x86/xpti: Don't leak TSS-adjacent percpu data via Meltdown

The XPTI work restricted the visibility of most of memory, but missed a few
aspects when it came to the TSS.

Given that the TSS is just an object in percpu data, the 4k mapping for it
created in setup_cpu_root_pgt() maps adjacent percpu data, making it all
leakable via Meltdown, even when XPTI is in use.

Furthermore, no care is taken to check that the TSS doesn't cross a page
boundary.  As it turns out, struct tss_struct is aligned on its size which
does prevent it straddling a page boundary.

Move the TSS into the page aligned percpu area, so no adjacent data can be
leaked.  Move the definition from setup.c to traps.c, which is a more
appropriate place for it to live.

Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>

Introduce / use struct tss_page.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
TBD: Especially with how the previous patch now works I'm unconvinced of
     the utility of the linker script alignment check. It in particular
     doesn't check the property we're after in this patch, i.e. the fact
     that there's nothing else in the same page.
NB: Sadly get_per_cpu_var() can't also be used on the "left" side of a
    #define.
---
v4:
 * Introduce / use struct tss_page.

v3:
 * Drop the remark about CET.  It is no longer accurate in the latest version
   of the CET spec.

v2:
 * Rebase over changes to include __aligned() within
   DEFINE_PER_CPU_PAGE_ALIGNED()
 * Drop now-unused xen/percpu.h from setup.c

--- a/xen/arch/x86/setup.c
+++ b/xen/arch/x86/setup.c
@@ -16,7 +16,6 @@
 #include <xen/domain_page.h>
 #include <xen/version.h>
 #include <xen/gdbstub.h>
-#include <xen/percpu.h>
 #include <xen/hypercall.h>
 #include <xen/keyhandler.h>
 #include <xen/numa.h>
@@ -100,8 +99,6 @@ unsigned long __read_mostly xen_phys_sta
 
 unsigned long __read_mostly xen_virt_end;
 
-DEFINE_PER_CPU(struct tss_struct, init_tss);
-
 char __section(".bss.stack_aligned") __aligned(STACK_SIZE)
     cpu0_stack[STACK_SIZE];
 
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -108,6 +108,12 @@ idt_entry_t __section(".bss.page_aligned
 /* Pointer to the IDT of every CPU. */
 idt_entry_t *idt_tables[NR_CPUS] __read_mostly;
 
+/*
+ * The TSS is smaller than a page, but we give it a full page to avoid
+ * adjacent per-cpu data leaking via Meltdown when XPTI is in use.
+ */
+DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_page, init_tss_page);
+
 bool (*ioemul_handle_quirk)(
     u8 opcode, char *io_emul_stub, struct cpu_user_regs *regs);
 
--- a/xen/arch/x86/xen.lds.S
+++ b/xen/arch/x86/xen.lds.S
@@ -368,6 +368,8 @@ ASSERT(IS_ALIGNED(__2M_rwdata_end,   SEC
 
 ASSERT(IS_ALIGNED(cpu0_stack, STACK_SIZE), "cpu0_stack misaligned")
 
+ASSERT(IS_ALIGNED(per_cpu__init_tss_page, PAGE_SIZE), "per_cpu(init_tss) misaligned")
+
 ASSERT(IS_ALIGNED(__init_begin, PAGE_SIZE), "__init_begin misaligned")
 ASSERT(IS_ALIGNED(__init_end,   PAGE_SIZE), "__init_end misaligned")
 
--- a/xen/include/asm-x86/processor.h
+++ b/xen/include/asm-x86/processor.h
@@ -411,7 +411,7 @@ static always_inline void __mwait(unsign
 #define IOBMP_BYTES             8192
 #define IOBMP_INVALID_OFFSET    0x8000
 
-struct __packed __cacheline_aligned tss_struct {
+struct __packed tss_struct {
     uint32_t :32;
     uint64_t rsp0, rsp1, rsp2;
     uint64_t :64;
@@ -425,6 +425,11 @@ struct __packed __cacheline_aligned tss_
     /* Pads the TSS to be cacheline-aligned (total size is 0x80). */
     uint8_t __cacheline_filler[24];
 };
+struct tss_page {
+    struct tss_struct __aligned(PAGE_SIZE) tss;
+};
+DECLARE_PER_CPU(struct tss_page, init_tss_page);
+#define per_cpu__init_tss get_per_cpu_var(init_tss_page.tss)
 
 #define IST_NONE 0UL
 #define IST_DF   1UL
@@ -463,7 +468,6 @@ static inline void disable_each_ist(idt_
 extern idt_entry_t idt_table[];
 extern idt_entry_t *idt_tables[];
 
-DECLARE_PER_CPU(struct tss_struct, init_tss);
 DECLARE_PER_CPU(root_pgentry_t *, root_pgt);
 
 extern void write_ptbase(struct vcpu *v);