patch above.
>
> Thanks,
> Mostafa
>
>>
>>
>> -aneesh
^ permalink raw reply
* Re: [PATCH v14 27/44] arm64: RMI: Set RIPAS of initial memslots
From: Aneesh Kumar K.V @ 2026-05-19 12:55 UTC (permalink / raw)
To: Suzuki K Poulose, Steven Price, kvm, kvmarm
Cc: Catalin Marinas, Marc Zyngier, Will Deacon, James Morse,
Oliver Upton, Zenghui Yu, linux-arm-kernel, linux-kernel,
Joey Gouly, Alexandru Elisei, Christoffer Dall, Fuad Tabba,
linux-coco, Ganapatrao Kulkarni, Gavin Shan, Shanker Donthineni,
Alper Gun, Emi Kisanuki, Vishal Annapurve, WeiLin.Chang,
Lorenzo.Pieralisi2
In-Reply-To: <6681f10b-0966-42e2-ae04-4e1aef47ec4d@arm.com>
Suzuki K Poulose <suzuki.poulose@arm.com> writes:
> On 19/05/2026 11:02, Aneesh Kumar K.V wrote:
>> Steven Price <steven.price@arm.com> writes:
>>
>>> The memory which the realm guest accesses must be set to RIPAS_RAM.
>>> Iterate over the memslots and set all gmem memslots to RIPAS_RAM.
>>>
>>> Signed-off-by: Steven Price <steven.price@arm.com>
>>> ---
>>
>> ...
>>
>>> +static int set_ripas_of_protected_regions(struct kvm *kvm)
>>> +{
>>> + struct kvm_memslots *slots;
>>> + struct kvm_memory_slot *memslot;
>>> + int idx, bkt;
>>> + int ret = 0;
>>> +
>>> + idx = srcu_read_lock(&kvm->srcu);
>>> +
>>> + slots = kvm_memslots(kvm);
>>> + kvm_for_each_memslot(memslot, bkt, slots) {
>>> + if (!kvm_slot_has_gmem(memslot))
>>> + continue;
>>> +
>>> + ret = realm_init_ipa_state(kvm, memslot->base_gfn,
>>> + memslot->npages);
>>> + if (ret)
>>> + break;
>>> + }
>>> + srcu_read_unlock(&kvm->srcu, idx);
>>> +
>>> + return ret;
>>> +}
>>> +
>>> int kvm_arm_rmi_populate(struct kvm *kvm,
>>> struct kvm_arm_rmi_populate *args)
>>> {
>>> @@ -890,6 +922,10 @@ int kvm_activate_realm(struct kvm *kvm)
>>> return ret;
>>> }
>>>
>>> + ret = set_ripas_of_protected_regions(kvm);
>>> + if (ret)
>>> + return ret;
>>> +
>>> ret = rmi_realm_activate(virt_to_phys(realm->rd));
>>> if (ret)
>>> return -ENXIO;
>>
>> relam guest already does.
>> for_each_mem_range(i, &start, &end) {
>> if (rsi_set_memory_range_protected_safe(start, end)) {
>> panic("Failed to set memory range to protected: %pa-%pa",
>> &start, &end);
>> }
>> }
>>
>> if so why is host required to do this ?
>
> Ideally this should be a call from the VMM (i.e., user). Irrespective of
> what the guest does (which the host has no knowledge about), the VMM/
> user is better aware of what to do for a given guest. We have done this
> implicitly in the KVM as a start, to keep the initial implementation
> simple. This could be moved out to the VMM as UABI, if there is
> sufficient demand for it.
>
> TL,DR: This should be a host/deployer decision, not the Guest. There
> may other guest OS, which do not do RIPAS_RAM early enough.
>
Are we suggesting that when the guest is running out of DRAM initialized
via rmi_rtt_data_map_init(), it may need to access memory outside that
range before it gets a chance to set the RIPAS as RAM?
Does that mean the guest now has to trust the host for that?
rmi_rtt_init_ripas() is not added to the measurement details, right?
-aneesh
^ permalink raw reply
* Re: [PATCH v14 27/44] arm64: RMI: Set RIPAS of initial memslots
From: Suzuki K Poulose @ 2026-05-19 13:06 UTC (permalink / raw)
To: Aneesh Kumar K.V, Steven Price, kvm, kvmarm
Cc: Catalin Marinas, Marc Zyngier, Will Deacon, James Morse,
Oliver Upton, Zenghui Yu, linux-arm-kernel, linux-kernel,
Joey Gouly, Alexandru Elisei, Christoffer Dall, Fuad Tabba,
linux-coco, Ganapatrao Kulkarni, Gavin Shan, Shanker Donthineni,
Alper Gun, Emi Kisanuki, Vishal Annapurve, WeiLin.Chang,
Lorenzo.Pieralisi2
In-Reply-To: <yq5ajyszsgmf.fsf@kernel.org>
On 19/05/2026 13:55, Aneesh Kumar K.V wrote:
> Suzuki K Poulose <suzuki.poulose@arm.com> writes:
>
>> On 19/05/2026 11:02, Aneesh Kumar K.V wrote:
>>> Steven Price <steven.price@arm.com> writes:
>>>
>>>> The memory which the realm guest accesses must be set to RIPAS_RAM.
>>>> Iterate over the memslots and set all gmem memslots to RIPAS_RAM.
>>>>
>>>> Signed-off-by: Steven Price <steven.price@arm.com>
>>>> ---
>>>
>>> ...
>>>
>>>> +static int set_ripas_of_protected_regions(struct kvm *kvm)
>>>> +{
>>>> + struct kvm_memslots *slots;
>>>> + struct kvm_memory_slot *memslot;
>>>> + int idx, bkt;
>>>> + int ret = 0;
>>>> +
>>>> + idx = srcu_read_lock(&kvm->srcu);
>>>> +
>>>> + slots = kvm_memslots(kvm);
>>>> + kvm_for_each_memslot(memslot, bkt, slots) {
>>>> + if (!kvm_slot_has_gmem(memslot))
>>>> + continue;
>>>> +
>>>> + ret = realm_init_ipa_state(kvm, memslot->base_gfn,
>>>> + memslot->npages);
>>>> + if (ret)
>>>> + break;
>>>> + }
>>>> + srcu_read_unlock(&kvm->srcu, idx);
>>>> +
>>>> + return ret;
>>>> +}
>>>> +
>>>> int kvm_arm_rmi_populate(struct kvm *kvm,
>>>> struct kvm_arm_rmi_populate *args)
>>>> {
>>>> @@ -890,6 +922,10 @@ int kvm_activate_realm(struct kvm *kvm)
>>>> return ret;
>>>> }
>>>>
>>>> + ret = set_ripas_of_protected_regions(kvm);
>>>> + if (ret)
>>>> + return ret;
>>>> +
>>>> ret = rmi_realm_activate(virt_to_phys(realm->rd));
>>>> if (ret)
>>>> return -ENXIO;
>>>
>>> relam guest already does.
>>> for_each_mem_range(i, &start, &end) {
>>> if (rsi_set_memory_range_protected_safe(start, end)) {
>>> panic("Failed to set memory range to protected: %pa-%pa",
>>> &start, &end);
>>> }
>>> }
>>>
>>> if so why is host required to do this ?
>>
>> Ideally this should be a call from the VMM (i.e., user). Irrespective of
>> what the guest does (which the host has no knowledge about), the VMM/
>> user is better aware of what to do for a given guest. We have done this
>> implicitly in the KVM as a start, to keep the initial implementation
>> simple. This could be moved out to the VMM as UABI, if there is
>> sufficient demand for it.
>>
>> TL,DR: This should be a host/deployer decision, not the Guest. There
>> may other guest OS, which do not do RIPAS_RAM early enough.
>>
>
> Are we suggesting that when the guest is running out of DRAM initialized
> via rmi_rtt_data_map_init(), it may need to access memory outside that
> range before it gets a chance to set the RIPAS as RAM?
It may. This was one of the review comments we got when we published
the Linux Guest patches. In fact, this is in the Linux booting
requirements. See :
Documentation/arch/arm64/booting.rst: Section 1
>
> Does that mean the guest now has to trust the host for that?
No, this has been the case. We added the code in Linux to convert memory
as a back stop. The worse could happens is Guest crashing, without it
having any secrets receving from the Remote entity.
> rmi_rtt_init_ripas() is not added to the measurement details, right?
It is not (at least for now). It doesn't matter for security much.
Suzuki
>
> -aneesh
^ permalink raw reply
* Re: [PATCH v4 04/13] dma: swiotlb: track pool encryption state and honor DMA_ATTR_CC_SHARED
From: Jason Gunthorpe @ 2026-05-19 13:29 UTC (permalink / raw)
To: Mostafa Saleh
Cc: Aneesh Kumar K.V, iommu, linux-arm-kernel, linux-kernel,
linux-coco, Robin Murphy, Marek Szyprowski, Will Deacon,
Marc Zyngier, Steven Price, Suzuki K Poulose, Catalin Marinas,
Jiri Pirko, Petr Tesarik, Alexey Kardashevskiy, Dan Williams,
Xu Yilun, linuxppc-dev, linux-s390, Madhavan Srinivasan,
Michael Ellerman, Nicholas Piggin, Christophe Leroy (CS GROUP),
Alexander Gordeev, Gerald Schaefer, Heiko Carstens, Vasily Gorbik,
Christian Borntraeger, Sven Schnelle, x86
In-Reply-To: <agxDxdxynp4KEovA@google.com>
On Tue, May 19, 2026 at 11:04:37AM +0000, Mostafa Saleh wrote:
> On Thu, May 14, 2026 at 08:13:25PM +0530, Aneesh Kumar K.V wrote:
> > >>
> > >> What I meant was that we need a generic way to identify a pKVM guest, so
> > >> that we can use it in the conditional above.
> > >
> > > I have this patch, with that I can boot with your series unmodified,
> > > but I will need to do more testing.
> > >
> >
> > Thanks, I can add this to the series once you complete the required testing.
> >
>
> I am still running more tests, but looking more into it. Setting
> force_dma_unencrypted() to true for pKVM guests is wrong, as the
> guest shouldn’t try to decrypt arbitrary memory as it can include
> sensitive information (for example in case of virtio sub-page
> allocation) and should strictly rely on the restricted-dma-pool
> for that.
??
Where does force_dma_unencrypted() cause arbitary memory passed into
the DMA API to be decrypted? That should never happen???
Jason
^ permalink raw reply
* Re: [PATCH v4 04/13] dma: swiotlb: track pool encryption state and honor DMA_ATTR_CC_SHARED
From: Jason Gunthorpe @ 2026-05-19 13:39 UTC (permalink / raw)
To: Mostafa Saleh
Cc: Aneesh Kumar K.V (Arm), iommu, linux-arm-kernel, linux-kernel,
linux-coco, Robin Murphy, Marek Szyprowski, Will Deacon,
Marc Zyngier, Steven Price, Suzuki K Poulose, Catalin Marinas,
Jiri Pirko, Petr Tesarik, Alexey Kardashevskiy, Dan Williams,
Xu Yilun, linuxppc-dev, linux-s390, Madhavan Srinivasan,
Michael Ellerman, Nicholas Piggin, Christophe Leroy (CS GROUP),
Alexander Gordeev, Gerald Schaefer, Heiko Carstens, Vasily Gorbik,
Christian Borntraeger, Sven Schnelle, x86
In-Reply-To: <agxETC1rXBUSkWYg@google.com>
On Tue, May 19, 2026 at 11:06:52AM +0000, Mostafa Saleh wrote:
> > > One other interesting case for device-passthrough is non-coherent
> > > devices which then require private pools for bouncing.
> >
> > Why does shared/private matter for bouncing? Why do you need to bounce
> > at all? Do cmo's not work in pkvm guests?
>
> At the moment, in iommu_dma_map_phys(), if a non coherent device
> tries to map an unaligned address or size it will be bounced.
Sure, that's fine.
> In pKVM, dma-iommu is used for assigned devices which operate on
> private memory, so bouncing that through the SWIOTLB would leak
> information from the guest as the SWIOTLB is decrypted.
Yes, a device that can do private access should not be using a shared
SWIOTLB, that should be part of the selection logic inside the SWIOTLB
stuff..
Jason
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox