[Xen-devel] [PATCH 1/4] x86: Begin to remove TSC mode PVRDTSCP

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

For more historical context, see
  c/s c17b36d5dc792cfdf59b6de0213b168bec0af8e8
  c/s 04656384a1b9714e43db850c51431008e23450d8

PVRDTSCP was an attempt to provide Xen-aware userspace with a stable monotonic
clock, and enough information for said userspace to cope with frequency
changes across migrate.  However, the PVRDTSCP infrastructure has resulted in
very tangled code, and non-architectural behaviour even in non-PVRDTSCP cases.

Seeing as the functionality has been replaced entirely by improvements in PV
clocks (including being plumbed into the VDSO nowadays), or alternatively by
hardware TSC scaling features, and no-one is aware of any remaining users of
this mode, take the opportunity to remove it.

For now, introduce an upper range check on the toolstack setting to
XEN_DOMCTL_settscinfo, and exclude TSC_MODE_PVRDTSCP from selection.
(Arguably, its a bug that this hypercall previously accepted any value and
turned into a nop).  This will catch and cleanly reject attempts to migrate in
a VM configured to use PVRDTSCP, rather than letting it run and have the wrong
timing mode.

Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
---
CC: Jan Beulich <JBeulich@xxxxxxxx>
CC: Wei Liu <wei.liu2@xxxxxxxxxx>
CC: Roger Pau Monné <roger.pau@xxxxxxxxxx>
CC: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
CC: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
---
 xen/arch/x86/domctl.c      |  3 ++-
 xen/arch/x86/time.c        | 35 -----------------------------------
 xen/include/asm-x86/time.h |  5 +----
 3 files changed, 3 insertions(+), 40 deletions(-)

diff --git a/xen/arch/x86/domctl.c b/xen/arch/x86/domctl.c
index 175a0c9..97ea5d8 100644
--- a/xen/arch/x86/domctl.c
+++ b/xen/arch/x86/domctl.c
@@ -970,7 +970,8 @@ long arch_do_domctl(
         break;
 
     case XEN_DOMCTL_settscinfo:
-        if ( d == currd ) /* no domain_pause() */
+        if ( d == currd || /* no domain_pause() */
+             domctl->u.tsc_info.tsc_mode > TSC_MODE_NEVER_EMULATE )
             ret = -EINVAL;
         else
         {
diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 24d4c27..3f095a2 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -2165,21 +2165,6 @@ void tsc_get_info(struct domain *d, uint32_t *tsc_mode,
         *elapsed_nsec = scale_delta(tsc, &d->arch.vtsc_to_ns);
         *gtsc_khz = enable_tsc_scaling ? d->arch.tsc_khz : cpu_khz;
         break;
-    case TSC_MODE_PVRDTSCP:
-        if ( d->arch.vtsc )
-        {
-            *elapsed_nsec = get_s_time() - d->arch.vtsc_offset;
-            *gtsc_khz = cpu_khz;
-        }
-        else
-        {
-            tsc = rdtsc();
-            *elapsed_nsec = scale_delta(tsc, &this_cpu(cpu_time).tsc_scale) -
-                            d->arch.vtsc_offset;
-            *gtsc_khz = enable_tsc_scaling ? d->arch.tsc_khz
-                                           : 0 /* ignored by tsc_set_info */;
-        }
-        break;
     }
 
     if ( (int64_t)*elapsed_nsec < 0 )
@@ -2208,8 +2193,6 @@ void tsc_set_info(struct domain *d,
 
     switch ( d->arch.tsc_mode = tsc_mode )
     {
-        bool enable_tsc_scaling;
-
     case TSC_MODE_DEFAULT:
     case TSC_MODE_ALWAYS_EMULATE:
         d->arch.vtsc_offset = get_s_time() - elapsed_nsec;
@@ -2235,24 +2218,6 @@ void tsc_set_info(struct domain *d,
         d->arch.vtsc = 1;
         d->arch.ns_to_vtsc = scale_reciprocal(d->arch.vtsc_to_ns);
         break;
-    case TSC_MODE_PVRDTSCP:
-        d->arch.vtsc = !boot_cpu_has(X86_FEATURE_RDTSCP) ||
-                       !host_tsc_is_safe();
-        enable_tsc_scaling = is_hvm_domain(d) && !d->arch.vtsc &&
-                             hvm_get_tsc_scaling_ratio(gtsc_khz ?: cpu_khz);
-        d->arch.tsc_khz = (enable_tsc_scaling && gtsc_khz) ? gtsc_khz : 
cpu_khz;
-        set_time_scale(&d->arch.vtsc_to_ns, d->arch.tsc_khz * 1000 );
-        d->arch.ns_to_vtsc = scale_reciprocal(d->arch.vtsc_to_ns);
-        if ( d->arch.vtsc )
-            d->arch.vtsc_offset = get_s_time() - elapsed_nsec;
-        else {
-            /* when using native TSC, offset is nsec relative to power-on
-             * of physical machine */
-            d->arch.vtsc_offset = scale_delta(rdtsc(),
-                                              &this_cpu(cpu_time).tsc_scale) -
-                                  elapsed_nsec;
-        }
-        break;
     }
     d->arch.incarnation = incarnation + 1;
     if ( is_hvm_domain(d) )
diff --git a/xen/include/asm-x86/time.h b/xen/include/asm-x86/time.h
index ce96ec9..070cdef 100644
--- a/xen/include/asm-x86/time.h
+++ b/xen/include/asm-x86/time.h
@@ -12,10 +12,7 @@
  *    2 = guest rdtsc always executed natively (no monotonicity/frequency
  *         guarantees); guest rdtscp emulated at native frequency if
  *         unsupported by h/w, else executed natively
- *    3 = same as 2, except xen manages TSC_AUX register so guest can
- *         determine when a restore/migration has occurred and assumes
- *         guest obtains/uses pvclock-like mechanism to adjust for
- *         monotonicity and frequency changes
+ *    3 = Removed, was PVRDTSCP.
  */
 #define TSC_MODE_DEFAULT          0
 #define TSC_MODE_ALWAYS_EMULATE   1
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel