On 24.10.19 13:40, Janosch Frank wrote:
From: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
Pin the guest pages when they are first accessed, instead of all at
the same time when starting the guest.
Signed-off-by: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
---
arch/s390/include/asm/gmap.h | 1 +
arch/s390/include/asm/uv.h | 6 +++++
arch/s390/kernel/uv.c | 20 ++++++++++++++
arch/s390/kvm/kvm-s390.c | 2 ++
arch/s390/kvm/pv.c | 51 ++++++++++++++++++++++++++++++------
5 files changed, 72 insertions(+), 8 deletions(-)
diff --git a/arch/s390/include/asm/gmap.h b/arch/s390/include/asm/gmap.h
index 99b3eedda26e..483f64427c0e 100644
--- a/arch/s390/include/asm/gmap.h
+++ b/arch/s390/include/asm/gmap.h
@@ -63,6 +63,7 @@ struct gmap {
struct gmap *parent;
unsigned long orig_asce;
unsigned long se_handle;
+ struct page **pinned_pages;
int edat_level;
bool removed;
bool initialized;
diff --git a/arch/s390/include/asm/uv.h b/arch/s390/include/asm/uv.h
index 99cdd2034503..9ce9363aee1c 100644
--- a/arch/s390/include/asm/uv.h
+++ b/arch/s390/include/asm/uv.h
@@ -298,6 +298,7 @@ static inline int uv_convert_from_secure(unsigned long paddr)
return -EINVAL;
}
+int kvm_s390_pv_pin_page(struct gmap *gmap, unsigned long gpa);
/*
* Requests the Ultravisor to make a page accessible to a guest
* (import). If it's brought in the first time, it will be cleared. If
@@ -317,6 +318,11 @@ static inline int uv_convert_to_secure(struct gmap *gmap, unsigned long gaddr)
.gaddr = gaddr
};
+ down_read(&gmap->mm->mmap_sem);
+ cc = kvm_s390_pv_pin_page(gmap, gaddr);
+ up_read(&gmap->mm->mmap_sem);
+ if (cc)
+ return cc;
cc = uv_call(0, (u64)&uvcb);
So, a theoretical question: Is any in-flight I/O from paging stopped
when we try to pin the pages? I am no export on paging, but the comment
from Christian ("you can actually fault in a page that is currently
under paging I/O.") made me wonder.
Let's assume you could have parallel I/O on that page. You would do a
uv_call() and convert the page to secure/unencrypted. The I/O can easily
stumble over that (now inaccessible) page and report an error.
Or is any such race not possible because we are using
get_user_pages_fast() vs. get_user_pages()? (then, I'd love to see a
comment regarding that in the patch description)
--
Thanks,
David / dhildenb
