Skip to contentSkip to conversation

RISC-V: KVM: Add minimal guest_memfd support

RFC PATCH Superseded Bingyu.Xian 1 participant 2 messages Jul 29 – Jul 30

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 RISC-V: KVM: Add minimal guest_memfd support

Add minimal guest_memfd support for RISC-V KVM, following the same pattern as x86 and arm64.

1. Select KVM_GUEST_MEMFD in Kconfig
2. In kvm_riscv_mmu_map(), check kvm_slot_has_gmem() early, before taking mmap_read_lock(). For gmem memslots, resolve the PFN via kvm_gmem_get_pfn(), then acquire kvm->mmu_lock and jump to G-stage mapping, bypassing vma_lookup()/__kvm_faultin_pfn(). The gmem PFN is resolved before mmap_read_lock() so the mmu_lock taken for the mapping is never held while waiting for mmap_lock, which would deadlock against mmu_notifier invalidation (mmu_lock -> mmap_lock).
3. Skip kvm_release_faultin_page() for gmem pages; their lifetime is managed by the gmem inode and invalidation notifiers.

Verified: defconfig + CONFIG_KVM_GUEST_MEMFD=y compiles cleanly.

Assisted-by YuanSheng: deepseek-v4-pro
---
 arch/riscv/kvm/Kconfig |  1 +
 arch/riscv/kvm/mmu.c   | 17 ++++++++++++++++-
 2 files changed, 17 insertions(+), 1 deletion(-)
arch/riscv/kvm/Kconfig +1 -0
index ec2cee0a39e0..77121100579f 100644
@@ -29,6 +29,7 @@ config KVM
 	select KVM_GENERIC_DIRTYLOG_READ_PROTECT
 	select KVM_GENERIC_HARDWARE_ENABLING
 	select KVM_MMIO
+	select KVM_GUEST_MEMFD
 	select VIRT_XFER_TO_GUEST_WORK
 	select SCHED_INFO
 	select GUEST_PERF_EVENTS if PERF_EVENTS
arch/riscv/kvm/mmu.c +16 -1
index d2a06a54be17..a9d61d0366e2 100644
@@ -557,6 +557,19 @@ int kvm_riscv_mmu_map(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
 	/* Setup initial state of output mapping */
 	memset(out_map, 0, sizeof(*out_map));
 
+	if (kvm_slot_has_gmem(memslot)) {
+		int max_order;
+
+		ret = kvm_gmem_get_pfn(kvm, memslot, gfn, &hfn, &page, &max_order);
+		if (ret)
+			return ret;
+
+		vma_pagesize = PAGE_SIZE;
+		writable = true;
+		write_lock(&kvm->mmu_lock);
+		goto map_gstage;
+	}
+
 	if (is_write && logging &&
 	    kvm_riscv_mmu_dirty_log_write_fault_fast(kvm, memslot, gpa, out_map))
 		return 0;
@@ -643,6 +656,7 @@ int kvm_riscv_mmu_map(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
 	if (!logging && !is_hugetlb && vma_pagesize == PAGE_SIZE)
 		vma_pagesize = transparent_hugepage_adjust(kvm, memslot, hva, &hfn, &gpa);
 
+map_gstage:
 	if (writable) {
 		mark_page_dirty_in_slot(kvm, memslot, gfn);
 		ret = kvm_riscv_gstage_map_page(&gstage, pcache, gpa, hfn << PAGE_SHIFT,
@@ -658,7 +672,8 @@ int kvm_riscv_mmu_map(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
 		kvm_err("Failed to map in G-stage\n");
 
 out_unlock:
-	kvm_release_faultin_page(kvm, page, ret, writable);
+	if (!kvm_slot_has_gmem(memslot))
+		kvm_release_faultin_page(kvm, page, ret, writable);
 	write_unlock(&kvm->mmu_lock);
 	return ret;
 }
Signature · 1 line
2.54.0
#1 lore raw

PATCH v2 1/1 RISC-V: KVM: Add minimal guest_memfd support

Add minimal guest_memfd support for RISC-V KVM, following the same

pattern as x86 and arm64.

1. Select KVM_GUEST_MEMFD in Kconfig

2. In kvm_riscv_mmu_map(), check kvm_slot_has_gmem() early, before

taking mmap_read_lock(). For gmem memslots, resolve the PFN via

kvm_gmem_get_pfn(), then acquire kvm->mmu_lock and jump to G-stage

mapping, bypassing vma_lookup()/__kvm_faultin_pfn(). The gmem PFN

is resolved before mmap_read_lock() so the mmu_lock taken for the

mapping is never held while waiting for mmap_lock, which would

deadlock against mmu_notifier invalidation (mmu_lock -> mmap_lock).

3. Skip kvm_release_faultin_page() for gmem pages; their lifetime is

managed by the gmem inode and invalidation notifier.

4. Skip hva lookup and validity check in gstage_page_fault() for gmem,

which does not have a userspace VMA backing it.

Changes since RFC v1:

- Add vcpu_exit.c hva lookup bypass for gmem slots

- Call kvm_prepare_memory_fault_exit() on gmem fault failure

- Respect KVM_MEM_READONLY for gmem slot writability

- Skip THP adjustment for gmem (only PAGE_SIZE supported)

- Skip dirty tracking for gmem (handled by gmem notifier)

- Replace goto with is_gmem variable for cleaner flow

Assisted-by YuanSheng: deepseek-v4-pro
---
 arch/riscv/kvm/Kconfig     |   1 +
 arch/riscv/kvm/mmu.c       | 134 +++++++++++++++++++++++--------------
 arch/riscv/kvm/vcpu_exit.c |  17 ++++-
 3 files changed, 99 insertions(+), 53 deletions(-)
arch/riscv/kvm/Kconfig +1 -0
index ec2cee0a39e0..77121100579f 100644
@@ -29,6 +29,7 @@ config KVM
 	select KVM_GENERIC_DIRTYLOG_READ_PROTECT
 	select KVM_GENERIC_HARDWARE_ENABLING
 	select KVM_MMIO
+	select KVM_GUEST_MEMFD
 	select VIRT_XFER_TO_GUEST_WORK
 	select SCHED_INFO
 	select GUEST_PERF_EVENTS if PERF_EVENTS
arch/riscv/kvm/mmu.c +84 -50

169 lines not on this page · raw message

arch/riscv/kvm/vcpu_exit.c +14 -3
index 28cf9b27bb07..5c381a4f9977 100644
@@ -26,10 +26,21 @@ static int gstage_page_fault(struct kvm_vcpu *vcpu, struct kvm_run *run,
 	fault_addr = ((gpa_t)trap->htval << 2) | (trap->stval & 0x3);
 	gfn = fault_addr >> PAGE_SHIFT;
 	memslot = gfn_to_memslot(vcpu->kvm, gfn);
-	hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable);
 
-	if (kvm_is_error_hva(hva) ||
-	    (trap->scause == EXC_STORE_GUEST_PAGE_FAULT && !writable)) {
+	/*
+	 * For guest_memfd-backed slots, bypass the hva lookup and validity check
+	 * since gmem does not have a userspace VMA backing it.
+	 */
+	if (kvm_slot_has_gmem(memslot)) {
+		hva = 0;
+		writable = !(memslot->flags & KVM_MEM_READONLY);
+	} else {
+		hva = gfn_to_hva_memslot_prot(memslot, gfn, &writable);
+	}
+
+	if (!kvm_slot_has_gmem(memslot) &&
+	    (kvm_is_error_hva(hva) ||
+	     (trap->scause == EXC_STORE_GUEST_PAGE_FAULT && !writable))) {
 		switch (trap->scause) {
 		case EXC_LOAD_GUEST_PAGE_FAULT:
 			return kvm_riscv_vcpu_mmio_load(vcpu, run,
Signature · 1 line
2.54.0