x86/sgx: Fix ELDU MAC failures
diff mbox series

Message ID 20190823020002.25550-1-sean.j.christopherson@intel.com
State New
Headers show
Series
  • x86/sgx: Fix ELDU MAC failures
Related show

Commit Message

Sean Christopherson Aug. 23, 2019, 2 a.m. UTC
Revert a change that directly propagates the page type from SECINFO into
encl_page->desc.  encl_page->desc is a packed value, jamming the SECINFO
page_type sets bits that are intended for other uses.

Specifically, bits 11:3 are used to store the VA offset when a page is
swapped out and are available for other uses when the page is resident
in the EPC.  To allow overloading bits 11:3, the VA offset is cleared at
ELDU and so is not explicitly cleared at EWB, e.g. the VA offset is OR'd
into encl_page->desc.  As a result, a subsequent ELDU gets a MAC failure
due to loading the wrong VA offset.

Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
---
 arch/x86/kernel/cpu/sgx/driver/ioctl.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Jarkko Sakkinen Aug. 23, 2019, 1:35 p.m. UTC | #1
On Thu, 2019-08-22 at 19:00 -0700, Sean Christopherson wrote:
> Revert a change that directly propagates the page type from SECINFO into
> encl_page->desc.  encl_page->desc is a packed value, jamming the SECINFO
> page_type sets bits that are intended for other uses.
> 
> Specifically, bits 11:3 are used to store the VA offset when a page is
> swapped out and are available for other uses when the page is resident
> in the EPC.  To allow overloading bits 11:3, the VA offset is cleared at
> ELDU and so is not explicitly cleared at EWB, e.g. the VA offset is OR'd
> into encl_page->desc.  As a result, a subsequent ELDU gets a MAC failure
> due to loading the wrong VA offset.
> 
> Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> ---
>  arch/x86/kernel/cpu/sgx/driver/ioctl.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kernel/cpu/sgx/driver/ioctl.c b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> index 85e36e530baf..355ce967a77f 100644
> --- a/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> +++ b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> @@ -140,7 +140,8 @@ static struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
>  		return ERR_PTR(-ENOMEM);
>  
>  	encl_page->desc = addr;
> -	encl_page->desc |= page_type;
> +	if (page_type == SGX_SECINFO_TCS)
> +		encl_page->desc |= SGX_ENCL_PAGE_TCS;
>  	encl_page->encl = encl;

Oops. How embrassing from my side. Please also ignore my original review
comment about unreadability. The parameter type was right and everything
was right. I was just blind for a while.

This is exactly how it should be e.g. no boolean parameter. I think with
most of this kind of APIs we should follow a patttern the API takes PT
and the function does whatever encoding we have.

/Jarkko
Jarkko Sakkinen Aug. 23, 2019, 1:39 p.m. UTC | #2
On Fri, 2019-08-23 at 16:35 +0300, Jarkko Sakkinen wrote:
> On Thu, 2019-08-22 at 19:00 -0700, Sean Christopherson wrote:
> > Revert a change that directly propagates the page type from SECINFO into
> > encl_page->desc.  encl_page->desc is a packed value, jamming the SECINFO
> > page_type sets bits that are intended for other uses.
> > 
> > Specifically, bits 11:3 are used to store the VA offset when a page is
> > swapped out and are available for other uses when the page is resident
> > in the EPC.  To allow overloading bits 11:3, the VA offset is cleared at
> > ELDU and so is not explicitly cleared at EWB, e.g. the VA offset is OR'd
> > into encl_page->desc.  As a result, a subsequent ELDU gets a MAC failure
> > due to loading the wrong VA offset.
> > 
> > Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> > ---
> >  arch/x86/kernel/cpu/sgx/driver/ioctl.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/arch/x86/kernel/cpu/sgx/driver/ioctl.c b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > index 85e36e530baf..355ce967a77f 100644
> > --- a/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > +++ b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > @@ -140,7 +140,8 @@ static struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
> >  		return ERR_PTR(-ENOMEM);
> >  
> >  	encl_page->desc = addr;
> > -	encl_page->desc |= page_type;
> > +	if (page_type == SGX_SECINFO_TCS)
> > +		encl_page->desc |= SGX_ENCL_PAGE_TCS;
> >  	encl_page->encl = encl;
> 
> Oops. How embrassing from my side. Please also ignore my original review
> comment about unreadability. The parameter type was right and everything
> was right. I was just blind for a while.
> 
> This is exactly how it should be e.g. no boolean parameter. I think with
> most of this kind of APIs we should follow a patttern the API takes PT
> and the function does whatever encoding we have.

In any case the issue is fixed now on my tree and I also rewrote commit
message for the driver.

/Jarkko
Jarkko Sakkinen Aug. 23, 2019, 1:54 p.m. UTC | #3
On Fri, 2019-08-23 at 16:39 +0300, Jarkko Sakkinen wrote:
> On Fri, 2019-08-23 at 16:35 +0300, Jarkko Sakkinen wrote:
> > On Thu, 2019-08-22 at 19:00 -0700, Sean Christopherson wrote:
> > > Revert a change that directly propagates the page type from SECINFO into
> > > encl_page->desc.  encl_page->desc is a packed value, jamming the SECINFO
> > > page_type sets bits that are intended for other uses.
> > > 
> > > Specifically, bits 11:3 are used to store the VA offset when a page is
> > > swapped out and are available for other uses when the page is resident
> > > in the EPC.  To allow overloading bits 11:3, the VA offset is cleared at
> > > ELDU and so is not explicitly cleared at EWB, e.g. the VA offset is OR'd
> > > into encl_page->desc.  As a result, a subsequent ELDU gets a MAC failure
> > > due to loading the wrong VA offset.
> > > 
> > > Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
> > > ---
> > >  arch/x86/kernel/cpu/sgx/driver/ioctl.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/arch/x86/kernel/cpu/sgx/driver/ioctl.c b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > > index 85e36e530baf..355ce967a77f 100644
> > > --- a/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > > +++ b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
> > > @@ -140,7 +140,8 @@ static struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
> > >  		return ERR_PTR(-ENOMEM);
> > >  
> > >  	encl_page->desc = addr;
> > > -	encl_page->desc |= page_type;
> > > +	if (page_type == SGX_SECINFO_TCS)
> > > +		encl_page->desc |= SGX_ENCL_PAGE_TCS;
> > >  	encl_page->encl = encl;
> > 
> > Oops. How embrassing from my side. Please also ignore my original review
> > comment about unreadability. The parameter type was right and everything
> > was right. I was just blind for a while.
> > 
> > This is exactly how it should be e.g. no boolean parameter. I think with
> > most of this kind of APIs we should follow a patttern the API takes PT
> > and the function does whatever encoding we have.
> 
> In any case the issue is fixed now on my tree and I also rewrote commit
> message for the driver.

The stress test that I've been using is temporarily broken ATM so could not
exercise the code path.

/Jarkko

Patch
diff mbox series

diff --git a/arch/x86/kernel/cpu/sgx/driver/ioctl.c b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
index 85e36e530baf..355ce967a77f 100644
--- a/arch/x86/kernel/cpu/sgx/driver/ioctl.c
+++ b/arch/x86/kernel/cpu/sgx/driver/ioctl.c
@@ -140,7 +140,8 @@  static struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
 		return ERR_PTR(-ENOMEM);
 
 	encl_page->desc = addr;
-	encl_page->desc |= page_type;
+	if (page_type == SGX_SECINFO_TCS)
+		encl_page->desc |= SGX_ENCL_PAGE_TCS;
 	encl_page->encl = encl;
 
 	/* Calculate maximum of the VM flags for the page. */