On 6/21/2017 2:16 AM, Thomas Gleixner wrote:

On Fri, 16 Jun 2017, Tom Lendacky wrote:

quoted

diff --git a/arch/x86/include/asm/mem_encrypt.h b/arch/x86/include/asm/mem_encrypt.h
index a105796..988b336 100644
--- a/arch/x86/include/asm/mem_encrypt.h
+++ b/arch/x86/include/asm/mem_encrypt.h

@@ -15,16 +15,24 @@
  
  #ifndef __ASSEMBLY__
  
+#include <linux/init.h>
+
  #ifdef CONFIG_AMD_MEM_ENCRYPT
  
  extern unsigned long sme_me_mask;
  
+void __init sme_enable(void);
+
  #else	/* !CONFIG_AMD_MEM_ENCRYPT */
  
  #define sme_me_mask	0UL
  
+static inline void __init sme_enable(void) { }
+
  #endif	/* CONFIG_AMD_MEM_ENCRYPT */
  
+unsigned long sme_get_me_mask(void);

Why is this an unconditional function? Isn't the mask simply 0 when the MEM
ENCRYPT support is disabled?

I made it unconditional because of the call from head_64.S. I can't make
use of the C level static inline function and since the mask is not a
variable if CONFIG_AMD_MEM_ENCRYPT is not configured (#defined to 0) I
can't reference the variable directly.

I could create a #define in head_64.S that changes this to load rax with
the variable if CONFIG_AMD_MEM_ENCRYPT is configured or a zero if it's
not or add a #ifdef at that point in the code directly. Thoughts on
that?

quoted

diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index 6225550..ef12729 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S

@@ -78,7 +78,29 @@ startup_64:
  	call	__startup_64
  	popq	%rsi
  
-	movq	$(early_top_pgt - __START_KERNEL_map), %rax
+	/*
+	 * Encrypt the kernel if SME is active.
+	 * The real_mode_data address is in %rsi and that register can be
+	 * clobbered by the called function so be sure to save it.
+	 */
+	push	%rsi
+	call	sme_encrypt_kernel
+	pop	%rsi

That does not make any sense. Neither the call to sme_encrypt_kernel() nor
the following call to sme_get_me_mask().

__startup_64() is already C code, so why can't you simply call that from
__startup_64() in C and return the mask from there?

I was trying to keep it explicit as to what was happening, but I can
move those calls into __startup_64(). I'll still need the call to
sme_get_me_mask() in the secondary_startup_64 path, though (depending on
your thoughts to the above response).

quoted

@@ -98,7 +120,20 @@ ENTRY(secondary_startup_64)
  	/* Sanitize CPU configuration */
  	call verify_cpu
  
-	movq	$(init_top_pgt - __START_KERNEL_map), %rax
+	/*
+	 * Get the SME encryption mask.
+	 *  The encryption mask will be returned in %rax so we do an ADD
+	 *  below to be sure that the encryption mask is part of the
+	 *  value that will stored in %cr3.
+	 *
+	 * The real_mode_data address is in %rsi and that register can be
+	 * clobbered by the called function so be sure to save it.
+	 */
+	push	%rsi
+	call	sme_get_me_mask
+	pop	%rsi

Do we really need a call here? The mask is established at this point, so
it's either 0 when the encryption stuff is not compiled in or it can be
retrieved from a variable which is accessible at this point.

Same as above, this can be updated based on the decided approach.

Thanks,
Tom

quoted

+
+	addq	$(init_top_pgt - __START_KERNEL_map), %rax
  1:
  
  	/* Enable PAE mode, PGE and LA57 */

Thanks,

	tglx