On Fri, Aug 30, 2024 at 03:11:34PM +0100, Marc Zyngier wrote:

On Fri, 30 Aug 2024 11:24:53 +0100,
Vincent Donnefort [off-list ref] wrote:

quoted

Hi Seb,

Thanks for the respin.

On Tue, Aug 27, 2024 at 08:45:47AM +0000, Sebastian Ene wrote:

quoted

While arch/*/mem/ptdump handles the kernel pagetable dumping code,
introduce KVM/ptdump to show the guest stage-2 pagetables. The
separation is necessary because most of the definitions from the
stage-2 pagetable reside in the KVM path and we will be invoking
functionality specific to KVM.

When a guest is created, register a new file entry under the guest
debugfs dir which allows userspace to show the contents of the guest
stage-2 pagetables when accessed.

Signed-off-by: Sebastian Ene <redacted>

I only have some nits, otherwise:

Reviewed-by: Vincent Donnefort <redacted>

quoted

---
 arch/arm64/include/asm/kvm_host.h |   6 +
 arch/arm64/kvm/Makefile           |   1 +
 arch/arm64/kvm/arm.c              |   1 +
 arch/arm64/kvm/ptdump.c           | 247 ++++++++++++++++++++++++++++++
 4 files changed, 255 insertions(+)
 create mode 100644 arch/arm64/kvm/ptdump.c

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index a33f5996ca9f..4acd589f086b 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h

@@ -1473,4 +1473,10 @@ void kvm_set_vm_id_reg(struct kvm *kvm, u32 reg, u64 val);
 		(pa + pi + pa3) == 1;					\
 	})
 
+#ifdef CONFIG_PTDUMP_STAGE2_DEBUGFS
+void kvm_s2_ptdump_create_debugfs(struct kvm *kvm);
+#else
+static inline void kvm_s2_ptdump_create_debugfs(struct kvm *kvm) {}
+#endif /* CONFIG_PTDUMP_STAGE2_DEBUGFS */
+
 #endif /* __ARM64_KVM_HOST_H__ */

diff --git a/arch/arm64/kvm/Makefile b/arch/arm64/kvm/Makefile
index 86a629aaf0a1..e4233b323a73 100644
--- a/arch/arm64/kvm/Makefile
+++ b/arch/arm64/kvm/Makefile

@@ -27,6 +27,7 @@ kvm-y += arm.o mmu.o mmio.o psci.o hypercalls.o pvtime.o \
 
 kvm-$(CONFIG_HW_PERF_EVENTS)  += pmu-emul.o pmu.o
 kvm-$(CONFIG_ARM64_PTR_AUTH)  += pauth.o
+kvm-$(CONFIG_PTDUMP_STAGE2_DEBUGFS) += ptdump.o
 
 always-y := hyp_constants.h hyp-constants.s
 

diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 9bef7638342e..b9fd928d3477 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c

@@ -228,6 +228,7 @@ vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 void kvm_arch_create_vm_debugfs(struct kvm *kvm)
 {
 	kvm_sys_regs_create_debugfs(kvm);
+	kvm_s2_ptdump_create_debugfs(kvm);
 }
 
 static void kvm_destroy_mpidr_data(struct kvm *kvm)

diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
new file mode 100644
index 000000000000..e72a928d4445
--- /dev/null
+++ b/arch/arm64/kvm/ptdump.c

@@ -0,0 +1,247 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Debug helper used to dump the stage-2 pagetables of the system and their
+ * associated permissions.
+ *
+ * Copyright (C) Google, 2024
+ * Author: Sebastian Ene <sebastianene@google.com>
+ */
+#include <linux/debugfs.h>
+#include <linux/kvm_host.h>
+#include <linux/seq_file.h>
+
+#include <asm/kvm_pgtable.h>
+#include <asm/kvm_host.h>

nit: I believe you wanted to follow the alphabetical order, if that is the case,
kvm_host.h then kvm_pgtable.h

quoted

+#include <asm/ptdump.h>
+
+

nit: don't think double empty are a rule, I would remove it.

quoted

+#define MARKERS_LEN		(2)

nit: The brackets are not necessary for MARKERS_LEN.

quoted

+#define KVM_PGTABLE_MAX_LEVELS	(KVM_PGTABLE_LAST_LEVEL + 1)
+
+struct kvm_ptdump_guest_state {
+	struct kvm		*kvm;
+	struct ptdump_pg_state	parser_state;
+	struct addr_marker	ipa_marker[MARKERS_LEN];
+	struct ptdump_pg_level	level[KVM_PGTABLE_MAX_LEVELS];
+	struct ptdump_range	range[MARKERS_LEN];
+};
+
+static const struct ptdump_prot_bits stage2_pte_bits[] = {
+	{
+		.mask	= PTE_VALID,
+		.val	= PTE_VALID,
+		.set	= " ",
+		.clear	= "F",

This is effectively never used because an invalid PTE is 0 and note_page() won't
print it. This probably can be removed?

Yeah, I can't see how we are going to trigger that one given that
PTE_VALID must be set, and that we only print something if the bit is
clear.

Seb?

When (pte_prot & mask == val) we print .set (which is empty), otherwise we
print .clear (This is from dump_prot). Invalid non-zero PTEs should be
printed in this case.

quoted

quoted

+	}, {
+		.mask	= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R | PTE_VALID,
+		.val	= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_R | PTE_VALID,
+		.set	= "R",
+		.clear	= " ",
+	}, {
+		.mask	= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W | PTE_VALID,
+		.val	= KVM_PTE_LEAF_ATTR_LO_S2_S2AP_W | PTE_VALID,
+		.set	= "W",
+		.clear	= " ",
+	}, {
+		.mask	= KVM_PTE_LEAF_ATTR_HI_S2_XN | PTE_VALID,
+		.val	= PTE_VALID,
+		.set	= " ",
+		.clear	= "X",
+	}, {
+		.mask	= KVM_PTE_LEAF_ATTR_LO_S2_AF | PTE_VALID,
+		.val	= KVM_PTE_LEAF_ATTR_LO_S2_AF | PTE_VALID,
+		.set	= "AF",
+		.clear	= "  ",
+	}, {
+		.mask	= PTE_TABLE_BIT | PTE_VALID,
+		.val	= PTE_VALID,
+		.set	= "BLK",
+		.clear	= "   ",
+	},
+};
+
+static int kvm_ptdump_visitor(const struct kvm_pgtable_visit_ctx *ctx,
+			      enum kvm_pgtable_walk_flags visit)
+{
+	struct ptdump_pg_state *st = ctx->arg;
+	struct ptdump_state *pt_st = &st->ptdump;
+
+	note_page(pt_st, ctx->addr, ctx->level, ctx->old);
+
+	return 0;
+}
+
+static int kvm_ptdump_build_levels(struct ptdump_pg_level *level, u32 start_lvl)
+{
+	u32 i;
+	u64 mask;
+
+	if (WARN_ON_ONCE(start_lvl >= KVM_PGTABLE_LAST_LEVEL))
+		return -EINVAL;
+
+	mask = 0;
+	for (i = 0; i < ARRAY_SIZE(stage2_pte_bits); i++)
+		mask |= stage2_pte_bits[i].mask;
+
+	for (i = start_lvl; i < KVM_PGTABLE_MAX_LEVELS; i++) {
+		snprintf(level[i].name, sizeof(level[i].name), "%d", i);

%u, i being unsigned.

quoted

+
+		level[i].num	= ARRAY_SIZE(stage2_pte_bits);
+		level[i].bits	= stage2_pte_bits;
+		level[i].mask	= mask;
+	}
+
+	return 0;
+}
+
+static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm *kvm)
+{
+	struct kvm_ptdump_guest_state *st;
+	struct kvm_s2_mmu *mmu = &kvm->arch.mmu;
+	struct kvm_pgtable *pgtable = mmu->pgt;
+	int ret;
+
+	st = kzalloc(sizeof(struct kvm_ptdump_guest_state), GFP_KERNEL_ACCOUNT);
+	if (!st)
+		return ERR_PTR(-ENOMEM);
+
+	ret = kvm_ptdump_build_levels(&st->level[0], pgtable->start_level);
+	if (ret) {
+		kfree(st);
+		return ERR_PTR(ret);
+	}
+
+	st->ipa_marker[0].name		= "Guest IPA";
+	st->ipa_marker[1].start_address = BIT(pgtable->ia_bits);
+	st->range[0].end		= BIT(pgtable->ia_bits);
+
+	st->kvm				= kvm;
+	st->parser_state = (struct ptdump_pg_state) {
+		.marker		= &st->ipa_marker[0],
+		.level		= -1,
+		.pg_level	= &st->level[0],
+		.ptdump.range	= &st->range[0],
+		.start_address	= 0,
+	};
+
+	return st;
+}
+
+static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
+{
+	int ret;
+	struct kvm_ptdump_guest_state *st = m->private;
+	struct kvm *kvm = st->kvm;
+	struct kvm_s2_mmu *mmu = &kvm->arch.mmu;
+	struct ptdump_pg_state *parser_state = &st->parser_state;
+	struct kvm_pgtable_walker walker = (struct kvm_pgtable_walker) {
+		.cb	= kvm_ptdump_visitor,
+		.arg	= parser_state,
+		.flags	= KVM_PGTABLE_WALK_LEAF,
+	};
+
+	parser_state->seq = m;
+
+	write_lock(&kvm->mmu_lock);
+	ret = kvm_pgtable_walk(mmu->pgt, 0, BIT(mmu->pgt->ia_bits), &walker);
+	write_unlock(&kvm->mmu_lock);
+
+	return ret;
+}
+
+static int kvm_ptdump_guest_open(struct inode *m, struct file *file)
+{
+	struct kvm *kvm = m->i_private;
+	struct kvm_ptdump_guest_state *st;
+	int ret;
+
+	if (!kvm_get_kvm_safe(kvm))
+		return -ENOENT;
+
+	st = kvm_ptdump_parser_create(kvm);
+	if (IS_ERR(st)) {
+		ret = PTR_ERR(st);
+		goto free_with_kvm_ref;
+	}
+
+	ret = single_open(file, kvm_ptdump_guest_show, st);
+	if (!ret)
+		return 0;
+
+	kfree(st);
+free_with_kvm_ref:

nit: I believe kfree understands IS_ERR() so you could have a simple "err:"
label covering all the error path.

I couldn't find such handling in kfree(). Could you point be to it?

quoted

quoted

+	kvm_put_kvm(kvm);
+	return ret;
+}
+
+static int kvm_ptdump_guest_close(struct inode *m, struct file *file)
+{
+	struct kvm *kvm = m->i_private;
+	void *st = ((struct seq_file *)file->private_data)->private;
+
+	kfree(st);
+	kvm_put_kvm(kvm);
+
+	return single_release(m, file);
+}
+
+static const struct file_operations kvm_ptdump_guest_fops = {
+	.open		= kvm_ptdump_guest_open,
+	.read		= seq_read,
+	.llseek		= seq_lseek,
+	.release	= kvm_ptdump_guest_close,
+};
+
+static int kvm_pgtable_debugfs_show(struct seq_file *m, void *unused)
+{
+	const struct file *file = m->file;
+	struct kvm_pgtable *pgtable = m->private;
+
+	if (!strcmp(file_dentry(file)->d_iname, "ipa_range"))

I really dislike this sort of construct, and I'd rather we pick the
correct callback by construction rather than relying on a string
comparison. See below for a suggestion.

Thanks for the feedback and for the suggestion. Let me try to re-write
this as indicated.

Thanks,
Seb

-- 
Without deviation from the norm, progress is not possible.