x86/amd: Work around CLFLUSH ordering on older parts

On pre-CLFLUSHOPT AMD CPUs, CLFLUSH is weakely ordered with everything,
including reads and writes to the address, and LFENCE/SFENCE instructions.

This creates a multitude of problematic corner cases, laid out in the manual.
Arrange to use MFENCE on both sides of the CLFLUSH to force proper ordering.

This is part of XSA-402.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>

xen/arch/x86/cpu/amd.c

@@ -812,6 +812,14 @@ static void cf_check init_amd(struct cpuinfo_x86 *c)
 	if (!cpu_has_lfence_dispatch)
 		__set_bit(X86_FEATURE_MFENCE_RDTSC, c->x86_capability);
 
+	/*
+	 * On pre-CLFLUSHOPT AMD CPUs, CLFLUSH is weakly ordered with
+	 * everything, including reads and writes to address, and
+	 * LFENCE/SFENCE instructions.
+	 */
+	if (!cpu_has_clflushopt)
+		setup_force_cpu_cap(X86_BUG_CLFLUSH_MFENCE);
+
 	switch(c->x86)
 	{
 	case 0xf ... 0x11:

xen/arch/x86/flushtlb.c

@@ -260,6 +260,13 @@ unsigned int flush_area_local(const void *va, unsigned int flags)
     return flags;
 }
 
+/*
+ * On pre-CLFLUSHOPT AMD CPUs, CLFLUSH is weakly ordered with everything,
+ * including reads and writes to address, and LFENCE/SFENCE instructions.
+ *
+ * This function only works safely after alternatives have run.  Luckily, at
+ * the time of writing, we don't flush the caches that early.
+ */
 void cache_flush(const void *addr, unsigned int size)
 {
     /*
@@ -269,6 +276,8 @@ void cache_flush(const void *addr, unsigned int size)
     unsigned int clflush_size = current_cpu_data.x86_clflush_size ?: 16;
     const void *end = addr + size;
 
+    alternative("", "mfence", X86_BUG_CLFLUSH_MFENCE);
+
     addr -= (unsigned long)addr & (clflush_size - 1);
     for ( ; addr < end; addr += clflush_size )
     {
@@ -284,7 +293,9 @@ void cache_flush(const void *addr, unsigned int size)
                        [p] "m" (*(const char *)(addr)));
     }
 
-    alternative("", "sfence", X86_FEATURE_CLFLUSHOPT);
+    alternative_2("",
+                  "sfence", X86_FEATURE_CLFLUSHOPT,
+                  "mfence", X86_BUG_CLFLUSH_MFENCE);
 }
 
 void cache_writeback(const void *addr, unsigned int size)

xen/arch/x86/include/asm/cpufeatures.h

@@ -47,6 +47,7 @@ XEN_CPUFEATURE(XEN_IBT,           X86_SYNTH(27)) /* Xen uses CET Indirect Branch
 
 #define X86_BUG_FPU_PTRS          X86_BUG( 0) /* (F)X{SAVE,RSTOR} doesn't save/restore FOP/FIP/FDP. */
 #define X86_BUG_NULL_SEG          X86_BUG( 1) /* NULL-ing a selector preserves the base and limit. */
+#define X86_BUG_CLFLUSH_MFENCE    X86_BUG( 2) /* MFENCE needed to serialise CLFLUSH */
 
 /* Total number of capability words, inc synth and bug words. */
 #define NCAPINTS (FSCAPINTS + X86_NR_SYNTH + X86_NR_BUG) /* N 32-bit words worth of info */