[PATCH v5 0/5] x86/time: PVCLOCK_TSC_STABLE_BIT support

[PATCH v5 0/5] x86/time: PVCLOCK_TSC_STABLE_BIT support

[Joao Martins]

Hey,

This is v5 on the pvclock TSC series addressing comments from previous
version, see individual patches for complete changelog.

This series is divided as follows:

 R      * Patch 1: Small refactor around init_platform_time to reuse
                   initialization code when switching to TSC.
 A      * Patch 2: Prerequisite for patch 3
 U      * Patch 3: Adds a new clocksource based on TSC
 R      * Patch 4: Implements the PVCLOCK_TSC_STABLE_BIT
 U      * Patch 5: Extend "tsc" param to relax monotonicity restriction
                   across sockets.

 [ A := Acked-by ;; R := Reviewed-by ;; U := Updated ]

The benefit of this series is two-fold:
 
 1. Provide the guarantee of monotonic results on xen own system time as seen
 by any cpu when using TSC as clocksource.

 2. Provide this same guarantee to guests and thus set the
 TSC_STABLE_BIT (both FreeBSD and Linux support it) which then allows guests to
 skip expensive monotonicity check between PV CPU time infos. Plus, on Linux
 specifically this also means that it could support vDSO which greatly increases
 performance (x10) for gettimeofday and clock_gettime since it would no
 longer need to do the system call to get a reliable snapshot of system time.
 For a reference on my laptop the speed of gettimeofday under xen pvclock is 
 ~2 Mops/sec (Million ops per sec) whereas with vDSO it's on the range
 of ~22 Mops/sec on <= 4.4 kernels and ~37 Mops on >= 4.5.
 
 Doing a long running time warp test for >2 weeks on a dual-socket Haswell
 machine and I haven't yet seen time going backwards (plus tests on older
 multi socket machines). Furthermore double checked migration to/from hosts
 with/without the bit while guest was running the time warp test and no issues
 occurred too.

Thanks!

Joao Martins (5):
  x86/time: refactor init_platform_time()
  x86/time: refactor read_platform_stime()
  x86/time: implement tsc as clocksource
  x86/time: implement PVCLOCK_TSC_STABLE_BIT
  x86/time: extend "tsc" param with "stable:socket"

 docs/misc/xen-command-line.markdown |  10 +-
 xen/arch/x86/platform_hypercall.c   |   3 +-
 xen/arch/x86/time.c                 | 230 +++++++++++++++++++++++++++++++-----
 xen/include/asm-x86/time.h          |   1 +
 4 files changed, 210 insertions(+), 34 deletions(-)

-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[PATCH v5 1/5] x86/time: refactor init_platform_time()

[Joao Martins]

And accomodate platform time source initialization in
try_platform_time(). This is a preparatory patch for deferring
TSC clocksource initialization to the stage where all CPUS are
up (verify_tsc_reliability init call).

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
---
 xen/arch/x86/time.c | 30 ++++++++++++++++++++----------
 1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 73e0f98..0c1ad45 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -576,6 +576,24 @@ static void resume_platform_timer(void)
     plt_stamp = plt_src.read_counter();
 }
 
+static s64 __init try_platform_timer(struct platform_timesource *pts)
+{
+    s64 rc = pts->init(pts);
+
+    if ( rc <= 0 )
+        return rc;
+
+    plt_mask = (u64)~0ull >> (64 - pts->counter_bits);
+
+    set_time_scale(&plt_scale, pts->frequency);
+
+    plt_overflow_period = scale_delta(
+        1ull << (pts->counter_bits - 1), &plt_scale);
+    plt_src = *pts;
+
+    return rc;
+}
+
 static u64 __init init_platform_timer(void)
 {
     static struct platform_timesource * __initdata plt_timers[] = {
@@ -593,7 +611,7 @@ static u64 __init init_platform_timer(void)
             pts = plt_timers[i];
             if ( !strcmp(opt_clocksource, pts->id) )
             {
-                rc = pts->init(pts);
+                rc = try_platform_timer(pts);
                 break;
             }
         }
@@ -609,21 +627,13 @@ static u64 __init init_platform_timer(void)
         for ( i = 0; i < ARRAY_SIZE(plt_timers); i++ )
         {
             pts = plt_timers[i];
-            if ( (rc = pts->init(pts)) > 0 )
+            if ( (rc = try_platform_timer(pts)) > 0 )
                 break;
         }
     }
 
     BUG_ON(rc <= 0);
 
-    plt_mask = (u64)~0ull >> (64 - pts->counter_bits);
-
-    set_time_scale(&plt_scale, pts->frequency);
-
-    plt_overflow_period = scale_delta(
-        1ull << (pts->counter_bits-1), &plt_scale);
-    plt_src = *pts;
-
     printk("Platform timer is %s %s\n",
            freq_string(pts->frequency), pts->name);
 
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[PATCH v5 2/5] x86/time: refactor read_platform_stime()

[Joao Martins]

To allow the caller to fetch the last read from the clocksource which
was used to calculate system_time. This is a prerequisite for a
subsequent patch that will use this last read.

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
---
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>

Changes since v4:
 - Add unlikely when callers provide a stamp argument
 - Add Acked-by from Jan

Changes since v3:
 - Add mention of this being a prerequisite to a later patch.
---
 xen/arch/x86/time.c | 22 +++++++++++++---------
 1 file changed, 13 insertions(+), 9 deletions(-)

diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 0c1ad45..6305a84 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -537,18 +537,22 @@ static void plt_overflow(void *unused)
     set_timer(&plt_overflow_timer, NOW() + plt_overflow_period);
 }
 
-static s_time_t read_platform_stime(void)
+static s_time_t read_platform_stime(u64 *stamp)
 {
-    u64 count;
+    u64 plt_counter, count;
     s_time_t stime;
 
     ASSERT(!local_irq_is_enabled());
 
     spin_lock(&platform_timer_lock);
-    count = plt_stamp64 + ((plt_src.read_counter() - plt_stamp) & plt_mask);
+    plt_counter = plt_src.read_counter();
+    count = plt_stamp64 + ((plt_counter - plt_stamp) & plt_mask);
     stime = __read_platform_stime(count);
     spin_unlock(&platform_timer_lock);
 
+    if ( unlikely(stamp) )
+        *stamp = plt_counter;
+
     return stime;
 }
 
@@ -661,7 +665,7 @@ void cstate_restore_tsc(void)
     if ( boot_cpu_has(X86_FEATURE_NONSTOP_TSC) )
         return;
 
-    write_tsc(stime2tsc(read_platform_stime()));
+    write_tsc(stime2tsc(read_platform_stime(NULL)));
 }
 
 /***************************************************************************
@@ -980,7 +984,7 @@ int cpu_frequency_change(u64 freq)
 
     local_irq_disable();
     /* Platform time /first/, as we may be delayed by platform_timer_lock. */
-    t->stamp.master_stime = read_platform_stime();
+    t->stamp.master_stime = read_platform_stime(NULL);
     curr_tsc = rdtsc_ordered();
     /* TSC-extrapolated time may be bogus after frequency change. */
     /*t->stamp.local_stime = get_s_time_fixed(curr_tsc);*/
@@ -1285,7 +1289,7 @@ static void time_calibration_tsc_rendezvous(void *_r)
 
             if ( r->master_stime == 0 )
             {
-                r->master_stime = read_platform_stime();
+                r->master_stime = read_platform_stime(NULL);
                 r->master_tsc_stamp = rdtsc_ordered();
             }
             atomic_inc(&r->semaphore);
@@ -1325,7 +1329,7 @@ static void time_calibration_std_rendezvous(void *_r)
     {
         while ( atomic_read(&r->semaphore) != (total_cpus - 1) )
             cpu_relax();
-        r->master_stime = read_platform_stime();
+        r->master_stime = read_platform_stime(NULL);
         smp_wmb(); /* write r->master_stime /then/ signal */
         atomic_inc(&r->semaphore);
     }
@@ -1364,7 +1368,7 @@ void time_latch_stamps(void)
     unsigned long flags;
 
     local_irq_save(flags);
-    ap_bringup_ref.master_stime = read_platform_stime();
+    ap_bringup_ref.master_stime = read_platform_stime(NULL);
     ap_bringup_ref.local_tsc = rdtsc_ordered();
     local_irq_restore(flags);
 
@@ -1382,7 +1386,7 @@ void init_percpu_time(void)
     t->tsc_scale = per_cpu(cpu_time, 0).tsc_scale;
 
     local_irq_save(flags);
-    now = read_platform_stime();
+    now = read_platform_stime(NULL);
     tsc = rdtsc_ordered();
     local_irq_restore(flags);
 
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[PATCH v5 3/5] x86/time: implement tsc as clocksource

[Joao Martins]

Recent x86/time changes improved a lot of the monotonicity in xen
timekeeping, making it much harder to observe time going backwards.
Although platform timer can't be expected to be perfectly in sync with
TSC and so get_s_time won't be guaranteed to always return
monotonically increasing values across cpus. This is the case in some
of the boxes I am testing with, observing sometimes ~100 warps (of
very few nanoseconds each) after a few hours.

This patch introduces support for using TSC as platform time source
which is the highest resolution time and most performant to get.
Though there are also several problems associated with its usage, and
there isn't a complete (and architecturally defined) guarantee that
all machines will provide reliable and monotonic TSC in all cases (I
believe Intel to be the only that can guarantee that?). For this reason
it's not used unless administrator changes "clocksource" boot option
to "tsc". Initializing TSC clocksource requires all CPUs up to have
the tsc reliability checks performed. init_xen_time is called before
all CPUs are up, so for example we would start with HPET (or ACPI,
PIT) at boot time, and switch later to TSC. The switch then happens on
verify_tsc_reliability initcall that is invoked when all CPUs are up.
When attempting to initialize TSC we also check for time warps and if
it has invariant TSC. Note that while we deem reliable a CONSTANT_TSC
with no deep C-states, it might not always be the case, so we're
conservative and allow TSC to be used as platform timer only with
invariant TSC. Additionally we check if CPU Hotplug isn't meant to be
performed on the host which will either be when max vcpus and
num_present_cpu are the same. This is because a newly hotplugged CPU
may not satisfy the condition of having all TSCs synchronized - so
when having tsc clocksource being used we allow offlining CPUs but not
onlining any ones back. Finally we prevent TSC from being used as
clocksource on multiple sockets because it isn't guaranteed to be
invariant. Further relaxing of this last requirement is added in a
separate patch, such that we allow vendors with such guarantee to use
TSC as clocksource. In case any of these conditions is not met, we
keep the clocksource that was previously initialized on init_xen_time.

Since b64438c7c ("x86/time: use correct (local) time stamp in
constant-TSC calibration fast path") updates to cpu time use local
stamps, which means platform timer is only used to seed the initial
cpu time. We further introduce a new rendezvous function
(nop_rendezvous) which doesn't require synchronization between master
and slave CPUS and just reads calibration_rendezvous struct and writes
it down the stime and stamp to the cpu_calibration struct to be used
later on. With clocksource=tsc there is no need to be in sync with
another clocksource, so we reseed the local/master stamps to be values
of TSC and update the platform time stamps accordingly. Time
calibration is set to 1sec after we switch to TSC, thus these stamps
are reseeded to also ensure monotonic returning values right after the
point we switch to TSC. This is to remove the possibility of having
inconsistent readings in this short period (i.e. until calibration
fires).

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
---
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>

Changes sinve v4:
 - Moved nop_rendezvous to this patch and along with its mention in the
 commit message.
 - Remove mention of HPET priority vs other clocksources.
 - Change "to avoid the possibility" into "to remove (...)"
 - Change TSC init messages to use warning log level and leave
 the last one as debug.
 - Change "Switched to Platform timer" message to same loglevel as the
 one in init_platform_timer()
 - Introduce "late" boolean function parameter in
   try_platform_timer_tail.
 - Do not split the if (checking for clocksource=tsc) into two lines.
 - Add brief comment about why TSC .counter_bits is not 64

Changes since v3:
 - Really fix "HPET switching to TSC" comment. Despite mentioned in the
 in previous version, the change wasn't there.
 - Remove parenthesis around the function call in init_platform_timer
 - Merge if on verify_tsc_reliability with opt_clocksource check
 - Removed comment above ".init = init_tsctimer"
 - Fixup docs updated into this patch.
 - Move host_tsc_is_clocksource() and CPU hotplug possibility check to this
 patch.
 - s/host_tsc_is_clocksource/clocksource_is_tsc
 - Use bool instead of bool_t
 - Add a comment above init_tsctimer() declaration mentioning the
   reliable TSC checks on verify_tsc_reliability(), under which the
   function is invoked.
 - Prevent clocksource=tsc on platforms with multiple-sockets. Further
 relaxing of this requirement is added in a separate patch, as
 extension of "tsc" boot parameter.
 - Removed control group to update cpu_time and do instead with
   on_selected_cpus to avoid any potential races.
 - Accomodate common path between init_xen_time and TSC switch into
 try_platform_timer_tail, such that finishing platform timer
 initialization is done in the same place (including platform timer
 overflow which was previously was removed in previous versions).
 - Changed TSC counter_bits 63 to avoid mishandling of TSC counter
 wrap-around in platform timer overflow timer.
 - Moved paragraph CPU Hotplug from last patch and add comment on
   commit message about multiple sockets TSC sync.
 - s/init_tsctimer/init_tsc/g to be consistent with other TSC platform
 timer functions.

Changes since v2:
 - Suggest "HPET switching to TSC" only as an example as otherwise it
 would be misleading on platforms not having one.
 - Change init_tsctimer to skip all the tests and assume it's called
 only on reliable TSC conditions and no warps observed. Tidy
 initialization on verify_tsc_reliability as suggested by Konrad.
 - CONSTANT_TSC and max_cstate <= 2 case removed and only allow tsc
   clocksource in invariant TSC boxes.
 - Prefer omit !=0 on init_platform_timer for tsc case.
 - Change comment on init_platform_timer.
 - Add comment on plt_tsc declaration.
 - Reinit CPU time for all online cpus instead of just CPU 0.
 - Use rdtsc_ordered() as opposed to rdtsc()
 - Remove tsc_freq variable and set plt_tsc clocksource frequency
 with the refined tsc calibration.
 - Rework a bit the commit message.

Changes since v1:
 - s/printk/printk(XENLOG_INFO
 - Remove extra space on inner brackets
 - Add missing space around brackets
 - Defer TSC initialization when all CPUs are up.

Changes since RFC:
 - Spelling fixes in the commit message.
 - Remove unused clocksource_is_tsc variable and introduce it instead
 on the patch that uses it.
 - Move plt_tsc from second to last in the available clocksources.
---
 docs/misc/xen-command-line.markdown |   6 +-
 xen/arch/x86/platform_hypercall.c   |   3 +-
 xen/arch/x86/time.c                 | 161 +++++++++++++++++++++++++++++++++---
 xen/include/asm-x86/time.h          |   1 +
 4 files changed, 159 insertions(+), 12 deletions(-)

diff --git a/docs/misc/xen-command-line.markdown b/docs/misc/xen-command-line.markdown
index 3a250cb..f92fb3f 100644
--- a/docs/misc/xen-command-line.markdown
+++ b/docs/misc/xen-command-line.markdown
@@ -264,9 +264,13 @@ minimum of 32M, subject to a suitably aligned and sized contiguous
 region of memory being available.
 
 ### clocksource
-> `= pit | hpet | acpi`
+> `= pit | hpet | acpi | tsc`
 
 If set, override Xen's default choice for the platform timer.
+Having TSC as platform timer requires being explicitly set. This is because
+TSC can only be safely used if CPU hotplug isn't performed on the system. In
+some platforms, "maxcpus" parameter may require further adjustment to the
+number of online cpus.
 
 ### cmci-threshold
 > `= <integer>`
diff --git a/xen/arch/x86/platform_hypercall.c b/xen/arch/x86/platform_hypercall.c
index 780f22d..0879e19 100644
--- a/xen/arch/x86/platform_hypercall.c
+++ b/xen/arch/x86/platform_hypercall.c
@@ -631,7 +631,8 @@ ret_t do_platform_op(XEN_GUEST_HANDLE_PARAM(xen_platform_op_t) u_xenpf_op)
         if ( ret )
             break;
 
-        if ( cpu >= nr_cpu_ids || !cpu_present(cpu) )
+        if ( cpu >= nr_cpu_ids || !cpu_present(cpu) ||
+             clocksource_is_tsc() )
         {
             ret = -EINVAL;
             break;
diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 6305a84..128e653 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -475,6 +475,55 @@ uint64_t ns_to_acpi_pm_tick(uint64_t ns)
 }
 
 /************************************************************
+ * PLATFORM TIMER 4: TSC
+ */
+
+/*
+ * Called in verify_tsc_reliability() under reliable TSC conditions
+ * thus reusing all the checks already performed there.
+ */
+static s64 __init init_tsc(struct platform_timesource *pts)
+{
+    u64 ret = pts->frequency;
+
+    if ( nr_cpu_ids != num_present_cpus() )
+    {
+        printk(XENLOG_WARNING "TSC: CPU Hotplug intended\n");
+        ret = 0;
+    }
+
+    if ( nr_sockets > 1 )
+    {
+        printk(XENLOG_WARNING "TSC: Not invariant across sockets\n");
+        ret = 0;
+    }
+
+    if ( !ret )
+        printk(XENLOG_DEBUG "TSC: Not setting it as clocksource\n");
+
+    return ret;
+}
+
+static u64 read_tsc(void)
+{
+    return rdtsc_ordered();
+}
+
+static struct platform_timesource __initdata plt_tsc =
+{
+    .id = "tsc",
+    .name = "TSC",
+    .read_counter = read_tsc,
+    /*
+     * Calculations for platform timer overflow assume u64 boundary.
+     * Hence we set to less than 64, such that the TSC wraparound is
+     * correctly checked and handled.
+     */
+    .counter_bits = 63,
+    .init = init_tsc,
+};
+
+/************************************************************
  * GENERIC PLATFORM TIMER INFRASTRUCTURE
  */
 
@@ -580,6 +629,21 @@ static void resume_platform_timer(void)
     plt_stamp = plt_src.read_counter();
 }
 
+static void __init reset_platform_timer(void)
+{
+    /* Deactivate any timers running */
+    kill_timer(&plt_overflow_timer);
+    kill_timer(&calibration_timer);
+
+    /* Reset counters and stamps */
+    spin_lock_irq(&platform_timer_lock);
+    plt_stamp = 0;
+    plt_stamp64 = 0;
+    platform_timer_stamp = 0;
+    stime_platform_stamp = 0;
+    spin_unlock_irq(&platform_timer_lock);
+}
+
 static s64 __init try_platform_timer(struct platform_timesource *pts)
 {
     s64 rc = pts->init(pts);
@@ -587,6 +651,10 @@ static s64 __init try_platform_timer(struct platform_timesource *pts)
     if ( rc <= 0 )
         return rc;
 
+    /* We have a platform timesource already so reset it */
+    if ( plt_src.counter_bits != 0 )
+        reset_platform_timer();
+
     plt_mask = (u64)~0ull >> (64 - pts->counter_bits);
 
     set_time_scale(&plt_scale, pts->frequency);
@@ -608,7 +676,8 @@ static u64 __init init_platform_timer(void)
     unsigned int i;
     s64 rc = -1;
 
-    if ( opt_clocksource[0] != '\0' )
+    /* clocksource=tsc is initialized via __initcalls (when CPUs are up). */
+    if ( (opt_clocksource[0] != '\0') && strcmp(opt_clocksource, "tsc") )
     {
         for ( i = 0; i < ARRAY_SIZE(plt_timers); i++ )
         {
@@ -1344,6 +1413,22 @@ static void time_calibration_std_rendezvous(void *_r)
     time_calibration_rendezvous_tail(r);
 }
 
+/*
+ * Rendezvous function used when clocksource is TSC and
+ * no CPU hotplug will be performed.
+ */
+static void time_calibration_nop_rendezvous(void *rv)
+{
+    const struct calibration_rendezvous *r = rv;
+    struct cpu_time_stamp *c = &this_cpu(cpu_calibration);
+
+    c->local_tsc    = r->master_tsc_stamp;
+    c->local_stime  = r->master_stime;
+    c->master_stime = r->master_stime;
+
+    raise_softirq(TIME_CALIBRATE_SOFTIRQ);
+}
+
 static void (*time_calibration_rendezvous_fn)(void *) =
     time_calibration_std_rendezvous;
 
@@ -1353,6 +1438,13 @@ static void time_calibration(void *unused)
         .semaphore = ATOMIC_INIT(0)
     };
 
+    if ( clocksource_is_tsc() )
+    {
+        local_irq_disable();
+        r.master_stime = read_platform_stime(&r.master_tsc_stamp);
+        local_irq_enable();
+    }
+
     cpumask_copy(&r.cpu_calibration_map, &cpu_online_map);
 
     /* @wait=1 because we must wait for all cpus before freeing @r. */
@@ -1467,6 +1559,31 @@ static void __init tsc_check_writability(void)
     disable_tsc_sync = 1;
 }
 
+static void __init reset_percpu_time(void *unused)
+{
+    struct cpu_time *t = &this_cpu(cpu_time);
+
+    t->stamp.local_tsc = boot_tsc_stamp;
+    t->stamp.local_stime = 0;
+    t->stamp.local_stime = get_s_time_fixed(boot_tsc_stamp);
+    t->stamp.master_stime = t->stamp.local_stime;
+}
+
+static void __init try_platform_timer_tail(bool late)
+{
+    init_timer(&plt_overflow_timer, plt_overflow, NULL, 0);
+    plt_overflow(NULL);
+
+    platform_timer_stamp = plt_stamp64;
+    stime_platform_stamp = NOW();
+
+    if ( !late )
+        init_percpu_time();
+
+    init_timer(&calibration_timer, time_calibration, NULL, 0);
+    set_timer(&calibration_timer, NOW() + EPOCH);
+}
+
 /* Late init function, after all cpus have booted */
 static int __init verify_tsc_reliability(void)
 {
@@ -1484,6 +1601,32 @@ static int __init verify_tsc_reliability(void)
             printk("TSC warp detected, disabling TSC_RELIABLE\n");
             setup_clear_cpu_cap(X86_FEATURE_TSC_RELIABLE);
         }
+        else if ( !strcmp(opt_clocksource, "tsc") &&
+                  (try_platform_timer(&plt_tsc) > 0) )
+        {
+            /*
+             * Platform timer has changed and CPU time will only be updated
+             * after we set again the calibration timer, which means we need to
+             * seed again each local CPU time. At this stage TSC is known to be
+             * reliable i.e. monotonically increasing across all CPUs so this
+             * lets us remove the skew between platform timer and TSC, since
+             * these are now effectively the same.
+             */
+            on_selected_cpus(&cpu_online_map, reset_percpu_time, NULL, 1);
+
+            /*
+             * We won't do CPU Hotplug and TSC clocksource is being used which
+             * means we have a reliable TSC, plus we don't sync with any other
+             * clocksource so no need for rendezvous.
+             */
+            time_calibration_rendezvous_fn = time_calibration_nop_rendezvous;
+
+            /* Finish platform timer switch. */
+            try_platform_timer_tail(true);
+
+            printk("Switched to Platform timer %s TSC\n",
+                   freq_string(plt_src.frequency));
+        }
     }
 
     return 0;
@@ -1509,15 +1652,7 @@ int __init init_xen_time(void)
     do_settime(get_cmos_time(), 0, NOW());
 
     /* Finish platform timer initialization. */
-    init_timer(&plt_overflow_timer, plt_overflow, NULL, 0);
-    plt_overflow(NULL);
-    platform_timer_stamp = plt_stamp64;
-    stime_platform_stamp = NOW();
-
-    init_percpu_time();
-
-    init_timer(&calibration_timer, time_calibration, NULL, 0);
-    set_timer(&calibration_timer, NOW() + EPOCH);
+    try_platform_timer_tail(false);
 
     return 0;
 }
@@ -1531,6 +1666,7 @@ void __init early_time_init(void)
 
     preinit_pit();
     tmp = init_platform_timer();
+    plt_tsc.frequency = tmp;
 
     set_time_scale(&t->tsc_scale, tmp);
     t->stamp.local_tsc = boot_tsc_stamp;
@@ -1779,6 +1915,11 @@ void pv_soft_rdtsc(struct vcpu *v, struct cpu_user_regs *regs, int rdtscp)
              (d->arch.tsc_mode == TSC_MODE_PVRDTSCP) ? d->arch.incarnation : 0;
 }
 
+bool clocksource_is_tsc(void)
+{
+    return plt_src.read_counter == read_tsc;
+}
+
 int host_tsc_is_safe(void)
 {
     return boot_cpu_has(X86_FEATURE_TSC_RELIABLE);
diff --git a/xen/include/asm-x86/time.h b/xen/include/asm-x86/time.h
index 971883a..6d704b4 100644
--- a/xen/include/asm-x86/time.h
+++ b/xen/include/asm-x86/time.h
@@ -69,6 +69,7 @@ void tsc_get_info(struct domain *d, uint32_t *tsc_mode, uint64_t *elapsed_nsec,
 
 void force_update_vcpu_system_time(struct vcpu *v);
 
+bool clocksource_is_tsc(void);
 int host_tsc_is_safe(void);
 void cpuid_time_leaf(uint32_t sub_idx, uint32_t *eax, uint32_t *ebx,
                      uint32_t *ecx, uint32_t *edx);
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[PATCH v5 4/5] x86/time: implement PVCLOCK_TSC_STABLE_BIT

[Joao Martins]

This patch proposes relying on host TSC synchronization and
passthrough to the guest, when running on a TSC-safe platform. On
time_calibration we retrieve the platform time in ns and the counter
read by the clocksource that was used to compute system time. We can
guarantee that on a platform with a constant and reliable TSC, that the
time read on vcpu B right after A is bigger independently of the VCPU
calibration error. Since pvclock time infos are monotonic as seen by any
vCPU set PVCLOCK_TSC_STABLE_BIT, which then enables usage of VDSO on
Linux.  IIUC, this is similar to how it's implemented on KVM. Add also a
comment regarding this bit changing and that guests are expected to
check this bit on every read.

Should note that I've yet to see time going backwards in a long running
test I ran for 2 weeks (in a dual socket machine), plus few other
tests I did on older platforms.

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
---
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>

Changes since v4:
 - Remove blanks from flag assignment.
 - Added "Reviewed-by" from Jan Beulich.
 - Move nop_rendezvous introduction (and the mention in the commit
   message) into the patch introducing clocksource=tsc.
 [I kept the Reviewed-by as there was no functional change but rather only
  moving things out (as from discussion in the other patch). I wasn't sure
  if I should have kept it, so if it was a wrong assumption in the protocol,
  let me know.]

Changes since v3:
 - Do not adjust time_calibration_rendezvous_tail for nop_rendezvous and
 instead set cpu_time_stamp directly on the rendezvous function.
 - Move CPU Hotplug checks into patch 2
 - Add a commit and code comment regarding guests cope with this bit
   changing on hosts.
 - s/host_tsc_is_clocksource/clocksource_is_tsc

Changes since v2:
 - Add XEN_ prefix to pvclock flags.
 - Adapter time_calibration_rendezvous_tail to have the case of setting master
 tsc/stime and use it for the nop_rendezvous.
 - Removed hotplug CPU option that was added in v1
 - Prevent online of CPUs when clocksource is tsc.
 - Remove use_tsc_stable_bit, since clocksource is only used to seed
 values. So instead we test if hotplug is possible, and prevent clocksource=tsc
 to be used.
 - Remove 1st paragrah of commit message since the behaviour described
   no longer applies since b64438c.

Changes since v1:
 - Change approach to skip std_rendezvous by introducing a
   nop_rendezvous
 - Change commit message reflecting the change above.
 - Use TSC_STABLE_BIT only if cpu hotplug isn't possible.
 - Add command line option to override it if no cpu hotplug is
 intended.
---
 xen/arch/x86/time.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index 128e653..d307d93 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -955,6 +955,14 @@ static void __update_vcpu_system_time(struct vcpu *v, int force)
     _u.tsc_timestamp = tsc_stamp;
     _u.system_time   = t->stamp.local_stime;
 
+    /*
+     * It's expected that domains cope with this bit changing on every
+     * pvclock read to check whether they can resort solely on this tuple
+     * or if it further requires monotonicity checks with other vcpus.
+     */
+    if ( clocksource_is_tsc() )
+        _u.flags |= XEN_PVCLOCK_TSC_STABLE_BIT;
+
     if ( is_hvm_domain(d) )
         _u.tsc_timestamp += v->arch.hvm_vcpu.cache_tsc_offset;
 
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

[PATCH v5 5/5] x86/time: extend "tsc" param with "stable:socket"

[Joao Martins]

Extend the "tsc" boot parameter is to further relax TSC restrictions and
allow it to be used on machines that guarantee reliable TSC across
sockets. This is up to board manufacturers and there's no way for the OS
to probe this property, therefore user needs to explicitly set this option.

Also make one style adjustment that is to remove the unnecessary
parenthesis around clearing TSC_RELIABLE.

Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
---
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>

Changes since v4:
 - Fix documentation spelling mistakes namely repeated use of
  "parameter"; matching option name ("stable:socket"); s/running
  under/running under/g; and wording from require to adjustment
  suggestion.
 - Use __initdata instead of __read_mostly
---
 docs/misc/xen-command-line.markdown |  6 ++++--
 xen/arch/x86/time.c                 | 11 ++++++++---
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/docs/misc/xen-command-line.markdown b/docs/misc/xen-command-line.markdown
index f92fb3f..962a18f 100644
--- a/docs/misc/xen-command-line.markdown
+++ b/docs/misc/xen-command-line.markdown
@@ -270,7 +270,9 @@ If set, override Xen's default choice for the platform timer.
 Having TSC as platform timer requires being explicitly set. This is because
 TSC can only be safely used if CPU hotplug isn't performed on the system. In
 some platforms, "maxcpus" parameter may require further adjustment to the
-number of online cpus.
+number of online cpus. When running on platforms that can guarantee a
+monotonic TSC across sockets you may want to adjust the "tsc" command line
+parameter to "stable:socket".
 
 ### cmci-threshold
 > `= <integer>`
@@ -1508,7 +1510,7 @@ pages) must also be specified via the tbuf\_size parameter.
 > `= <integer>`
 
 ### tsc
-> `= unstable | skewed`
+> `= unstable | skewed | stable:socket`
 
 ### ucode
 > `= [<integer> | scan]`
diff --git a/xen/arch/x86/time.c b/xen/arch/x86/time.c
index d307d93..dda89d8 100644
--- a/xen/arch/x86/time.c
+++ b/xen/arch/x86/time.c
@@ -477,6 +477,10 @@ uint64_t ns_to_acpi_pm_tick(uint64_t ns)
 /************************************************************
  * PLATFORM TIMER 4: TSC
  */
+static unsigned int __initdata tsc_flags;
+
+/* TSC is reliable across sockets */
+#define TSC_RELIABLE_SOCKET (1 << 0)
 
 /*
  * Called in verify_tsc_reliability() under reliable TSC conditions
@@ -492,7 +496,7 @@ static s64 __init init_tsc(struct platform_timesource *pts)
         ret = 0;
     }
 
-    if ( nr_sockets > 1 )
+    if ( nr_sockets > 1 && !(tsc_flags & TSC_RELIABLE_SOCKET) )
     {
         printk(XENLOG_WARNING "TSC: Not invariant across sockets\n");
         ret = 0;
@@ -1855,6 +1859,7 @@ int hwdom_pit_access(struct ioreq *ioreq)
 /*
  * tsc=unstable: Override all tests; assume TSC is unreliable.
  * tsc=skewed: Assume TSCs are individually reliable, but skewed across CPUs.
+ * tsc=stable:socket: Assume TSCs are reliable across sockets.
  */
 static void __init tsc_parse(const char *s)
 {
@@ -1865,9 +1870,9 @@ static void __init tsc_parse(const char *s)
         setup_clear_cpu_cap(X86_FEATURE_TSC_RELIABLE);
     }
     else if ( !strcmp(s, "skewed") )
-    {
         setup_clear_cpu_cap(X86_FEATURE_TSC_RELIABLE);
-    }
+    else if ( !strcmp(s, "stable:socket") )
+        tsc_flags |= TSC_RELIABLE_SOCKET;
 }
 custom_param("tsc", tsc_parse);
 
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [PATCH v5 3/5] x86/time: implement tsc as clocksource

[Jan Beulich]

>>> On 23.09.16 at 12:42, <joao.m.martins@oracle.com> wrote:
> Recent x86/time changes improved a lot of the monotonicity in xen
> timekeeping, making it much harder to observe time going backwards.
> Although platform timer can't be expected to be perfectly in sync with
> TSC and so get_s_time won't be guaranteed to always return
> monotonically increasing values across cpus. This is the case in some
> of the boxes I am testing with, observing sometimes ~100 warps (of
> very few nanoseconds each) after a few hours.
> 
> This patch introduces support for using TSC as platform time source
> which is the highest resolution time and most performant to get.
> Though there are also several problems associated with its usage, and
> there isn't a complete (and architecturally defined) guarantee that
> all machines will provide reliable and monotonic TSC in all cases (I
> believe Intel to be the only that can guarantee that?). For this reason
> it's not used unless administrator changes "clocksource" boot option
> to "tsc". Initializing TSC clocksource requires all CPUs up to have
> the tsc reliability checks performed. init_xen_time is called before
> all CPUs are up, so for example we would start with HPET (or ACPI,
> PIT) at boot time, and switch later to TSC. The switch then happens on
> verify_tsc_reliability initcall that is invoked when all CPUs are up.
> When attempting to initialize TSC we also check for time warps and if
> it has invariant TSC. Note that while we deem reliable a CONSTANT_TSC
> with no deep C-states, it might not always be the case, so we're
> conservative and allow TSC to be used as platform timer only with
> invariant TSC. Additionally we check if CPU Hotplug isn't meant to be
> performed on the host which will either be when max vcpus and
> num_present_cpu are the same. This is because a newly hotplugged CPU
> may not satisfy the condition of having all TSCs synchronized - so
> when having tsc clocksource being used we allow offlining CPUs but not
> onlining any ones back. Finally we prevent TSC from being used as
> clocksource on multiple sockets because it isn't guaranteed to be
> invariant. Further relaxing of this last requirement is added in a
> separate patch, such that we allow vendors with such guarantee to use
> TSC as clocksource. In case any of these conditions is not met, we
> keep the clocksource that was previously initialized on init_xen_time.
> 
> Since b64438c7c ("x86/time: use correct (local) time stamp in
> constant-TSC calibration fast path") updates to cpu time use local
> stamps, which means platform timer is only used to seed the initial
> cpu time. We further introduce a new rendezvous function
> (nop_rendezvous) which doesn't require synchronization between master
> and slave CPUS and just reads calibration_rendezvous struct and writes
> it down the stime and stamp to the cpu_calibration struct to be used
> later on. With clocksource=tsc there is no need to be in sync with
> another clocksource, so we reseed the local/master stamps to be values
> of TSC and update the platform time stamps accordingly. Time
> calibration is set to 1sec after we switch to TSC, thus these stamps
> are reseeded to also ensure monotonic returning values right after the
> point we switch to TSC. This is to remove the possibility of having
> inconsistent readings in this short period (i.e. until calibration
> fires).
> 
> Signed-off-by: Joao Martins <joao.m.martins@oracle.com>

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

with one further adjustment (no need to re-send):

> --- a/docs/misc/xen-command-line.markdown
> +++ b/docs/misc/xen-command-line.markdown
> @@ -264,9 +264,13 @@ minimum of 32M, subject to a suitably aligned and sized contiguous
>  region of memory being available.
>  
>  ### clocksource
> -> `= pit | hpet | acpi`
> +> `= pit | hpet | acpi | tsc`
>  
>  If set, override Xen's default choice for the platform timer.
> +Having TSC as platform timer requires being explicitly set. This is because
> +TSC can only be safely used if CPU hotplug isn't performed on the system. In
> +some platforms, "maxcpus" parameter may require further adjustment to the
> +number of online cpus.

I think I'll re-write this last sentence to "On some platforms,
the "maxcpus" option may need to be used to further adjust the
number of allowed CPUs."

Jan

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel

Re: [PATCH v5 5/5] x86/time: extend "tsc" param with "stable:socket"

[Jan Beulich]

>>> On 23.09.16 at 12:42, <joao.m.martins@oracle.com> wrote:
> Extend the "tsc" boot parameter is to further relax TSC restrictions and
> allow it to be used on machines that guarantee reliable TSC across
> sockets. This is up to board manufacturers and there's no way for the OS
> to probe this property, therefore user needs to explicitly set this option.
> 
> Also make one style adjustment that is to remove the unnecessary
> parenthesis around clearing TSC_RELIABLE.
> 
> Signed-off-by: Joao Martins <joao.m.martins@oracle.com>

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


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel