[v2,11/11] arm64: ptdump: Add support for guest stage-2 pagetables dumping
Commit Message
Register a debugfs file on guest creation to be able to view their
second translation tables with ptdump. This assumes that the host is in
control of the guest stage-2 and has direct access to the pagetables.
Signed-off-by: Sebastian Ene <sebastianene@google.com>
---
arch/arm64/include/asm/ptdump.h | 21 +++++++--
arch/arm64/kvm/mmu.c | 3 ++
arch/arm64/mm/ptdump.c | 84 +++++++++++++++++++++++++++++++++
arch/arm64/mm/ptdump_debugfs.c | 5 +-
4 files changed, 108 insertions(+), 5 deletions(-)
Comments
On Thu, Oct 19, 2023 at 02:40:33PM +0000, Sebastian Ene wrote:
> Register a debugfs file on guest creation to be able to view their
> second translation tables with ptdump. This assumes that the host is in
> control of the guest stage-2 and has direct access to the pagetables.
What about pKVM? The walker you wrote for the host stage-2 should be
reusable in that case?
>
> Signed-off-by: Sebastian Ene <sebastianene@google.com>
> ---
> arch/arm64/include/asm/ptdump.h | 21 +++++++--
> arch/arm64/kvm/mmu.c | 3 ++
> arch/arm64/mm/ptdump.c | 84 +++++++++++++++++++++++++++++++++
> arch/arm64/mm/ptdump_debugfs.c | 5 +-
> 4 files changed, 108 insertions(+), 5 deletions(-)
>
> diff --git a/arch/arm64/include/asm/ptdump.h b/arch/arm64/include/asm/ptdump.h
> index 35b883524462..be86244d532b 100644
> --- a/arch/arm64/include/asm/ptdump.h
> +++ b/arch/arm64/include/asm/ptdump.h
> @@ -5,6 +5,8 @@
> #ifndef __ASM_PTDUMP_H
> #define __ASM_PTDUMP_H
>
> +#include <asm/kvm_pgtable.h>
> +
> #ifdef CONFIG_PTDUMP_CORE
>
> #include <linux/mm_types.h>
> @@ -30,14 +32,27 @@ struct ptdump_info {
> void ptdump_walk(struct seq_file *s, struct ptdump_info *info);
> #ifdef CONFIG_PTDUMP_DEBUGFS
> #define EFI_RUNTIME_MAP_END DEFAULT_MAP_WINDOW_64
> -void __init ptdump_debugfs_register(struct ptdump_info *info, const char *name);
> +struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
> + const char *name);
> #else
> -static inline void ptdump_debugfs_register(struct ptdump_info *info,
> - const char *name) { }
> +static inline struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
> + const char *name)
> +{
> + return NULL;
> +}
> #endif
> void ptdump_check_wx(void);
> #endif /* CONFIG_PTDUMP_CORE */
>
> +#ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
> +void ptdump_register_guest_stage2(struct kvm_pgtable *pgt, void *lock);
> +void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt);
> +#else
> +static inline void ptdump_register_guest_stage2(struct kvm_pgtable *pgt,
> + void *lock) { }
> +static inline void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt) { }
> +#endif /* CONFIG_NVHE_EL2_PTDUMP_DEBUGFS */
I believe this should be compatible with VHE as well, that option should be
renamed.
> +
> #ifdef CONFIG_DEBUG_WX
> #define debug_checkwx() ptdump_check_wx()
> #else
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 482280fe22d7..e47988dba34d 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -11,6 +11,7 @@
> #include <linux/sched/signal.h>
> #include <trace/events/kvm.h>
> #include <asm/pgalloc.h>
> +#include <asm/ptdump.h>
> #include <asm/cacheflush.h>
> #include <asm/kvm_arm.h>
> #include <asm/kvm_mmu.h>
> @@ -908,6 +909,7 @@ int kvm_init_stage2_mmu(struct kvm *kvm, struct kvm_s2_mmu *mmu, unsigned long t
> if (err)
> goto out_free_pgtable;
>
> + ptdump_register_guest_stage2(pgt, &kvm->mmu_lock);
> mmu->last_vcpu_ran = alloc_percpu(typeof(*mmu->last_vcpu_ran));
> if (!mmu->last_vcpu_ran) {
> err = -ENOMEM;
> @@ -1021,6 +1023,7 @@ void kvm_free_stage2_pgd(struct kvm_s2_mmu *mmu)
> write_unlock(&kvm->mmu_lock);
>
> if (pgt) {
> + ptdump_unregister_guest_stage2(pgt);
> kvm_pgtable_stage2_destroy(pgt);
> kfree(pgt);
> }
> diff --git a/arch/arm64/mm/ptdump.c b/arch/arm64/mm/ptdump.c
> index 4687840dcb69..facfb15468f5 100644
> --- a/arch/arm64/mm/ptdump.c
> +++ b/arch/arm64/mm/ptdump.c
> @@ -26,6 +26,7 @@
> #include <asm/ptdump.h>
> #include <asm/kvm_pkvm.h>
> #include <asm/kvm_pgtable.h>
> +#include <asm/kvm_host.h>
>
>
> enum address_markers_idx {
> @@ -543,6 +544,22 @@ void ptdump_check_wx(void)
> #ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
> static struct ptdump_info stage2_kernel_ptdump_info;
>
> +#define GUEST_NAME_LEN (32U)
> +
> +struct ptdump_registered_guest {
> + struct list_head reg_list;
> + struct ptdump_info info;
> + struct mm_struct mem;
> + struct kvm_pgtable_snapshot snapshot;
> + struct dentry *dentry;
> + rwlock_t *lock;
> + char reg_name[GUEST_NAME_LEN];
> +};
> +
> +static LIST_HEAD(ptdump_guest_list);
> +static DEFINE_MUTEX(ptdump_list_lock);
> +static u16 guest_no;
This is not robust enough: If 1 VM starts then 65535 others which are killed.
guest_no overflows. The next number is 0 which is already taken.
Linux has and ID allocation to solve this problem, but I don't think this is
necessary anyway. This should simply reuse the struct kvm->debugfs_dentry.
Also probably most of the informations contained in ptdump_registered_guest can
be found in struct kvm. The debugfs should then probably simply take struct kvm
for the private argument.
> +
> static phys_addr_t ptdump_host_pa(void *addr)
> {
> return __pa(addr);
> @@ -740,6 +757,73 @@ static void stage2_ptdump_walk(struct seq_file *s, struct ptdump_info *info)
>
> kvm_pgtable_walk(pgtable, start_ipa, end_ipa, &walker);
> }
[...]
On Fri, Oct 20, 2023 at 09:40:06AM +0100, Vincent Donnefort wrote:
> On Thu, Oct 19, 2023 at 02:40:33PM +0000, Sebastian Ene wrote:
> > Register a debugfs file on guest creation to be able to view their
> > second translation tables with ptdump. This assumes that the host is in
> > control of the guest stage-2 and has direct access to the pagetables.
>
> What about pKVM? The walker you wrote for the host stage-2 should be
> reusable in that case?
>
Yes, when pKVM will be ready upstream the walker which duplicates the
pagetables for the host will be re-used for the guests. We will have to
add a separate HVC for this which receives as an argument the guest
vmid.
> >
> > Signed-off-by: Sebastian Ene <sebastianene@google.com>
> > ---
> > arch/arm64/include/asm/ptdump.h | 21 +++++++--
> > arch/arm64/kvm/mmu.c | 3 ++
> > arch/arm64/mm/ptdump.c | 84 +++++++++++++++++++++++++++++++++
> > arch/arm64/mm/ptdump_debugfs.c | 5 +-
> > 4 files changed, 108 insertions(+), 5 deletions(-)
> >
> > diff --git a/arch/arm64/include/asm/ptdump.h b/arch/arm64/include/asm/ptdump.h
> > index 35b883524462..be86244d532b 100644
> > --- a/arch/arm64/include/asm/ptdump.h
> > +++ b/arch/arm64/include/asm/ptdump.h
> > @@ -5,6 +5,8 @@
> > #ifndef __ASM_PTDUMP_H
> > #define __ASM_PTDUMP_H
> >
> > +#include <asm/kvm_pgtable.h>
> > +
> > #ifdef CONFIG_PTDUMP_CORE
> >
> > #include <linux/mm_types.h>
> > @@ -30,14 +32,27 @@ struct ptdump_info {
> > void ptdump_walk(struct seq_file *s, struct ptdump_info *info);
> > #ifdef CONFIG_PTDUMP_DEBUGFS
> > #define EFI_RUNTIME_MAP_END DEFAULT_MAP_WINDOW_64
> > -void __init ptdump_debugfs_register(struct ptdump_info *info, const char *name);
> > +struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
> > + const char *name);
> > #else
> > -static inline void ptdump_debugfs_register(struct ptdump_info *info,
> > - const char *name) { }
> > +static inline struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
> > + const char *name)
> > +{
> > + return NULL;
> > +}
> > #endif
> > void ptdump_check_wx(void);
> > #endif /* CONFIG_PTDUMP_CORE */
> >
> > +#ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
> > +void ptdump_register_guest_stage2(struct kvm_pgtable *pgt, void *lock);
> > +void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt);
> > +#else
> > +static inline void ptdump_register_guest_stage2(struct kvm_pgtable *pgt,
> > + void *lock) { }
> > +static inline void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt) { }
> > +#endif /* CONFIG_NVHE_EL2_PTDUMP_DEBUGFS */
>
> I believe this should be compatible with VHE as well, that option should be
> renamed.
>
Good point, I will rename this.
> > +
> > #ifdef CONFIG_DEBUG_WX
> > #define debug_checkwx() ptdump_check_wx()
> > #else
> > diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> > index 482280fe22d7..e47988dba34d 100644
> > --- a/arch/arm64/kvm/mmu.c
> > +++ b/arch/arm64/kvm/mmu.c
> > @@ -11,6 +11,7 @@
> > #include <linux/sched/signal.h>
> > #include <trace/events/kvm.h>
> > #include <asm/pgalloc.h>
> > +#include <asm/ptdump.h>
> > #include <asm/cacheflush.h>
> > #include <asm/kvm_arm.h>
> > #include <asm/kvm_mmu.h>
> > @@ -908,6 +909,7 @@ int kvm_init_stage2_mmu(struct kvm *kvm, struct kvm_s2_mmu *mmu, unsigned long t
> > if (err)
> > goto out_free_pgtable;
> >
> > + ptdump_register_guest_stage2(pgt, &kvm->mmu_lock);
> > mmu->last_vcpu_ran = alloc_percpu(typeof(*mmu->last_vcpu_ran));
> > if (!mmu->last_vcpu_ran) {
> > err = -ENOMEM;
> > @@ -1021,6 +1023,7 @@ void kvm_free_stage2_pgd(struct kvm_s2_mmu *mmu)
> > write_unlock(&kvm->mmu_lock);
> >
> > if (pgt) {
> > + ptdump_unregister_guest_stage2(pgt);
> > kvm_pgtable_stage2_destroy(pgt);
> > kfree(pgt);
> > }
> > diff --git a/arch/arm64/mm/ptdump.c b/arch/arm64/mm/ptdump.c
> > index 4687840dcb69..facfb15468f5 100644
> > --- a/arch/arm64/mm/ptdump.c
> > +++ b/arch/arm64/mm/ptdump.c
> > @@ -26,6 +26,7 @@
> > #include <asm/ptdump.h>
> > #include <asm/kvm_pkvm.h>
> > #include <asm/kvm_pgtable.h>
> > +#include <asm/kvm_host.h>
> >
> >
> > enum address_markers_idx {
> > @@ -543,6 +544,22 @@ void ptdump_check_wx(void)
> > #ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
> > static struct ptdump_info stage2_kernel_ptdump_info;
> >
> > +#define GUEST_NAME_LEN (32U)
> > +
> > +struct ptdump_registered_guest {
> > + struct list_head reg_list;
> > + struct ptdump_info info;
> > + struct mm_struct mem;
> > + struct kvm_pgtable_snapshot snapshot;
> > + struct dentry *dentry;
> > + rwlock_t *lock;
> > + char reg_name[GUEST_NAME_LEN];
> > +};
> > +
> > +static LIST_HEAD(ptdump_guest_list);
> > +static DEFINE_MUTEX(ptdump_list_lock);
> > +static u16 guest_no;
>
> This is not robust enough: If 1 VM starts then 65535 others which are killed.
> guest_no overflows. The next number is 0 which is already taken.
>
Yes, I guess this should be improved. In the case you described we won't
register any debugfs file because of the name clash.
> Linux has and ID allocation to solve this problem, but I don't think this is
> necessary anyway. This should simply reuse the struct kvm->debugfs_dentry.
>
> Also probably most of the informations contained in ptdump_registered_guest can
> be found in struct kvm. The debugfs should then probably simply take struct kvm
> for the private argument.
>
I would prefer to keep it as a separate struct here as it gives some
flexibility if we need to extend it for guests pKVM support. I think we
can drop the struct mm_struct from here.
Thanks,
Sebastian
> > +
> > static phys_addr_t ptdump_host_pa(void *addr)
> > {
> > return __pa(addr);
> > @@ -740,6 +757,73 @@ static void stage2_ptdump_walk(struct seq_file *s, struct ptdump_info *info)
> >
> > kvm_pgtable_walk(pgtable, start_ipa, end_ipa, &walker);
> > }
>
> [...]
@@ -5,6 +5,8 @@
#ifndef __ASM_PTDUMP_H
#define __ASM_PTDUMP_H
+#include <asm/kvm_pgtable.h>
+
#ifdef CONFIG_PTDUMP_CORE
#include <linux/mm_types.h>
@@ -30,14 +32,27 @@ struct ptdump_info {
void ptdump_walk(struct seq_file *s, struct ptdump_info *info);
#ifdef CONFIG_PTDUMP_DEBUGFS
#define EFI_RUNTIME_MAP_END DEFAULT_MAP_WINDOW_64
-void __init ptdump_debugfs_register(struct ptdump_info *info, const char *name);
+struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
+ const char *name);
#else
-static inline void ptdump_debugfs_register(struct ptdump_info *info,
- const char *name) { }
+static inline struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
+ const char *name)
+{
+ return NULL;
+}
#endif
void ptdump_check_wx(void);
#endif /* CONFIG_PTDUMP_CORE */
+#ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
+void ptdump_register_guest_stage2(struct kvm_pgtable *pgt, void *lock);
+void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt);
+#else
+static inline void ptdump_register_guest_stage2(struct kvm_pgtable *pgt,
+ void *lock) { }
+static inline void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt) { }
+#endif /* CONFIG_NVHE_EL2_PTDUMP_DEBUGFS */
+
#ifdef CONFIG_DEBUG_WX
#define debug_checkwx() ptdump_check_wx()
#else
@@ -11,6 +11,7 @@
#include <linux/sched/signal.h>
#include <trace/events/kvm.h>
#include <asm/pgalloc.h>
+#include <asm/ptdump.h>
#include <asm/cacheflush.h>
#include <asm/kvm_arm.h>
#include <asm/kvm_mmu.h>
@@ -908,6 +909,7 @@ int kvm_init_stage2_mmu(struct kvm *kvm, struct kvm_s2_mmu *mmu, unsigned long t
if (err)
goto out_free_pgtable;
+ ptdump_register_guest_stage2(pgt, &kvm->mmu_lock);
mmu->last_vcpu_ran = alloc_percpu(typeof(*mmu->last_vcpu_ran));
if (!mmu->last_vcpu_ran) {
err = -ENOMEM;
@@ -1021,6 +1023,7 @@ void kvm_free_stage2_pgd(struct kvm_s2_mmu *mmu)
write_unlock(&kvm->mmu_lock);
if (pgt) {
+ ptdump_unregister_guest_stage2(pgt);
kvm_pgtable_stage2_destroy(pgt);
kfree(pgt);
}
@@ -26,6 +26,7 @@
#include <asm/ptdump.h>
#include <asm/kvm_pkvm.h>
#include <asm/kvm_pgtable.h>
+#include <asm/kvm_host.h>
enum address_markers_idx {
@@ -543,6 +544,22 @@ void ptdump_check_wx(void)
#ifdef CONFIG_NVHE_EL2_PTDUMP_DEBUGFS
static struct ptdump_info stage2_kernel_ptdump_info;
+#define GUEST_NAME_LEN (32U)
+
+struct ptdump_registered_guest {
+ struct list_head reg_list;
+ struct ptdump_info info;
+ struct mm_struct mem;
+ struct kvm_pgtable_snapshot snapshot;
+ struct dentry *dentry;
+ rwlock_t *lock;
+ char reg_name[GUEST_NAME_LEN];
+};
+
+static LIST_HEAD(ptdump_guest_list);
+static DEFINE_MUTEX(ptdump_list_lock);
+static u16 guest_no;
+
static phys_addr_t ptdump_host_pa(void *addr)
{
return __pa(addr);
@@ -740,6 +757,73 @@ static void stage2_ptdump_walk(struct seq_file *s, struct ptdump_info *info)
kvm_pgtable_walk(pgtable, start_ipa, end_ipa, &walker);
}
+
+static void guest_stage2_ptdump_walk(struct seq_file *s,
+ struct ptdump_info *info)
+{
+ struct kvm_pgtable_snapshot *snapshot = info->priv;
+ struct ptdump_registered_guest *guest;
+
+ guest = container_of(snapshot, struct ptdump_registered_guest,
+ snapshot);
+ read_lock(guest->lock);
+ stage2_ptdump_walk(s, info);
+ read_unlock(guest->lock);
+}
+
+void ptdump_register_guest_stage2(struct kvm_pgtable *pgt, void *lock)
+{
+ struct ptdump_registered_guest *guest;
+ struct dentry *d;
+
+ if (pgt == NULL || lock == NULL)
+ return;
+
+ guest = kzalloc(sizeof(struct ptdump_registered_guest), GFP_KERNEL);
+ if (!guest)
+ return;
+
+ memcpy(&guest->snapshot.pgtable, pgt, sizeof(struct kvm_pgtable));
+ guest->info = (struct ptdump_info) {
+ .ptdump_walk = guest_stage2_ptdump_walk,
+ .priv = &guest->snapshot
+ };
+
+ mutex_init(&guest->info.file_lock);
+ guest->lock = lock;
+ mutex_lock(&ptdump_list_lock);
+ snprintf(guest->reg_name, GUEST_NAME_LEN,
+ "%u_guest_stage2_page_tables", guest_no++);
+ d = ptdump_debugfs_register(&guest->info, guest->reg_name);
+ if (!d) {
+ mutex_unlock(&ptdump_list_lock);
+ goto free_entry;
+ }
+
+ guest->dentry = d;
+ list_add(&guest->reg_list, &ptdump_guest_list);
+ mutex_unlock(&ptdump_list_lock);
+ return;
+
+free_entry:
+ kfree(guest);
+}
+
+void ptdump_unregister_guest_stage2(struct kvm_pgtable *pgt)
+{
+ struct ptdump_registered_guest *guest;
+
+ mutex_lock(&ptdump_list_lock);
+ list_for_each_entry(guest, &ptdump_guest_list, reg_list) {
+ if (guest->snapshot.pgtable.pgd == pgt->pgd) {
+ list_del(&guest->reg_list);
+ debugfs_remove(guest->dentry);
+ kfree(guest);
+ break;
+ }
+ }
+ mutex_unlock(&ptdump_list_lock);
+}
#endif /* CONFIG_NVHE_EL2_PTDUMP_DEBUGFS */
static void __init ptdump_register_host_stage2(void)
@@ -49,7 +49,8 @@ static const struct file_operations ptdump_fops = {
.release = ptdump_release,
};
-void __init ptdump_debugfs_register(struct ptdump_info *info, const char *name)
+struct dentry *ptdump_debugfs_register(struct ptdump_info *info,
+ const char *name)
{
- debugfs_create_file(name, 0400, NULL, info, &ptdump_fops);
+ return debugfs_create_file(name, 0400, NULL, info, &ptdump_fops);
}