Re: [PATCH v8 4/6] KVM: arm64: Register ptdump with debugfs on guest creation

[Marc Zyngier <maz@kernel.org>]

On Fri, 16 Aug 2024 13:39:04 +0100,
Sebastian Ene <sebastianene@google.com> wrote:
> 
> While arch/*/mem/ptdump handles the kernel pagetable dumping code,
> introduce KVM/ptdump which shows 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.

Drop the ** emphasis.

> 
> 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 <sebastianene@google.com>
> ---
>  arch/arm64/kvm/Kconfig      | 14 ++++++
>  arch/arm64/kvm/Makefile     |  1 +
>  arch/arm64/kvm/arm.c        |  2 +
>  arch/arm64/kvm/kvm_ptdump.h | 20 ++++++++
>  arch/arm64/kvm/ptdump.c     | 91 +++++++++++++++++++++++++++++++++++++
>  5 files changed, 128 insertions(+)
>  create mode 100644 arch/arm64/kvm/kvm_ptdump.h
>  create mode 100644 arch/arm64/kvm/ptdump.c
> 
> diff --git a/arch/arm64/kvm/Kconfig b/arch/arm64/kvm/Kconfig
> index 8304eb342be9..fcc41e58ede6 100644
> --- a/arch/arm64/kvm/Kconfig
> +++ b/arch/arm64/kvm/Kconfig
> @@ -66,4 +66,18 @@ config PROTECTED_NVHE_STACKTRACE
>  
>  	  If unsure, or not using protected nVHE (pKVM), say N.
>  
> +config PTDUMP_STAGE2_DEBUGFS
> +       bool "Present the stage-2 pagetables to debugfs"
> +       depends on KVM
> +       select PTDUMP_CORE

This looks wrong. Looking at PTDUMP_DEBUGFS, it has the following
constraints:

        depends on DEBUG_KERNEL
        depends on DEBUG_FS
        depends on GENERIC_PTDUMP
        select PTDUMP_CORE

I don't see why the Stage-2 version should have anything different.

> +       default n
> +       help
> +         Say Y here if you want to show the stage-2 kernel pagetables
> +         layout in a debugfs file. This information is only useful for kernel developers
> +         who are working in architecture specific areas of the kernel.
> +         It is probably not a good idea to enable this feature in a production
> +         kernel.

nit: try to keep the formatting within 80 columns.

More importantly, I find it very strange to expose the configuration
option so early in the series, while the support code isn't there yet.
You can perfectly introduce code that is conditional on a config
option and only add it at the end.

> +
> +         If in doubt, say N.
> +
>  endif # VIRTUALIZATION
> 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..60fed2146763 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -45,6 +45,7 @@
>  #include <kvm/arm_hypercalls.h>
>  #include <kvm/arm_pmu.h>
>  #include <kvm/arm_psci.h>
> +#include <kvm_ptdump.h>
>  
>  static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
>  
> @@ -228,6 +229,7 @@ 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_ptdump_guest_register(kvm);

Consider using a name that is homogeneous with what we already have
(kvm_s2_ptdump_create_debugfs?).

>  }
>  
>  static void kvm_destroy_mpidr_data(struct kvm *kvm)
> diff --git a/arch/arm64/kvm/kvm_ptdump.h b/arch/arm64/kvm/kvm_ptdump.h
> new file mode 100644
> index 000000000000..0a62b0e2908c
> --- /dev/null
> +++ b/arch/arm64/kvm/kvm_ptdump.h
> @@ -0,0 +1,20 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Copyright (C) Google, 2024
> + * Author: Sebastian Ene <sebastianene@google.com>
> + */
> +
> +#ifndef __KVM_PTDUMP_H
> +#define __KVM_PTDUMP_H
> +
> +#include <linux/kvm_host.h>
> +#include <asm/ptdump.h>
> +
> +
> +#ifdef CONFIG_PTDUMP_STAGE2_DEBUGFS
> +void kvm_ptdump_guest_register(struct kvm *kvm);
> +#else
> +static inline void kvm_ptdump_guest_register(struct kvm *kvm) {}
> +#endif /* CONFIG_PTDUMP_STAGE2_DEBUGFS */
> +
> +#endif /* __KVM_PTDUMP_H */

Please don't add new include files that contain so little stuff. These
things may as well be added either to asm/kvm_host.h or asm/ptdump.h.

> diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
> new file mode 100644
> index 000000000000..52483d56be2e
> --- /dev/null
> +++ b/arch/arm64/kvm/ptdump.c
> @@ -0,0 +1,91 @@
> +// 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 <kvm_ptdump.h>
> +
> +
> +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_show_common(struct seq_file *m,

What does "common" mean here? You have exactly *one* caller, so why
isn't that inlined in kvm_ptdump_guest_show()?

> +				  struct kvm_pgtable *pgtable,
> +				  struct ptdump_pg_state *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->level = -1;
> +	parser_state->start_address = 0;
> +
> +	return kvm_pgtable_walk(pgtable, 0, BIT(pgtable->ia_bits), &walker);
> +}
> +
> +static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
> +{
> +	struct kvm *kvm = m->private;
> +	struct kvm_s2_mmu *mmu = &kvm->arch.mmu;
> +	struct ptdump_pg_state parser_state = {0};

nit: the common idiom is "parser_state = {}".

> +	int ret;
> +
> +	write_lock(&kvm->mmu_lock);
> +	ret = kvm_ptdump_show_common(m, mmu->pgt, &parser_state);
> +	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;
> +	int ret;
> +
> +	if (!kvm_get_kvm_safe(kvm))
> +		return -ENOENT;
> +
> +	ret = single_open(file, kvm_ptdump_guest_show, m->i_private);
> +	if (ret < 0)
> +		kvm_put_kvm(kvm);
> +
> +	return ret;
> +}
> +
> +static int kvm_ptdump_guest_close(struct inode *m, struct file *file)
> +{
> +	struct kvm *kvm = m->i_private;
> +
> +	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,
> +};
> +
> +void kvm_ptdump_guest_register(struct kvm *kvm)
> +{
> +	debugfs_create_file("stage2_page_tables", 0400, kvm->debugfs_dentry,
> +			    kvm, &kvm_ptdump_guest_fops);
> +}
> -- 
> 2.46.0.184.g6999bdac58-goog

Overall, I have a hard time understanding what this does. It walks the
page tables, but doesn't do anything useful.

I have the feeling that this patch would be better squashed together
with patch #5, since it significantly reworks what patch #4 does.

Thanks,

	M.

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