[30/30] x86, kaiser, xen: Dynamically disable KAISER when running under Xen PV
diff mbox series

Message ID 20171108194742.8CD79E09@viggo.jf.intel.com
State New, archived
Headers show
Series
  • KAISER: unmap most of the kernel from userspace page tables
Related show

Commit Message

Dave Hansen Nov. 8, 2017, 7:47 p.m. UTC
From: Dave Hansen <dave.hansen@linux.intel.com>

If you paravirtualize the MMU, you can not use KAISER.  This boils down
to the fact that KAISER needs to do CR3 writes in places that it is not
feasible to do real hypercalls.

If we detect that Xen PV is in use, do not do the KAISER CR3 switches.

I don't think this too bug of a deal for Xen.  I was under the
impression that the Xen guest kernel and Xen guest userspace didn't
share an address space *anyway* so Xen PV is not normally even exposed
to the kinds of things that KAISER protects against.

This allows KAISER=y kernels to deployed in environments that also
require PARAVIRT=y.

Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Moritz Lipp <moritz.lipp@iaik.tugraz.at>
Cc: Daniel Gruss <daniel.gruss@iaik.tugraz.at>
Cc: Michael Schwarz <michael.schwarz@iaik.tugraz.at>
Cc: Richard Fellner <richard.fellner@student.tugraz.at>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Kees Cook <keescook@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: x86@kernel.org
Cc: Juergen Gross <jgross@suse.com>
---

 b/arch/x86/mm/kaiser.c |   24 ++++++++++++++++++++++--
 b/security/Kconfig     |    2 +-
 2 files changed, 23 insertions(+), 3 deletions(-)

Comments

Juergen Gross Nov. 9, 2017, 3:01 p.m. UTC | #1
On 08/11/17 20:47, Dave Hansen wrote:
> From: Dave Hansen <dave.hansen@linux.intel.com>
> 
> If you paravirtualize the MMU, you can not use KAISER.  This boils down
> to the fact that KAISER needs to do CR3 writes in places that it is not
> feasible to do real hypercalls.
> 
> If we detect that Xen PV is in use, do not do the KAISER CR3 switches.
> 
> I don't think this too bug of a deal for Xen.  I was under the
> impression that the Xen guest kernel and Xen guest userspace didn't
> share an address space *anyway* so Xen PV is not normally even exposed
> to the kinds of things that KAISER protects against.
> 
> This allows KAISER=y kernels to deployed in environments that also
> require PARAVIRT=y.
> 
> Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>

Acked-by: Juergen Gross <jgross@suse.com>


Juergen

Patch
diff mbox series

diff -puN arch/x86/mm/kaiser.c~kaiser-disable-for-xen-pv arch/x86/mm/kaiser.c
--- a/arch/x86/mm/kaiser.c~kaiser-disable-for-xen-pv	2017-11-08 10:46:16.913681276 -0800
+++ b/arch/x86/mm/kaiser.c	2017-11-08 10:46:16.918681276 -0800
@@ -31,8 +31,20 @@ 
 #include <asm/tlbflush.h>
 #include <asm/desc.h>
 
+/*
+ * We need a two-stage enable/disable.  One (kaiser_enabled) to stop
+ * the ongoing work that keeps KAISER from being disabled (like PGD
+ * poisoning) and another (kaiser_asm_do_switch) that we set when it
+ * is completely safe to run without doing KAISER switches.
+ */
+int kaiser_enabled;
+
+/*
+ * Sized and aligned so that we can easily map it out to userspace
+ * for use before we have done the assembly CR3 switching.
+ */
 __aligned(PAGE_SIZE)
-unsigned long kaiser_asm_do_switch[PAGE_SIZE/sizeof(unsigned long)] = { 1 };
+unsigned long kaiser_asm_do_switch[PAGE_SIZE/sizeof(unsigned long)];
 
 /*
  * At runtime, the only things we map are some things for CPU
@@ -404,6 +416,15 @@  void __init kaiser_init(void)
 	kaiser_add_user_map_ptrs_early(__irqentry_text_start,
 				       __irqentry_text_end,
 				       __PAGE_KERNEL_RX | _PAGE_GLOBAL);
+
+	if (cpu_feature_enabled(X86_FEATURE_XENPV)) {
+		pr_info("x86/kaiser: Xen PV detected, disabling "
+			"KAISER protection\n");
+	} else {
+		pr_info("x86/kaiser: Unmapping kernel while in userspace\n");
+		kaiser_asm_do_switch[0] = 1;
+		kaiser_enabled = 1;
+	}
 }
 
 int kaiser_add_mapping(unsigned long addr, unsigned long size,
@@ -454,7 +475,6 @@  void kaiser_remove_mapping(unsigned long
 	__native_flush_tlb_global();
 }
 
-int kaiser_enabled = 1;
 static ssize_t kaiser_enabled_read_file(struct file *file, char __user *user_buf,
 			     size_t count, loff_t *ppos)
 {
diff -puN security/Kconfig~kaiser-disable-for-xen-pv security/Kconfig
--- a/security/Kconfig~kaiser-disable-for-xen-pv	2017-11-08 10:46:16.914681276 -0800
+++ b/security/Kconfig	2017-11-08 10:46:16.918681276 -0800
@@ -56,7 +56,7 @@  config SECURITY_NETWORK
 
 config KAISER
 	bool "Remove the kernel mapping in user mode"
-	depends on X86_64 && SMP && !PARAVIRT
+	depends on X86_64 && SMP
 	help
 	  This feature reduces the number of hardware side channels by
 	  ensuring that the majority of kernel addresses are not mapped