From: <security@xenproject.org>
Subject: vtd: optimize CPU cache sync

Some VT-d IOMMUs are non-coherent, which requires a cache write back
in order for the changes made by the CPU to be visible to the IOMMU.
This cache write back was unconditionally done using clflush, but there are
other more efficient instructions to do so, hence implement support
for them using the alternative framework.

This is part of XSA-321.

Reviewed-by: Jan Beulich <jbeulich@suse.com>

--- a/xen/drivers/passthrough/vtd/extern.h
+++ b/xen/drivers/passthrough/vtd/extern.h
@@ -68,7 +68,6 @@ int __must_check qinval_device_iotlb_syn
                                           u16 did, u16 size, u64 addr);
 
 unsigned int get_cache_line_size(void);
-void cacheline_flush(char *);
 void flush_all_cache(void);
 
 uint64_t alloc_pgtable_maddr(unsigned long npages, nodeid_t node);
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -31,6 +31,7 @@
 #include <xen/pci_regs.h>
 #include <xen/keyhandler.h>
 #include <asm/msi.h>
+#include <asm/nops.h>
 #include <asm/irq.h>
 #include <asm/hvm/vmx/vmx.h>
 #include <asm/p2m.h>
@@ -154,7 +155,42 @@ static void sync_cache(const void *addr,
 
     addr -= (unsigned long)addr & (clflush_size - 1);
     for ( ; addr < end; addr += clflush_size )
-        cacheline_flush((char *)addr);
+/*
+ * The arguments to a macro must not include preprocessor directives. Doing so
+ * results in undefined behavior, so we have to create some defines here in
+ * order to avoid it.
+ */
+#if defined(HAVE_AS_CLWB)
+# define CLWB_ENCODING "clwb %[p]"
+#elif defined(HAVE_AS_XSAVEOPT)
+# define CLWB_ENCODING "data16 xsaveopt %[p]" /* clwb */
+#else
+# define CLWB_ENCODING ".byte 0x66, 0x0f, 0xae, 0x30" /* clwb (%%rax) */
+#endif
+
+#define BASE_INPUT(addr) [p] "m" (*(const char *)(addr))
+#if defined(HAVE_AS_CLWB) || defined(HAVE_AS_XSAVEOPT)
+# define INPUT BASE_INPUT
+#else
+# define INPUT(addr) "a" (addr), BASE_INPUT(addr)
+#endif
+        /*
+         * Note regarding the use of NOP_DS_PREFIX: it's faster to do a clflush
+         * + prefix than a clflush + nop, and hence the prefix is added instead
+         * of letting the alternative framework fill the gap by appending nops.
+         */
+        alternative_io_2(".byte " __stringify(NOP_DS_PREFIX) "; clflush %[p]",
+                         "data16 clflush %[p]", /* clflushopt */
+                         X86_FEATURE_CLFLUSHOPT,
+                         CLWB_ENCODING,
+                         X86_FEATURE_CLWB, /* no outputs */,
+                         INPUT(addr));
+#undef INPUT
+#undef BASE_INPUT
+#undef CLWB_ENCODING
+
+    alternative_2("", "sfence", X86_FEATURE_CLFLUSHOPT,
+                      "sfence", X86_FEATURE_CLWB);
 }
 
 /* Allocate page table, return its machine address */
--- a/xen/drivers/passthrough/vtd/x86/vtd.c
+++ b/xen/drivers/passthrough/vtd/x86/vtd.c
@@ -51,11 +51,6 @@ unsigned int get_cache_line_size(void)
     return ((cpuid_ebx(1) >> 8) & 0xff) * 8;
 }
 
-void cacheline_flush(char * addr)
-{
-    clflush(addr);
-}
-
 void flush_all_cache()
 {
     wbinvd();