kvm: arm64: allow the VM to select DEVICE_* and NORMAL_NC for IO memory
Subsystems (6)
v7 cover kvm: arm64: allow the VM to select DEVICE_* and NORMAL_NC for IO memory
From: Ankit Agrawal <[email protected]>
Currently, KVM for ARM64 maps at stage 2 memory that is considered device with DEVICE_nGnRE memory attributes; this setting overrides (per ARM architecture [1]) any device MMIO mapping present at stage 1, resulting in a set-up whereby a guest operating system cannot determine device MMIO mapping memory attributes on its own but it is always overridden by the KVM stage 2 default.
This set-up does not allow guest operating systems to select device memory attributes independently from KVM stage-2 mappings (refer to [1], "Combining stage 1 and stage 2 memory type attributes"), which turns out to be an issue in that guest operating systems (e.g. Linux) may request to map devices MMIO regions with memory attributes that guarantee better performance (e.g. gathering attribute - that for some devices can generate larger PCIe memory writes TLPs) and specific operations (e.g. unaligned transactions) such as the NormalNC memory type.
The default device stage 2 mapping was chosen in KVM for ARM64 since it was considered safer (i.e. it would not allow guests to trigger uncontained failures ultimately crashing the machine) but this turned out to be asynchronous (SError) defeating the purpose.
For these reasons, relax the KVM stage 2 device memory attributes from DEVICE_nGnRE to Normal-NC.
Generalizing to other devices may be problematic, however. E.g. GICv2 VCPU interface, which is effectively a shared peripheral, can allow a guest to affect another guest's interrupt distribution. Hence limit the change to VFIO PCI as caution. This is achieved by making the VFIO PCI core module set a flag that is tested by KVM to activate the code. This could be extended to other devices in the future once that is deemed safe.
[1] section D8.5 - DDI0487J_a_a-profile_architecture_reference_manual.pdf
Applied over v6.8-rc2.
History
=======
v6 -> v7
- Changed VM_VFIO_ALLOW_WC to VM_ALLOW_ANY_UNCACHED based on suggestion from Alex Williamson.
- Refactored stage2_set_prot_attr() based on Will's suggestion to reorganize the switch cases. Also updated the case to return -EINVAL when both KVM_PGTABLE_PROT_DEVICE and KVM_PGTABLE_PROT_NORMAL_NC set.
- Fixed nits pointed by Oliver and Catalin.
v5 -> v6
- Rebased to v6.8-rc2
v4 -> v5
- Moved the cover letter description text to patch 1/4.
- Cleaned up stage2_set_prot_attr() based on Marc Zyngier suggestions.
- Moved the mm header file changes to a separate patch.
- Rebased to v6.7-rc3.
v3 -> v4
- Moved the vfio-pci change to use the VM_VFIO_ALLOW_WC into separate patch.
- Added check to warn on the case NORMAL_NC and DEVICE are set simultaneously.
- Fixed miscellaneous nitpicks suggested in v3.
v2 -> v3
- Added a new patch (and converted to patch series) suggested by Catalin Marinas to ensure the code changes are restricted to VFIO PCI devices.
- Introduced VM_VFIO_ALLOW_WC flag for VFIO PCI to communicate with VMM.
- Reverted GIC mapping to DEVICE.
v1 -> v2
- Updated commit log to the one posted by
Lorenzo Pieralisi <[email protected]> (Thanks!)
- Added new flag to represent the NORMAL_NC setting. Updated stage2_set_prot_attr() to handle new flag.
v6 Link:
https://lore.kernel.org/all/[email protected]/
Ankit Agrawal (4):
kvm: arm64: introduce new flag for non-cacheable IO memory
mm: introduce new flag to indicate wc safe
kvm: arm64: set io memory s2 pte as normalnc for vfio pci device
vfio: convey kvm that the vfio-pci device is wc safe
arch/arm64/include/asm/kvm_pgtable.h | 2 ++ arch/arm64/include/asm/memory.h | 2 ++ arch/arm64/kvm/hyp/pgtable.c | 24 +++++++++++++++++++----- arch/arm64/kvm/mmu.c | 14 ++++++++++---- drivers/vfio/pci/vfio_pci_core.c | 6 +++++- include/linux/mm.h | 14 ++++++++++++++ 6 files changed, 52 insertions(+), 10 deletions(-)
Signature · 1 line
2.34.1
PATCH v7 1/4 kvm: arm64: introduce new flag for non-cacheable IO memory
From: Ankit Agrawal <[email protected]>
Currently, KVM for ARM64 maps at stage 2 memory that is considered device (i.e. it is not RAM) with DEVICE_nGnRE memory attributes; this setting overrides (as per the ARM architecture [1]) any device MMIO mapping present at stage 1, resulting in a set-up whereby a guest operating system cannot determine device MMIO mapping memory attributes on its own but it is always overridden by the KVM stage 2 default.
This set-up does not allow guest operating systems to select device memory attributes independently from KVM stage-2 mappings (refer to [1], "Combining stage 1 and stage 2 memory type attributes"), which turns out to be an issue in that guest operating systems (e.g. Linux) may request to map devices MMIO regions with memory attributes that guarantee better performance (e.g. gathering attribute - that for some devices can generate larger PCIe memory writes TLPs) and specific operations (e.g. unaligned transactions) such as the NormalNC memory type.
The default device stage 2 mapping was chosen in KVM for ARM64 since it was considered safer (i.e. it would not allow guests to trigger uncontained failures ultimately crashing the machine) but this turned out to be asynchronous (SError) defeating the purpose.
Failures containability is a property of the platform and is independent from the memory type used for MMIO device memory mappings.
Actually, DEVICE_nGnRE memory type is even more problematic than Normal-NC memory type in terms of faults containability in that e.g. aborts triggered on DEVICE_nGnRE loads cannot be made, architecturally, synchronous (i.e. that would imply that the processor should issue at most 1 load transaction at a time - it cannot pipeline them - otherwise the synchronous abort semantics would break the no-speculation attribute attached to DEVICE_XXX memory).
This means that regardless of the combined stage1+stage2 mappings a platform is safe if and only if device transactions cannot trigger uncontained failures and that in turn relies on platform capabilities and the device type being assigned (i.e. PCIe AER/DPC error containment and RAS architecture[3]); therefore the default KVM device stage 2 memory attributes play no role in making device assignment safer for a given platform (if the platform design adheres to design guidelines outlined in [3]) and therefore can be relaxed.
For all these reasons, relax the KVM stage 2 device memory attributes from DEVICE_nGnRE to Normal-NC.
The NormalNC was chosen over a different Normal memory type default at stage-2 (e.g. Normal Write-through) to avoid cache allocation/snooping.
Relaxing S2 KVM device MMIO mappings to Normal-NC is not expected to trigger any issue on guest device reclaim use cases either (i.e. device MMIO unmap followed by a device reset) at least for PCIe devices, in that in PCIe a device reset is architected and carried out through PCI config space transactions that are naturally ordered with respect to MMIO transactions according to the PCI ordering rules.
Having Normal-NC S2 default puts guests in control (thanks to stage1+stage2 combined memory attributes rules [1]) of device MMIO regions memory mappings, according to the rules described in [1] and summarized here ([(S1) - stage1], [(S2) - stage 2]):
S1 | S2 | Result NORMAL-WB | NORMAL-NC | NORMAL-NC NORMAL-WT | NORMAL-NC | NORMAL-NC NORMAL-NC | NORMAL-NC | NORMAL-NC DEVICE<attr> | NORMAL-NC | DEVICE<attr>
It is worth noting that currently, to map devices MMIO space to user space in a device pass-through use case the VFIO framework applies memory attributes derived from pgprot_noncached() settings applied to VMAs, which result in device-nGnRnE memory attributes for the stage-1 VMM mappings.
This means that a userspace mapping for device MMIO space carried out with the current VFIO framework and a guest OS mapping for the same MMIO space may result in a mismatched alias as described in [2].
Defaulting KVM device stage-2 mappings to Normal-NC attributes does not change anything in this respect, in that the mismatched aliases would only affect (refer to [2] for a detailed explanation) ordering between the userspace and GuestOS mappings resulting stream of transactions (i.e. it does not cause loss of property for either stream of transactions on its own), which is harmless given that the userspace and GuestOS access to the device is carried out through independent transactions streams.
A Normal-NC flag is not present today. So add a new kvm_pgtable_prot (KVM_PGTABLE_PROT_NORMAL_NC) flag for it, along with its corresponding PTE value 0x5 (0b101) determined from [1].
Lastly, adapt the stage2 PTE property setter function (stage2_set_prot_attr) to handle the NormalNC attribute.
[1] section D8.5.5 - DDI0487J_a_a-profile_architecture_reference_manual.pdf
[2] section B2.8 - DDI0487J_a_a-profile_architecture_reference_manual.pdf
[3] sections 1.7.7.3/1.8.5.2/appendix C - DEN0029H_SBSA_7.1.pdf
--- arch/arm64/include/asm/kvm_pgtable.h | 2 ++ arch/arm64/include/asm/memory.h | 2 ++ arch/arm64/kvm/hyp/pgtable.c | 24 +++++++++++++++++++----- 3 files changed, 23 insertions(+), 5 deletions(-)
arch/arm64/include/asm/kvm_pgtable.h +2 -0
@@ -197,6 +197,7 @@ enum kvm_pgtable_stage2_flags {* @KVM_PGTABLE_PROT_W: Write permission. * @KVM_PGTABLE_PROT_R: Read permission. * @KVM_PGTABLE_PROT_DEVICE: Device attributes. +* @KVM_PGTABLE_PROT_NORMAL_NC: Normal noncacheable attributes. * @KVM_PGTABLE_PROT_SW0: Software bit 0. * @KVM_PGTABLE_PROT_SW1: Software bit 1. * @KVM_PGTABLE_PROT_SW2: Software bit 2.
@@ -208,6 +209,7 @@ enum kvm_pgtable_prot {KVM_PGTABLE_PROT_R =BIT (2 ),KVM_PGTABLE_PROT_DEVICE =BIT (3 ), +KVM_PGTABLE_PROT_NORMAL_NC =BIT (4 ),KVM_PGTABLE_PROT_SW0 =BIT (55 ), KVM_PGTABLE_PROT_SW1 = BIT(56),
arch/arm64/include/asm/memory.h +2 -0
@@ -173,6 +173,7 @@* Memory types for Stage-2 translation */ #define MT_S2_NORMAL 0xf +#define MT_S2_NORMAL_NC 0x5#define MT_S2_DEVICE_nGnRE 0x1/*
@@ -180,6 +181,7 @@* Stage-2 enforces Normal-WB and Device-nGnRE */ #define MT_S2_FWB_NORMAL 6 +#define MT_S2_FWB_NORMAL_NC 5#define MT_S2_FWB_DEVICE_nGnRE 1#ifdef CONFIG_ARM64_4K_PAGES
arch/arm64/kvm/hyp/pgtable.c +19 -5
@@ -717,15 +717,29 @@ void kvm_tlb_flush_vmid_range(struct kvm_s2_mmu *mmu,static int stage2_set_prot_attr (struct kvm_pgtable *pgt,enum kvm_pgtable_prot prot,kvm_pte_t *ptep) { -bool device = prot &KVM_PGTABLE_PROT_DEVICE ; -kvm_pte_t attr = device ?KVM_S2_MEMATTR (pgt, DEVICE_nGnRE) : -KVM_S2_MEMATTR (pgt,NORMAL ); +kvm_pte_t attr;u32 sh =KVM_PTE_LEAF_ATTR_LO_S2_SH_IS ; +switch (prot & (KVM_PGTABLE_PROT_DEVICE | +KVM_PGTABLE_PROT_NORMAL_NC )) { +case KVM_PGTABLE_PROT_DEVICE |KVM_PGTABLE_PROT_NORMAL_NC : +return -EINVAL ; +case KVM_PGTABLE_PROT_DEVICE : +if (prot &KVM_PGTABLE_PROT_X ) +return -EINVAL ; + attr =KVM_S2_MEMATTR (pgt, DEVICE_nGnRE); +break ; +case KVM_PGTABLE_PROT_NORMAL_NC : +if (prot &KVM_PGTABLE_PROT_X ) +return -EINVAL ; + attr =KVM_S2_MEMATTR (pgt,NORMAL_NC ); +break ; +default : + attr =KVM_S2_MEMATTR (pgt,NORMAL ); + } +if (!(prot &KVM_PGTABLE_PROT_X )) attr |=KVM_PTE_LEAF_ATTR_HI_S2_XN ; -else if (device) -return -EINVAL ;if (prot &KVM_PGTABLE_PROT_R ) attr |=KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R ;
Signature · 1 line
2.34.1
PATCH v7 2/4 mm: introduce new flag to indicate wc safe
In reply to [email protected]From: Ankit Agrawal <[email protected]>
Generalizing S2 setting from DEVICE_nGnRE to NormalNc for non PCI devices may be problematic. E.g. GICv2 vCPU interface, which is effectively a shared peripheral, can allow a guest to affect another guest's interrupt distribution. The issue may be solved by limiting the relaxation to mappings that have a user VMA. Still there is insufficient information and uncertainity in the behavior of non PCI drivers.
Add a new flag VM_ALLOW_ANY_UNCACHED to indicate KVM that the device is WC capable and these S2 changes can be extended to it. KVM can use this flag to activate the code.
--- include/linux/mm.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
include/linux/mm.h +14 -0
@@ -391,6 +391,20 @@ extern unsigned int kobjsize(const void *objp);# define VM_UFFD_MINOR VM_NONE#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ +/* +* This flag is used to connect VFIO to arch specific KVM code. It +* indicates that the memory under this VMA is safe for use with any +* non-cachable memory type inside KVM. Some VFIO devices, on some +* platforms, are thought to be unsafe and can cause machine crashes +* if KVM does not lock down the memory type. +*/ +#ifdef CONFIG_64BIT +#define VM_ALLOW_ANY_UNCACHED_BIT 39 +#define VM_ALLOW_ANY_UNCACHED BIT(VM_ALLOW_ANY_UNCACHED_BIT) +#else +#define VM_ALLOW_ANY_UNCACHED VM_NONE +#endif +/* Bits set in the VMA until the stack is in its final location */ #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
Signature · 1 line
2.34.1
Re: [PATCH v7 2/4] mm: introduce new flag to indicate wc safe
Quoted 8 lines · On 11.02.24 18:47, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
Generalizing S2 setting from
DEVICE_nGnREto NormalNc for non PCI devices may be problematic. E.g. GICv2 vCPU interface, which is effectively a shared peripheral, can allow a guest to affect another guest's interrupt distribution. The issue may be solved by limiting the relaxation to mappings that have a user VMA. Still there is insufficient information and uncertainity in the behavior of
s/uncertainity/uncertainty/
Quoted 6 lines
non PCI drivers.
Add a new flag
VM_ALLOW_ANY_UNCACHEDto indicate KVM that the device is WC capable and these S2 changes can be extended to it. KVM can use this flag to activate the code.
MM people will stumble only over this commit at some point, looking for details. It might make sense to add a bit more details on the underlying problem (user space tables vs. stage-1 vs. stage-2) and why we want to have a different mapping in user space compared to stage-1.
Then, describe that the VMA flag was found to be the simplest and cleanest way to communicate this information from VFIO to KVM.
Quoted 22 lines
Suggested-by Catalin Marinas <[email protected]>Reviewed-by Jason Gunthorpe <[email protected]>Signed-off-by Ankit Agrawal <[email protected]>--- include/linux/mm.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+)include/linux/mm.h +7
-0@@ -391,6 +391,20 @@ extern unsigned int kobjsize(const void *objp);# define VM_UFFD_MINOR VM_NONE#endif /* CONFIG_HAVE_ARCH_USERFAULTFD_MINOR */ +/* +* This flag is used to connect VFIO to arch specific KVM code. It +* indicates that the memory under this VMA is safe for use with any +* non-cachable memory type inside KVM. Some VFIO devices, on some +* platforms, are thought to be unsafe and can cause machine crashes +* if KVM does not lock down the memory type. +*/
include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 410
Quoted 10 lines
include/linux/mm.h +7
-0+#ifdef CONFIG_64BIT +#define VM_ALLOW_ANY_UNCACHED_BIT 39 +#define VM_ALLOW_ANY_UNCACHED BIT(VM_ALLOW_ANY_UNCACHED_BIT) +#else +#define VM_ALLOW_ANY_UNCACHED VM_NONE +#endif +/* Bits set in the VMA until the stack is in its final location */ #define VM_STACK_INCOMPLETE_SETUP (VM_RAND_READ | VM_SEQ_READ | VM_STACK_EARLY)
It's not perfect (very VFIO <-> KVM specific right now, VMA flags feel a bit wrong), but it certainly easier and cleaner than any alternatives I could think of.
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH v7 2/4] mm: introduce new flag to indicate wc safe
Quoted 3 lines
Quoted 1 line
insufficient information and uncertainity in the behavior of
s/uncertainity/uncertainty/
Ack.
Quoted 7 lines
Quoted 6 lines
non PCI drivers.
Add a new flag
VM_ALLOW_ANY_UNCACHEDto indicate KVM that the device is WC capable and these S2 changes can be extended to it. KVM can use this flag to activate the code.
Quoted 7 lines
MM people will stumble only over this commit at some point, looking for details. It might make sense to add a bit more details on the underlying problem (user space tables vs. stage-1 vs. stage-2) and why we want to have a different mapping in user space compared to stage-1.
Then, describe that the VMA flag was found to be the simplest and cleanest way to communicate this information from VFIO to KVM.
Okay, I'll work on the commit message and describe in more details in the next version.
Quoted 22 lines
Quoted 14 lines
include/linux/mm.h +14
-0+/* +* This flag is used to connect VFIO to arch specific KVM code. It +* indicates that the memory under this VMA is safe for use with any +* non-cachable memory type inside KVM. Some VFIO devices, on some +* platforms, are thought to be unsafe and can cause machine crashes +* if KVM does not lock down the memory type. +*/ +#ifdef CONFIG_64BIT +#define VM_ALLOW_ANY_UNCACHED_BIT 39 +#define VM_ALLOW_ANY_UNCACHED BIT(VM_ALLOW_ANY_UNCACHED_BIT) +#else +#define VM_ALLOW_ANY_UNCACHED VM_NONE +#endif +Quoted 3 lines
/* Bits set in the VMA until the stack is in its final location */ #define
VM_STACK_INCOMPLETE_SETUP(VM_RAND_READ|VM_SEQ_READ|VM_STACK_EARLY)It's not perfect (very VFIO <-> KVM specific right now, VMA flags feel a bit wrong), but it certainly easier and cleaner than any alternatives I could think of.
Quoted 1 line
Thanks! FWIW, it was found the cleanest way to restrict the changes to vfio-pci.
PATCH v7 3/4 kvm: arm64: set io memory s2 pte as normalnc for vfio pci device
In reply to [email protected]From: Ankit Agrawal <[email protected]>
To provide VM with the ability to get device IO memory with NormalNC property, map device MMIO in KVM for ARM64 at stage2 as NormalNC. Having NormalNC S2 default puts guests in control (based on [1], "Combining stage 1 and stage 2 memory type attributes") of device MMIO regions memory mappings. The rules are summarized below:
([(S1) - stage1], [(S2) - stage 2])
S1 | S2 | Result NORMAL-WB | NORMAL-NC | NORMAL-NC NORMAL-WT | NORMAL-NC | NORMAL-NC NORMAL-NC | NORMAL-NC | NORMAL-NC DEVICE<attr> | NORMAL-NC | DEVICE<attr>
Still this cannot be generalized to non PCI devices such as GICv2. There is insufficient information and uncertainity in the behavior of non PCI driver. A driver must indicate support using the new flag VM_ALLOW_ANY_UNCACHED.
Adapt KVM to make use of the flag VM_ALLOW_ANY_UNCACHED as indicator to activate the S2 setting to NormalNc.
[1] section D8.5.5 of DDI0487J_a_a-profile_architecture_reference_manual.pdf
--- arch/arm64/kvm/mmu.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-)
arch/arm64/kvm/mmu.c +10 -4
@@ -1381,7 +1381,7 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,int ret =0 ;bool write_fault, writable, force_pte = false;bool exec_fault, mte_allowed; -bool device = false; +bool device = false, vfio_allow_any_uc = false;unsigned long mmu_seq;struct kvm *kvm = vcpu->kvm;struct kvm_mmu_memory_cache *memcache = &vcpu->arch.mmu_page_cache;
@@ -1472,6 +1472,8 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa, gfn = fault_ipa >>PAGE_SHIFT ; mte_allowed =kvm_vma_mte_allowed (vma); + vfio_allow_any_uc = vma->vm_flags &VM_ALLOW_ANY_UNCACHED ; +/* Don't use the VMA after the unlock -- it may have vanished */ vma =NULL ;
@@ -1557,10 +1559,14 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,if (exec_fault) prot |=KVM_PGTABLE_PROT_X ; -if (device) - prot |=KVM_PGTABLE_PROT_DEVICE ; -else if (cpus_have_final_cap (ARM64_HAS_CACHE_DIC )) +if (device) { +if (vfio_allow_any_uc) + prot |=KVM_PGTABLE_PROT_NORMAL_NC ; +else + prot |=KVM_PGTABLE_PROT_DEVICE ; + }else if (cpus_have_final_cap (ARM64_HAS_CACHE_DIC )) { prot |=KVM_PGTABLE_PROT_X ; + }/* * Under the premise of getting a FSC_PERM fault, we just need to relax
Signature · 1 line
2.34.1
PATCH v7 4/4 vfio: convey kvm that the vfio-pci device is wc safe
In reply to [email protected]From: Ankit Agrawal <[email protected]>
The code to map the MMIO in S2 as NormalNC is enabled when conveyed that the device is WC safe using a new flag VM_ALLOW_ANY_UNCACHED.
Make vfio-pci set the VM_ALLOW_ANY_UNCACHED flag.
This could be extended to other devices in the future once that is deemed safe.
--- drivers/vfio/pci/vfio_pci_core.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/vfio/pci/vfio_pci_core.c +5 -1
@@ -1862,8 +1862,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma/* * See remap_pfn_range(), called from vfio_pci_fault() but we can't * change vm_flags within the fault handler. Set them now. +* +* Set an additional flag VM_ALLOW_ANY_UNCACHED to convey kvm that +* the device is wc safe. */ -vm_flags_set (vma,VM_IO |VM_PFNMAP |VM_DONTEXPAND |VM_DONTDUMP ); +vm_flags_set (vma,VM_ALLOW_ANY_UNCACHED |VM_IO |VM_PFNMAP | +VM_DONTEXPAND |VM_DONTDUMP ); vma->vm_ops = &vfio_pci_mmap_ops;return 0 ;
Signature · 1 line
2.34.1
Re: [PATCH v7 4/4] vfio: convey kvm that the vfio-pci device is wc safe
Quoted 9 lines · On 11.02.24 18:47, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
The code to map the MMIO in S2 as NormalNC is enabled when conveyed that the device is WC safe using a new flag
VM_ALLOW_ANY_UNCACHED.Make vfio-pci set the
VM_ALLOW_ANY_UNCACHEDflag.This could be extended to other devices in the future once that is deemed safe.
Maybe add some more details how one could make a decision whether it would be safe (either here or in patch #2).
Quoted 17 lines
Suggested-by Catalin Marinas <[email protected]>Acked-by Jason Gunthorpe <[email protected]>Acked-by Catalin Marinas <[email protected]>Signed-off-by Ankit Agrawal <[email protected]>--- drivers/vfio/pci/vfio_pci_core.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)drivers/vfio/pci/vfio_pci_core.c +0
-0@@ -1862,8 +1862,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma/* * See remap_pfn_range(), called from vfio_pci_fault() but we can't * change vm_flags within the fault handler. Set them now.
drivers/vfio/pci/vfio_pci_core.c int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma · line 1873
Quoted 10 lines
drivers/vfio/pci/vfio_pci_core.c +5
-1+* +* Set an additional flag VM_ALLOW_ANY_UNCACHED to convey kvm that +* the device is wc safe. */ -vm_flags_set (vma,VM_IO |VM_PFNMAP |VM_DONTEXPAND |VM_DONTDUMP ); +vm_flags_set (vma,VM_ALLOW_ANY_UNCACHED |VM_IO |VM_PFNMAP | +VM_DONTEXPAND |VM_DONTDUMP ); vma->vm_ops = &vfio_pci_mmap_ops;return 0 ;
Signature · 3 lines
Cheers, David / dhildenb
Re: [PATCH v7 4/4] vfio: convey kvm that the vfio-pci device is wc safe
In reply to [email protected]Quoted 23 lines · On Sun, 11 Feb 2024 23:17:05 +0530 <[email protected]> wrote:
From: Ankit Agrawal <[email protected]>
The code to map the MMIO in S2 as NormalNC is enabled when conveyed that the device is WC safe using a new flag
VM_ALLOW_ANY_UNCACHED.Make vfio-pci set the
VM_ALLOW_ANY_UNCACHEDflag.This could be extended to other devices in the future once that is deemed safe.
Suggested-by Catalin Marinas <[email protected]>Acked-by Jason Gunthorpe <[email protected]>Acked-by Catalin Marinas <[email protected]>Signed-off-by Ankit Agrawal <[email protected]>--- drivers/vfio/pci/vfio_pci_core.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)drivers/vfio/pci/vfio_pci_core.c +0
-0
drivers/vfio/pci/vfio_pci_core.c int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma · line 1868
Quoted 7 lines
drivers/vfio/pci/vfio_pci_core.c +3
-0@@ -1862,8 +1862,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma/* * See remap_pfn_range(), called from vfio_pci_fault() but we can't * change vm_flags within the fault handler. Set them now. +* +* Set an additional flag VM_ALLOW_ANY_UNCACHED to convey kvm that +* the device is wc safe. */
That's a pretty superficial comment. Check that this is accurate, but maybe something like:
The VM_ALLOW_ANY_UNCACHED flag is implemented for ARM64, allowing stage 2 device mapping attributes to use Normal-NC rather than DEVICE_nGnRE, which allows guest mappings supporting combining attributes (WC). This attribute has potential risks with the GICv2 VCPU interface, but is expected to be safe for vfio-pci use cases.
And specifically, I think these other devices that may be problematic as described in the cover letter is a warning against use for vfio-platform, is that correct?
Thanks,
Alex
Quoted 6 lines
drivers/vfio/pci/vfio_pci_core.c +2
-1-vm_flags_set (vma,VM_IO |VM_PFNMAP |VM_DONTEXPAND |VM_DONTDUMP ); +vm_flags_set (vma,VM_ALLOW_ANY_UNCACHED |VM_IO |VM_PFNMAP | +VM_DONTEXPAND |VM_DONTDUMP ); vma->vm_ops = &vfio_pci_mmap_ops;return 0 ;
Re: [PATCH v7 4/4] vfio: convey kvm that the vfio-pci device is wc safe
Quoted 14 lines · Alex Williamson
drivers/vfio/pci/vfio_pci_core.c int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma · line 1868
Quoted 10 lines
drivers/vfio/pci/vfio_pci_core.c +3
-0@@ -1862,8 +1862,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma/* * See remap_pfn_range(), called from vfio_pci_fault() but we can't * change vm_flags within the fault handler. Set them now. +* +* Set an additional flag VM_ALLOW_ANY_UNCACHED to convey kvm that +* the device is wc safe. */ That's a pretty superficial comment. Check that this is accurate, but maybe something like:
Quoted 2 lines
The
VM_ALLOW_ANY_UNCACHEDflag is implemented for ARM64, allowing stage 2 device mapping attributes to use Normal-NC
^^^^
Quoted 4 lines
rather than
DEVICE_nGnRE, which allows guest mappings supporting combining attributes (WC). This attribute has potential risks with the GICv2 VCPU interface, but is expected to be safe for vfio-pci use cases.
Sure, if you want to elaborate more
The VM_ALLOW_ANY_UNCACHED flag is implemented for ARM64, allowing KVM stage 2 device mapping attributes to use Normal-NC rather than DEVICE_nGnRE, which allows guest mappings supporting combining attributes (WC). ARM does not architecturally guarentee this is safe, and indeed some MMIO regions like the GICv2 VCPU interface can trigger uncontained faults if Normal-NC is used.
Even worse we expect there are platforms where even DEVICE_nGnRE can allow uncontained faults in conercases. Unfortunately existing ARM IP requires platform integration to take responsibility to prevent this.
To safely use VFIO in KVM the platform must guarantee full safety in the guest where no action taken against a MMIO mapping can trigger an uncontainer failure. We belive that most VFIO PCI platforms support this for both mapping types, at least in common flows, based on some expectations of how PCI IP is integrated. This can be enabled more broadly, for instance into vfio-platform drivers, but only after the platform vendor completes auditing for safety.
Quoted 3 lines
And specifically, I think these other devices that may be problematic as described in the cover letter is a warning against use for vfio-platform, is that correct?
Maybe more like "we have a general consensus that vfio-pci is likely safe due to how PCI IP is typically integrated, but it is much less obvious for other VFIO bus types. As there is no known WC user for vfio-platform drivers be conservative and do not enable it."
Jason
Re: [PATCH v7 4/4] vfio: convey kvm that the vfio-pci device is wc safe
Quoted 34 lines · On Mon, 12 Feb 2024 13:20:01 -0400 Jason Gunthorpe <[email protected]> wrote:
Quoted 14 lines · Alex Williamson
drivers/vfio/pci/vfio_pci_core.c int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma · line 1868
Quoted 10 lines
drivers/vfio/pci/vfio_pci_core.c +3
-0@@ -1862,8 +1862,12 @@ int vfio_pci_core_mmap(struct vfio_device *core_vdev, struct vm_area_struct *vma/* * See remap_pfn_range(), called from vfio_pci_fault() but we can't * change vm_flags within the fault handler. Set them now. +* +* Set an additional flag VM_ALLOW_ANY_UNCACHED to convey kvm that +* the device is wc safe. */ That's a pretty superficial comment. Check that this is accurate, but maybe something like:
Quoted 2 lines
The
VM_ALLOW_ANY_UNCACHEDflag is implemented for ARM64, allowing stage 2 device mapping attributes to use Normal-NC^^^^Quoted 4 lines
rather than
DEVICE_nGnRE, which allows guest mappings supporting combining attributes (WC). This attribute has potential risks with the GICv2 VCPU interface, but is expected to be safe for vfio-pci use cases.Sure, if you want to elaborate more
The
VM_ALLOW_ANY_UNCACHEDflag is implemented for ARM64, allowing KVM stage 2 device mapping attributes to use Normal-NC rather thanDEVICE_nGnRE, which allows guest mappings supporting combining attributes (WC). ARM does not architecturally guarentee this is safe, and indeed some MMIO regions like the GICv2 VCPU interface can trigger uncontained faults if Normal-NC is used.
Quoted 2 lines
Even worse we expect there are platforms where even
DEVICE_nGnREcan allow uncontained faults in conercases. Unfortunately existing ARM
^^^^^^^^^^
*corner cases
Quoted 11 lines
IP requires platform integration to take responsibility to prevent this.
To safely use VFIO in KVM the platform must guarantee full safety in the guest where no action taken against a MMIO mapping can trigger an uncontainer failure. We belive that most VFIO PCI platforms support this for both mapping types, at least in common flows, based on some expectations of how PCI IP is integrated. This can be enabled more broadly, for instance into vfio-platform drivers, but only after the platform vendor completes auditing for safety.
I like it, please incorporate into the next version.
Quoted 8 lines
Quoted 3 lines
And specifically, I think these other devices that may be problematic as described in the cover letter is a warning against use for vfio-platform, is that correct?
Maybe more like "we have a general consensus that vfio-pci is likely safe due to how PCI IP is typically integrated, but it is much less obvious for other VFIO bus types. As there is no known WC user for vfio-platform drivers be conservative and do not enable it."
Ok. Thanks for the clarification.
Alex
Re: [PATCH v7 4/4] vfio: convey kvm that the vfio-pci device is wc safe
Quoted 12 lines
Quoted 11 lines
IP requires platform integration to take responsibility to prevent this.
To safely use VFIO in KVM the platform must guarantee full safety in the guest where no action taken against a MMIO mapping can trigger an uncontainer failure. We belive that most VFIO PCI platforms support this for both mapping types, at least in common flows, based on some expectations of how PCI IP is integrated. This can be enabled more broadly, for instance into vfio-platform drivers, but only after the platform vendor completes auditing for safety.
Quoted 1 line
I like it, please incorporate into the next version.
Yes, will fix the typos and add it.
Quoted 2 lines · On 11.02.24 18:47, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
Hi,
Quoted 25 lines
Currently, KVM for ARM64 maps at stage 2 memory that is considered device with
DEVICE_nGnREmemory attributes; this setting overrides (per ARM architecture [1]) any device MMIO mapping present at stage 1, resulting in a set-up whereby a guest operating system cannot determine device MMIO mapping memory attributes on its own but it is always overridden by the KVM stage 2 default.This set-up does not allow guest operating systems to select device memory attributes independently from KVM stage-2 mappings (refer to [1], "Combining stage 1 and stage 2 memory type attributes"), which turns out to be an issue in that guest operating systems (e.g. Linux) may request to map devices MMIO regions with memory attributes that guarantee better performance (e.g. gathering attribute - that for some devices can generate larger PCIe memory writes TLPs) and specific operations (e.g. unaligned transactions) such as the NormalNC memory type.
The default device stage 2 mapping was chosen in KVM for ARM64 since it was considered safer (i.e. it would not allow guests to trigger uncontained failures ultimately crashing the machine) but this turned out to be asynchronous (SError) defeating the purpose.
For these reasons, relax the KVM stage 2 device memory attributes from
DEVICE_nGnREto Normal-NC.
Quoted 7 lines
Generalizing to other devices may be problematic, however. E.g. GICv2 VCPU interface, which is effectively a shared peripheral, can allow a guest to affect another guest's interrupt distribution. Hence limit the change to VFIO PCI as caution. This is achieved by making the VFIO PCI core module set a flag that is tested by KVM to activate the code. This could be extended to other devices in the future once that is deemed safe.
I still have to digest some of the stuff I learned about this issue, please bear with me :)
(1) PCI BARs might contain mixtures of RAM and MMIO, the exact locations/semantics within a BAR are only really known to the actual device driver.
We must not unconditionally map PFNs "the wrong way", because it can have undesired side effects. Side effects might include read-speculation, that can be very problematic with MMIO regions.
The safe way (for the host) is DEVICE_nGnRE. But that is actually problematic for performance (where we want WC?) and unaligned accesses (where we want NC?).
We can trigger both cases right now inside VMs, where we want the device driver to actually make the decision.
(2) For a VM, that device driver lives inside the VM, for DPDK and friends, it lives in user space. They have this information.
We only focus here on optimizing (fixing?) the mapping for VMs, DPDK is out of the picture. So we want to allow the VM to achieve a WC/NC mapping by using a relaxed (NC) mapping in stage-1. Whatever is set in stage-2 wins.
(3) vfio knows whether using WC (and NC?) could be problematic, and must forbid it, if that is the case. There are cases where we could otherwise cause harm (bring down the host?). We must keep mapping the memory as DEVICE_nGnRE when in doubt.
Now, what the new mmap() flag does is tell the world "using the wrong mapping type cannot bring down the host", and KVM uses that to use a different mapping type (NC) in stage-1 as setup by vfio in the user space page tables.
I was trying to find ways of avoiding a mmap() flag and was hoping that we could just use a PTE bit that does not have semantics in VM_PFNMAP mappings. Unfortunately, arm64 does not support uffd-wp, which I had in mind, so it's not that easy.
Further, I was wondering if there would be a way to let DPDK similarly benefit, because it looks like we are happily ignoring that (I was told they apply some hacks to work around that).
In essence, user space knows how it will consume that memory: QEMU wants to mmap() it only to get it into stage-1 and not access it via the user page tables. DPDK wants to mmap() it to actually access it from user space.
So I am curious, is the following problematic, and why:
(a) User space tells VFIO which parts of a BAR it would like to have mapped differently. For QEMU, this would mean, requesting a NC mapping for the whole BAR. For DPDK, it could mean requesting different types for parts of a BAR.
(b) VFIO decides if it is safe to use a relaxed mapping. If in doubt, it falls back to existing (legacy) handling -- DEVICE_nGnRE.
(c) KVM simply uses the existing mapping type instead of diverging from the one in the user space mapping.
That would mean, that we would map NC already in QEMU. I wonder if that could be a problem with read speculation, even if QEMU never really accesses that mmap'ed region.
Something like that would of course require user space changes. Handling it without such changes (ignoring DPDK of course) would require some information exchange between KVM and vfio, like the mmap flag proposed.
Signature · 3 lines
Cheers, David / dhildenb
Quoted 6 lines · David Hildenbrand
I still have to digest some of the stuff I learned about this issue, please bear with me :)
(1) PCI BARs might contain mixtures of RAM and MMIO, the exact locations/semantics within a BAR are only really known to the actual device driver.
Nit: Not RAM and MMIO but different kinds of MMIO that have different access patterns. The conclusion is correct.
Quoted 3 lines
We must not unconditionally map PFNs "the wrong way", because it can have undesired side effects. Side effects might include read-speculation, that can be very problematic with MMIO regions.
It is worse that some hand wavey "side effect". If you map memory with NORMAL_NC (ie for write combining) then writel() doesn't work correctly at all.
The memory must be mapped according to which kernel APIs the actual driver in the VM will use. writel() vs __iowrite64_copy().
Quoted 2 lines
We can trigger both cases right now inside VMs, where we want the device driver to actually make the decision.
Yes
Quoted 2 lines
(2) For a VM, that device driver lives inside the VM, for DPDK and friends, it lives in user space. They have this information.
Yes
Quoted 2 lines
We only focus here on optimizing (fixing?) the mapping for VMs, DPDK is out of the picture.
DPDK will be solved through some VFIO ioctl, we know how to do it, just nobody has cared enough to do it.
Quoted 2 lines
So we want to allow the VM to achieve a WC/NC mapping by using a relaxed (NC) mapping in stage-1. Whatever is set in stage-2 wins.
Yes
Quoted 5 lines
(3) vfio knows whether using WC (and NC?) could be problematic, and must forbid it, if that is the case. There are cases where we could otherwise cause harm (bring down the host?). We must keep mapping the memory as
DEVICE_nGnREwhen in doubt.
Yes, there is an unspecific fear that on ARM platforms using NORMAL_NC in the wrong way can trigger a catastrophic error and kill the host. There is no way to know if the platform has this bug, so the agreement was to be conservative and only allow it for vfio-pci, based on some specific details of how PCI has to be implemented and ARM guidance on PCI integration..
Quoted 4 lines
Now, what the new
mmap()flag does is tell the world "using the wrong mapping type cannot bring down the host", and KVM uses that to use a different mapping type (NC) in stage-1 as setup by vfio in the user space page tables.
The inverse meaning, we assume VMAs with the flag can bring down the host, but yes.
Quoted 4 lines
I was trying to find ways of avoiding a
mmap()flag and was hoping that we could just use a PTE bit that does not have semantics inVM_PFNMAPmappings. Unfortunately, arm64 does not support uffd-wp, which I had in mind, so it's not that easy.
Seems like a waste of a valuable PTE bit to me.
Quoted 3 lines
Further, I was wondering if there would be a way to let DPDK similarly benefit, because it looks like we are happily ignoring that (I was told they apply some hacks to work around that).
dpdk doesn't need the VMA bit, we know how to solve it with vfio ioctls, it is very straightforward. dpdk just does a ioctl & mmap and VFIO will create a vma with pgprote_writecombine(). Completely trivial, the only nasty bit is fitting this into the VFIO uAPI.
Quoted 4 lines
(a) User space tells VFIO which parts of a BAR it would like to have mapped differently. For QEMU, this would mean, requesting a NC mapping for the whole BAR. For DPDK, it could mean requesting different types for parts of a BAR.
We don't want to have have the memory mapped as NC in qemu. As I said above if it is mapped NC then writel() doesn't work. We can't have conflicting mappings that go toward NC when the right answer is DEVICE.
writel() on NC will malfunction.
__iowrite64_copy() on DEVICE will be functionally correct but slower.
The S2 mapping that KVM creates is special because it doesn't actually map it once the VM kernel gets started. The VM kernel always supplies a S1 table that sets the correct type.
So if qemu has DEVICE, the S2 has NC and the VM's S1 has DEVICE then the mapping is realiably made to be DEVICE. The hidden S2 doesn't cause a problem.
Quoted 3 lines
That would mean, that we would map NC already in QEMU. I wonder if that could be a problem with read speculation, even if QEMU never really accesses that mmap'ed region.
Also correct.
Further, qemu may need to do emulation for MMIO in various cases and the qemu logic for this requires a DEVICE mapping or the emulation will malfunction.
Using NC in qemu is off the table.
Jason
Hi Jason,
Thanks for all the details (some might be valuable to document in more detail, but I'm not that experienced with all of the mapping types on arm64, so it might "just be me").
Quoted 15 lines
It is worse that some hand wavey "side effect". If you map memory with
NORMAL_NC(ie for write combining) thenwritel()doesn't work correctly at all.The memory must be mapped according to which kernel APIs the actual driver in the VM will use.
writel()vs__iowrite64_copy().Quoted 2 lines
We can trigger both cases right now inside VMs, where we want the device driver to actually make the decision.
Yes
Quoted 2 lines
(2) For a VM, that device driver lives inside the VM, for DPDK and friends, it lives in user space. They have this information.
Quoted 7 lines
Yes
Quoted 2 lines
We only focus here on optimizing (fixing?) the mapping for VMs, DPDK is out of the picture.
DPDK will be solved through some VFIO ioctl, we know how to do it, just nobody has cared enough to do it.
Good!
Quoted 19 lines
Quoted 3 lines
So we want to allow the VM to achieve a WC/NC mapping by using a relaxed (NC) mapping in stage-1. Whatever is set in stage-2 wins.
Yes
Quoted 5 lines
(3) vfio knows whether using WC (and NC?) could be problematic, and must forbid it, if that is the case. There are cases where we could otherwise cause harm (bring down the host?). We must keep mapping the memory as
DEVICE_nGnREwhen in doubt.Yes, there is an unspecific fear that on ARM platforms using
NORMAL_NCin the wrong way can trigger a catastrophic error and kill the host. There is no way to know if the platform has this bug, so the agreement was to be conservative and only allow it for vfio-pci, based on some specific details of how PCI has to be implemented and ARM guidance on PCI integration..
Quoted 7 lines
Quoted 4 lines
Now, what the new
mmap()flag does is tell the world "using the wrong mapping type cannot bring down the host", and KVM uses that to use a different mapping type (NC) in stage-1 as setup by vfio in the user space page tables.The inverse meaning, we assume VMAs with the flag can bring down the host, but yes.
Got it, will have a closer look at the patch soon.
Quoted 7 lines
Quoted 5 lines
I was trying to find ways of avoiding a
mmap()flag and was hoping that we could just use a PTE bit that does not have semantics inVM_PFNMAPmappings. Unfortunately, arm64 does not support uffd-wp, which I had in mind, so it's not that easy.Seems like a waste of a valuable PTE bit to me.
It would rather have been "it's already unused there, so let's reuse it". But there was no such low-hanging gruit.
Quoted 9 lines
Quoted 4 lines
Further, I was wondering if there would be a way to let DPDK similarly benefit, because it looks like we are happily ignoring that (I was told they apply some hacks to work around that).
dpdk doesn't need the VMA bit, we know how to solve it with vfio ioctls, it is very straightforward. dpdk just does a ioctl & mmap and VFIO will create a vma with
pgprote_writecombine(). Completely trivial, the only nasty bit is fitting this into the VFIO uAPI.
That's what I thought.
Quoted 10 lines
Quoted 5 lines
(a) User space tells VFIO which parts of a BAR it would like to have mapped differently. For QEMU, this would mean, requesting a NC mapping for the whole BAR. For DPDK, it could mean requesting different types for parts of a BAR.
We don't want to have have the memory mapped as NC in qemu. As I said above if it is mapped NC then
writel()doesn't work. We can't have conflicting mappings that go toward NC when the right answer is DEVICE.
I was wondering who would trigger that, but as I read below it could be MMIO emulation.
Quoted 17 lines
writel()on NC will malfunction.
__iowrite64_copy()on DEVICE will be functionally correct but slower.The S2 mapping that KVM creates is special because it doesn't actually map it once the VM kernel gets started. The VM kernel always supplies a S1 table that sets the correct type.
So if qemu has DEVICE, the S2 has NC and the VM's S1 has DEVICE then the mapping is realiably made to be DEVICE. The hidden S2 doesn't cause a problem.
Quoted 3 lines
That would mean, that we would map NC already in QEMU. I wonder if that could be a problem with read speculation, even if QEMU never really accesses that mmap'ed region.
Quoted 7 lines
Also correct.
Further, qemu may need to do emulation for MMIO in various cases and the qemu logic for this requires a DEVICE mapping or the emulation will malfunction.
Using NC in qemu is off the table.
Good, thanks for the details, all makes sense to me.
Signature · 3 lines
Cheers, David / dhildenb
Quoted 19 lines · On Sun, Feb 11, 2024 at 11:17:01PM +0530, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
Currently, KVM for ARM64 maps at stage 2 memory that is considered device with
DEVICE_nGnREmemory attributes; this setting overrides (per ARM architecture [1]) any device MMIO mapping present at stage 1, resulting in a set-up whereby a guest operating system cannot determine device MMIO mapping memory attributes on its own but it is always overridden by the KVM stage 2 default.This set-up does not allow guest operating systems to select device memory attributes independently from KVM stage-2 mappings (refer to [1], "Combining stage 1 and stage 2 memory type attributes"), which turns out to be an issue in that guest operating systems (e.g. Linux) may request to map devices MMIO regions with memory attributes that guarantee better performance (e.g. gathering attribute - that for some devices can generate larger PCIe memory writes TLPs) and specific operations (e.g. unaligned transactions) such as the NormalNC memory type.
Quoted 7 lines
The default device stage 2 mapping was chosen in KVM for ARM64 since it was considered safer (i.e. it would not allow guests to trigger uncontained failures ultimately crashing the machine) but this turned out to be asynchronous (SError) defeating the purpose.
For these reasons, relax the KVM stage 2 device memory attributes from
DEVICE_nGnREto Normal-NC.
Hi Ankit,
Thanks for being responsive in respinning the series according to the feedback. I think we're pretty close here, but it'd be good to address the comment / changelog feedback as well.
Can you respin this once more? Hopefully we can get this stuff soaking in -next thereafter.
Signature · 2 lines
Thanks, Oliver
Quoted 9 lines
Quoted 8 lines
The default device stage 2 mapping was chosen in KVM for ARM64 since it was considered safer (i.e. it would not allow guests to trigger uncontained failures ultimately crashing the machine) but this turned out to be asynchronous (SError) defeating the purpose.
For these reasons, relax the KVM stage 2 device memory attributes from
DEVICE_nGnREto Normal-NC.
Quoted 8 lines
Hi Ankit,
Thanks for being responsive in respinning the series according to the feedback. I think we're pretty close here, but it'd be good to address the comment / changelog feedback as well.
Can you respin this once more? Hopefully we can get this stuff soaking in -next thereafter.
Hi Oliver, yes I am planning to refresh it in the next few days after incorporating the comments.