Previously, when a protected VM was rebooted or when it was shut down,
its memory was made unprotected, and then the protected VM itself was
destroyed. Looping over the whole address space can take some time,
considering the overhead of the various Ultravisor Calls (UVCs). This
means that a reboot or a shutdown would take a potentially long amount
of time, depending on the amount of used memory.
This patchseries implements a deferred destroy mechanism for protected
guests. When a protected guest is destroyed, its memory can be cleared
in background, allowing the guest to restart or terminate significantly
faster than before.
There are 2 possibilities when a protected VM is torn down:
* it still has an address space associated (reboot case)
* it does not have an address space anymore (shutdown case)
For the reboot case, two new commands are available for the
KVM_S390_PV_COMMAND:
KVM_PV_ASYNC_DISABLE_PREPARE: prepares the current protected VM for
asynchronous teardown. The current VM will then continue immediately
as non-protected. If a protected VM had already been set aside without
starting the teardown process, this call will fail. In this case the
userspace process should issue a normal KVM_PV_DISABLE
KVM_PV_ASYNC_DISABLE: tears down the protected VM previously set aside
for asychronous teardown. This PV command should ideally be issued by
userspace from a separate thread. If a fatal signal is received (or
the process terminates naturally), the command will terminate
immediately without completing.
The idea is that userspace should first issue the
KVM_PV_ASYNC_DISABLE_PREPARE command, and in case of success, create a
new thread and issue KVM_PV_ASYNC_DISABLE from there. This also allows
for proper accounting of the CPU time needed for the asynchronous
teardown.
This means that the same address space can have memory belonging to
more than one protected guest, although only one will be running, the
others will in fact not even have any CPUs.
The shutdown case should be dealt with in userspace (e.g. using
clone(CLONE_VM)).
A module parameter is also provided to disable the new functionality,
which is otherwise enabled by default. This should not be an issue
since the new functionality is opt-in anyway. This is mainly thought to
aid debugging.
v7->v8
* switched patches 8 and 9
* improved comments, documentation and patch descriptions
* remove mm notifier when the struct kvm is torn down
* removed useless locks in the mm notifier
* use _ASCE_ORIGIN instead of PAGE_MASK for ASCEs
* cleanup of some compiler warnings
* remove some harmless but useless duplicate code
* the last parameter of __s390_uv_destroy_range is now bool
* rename the KVM capability to KVM_CAP_S390_PROTECTED_ASYNC_DISABLE
v6->v7
* moved INIT_LIST_HEAD inside spinlock in patch 1
* improved commit messages in patch 2
* added missing locks in patch 3
* added and expanded some comments in patch 11
* rebased
v5->v6
* completely reworked the series
* removed kernel thread for asynchronous teardown
* added new commands to KVM_S390_PV_COMMAND ioctl
v4->v5
* fixed and improved some patch descriptions
* added some comments to better explain what's going on
* use vma_lookup instead of find_vma
* rename is_protected to protected_count since now it's used as a counter
v3->v4
* added patch 2
* split patch 3
* removed the shutdown part -- will be a separate patchseries
* moved the patch introducing the module parameter
v2->v3
* added definitions for CC return codes for the UVC instruction
* improved make_secure_pte:
- renamed rc to cc
- added comments to explain why returning -EAGAIN is ok
* fixed kvm_s390_pv_replace_asce and kvm_s390_pv_remove_old_asce:
- renamed
- added locking
- moved to gmap.c
* do proper error management in do_secure_storage_access instead of
trying again hoping to get a different exception
* fix outdated patch descriptions
v1->v2
* rebased on a more recent kernel
* improved/expanded some patch descriptions
* improves/expanded some comments
* added patch 1, which prevents stall notification when the system is
under heavy load.
* rename some members of struct deferred_priv to improve readability
* avoid an use-after-free bug of the struct mm in case of shutdown
* add missing return when lazy destroy is disabled
* add support for OOM notifier
Claudio Imbrenda (17):
KVM: s390: pv: leak the topmost page table when destroy fails
KVM: s390: pv: handle secure storage violations for protected guests
KVM: s390: pv: handle secure storage exceptions for normal guests
KVM: s390: pv: refactor s390_reset_acc
KVM: s390: pv: usage counter instead of flag
KVM: s390: pv: add export before import
KVM: s390: pv: module parameter to fence lazy destroy
KVM: s390: pv: clear the state without memset
KVM: s390: pv: Add kvm_s390_cpus_from_pv to kvm-s390.h and add
documentation
KVM: s390: pv: add mmu_notifier
s390/mm: KVM: pv: when tearing down, try to destroy protected pages
KVM: s390: pv: refactoring of kvm_s390_pv_deinit_vm
KVM: s390: pv: cleanup leftover protected VMs if needed
KVM: s390: pv: asynchronous destroy for reboot
KVM: s390: pv: api documentation for asynchronous destroy
KVM: s390: pv: add KVM_CAP_S390_PROTECTED_ASYNC_DISABLE
KVM: s390: pv: avoid export before import if possible
Documentation/virt/kvm/api.rst | 25 ++-
arch/s390/include/asm/gmap.h | 39 +++-
arch/s390/include/asm/kvm_host.h | 4 +
arch/s390/include/asm/mmu.h | 2 +-
arch/s390/include/asm/mmu_context.h | 2 +-
arch/s390/include/asm/pgtable.h | 20 ++-
arch/s390/include/asm/uv.h | 1 +
arch/s390/kernel/uv.c | 64 +++++++
arch/s390/kvm/kvm-s390.c | 64 ++++++-
arch/s390/kvm/kvm-s390.h | 3 +
arch/s390/kvm/pv.c | 264 ++++++++++++++++++++++++++--
arch/s390/mm/fault.c | 23 ++-
arch/s390/mm/gmap.c | 158 ++++++++++++++---
include/uapi/linux/kvm.h | 3 +
14 files changed, 623 insertions(+), 49 deletions(-)
--
2.34.1
With upcoming patches, normal guests might touch secure pages.
This patch extends the existing exception handler to convert the pages
to non secure also when the exception is triggered by a normal guest.
This can happen for example when a secure guest reboots; the first
stage of a secure guest is non secure, and in general a secure guest
can reboot into non-secure mode.
If the secure memory of the previous boot has not been cleared up
completely yet (which will be allowed to happen in an upcoming patch),
a non-secure guest might touch secure memory, which will need to be
handled properly.
Signed-off-by: Claudio Imbrenda <[email protected]>
Reviewed-by: Janosch Frank <[email protected]>
---
arch/s390/mm/fault.c | 13 ++++++++++++-
1 file changed, 12 insertions(+), 1 deletion(-)
diff --git a/arch/s390/mm/fault.c b/arch/s390/mm/fault.c
index 47b52e5384f8..54571a88a641 100644
--- a/arch/s390/mm/fault.c
+++ b/arch/s390/mm/fault.c
@@ -770,6 +770,7 @@ void do_secure_storage_access(struct pt_regs *regs)
struct vm_area_struct *vma;
struct mm_struct *mm;
struct page *page;
+ struct gmap *gmap;
int rc;
/*
@@ -799,6 +800,17 @@ void do_secure_storage_access(struct pt_regs *regs)
}
switch (get_fault_type(regs)) {
+ case GMAP_FAULT:
+ mm = current->mm;
+ gmap = (struct gmap *)S390_lowcore.gmap;
+ mmap_read_lock(mm);
+ addr = __gmap_translate(gmap, addr);
+ mmap_read_unlock(mm);
+ if (IS_ERR_VALUE(addr)) {
+ do_fault_error(regs, VM_ACCESS_FLAGS, VM_FAULT_BADMAP);
+ break;
+ }
+ fallthrough;
case USER_FAULT:
mm = current->mm;
mmap_read_lock(mm);
@@ -827,7 +839,6 @@ void do_secure_storage_access(struct pt_regs *regs)
if (rc)
BUG();
break;
- case GMAP_FAULT:
default:
do_fault_error(regs, VM_READ | VM_WRITE, VM_FAULT_BADMAP);
WARN_ON_ONCE(1);
--
2.34.1
If the appropriate UV feature bit is set, there is no need to perform
an export before import.
Signed-off-by: Claudio Imbrenda <[email protected]>
---
arch/s390/kernel/uv.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/s390/kernel/uv.c b/arch/s390/kernel/uv.c
index e358b8bd864b..43393568f844 100644
--- a/arch/s390/kernel/uv.c
+++ b/arch/s390/kernel/uv.c
@@ -236,7 +236,8 @@ static int make_secure_pte(pte_t *ptep, unsigned long addr,
static bool should_export_before_import(struct uv_cb_header *uvcb, struct mm_struct *mm)
{
- return uvcb->cmd != UVC_CMD_UNPIN_PAGE_SHARED &&
+ return !test_bit_inv(BIT_UV_FEAT_MISC, &uv_info.uv_feature_indications) &&
+ uvcb->cmd != UVC_CMD_UNPIN_PAGE_SHARED &&
atomic_read(&mm->context.protected_count) > 1;
}
--
2.34.1
Add the module parameter "lazy_destroy", to allow the asynchronous destroy
mechanism to be switched off. This might be useful for debugging purposes.
The parameter is enabled by default.
Signed-off-by: Claudio Imbrenda <[email protected]>
Reviewed-by: Janosch Frank <[email protected]>
---
arch/s390/kvm/kvm-s390.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 2296b1ff1e02..702696189505 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -207,6 +207,11 @@ unsigned int diag9c_forwarding_hz;
module_param(diag9c_forwarding_hz, uint, 0644);
MODULE_PARM_DESC(diag9c_forwarding_hz, "Maximum diag9c forwarding per second, 0 to turn off");
+/* allow asynchronous deinit for protected guests */
+static int lazy_destroy = 1;
+module_param(lazy_destroy, int, 0444);
+MODULE_PARM_DESC(lazy_destroy, "Asynchronous destroy for protected guests");
+
/*
* For now we handle at most 16 double words as this is what the s390 base
* kernel handles and stores in the prefix page. If we ever need to go beyond
--
2.34.1
Due to upcoming changes, it will be possible to temporarily have
multiple protected VMs in the same address space, although only one
will be actually active.
In that scenario, it is necessary to perform an export of every page
that is to be imported, since the hardware does not allow a page
belonging to a protected guest to be imported into a different
protected guest.
This also applies to pages that are shared, and thus accessible by the
host.
Signed-off-by: Claudio Imbrenda <[email protected]>
Reviewed-by: Janosch Frank <[email protected]>
---
arch/s390/kernel/uv.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/arch/s390/kernel/uv.c b/arch/s390/kernel/uv.c
index 2754471cc789..e358b8bd864b 100644
--- a/arch/s390/kernel/uv.c
+++ b/arch/s390/kernel/uv.c
@@ -234,6 +234,12 @@ static int make_secure_pte(pte_t *ptep, unsigned long addr,
return uvcb->rc == 0x10a ? -ENXIO : -EINVAL;
}
+static bool should_export_before_import(struct uv_cb_header *uvcb, struct mm_struct *mm)
+{
+ return uvcb->cmd != UVC_CMD_UNPIN_PAGE_SHARED &&
+ atomic_read(&mm->context.protected_count) > 1;
+}
+
/*
* Requests the Ultravisor to make a page accessible to a guest.
* If it's brought in the first time, it will be cleared. If
@@ -277,6 +283,8 @@ int gmap_make_secure(struct gmap *gmap, unsigned long gaddr, void *uvcb)
lock_page(page);
ptep = get_locked_pte(gmap->mm, uaddr, &ptelock);
+ if (should_export_before_import(uvcb, gmap->mm))
+ uv_convert_from_secure(page_to_phys(page));
rc = make_secure_pte(ptep, uaddr, page, uvcb);
pte_unmap_unlock(ptep, ptelock);
unlock_page(page);
--
2.34.1
Add documentation for the new commands added to the KVM_S390_PV_COMMAND
ioctl.
Signed-off-by: Claudio Imbrenda <[email protected]>
---
Documentation/virt/kvm/api.rst | 25 ++++++++++++++++++++++---
1 file changed, 22 insertions(+), 3 deletions(-)
diff --git a/Documentation/virt/kvm/api.rst b/Documentation/virt/kvm/api.rst
index 9f3172376ec3..52ba1c52ae3c 100644
--- a/Documentation/virt/kvm/api.rst
+++ b/Documentation/virt/kvm/api.rst
@@ -5010,11 +5010,13 @@ KVM_PV_ENABLE
===== =============================
KVM_PV_DISABLE
-
Deregister the VM from the Ultravisor and reclaim the memory that
had been donated to the Ultravisor, making it usable by the kernel
- again. All registered VCPUs are converted back to non-protected
- ones.
+ again. All registered VCPUs are converted back to non-protected
+ ones. If a previous VM had been prepared for asynchonous teardown
+ with KVM_PV_ASYNC_DISABLE_PREPARE and not actually torn down with
+ KVM_PV_ASYNC_DISABLE, it will be torn down in this call together with
+ the current VM.
KVM_PV_VM_SET_SEC_PARMS
Pass the image header from VM memory to the Ultravisor in
@@ -5027,6 +5029,23 @@ KVM_PV_VM_VERIFY
Verify the integrity of the unpacked image. Only if this succeeds,
KVM is allowed to start protected VCPUs.
+KVM_PV_ASYNC_DISABLE_PREPARE
+ Prepare the current protected VM for asynchronous teardown. Most
+ resources used by the current protected VM will be set aside for a
+ subsequent asynchronous teardown. The current protected VM will then
+ resume execution immediately as non-protected. If a protected VM had
+ already been prepared without starting the asynchronous teardown process,
+ this call will fail. In that case, the userspace process should issue a
+ normal KVM_PV_DISABLE.
+
+KVM_PV_ASYNC_DISABLE
+ Tear down the protected VM previously prepared for asynchronous teardown.
+ The resources that had been set aside will be freed asynchronously during
+ the execution of this command.
+ This PV command should ideally be issued by userspace from a separate
+ thread. If a fatal signal is received (or the process terminates
+ naturally), the command will terminate immediately without completing.
+
4.126 KVM_X86_SET_MSR_FILTER
----------------------------
--
2.34.1
Each secure guest must have a unique ASCE (address space control
element); we must avoid that new guests use the same page for their
ASCE, to avoid errors.
Since the ASCE mostly consists of the address of the topmost page table
(plus some flags), we must not return that memory to the pool unless
the ASCE is no longer in use.
Only a successful Destroy Secure Configuration UVC will make the ASCE
reusable again.
If the Destroy Configuration UVC fails, the ASCE cannot be reused for a
secure guest (either for the ASCE or for other memory areas). To avoid
a collision, it must not be used again. This is a permanent error and
the page becomes in practice unusable, so we set it aside and leak it.
On failure we already leak other memory that belongs to the ultravisor
(i.e. the variable and base storage for a guest) and not leaking the
topmost page table was an oversight.
This error (and thus the leakage) should not happen unless the hardware
is broken or KVM has some unknown serious bug.
Signed-off-by: Claudio Imbrenda <[email protected]>
Fixes: 29b40f105ec8d55 ("KVM: s390: protvirt: Add initial vm and cpu lifecycle handling")
---
arch/s390/include/asm/gmap.h | 2 +
arch/s390/kvm/pv.c | 9 +++--
arch/s390/mm/gmap.c | 71 ++++++++++++++++++++++++++++++++++++
3 files changed, 79 insertions(+), 3 deletions(-)
diff --git a/arch/s390/include/asm/gmap.h b/arch/s390/include/asm/gmap.h
index 40264f60b0da..746e18bf8984 100644
--- a/arch/s390/include/asm/gmap.h
+++ b/arch/s390/include/asm/gmap.h
@@ -148,4 +148,6 @@ void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long dirty_bitmap[4],
unsigned long gaddr, unsigned long vmaddr);
int gmap_mark_unmergeable(void);
void s390_reset_acc(struct mm_struct *mm);
+void s390_remove_old_asce(struct gmap *gmap);
+int s390_replace_asce(struct gmap *gmap);
#endif /* _ASM_S390_GMAP_H */
diff --git a/arch/s390/kvm/pv.c b/arch/s390/kvm/pv.c
index 7f7c0d6af2ce..3c59ef763dde 100644
--- a/arch/s390/kvm/pv.c
+++ b/arch/s390/kvm/pv.c
@@ -166,10 +166,13 @@ int kvm_s390_pv_deinit_vm(struct kvm *kvm, u16 *rc, u16 *rrc)
atomic_set(&kvm->mm->context.is_protected, 0);
KVM_UV_EVENT(kvm, 3, "PROTVIRT DESTROY VM: rc %x rrc %x", *rc, *rrc);
WARN_ONCE(cc, "protvirt destroy vm failed rc %x rrc %x", *rc, *rrc);
- /* Inteded memory leak on "impossible" error */
- if (!cc)
+ /* Intended memory leak on "impossible" error */
+ if (!cc) {
kvm_s390_pv_dealloc_vm(kvm);
- return cc ? -EIO : 0;
+ return 0;
+ }
+ s390_replace_asce(kvm->arch.gmap);
+ return -EIO;
}
int kvm_s390_pv_init_vm(struct kvm *kvm, u16 *rc, u16 *rrc)
diff --git a/arch/s390/mm/gmap.c b/arch/s390/mm/gmap.c
index dfee0ebb2fac..e78857d6942a 100644
--- a/arch/s390/mm/gmap.c
+++ b/arch/s390/mm/gmap.c
@@ -2714,3 +2714,74 @@ void s390_reset_acc(struct mm_struct *mm)
mmput(mm);
}
EXPORT_SYMBOL_GPL(s390_reset_acc);
+
+/**
+ * s390_remove_old_asce - Remove the topmost level of page tables from the
+ * list of page tables of the gmap.
+ * @gmap the gmap whose table is to be removed
+ *
+ * This means that it will not be freed when the VM is torn down, and needs
+ * to be handled separately by the caller, unless an intentional leak is
+ * intended. Notice that this function will only remove the page from the
+ * list, the page will still be used as a top level page table (and ASCE).
+ */
+void s390_remove_old_asce(struct gmap *gmap)
+{
+ struct page *old;
+
+ old = virt_to_page(gmap->table);
+ spin_lock(&gmap->guest_table_lock);
+ list_del(&old->lru);
+ /*
+ * in case the ASCE needs to be "removed" multiple times, for example
+ * if the VM is rebooted into secure mode several times
+ * concurrently.
+ */
+ INIT_LIST_HEAD(&old->lru);
+ spin_unlock(&gmap->guest_table_lock);
+}
+EXPORT_SYMBOL_GPL(s390_remove_old_asce);
+
+/**
+ * s390_replace_asce - Try to replace the current ASCE of a gmap with
+ * another equivalent one.
+ * @gmap the gmap
+ *
+ * If the allocation of the new top level page table fails, the ASCE is not
+ * replaced.
+ * In any case, the old ASCE is always removed from the list. Therefore the
+ * caller has to make sure to save a pointer to it beforehands, unless an
+ * intentional leak is intended.
+ */
+int s390_replace_asce(struct gmap *gmap)
+{
+ unsigned long asce;
+ struct page *page;
+ void *table;
+
+ s390_remove_old_asce(gmap);
+
+ page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
+ if (!page)
+ return -ENOMEM;
+ table = page_to_virt(page);
+ memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
+
+ /*
+ * The caller has to deal with the old ASCE, but here we make sure
+ * the new one is properly added to the list of page tables, so that
+ * it will be freed when the VM is torn down.
+ */
+ spin_lock(&gmap->guest_table_lock);
+ list_add(&page->lru, &gmap->crst_list);
+ spin_unlock(&gmap->guest_table_lock);
+
+ /* Set new table origin while preserving existing ASCE control bits */
+ asce = (gmap->asce & _ASCE_ORIGIN) | __pa(table);
+ WRITE_ONCE(gmap->asce, asce);
+ WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
+ WRITE_ONCE(gmap->table, table);
+
+ return 0;
+}
+EXPORT_SYMBOL_GPL(s390_replace_asce);
--
2.34.1
Until now, destroying a protected guest was an entirely synchronous
operation that could potentially take a very long time, depending on
the size of the guest, due to the time needed to clean up the address
space from protected pages.
This patch implements an asynchronous destroy mechanism, that allows a
protected guest to reboot significantly faster than previously.
This is achieved by clearing the pages of the old guest in background.
In case of reboot, the new guest will be able to run in the same
address space almost immediately.
The old protected guest is then only destroyed when all of its memory has
been destroyed or otherwise made non protected.
Two new PV commands are added for the KVM_S390_PV_COMMAND ioctl:
KVM_PV_ASYNC_DISABLE_PREPARE: prepares the current protected VM for
asynchronous teardown. The current VM will then continue immediately
as non-protected. If a protected VM had already been set aside without
starting the teardown process, this call will fail.
KVM_PV_ASYNC_DISABLE: tears down the protected VM previously set aside
for asynchronous teardown. This PV command should ideally be issued by
userspace from a separate thread. If a fatal signal is received (or the
process terminates naturally), the command will terminate immediately
without completing.
Leftover protected VMs are cleaned up when a KVM VM is torn down
normally (either via IOCTL or when the process terminates); this
cleanup has been implemented in a previous patch.
Signed-off-by: Claudio Imbrenda <[email protected]>
---
arch/s390/kvm/kvm-s390.c | 24 ++++++++
arch/s390/kvm/kvm-s390.h | 2 +
arch/s390/kvm/pv.c | 126 +++++++++++++++++++++++++++++++++++++++
include/uapi/linux/kvm.h | 2 +
4 files changed, 154 insertions(+)
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 3637f556ff33..2453d2d90d6c 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -2285,6 +2285,30 @@ static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
set_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
break;
}
+ case KVM_PV_ASYNC_DISABLE_PREPARE:
+ r = -EINVAL;
+ if (!kvm_s390_pv_is_protected(kvm) || !lazy_destroy)
+ break;
+
+ r = kvm_s390_cpus_from_pv(kvm, &cmd->rc, &cmd->rrc);
+ /*
+ * If a CPU could not be destroyed, destroy VM will also fail.
+ * There is no point in trying to destroy it. Instead return
+ * the rc and rrc from the first CPU that failed destroying.
+ */
+ if (r)
+ break;
+ r = kvm_s390_pv_deinit_vm_async_prepare(kvm, &cmd->rc, &cmd->rrc);
+
+ /* no need to block service interrupts any more */
+ clear_bit(IRQ_PEND_EXT_SERVICE, &kvm->arch.float_int.masked_irqs);
+ break;
+ case KVM_PV_ASYNC_DISABLE:
+ r = -EINVAL;
+ if (!kvm->arch.pv.async_deinit)
+ break;
+ r = kvm_s390_pv_deinit_vm_async(kvm, &cmd->rc, &cmd->rrc);
+ break;
case KVM_PV_DISABLE: {
r = -EINVAL;
if (!kvm_s390_pv_is_protected(kvm))
diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
index 9276d910631b..be53c7750248 100644
--- a/arch/s390/kvm/kvm-s390.h
+++ b/arch/s390/kvm/kvm-s390.h
@@ -234,6 +234,8 @@ static inline unsigned long kvm_s390_get_gfn_end(struct kvm_memslots *slots)
/* implemented in pv.c */
int kvm_s390_pv_destroy_cpu(struct kvm_vcpu *vcpu, u16 *rc, u16 *rrc);
int kvm_s390_pv_create_cpu(struct kvm_vcpu *vcpu, u16 *rc, u16 *rrc);
+int kvm_s390_pv_deinit_vm_async_prepare(struct kvm *kvm, u16 *rc, u16 *rrc);
+int kvm_s390_pv_deinit_vm_async(struct kvm *kvm, u16 *rc, u16 *rrc);
int kvm_s390_pv_deinit_vm(struct kvm *kvm, u16 *rc, u16 *rrc);
int kvm_s390_pv_init_vm(struct kvm *kvm, u16 *rc, u16 *rrc);
int kvm_s390_pv_set_sec_parms(struct kvm *kvm, void *hdr, u64 length, u16 *rc,
diff --git a/arch/s390/kvm/pv.c b/arch/s390/kvm/pv.c
index 56412617dd01..5111f1fc64ab 100644
--- a/arch/s390/kvm/pv.c
+++ b/arch/s390/kvm/pv.c
@@ -262,6 +262,132 @@ int kvm_s390_pv_deinit_vm(struct kvm *kvm, u16 *rc, u16 *rrc)
return cc ? -EIO : 0;
}
+/**
+ * kvm_s390_clear_2g - Clear the first 2GB of guest memory.
+ * @kvm the VM whose memory is to be cleared.
+ * Clear the first 2GB of guest memory, to avoid prefix issues after reboot.
+ */
+static void kvm_s390_clear_2g(struct kvm *kvm)
+{
+ struct kvm_memory_slot *slot;
+ unsigned long lim;
+ int srcu_idx;
+
+ srcu_idx = srcu_read_lock(&kvm->srcu);
+
+ slot = gfn_to_memslot(kvm, 0);
+ /* Clear all slots that are completely below 2GB */
+ while (slot && slot->base_gfn + slot->npages < SZ_2G / PAGE_SIZE) {
+ lim = slot->userspace_addr + slot->npages * PAGE_SIZE;
+ s390_uv_destroy_range(kvm->mm, slot->userspace_addr, lim);
+ slot = gfn_to_memslot(kvm, slot->base_gfn + slot->npages);
+ }
+ /* Last slot crosses the 2G boundary, clear only up to 2GB */
+ if (slot && slot->base_gfn < SZ_2G / PAGE_SIZE) {
+ lim = slot->userspace_addr + SZ_2G - slot->base_gfn * PAGE_SIZE;
+ s390_uv_destroy_range(kvm->mm, slot->userspace_addr, lim);
+ }
+
+ srcu_read_unlock(&kvm->srcu, srcu_idx);
+}
+
+/**
+ * kvm_s390_pv_deinit_vm_async_prepare - Prepare a protected VM for
+ * asynchronous teardown.
+ * @kvm the VM
+ * @rc return value for the RC field of the UVCB
+ * @rrc return value for the RRC field of the UVCB
+ *
+ * Prepare the protected VM for asynchronous teardown. The VM will be able
+ * to continue immediately as a non-secure VM, and the information needed to
+ * properly tear down the protected VM is set aside. If another protected VM
+ * was already set aside without starting a teardown, the function will
+ * fail.
+ *
+ * Context: kvm->lock needs to be held
+ *
+ * Return: 0 in case of success, -EINVAL if another protected VM was already set
+ * aside, -ENOMEM if the system ran out of memory.
+ */
+int kvm_s390_pv_deinit_vm_async_prepare(struct kvm *kvm, u16 *rc, u16 *rrc)
+{
+ struct deferred_priv *priv;
+
+ /*
+ * If an asynchronous deinitialization is already pending, refuse.
+ * A synchronous deinitialization has to be performed instead.
+ */
+ if (kvm->arch.pv.async_deinit)
+ return -EINVAL;
+ priv = kmalloc(sizeof(*priv), GFP_KERNEL | __GFP_ZERO);
+ if (!priv)
+ return -ENOMEM;
+
+ priv->stor_var = kvm->arch.pv.stor_var;
+ priv->stor_base = kvm->arch.pv.stor_base;
+ priv->handle = kvm_s390_pv_get_handle(kvm);
+ priv->old_table = (unsigned long)kvm->arch.gmap->table;
+ WRITE_ONCE(kvm->arch.gmap->guest_handle, 0);
+ if (s390_replace_asce(kvm->arch.gmap)) {
+ kfree(priv);
+ return -ENOMEM;
+ }
+
+ kvm_s390_clear_2g(kvm);
+ kvm_s390_clear_pv_state(kvm);
+ kvm->arch.pv.async_deinit = priv;
+
+ *rc = 1;
+ *rrc = 42;
+ return 0;
+}
+
+/**
+ * kvm_s390_pv_deinit_vm_async - Perform an asynchronous teardown of a
+ * protected VM.
+ * @kvm the VM previously associated with the protected VM
+ * @rc return value for the RC field of the UVCB
+ * @rrc return value for the RRC field of the UVCB
+ *
+ * Tear down the protected VM that had previously been set aside using
+ * kvm_s390_pv_deinit_vm_async_prepare.
+ *
+ * Context: kvm->lock needs to be held
+ *
+ * Return: 0 in case of success, -EINVAL if no protected VM had been
+ * prepared for asynchronous teardowm, -EIO in case of other errors.
+ */
+int kvm_s390_pv_deinit_vm_async(struct kvm *kvm, u16 *rc, u16 *rrc)
+{
+ struct deferred_priv *p = kvm->arch.pv.async_deinit;
+ int ret = 0;
+
+ if (!p)
+ return -EINVAL;
+ kvm->arch.pv.async_deinit = NULL;
+ mutex_unlock(&kvm->lock);
+
+ /* When a fatal signal is received, stop immediately */
+ if (s390_uv_destroy_range_interruptible(kvm->mm, 0, TASK_SIZE_MAX))
+ goto done;
+ if (kvm_s390_pv_cleanup_deferred(kvm, p))
+ ret = -EIO;
+ else
+ atomic_dec(&kvm->mm->context.protected_count);
+ kfree(p);
+ p = NULL;
+done:
+ /* The caller expects the lock to be held */
+ mutex_lock(&kvm->lock);
+ /*
+ * p is not NULL if we aborted because of a fatal signal, in which
+ * case queue the leftover for later cleanup.
+ */
+ if (p)
+ list_add(&p->list, &kvm->arch.pv.need_cleanup);
+ return ret;
+}
+
static void kvm_s390_pv_mmu_notifier_release(struct mmu_notifier *subscription,
struct mm_struct *mm)
{
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index 507ee1f2aa96..d150610e7a4b 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -1644,6 +1644,8 @@ enum pv_cmd_id {
KVM_PV_VERIFY,
KVM_PV_PREP_RESET,
KVM_PV_UNSHARE_ALL,
+ KVM_PV_ASYNC_DISABLE_PREPARE,
+ KVM_PV_ASYNC_DISABLE,
};
struct kvm_pv_cmd {
--
2.34.1
When ptep_get_and_clear_full is called for a mm teardown, we will now
attempt to destroy the secure pages. This will be faster than export.
In case it was not a teardown, or if for some reason the destroy page
UVC failed, we try with an export page, like before.
Signed-off-by: Claudio Imbrenda <[email protected]>
Acked-by: Janosch Frank <[email protected]>
---
arch/s390/include/asm/pgtable.h | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)
diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
index 23ca0d8e058a..72544a1b4a68 100644
--- a/arch/s390/include/asm/pgtable.h
+++ b/arch/s390/include/asm/pgtable.h
@@ -1118,9 +1118,21 @@ static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
} else {
res = ptep_xchg_lazy(mm, addr, ptep, __pte(_PAGE_INVALID));
}
- /* At this point the reference through the mapping is still present */
- if (mm_is_protected(mm) && pte_present(res))
- uv_convert_owned_from_secure(pte_val(res) & PAGE_MASK);
+ /* Nothing to do */
+ if (!mm_is_protected(mm) || !pte_present(res))
+ return res;
+ /*
+ * At this point the reference through the mapping is still present.
+ * The notifier should have destroyed all protected vCPUs at this
+ * point, so the destroy should be successful.
+ */
+ if (full && !uv_destroy_owned_page(pte_val(res) & PAGE_MASK))
+ return res;
+ /*
+ * But if something went wrong and the pages could not be destroyed,
+ * the slower export is used as fallback instead.
+ */
+ uv_convert_owned_from_secure(pte_val(res) & PAGE_MASK);
return res;
}
--
2.34.1
On 3/2/22 19:11, Claudio Imbrenda wrote:
> Each secure guest must have a unique ASCE (address space control
> element); we must avoid that new guests use the same page for their
> ASCE, to avoid errors.
>
> Since the ASCE mostly consists of the address of the topmost page table
> (plus some flags), we must not return that memory to the pool unless
> the ASCE is no longer in use.
>
> Only a successful Destroy Secure Configuration UVC will make the ASCE
> reusable again.
>
> If the Destroy Configuration UVC fails, the ASCE cannot be reused for a
> secure guest (either for the ASCE or for other memory areas). To avoid
> a collision, it must not be used again. This is a permanent error and
> the page becomes in practice unusable, so we set it aside and leak it.
> On failure we already leak other memory that belongs to the ultravisor
> (i.e. the variable and base storage for a guest) and not leaking the
> topmost page table was an oversight.
>
> This error (and thus the leakage) should not happen unless the hardware
> is broken or KVM has some unknown serious bug.
>
> Signed-off-by: Claudio Imbrenda <[email protected]>
> Fixes: 29b40f105ec8d55 ("KVM: s390: protvirt: Add initial vm and cpu lifecycle handling")
> ---
> arch/s390/include/asm/gmap.h | 2 +
> arch/s390/kvm/pv.c | 9 +++--
> arch/s390/mm/gmap.c | 71 ++++++++++++++++++++++++++++++++++++
> 3 files changed, 79 insertions(+), 3 deletions(-)
>
[...]
> +/**
> + * s390_replace_asce - Try to replace the current ASCE of a gmap with
> + * another equivalent one.
> + * @gmap the gmap
> + *
> + * If the allocation of the new top level page table fails, the ASCE is not
> + * replaced.
> + * In any case, the old ASCE is always removed from the list. Therefore the
> + * caller has to make sure to save a pointer to it beforehands, unless an
> + * intentional leak is intended.
> + */
> +int s390_replace_asce(struct gmap *gmap)
> +{
> + unsigned long asce;
> + struct page *page;
> + void *table;
> +
> + s390_remove_old_asce(gmap);
> +
> + page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
> + if (!page)
> + return -ENOMEM;
> + table = page_to_virt(page);
> + memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
Is concurrent modification of *gmap->table possible during the copy?
> +
> + /*
> + * The caller has to deal with the old ASCE, but here we make sure
> + * the new one is properly added to the list of page tables, so that
> + * it will be freed when the VM is torn down.
> + */
> + spin_lock(&gmap->guest_table_lock);
> + list_add(&page->lru, &gmap->crst_list);
> + spin_unlock(&gmap->guest_table_lock);
> +
> + /* Set new table origin while preserving existing ASCE control bits */
> + asce = (gmap->asce & _ASCE_ORIGIN) | __pa(table);
> + WRITE_ONCE(gmap->asce, asce);
> + WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
> + WRITE_ONCE(gmap->table, table);
> +
> + return 0;
> +}
> +EXPORT_SYMBOL_GPL(s390_replace_asce);
On 3/3/22 16:05, Claudio Imbrenda wrote:
> On Thu, 3 Mar 2022 15:40:42 +0100
> Janis Schoetterl-Glausch <[email protected]> wrote:
>
>> On 3/2/22 19:11, Claudio Imbrenda wrote:
>>> Each secure guest must have a unique ASCE (address space control
>>> element); we must avoid that new guests use the same page for their
>>> ASCE, to avoid errors.
>>>
>>> Since the ASCE mostly consists of the address of the topmost page table
>>> (plus some flags), we must not return that memory to the pool unless
>>> the ASCE is no longer in use.
>>>
>>> Only a successful Destroy Secure Configuration UVC will make the ASCE
>>> reusable again.
>>>
>>> If the Destroy Configuration UVC fails, the ASCE cannot be reused for a
>>> secure guest (either for the ASCE or for other memory areas). To avoid
>>> a collision, it must not be used again. This is a permanent error and
>>> the page becomes in practice unusable, so we set it aside and leak it.
>>> On failure we already leak other memory that belongs to the ultravisor
>>> (i.e. the variable and base storage for a guest) and not leaking the
>>> topmost page table was an oversight.
>>>
>>> This error (and thus the leakage) should not happen unless the hardware
>>> is broken or KVM has some unknown serious bug.
>>>
>>> Signed-off-by: Claudio Imbrenda <[email protected]>
>>> Fixes: 29b40f105ec8d55 ("KVM: s390: protvirt: Add initial vm and cpu lifecycle handling")
>>> ---
>>> arch/s390/include/asm/gmap.h | 2 +
>>> arch/s390/kvm/pv.c | 9 +++--
>>> arch/s390/mm/gmap.c | 71 ++++++++++++++++++++++++++++++++++++
>>> 3 files changed, 79 insertions(+), 3 deletions(-)
>>>
>> [...]
>>
>>> +/**
>>> + * s390_replace_asce - Try to replace the current ASCE of a gmap with
>>> + * another equivalent one.
>>> + * @gmap the gmap
>>> + *
>>> + * If the allocation of the new top level page table fails, the ASCE is not
>>> + * replaced.
>>> + * In any case, the old ASCE is always removed from the list. Therefore the
>>> + * caller has to make sure to save a pointer to it beforehands, unless an
>>> + * intentional leak is intended.
>>> + */
>>> +int s390_replace_asce(struct gmap *gmap)
>>> +{
>>> + unsigned long asce;
>>> + struct page *page;
>>> + void *table;
>>> +
>>> + s390_remove_old_asce(gmap);
>>> +
>>> + page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
>>> + if (!page)
>>> + return -ENOMEM;
>>> + table = page_to_virt(page);
>>> + memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
>>
>> Is concurrent modification of *gmap->table possible during the copy?
>
> that would only be possible if the guest touches memory in such way
> that the table needs to be changed.
>
> this function is only called when the guest is not running (e.g. during
> reboot), so nobody should touch the table
Is that asserted?
I guess if modifications to the table are block concurrent and entries are only
freed when the vm is destroyed you cannot intentionally do any funny business.
>
>>
>>> +
>>> + /*
>>> + * The caller has to deal with the old ASCE, but here we make sure
>>> + * the new one is properly added to the list of page tables, so that
>>> + * it will be freed when the VM is torn down.
>>> + */
>>> + spin_lock(&gmap->guest_table_lock);
>>> + list_add(&page->lru, &gmap->crst_list);
>>> + spin_unlock(&gmap->guest_table_lock);
>>> +
>>> + /* Set new table origin while preserving existing ASCE control bits */
>>> + asce = (gmap->asce & _ASCE_ORIGIN) | __pa(table);
>>> + WRITE_ONCE(gmap->asce, asce);
>>> + WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
>>> + WRITE_ONCE(gmap->table, table);
>>> +
>>> + return 0;
>>> +}
>>> +EXPORT_SYMBOL_GPL(s390_replace_asce);
>>
>
On Thu, 3 Mar 2022 15:40:42 +0100
Janis Schoetterl-Glausch <[email protected]> wrote:
> On 3/2/22 19:11, Claudio Imbrenda wrote:
> > Each secure guest must have a unique ASCE (address space control
> > element); we must avoid that new guests use the same page for their
> > ASCE, to avoid errors.
> >
> > Since the ASCE mostly consists of the address of the topmost page table
> > (plus some flags), we must not return that memory to the pool unless
> > the ASCE is no longer in use.
> >
> > Only a successful Destroy Secure Configuration UVC will make the ASCE
> > reusable again.
> >
> > If the Destroy Configuration UVC fails, the ASCE cannot be reused for a
> > secure guest (either for the ASCE or for other memory areas). To avoid
> > a collision, it must not be used again. This is a permanent error and
> > the page becomes in practice unusable, so we set it aside and leak it.
> > On failure we already leak other memory that belongs to the ultravisor
> > (i.e. the variable and base storage for a guest) and not leaking the
> > topmost page table was an oversight.
> >
> > This error (and thus the leakage) should not happen unless the hardware
> > is broken or KVM has some unknown serious bug.
> >
> > Signed-off-by: Claudio Imbrenda <[email protected]>
> > Fixes: 29b40f105ec8d55 ("KVM: s390: protvirt: Add initial vm and cpu lifecycle handling")
> > ---
> > arch/s390/include/asm/gmap.h | 2 +
> > arch/s390/kvm/pv.c | 9 +++--
> > arch/s390/mm/gmap.c | 71 ++++++++++++++++++++++++++++++++++++
> > 3 files changed, 79 insertions(+), 3 deletions(-)
> >
> [...]
>
> > +/**
> > + * s390_replace_asce - Try to replace the current ASCE of a gmap with
> > + * another equivalent one.
> > + * @gmap the gmap
> > + *
> > + * If the allocation of the new top level page table fails, the ASCE is not
> > + * replaced.
> > + * In any case, the old ASCE is always removed from the list. Therefore the
> > + * caller has to make sure to save a pointer to it beforehands, unless an
> > + * intentional leak is intended.
> > + */
> > +int s390_replace_asce(struct gmap *gmap)
> > +{
> > + unsigned long asce;
> > + struct page *page;
> > + void *table;
> > +
> > + s390_remove_old_asce(gmap);
> > +
> > + page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
> > + if (!page)
> > + return -ENOMEM;
> > + table = page_to_virt(page);
> > + memcpy(table, gmap->table, 1UL << (CRST_ALLOC_ORDER + PAGE_SHIFT));
>
> Is concurrent modification of *gmap->table possible during the copy?
that would only be possible if the guest touches memory in such way
that the table needs to be changed.
this function is only called when the guest is not running (e.g. during
reboot), so nobody should touch the table
>
> > +
> > + /*
> > + * The caller has to deal with the old ASCE, but here we make sure
> > + * the new one is properly added to the list of page tables, so that
> > + * it will be freed when the VM is torn down.
> > + */
> > + spin_lock(&gmap->guest_table_lock);
> > + list_add(&page->lru, &gmap->crst_list);
> > + spin_unlock(&gmap->guest_table_lock);
> > +
> > + /* Set new table origin while preserving existing ASCE control bits */
> > + asce = (gmap->asce & _ASCE_ORIGIN) | __pa(table);
> > + WRITE_ONCE(gmap->asce, asce);
> > + WRITE_ONCE(gmap->mm->context.gmap_asce, asce);
> > + WRITE_ONCE(gmap->table, table);
> > +
> > + return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(s390_replace_asce);
>