[PATCH v3 0/6] Support calling functions on dedicated physical cpu

[PATCH v3 0/6] Support calling functions on dedicated physical cpu

[Juergen Gross]

Some hardware (e.g. Dell Studio laptops) require special functions to
be called on physical cpu 0 in order to avoid occasional hangs. When
running as dom0 under Xen this could be achieved only via special boot
parameters (vcpu pinning) limiting the hypervisor in it's scheduling
decisions.

This patch series is adding a generic function to be able to temporarily
pin a (virtual) cpu to a dedicated physical cpu for executing above
mentioned functions on that specific cpu. The drivers (dcdbas and i8k)
requiring this functionality are modified accordingly.

Changes in V3:
- use get_cpu()/put_cpu() as suggested by David Vrabel

Changes in V2:
- instead of manipulating the allowed set of cpus use cpu specific
  workqueue as requested by Peter Zijlstra
- add include/linux/hypervisor.h to hide architecture specific stuff
  from generic kernel code

Juergen Gross (6):
  xen: sync xen header
  smp: add function to execute a function synchronously on a physical
    cpu
  dcdbas: make use of smp_call_sync_on_phys_cpu()
  hwmon: use smp_call_sync_on_phys_cpu() for dell-smm i8k
  virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()
  xen: add xen_pin_vcpu() to support calling functions on a dedicated
    pcpu

 MAINTAINERS                       |   1 +
 arch/x86/include/asm/hypervisor.h |   9 ++++
 arch/x86/xen/enlighten.c          |  40 +++++++++++++++
 drivers/firmware/dcdbas.c         |  46 ++++++++----------
 drivers/hwmon/dell-smm-hwmon.c    |  27 +++++-----
 include/linux/hypervisor.h        |  17 +++++++
 include/linux/smp.h               |   2 +
 include/xen/interface/sched.h     | 100 +++++++++++++++++++++++++++++++-------
 kernel/smp.c                      |  50 +++++++++++++++++++
 kernel/up.c                       |  18 +++++++
 10 files changed, 251 insertions(+), 59 deletions(-)
 create mode 100644 include/linux/hypervisor.h

-- 
2.6.2

[PATCH v3 1/6] xen: sync xen header

[Juergen Gross]

Import the actual version of include/xen/interface/sched.h from Xen.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
 include/xen/interface/sched.h | 100 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 82 insertions(+), 18 deletions(-)

diff --git a/include/xen/interface/sched.h b/include/xen/interface/sched.h
index f184909..a4c4d73 100644
--- a/include/xen/interface/sched.h
+++ b/include/xen/interface/sched.h
@@ -3,6 +3,24 @@
  *
  * Scheduler state interactions
  *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ *
  * Copyright (c) 2005, Keir Fraser <keir@xensource.com>
  */
 
@@ -12,18 +30,30 @@
 #include <xen/interface/event_channel.h>
 
 /*
+ * Guest Scheduler Operations
+ *
+ * The SCHEDOP interface provides mechanisms for a guest to interact
+ * with the scheduler, including yield, blocking and shutting itself
+ * down.
+ */
+
+/*
  * The prototype for this hypercall is:
- *  long sched_op_new(int cmd, void *arg)
+ * long HYPERVISOR_sched_op(enum sched_op cmd, void *arg, ...)
+ *
  * @cmd == SCHEDOP_??? (scheduler operation).
  * @arg == Operation-specific extra argument(s), as described below.
+ * ...  == Additional Operation-specific extra arguments, described below.
  *
- * **NOTE**:
- * Versions of Xen prior to 3.0.2 provide only the following legacy version
+ * Versions of Xen prior to 3.0.2 provided only the following legacy version
  * of this hypercall, supporting only the commands yield, block and shutdown:
  *  long sched_op(int cmd, unsigned long arg)
  * @cmd == SCHEDOP_??? (scheduler operation).
  * @arg == 0               (SCHEDOP_yield and SCHEDOP_block)
  *      == SHUTDOWN_* code (SCHEDOP_shutdown)
+ *
+ * This legacy version is available to new guests as:
+ * long HYPERVISOR_sched_op_compat(enum sched_op cmd, unsigned long arg)
  */
 
 /*
@@ -44,12 +74,17 @@
 /*
  * Halt execution of this domain (all VCPUs) and notify the system controller.
  * @arg == pointer to sched_shutdown structure.
+ *
+ * If the sched_shutdown_t reason is SHUTDOWN_suspend then
+ * x86 PV guests must also set RDX (EDX for 32-bit guests) to the MFN
+ * of the guest's start info page.  RDX/EDX is the third hypercall
+ * argument.
+ *
+ * In addition, which reason is SHUTDOWN_suspend this hypercall
+ * returns 1 if suspend was cancelled or the domain was merely
+ * checkpointed, and 0 if it is resuming in a new domain.
  */
 #define SCHEDOP_shutdown    2
-struct sched_shutdown {
-    unsigned int reason; /* SHUTDOWN_* */
-};
-DEFINE_GUEST_HANDLE_STRUCT(sched_shutdown);
 
 /*
  * Poll a set of event-channel ports. Return when one or more are pending. An
@@ -57,12 +92,6 @@ DEFINE_GUEST_HANDLE_STRUCT(sched_shutdown);
  * @arg == pointer to sched_poll structure.
  */
 #define SCHEDOP_poll        3
-struct sched_poll {
-    GUEST_HANDLE(evtchn_port_t) ports;
-    unsigned int nr_ports;
-    uint64_t timeout;
-};
-DEFINE_GUEST_HANDLE_STRUCT(sched_poll);
 
 /*
  * Declare a shutdown for another domain. The main use of this function is
@@ -71,15 +100,11 @@ DEFINE_GUEST_HANDLE_STRUCT(sched_poll);
  * @arg == pointer to sched_remote_shutdown structure.
  */
 #define SCHEDOP_remote_shutdown        4
-struct sched_remote_shutdown {
-    domid_t domain_id;         /* Remote domain ID */
-    unsigned int reason;       /* SHUTDOWN_xxx reason */
-};
 
 /*
  * Latch a shutdown code, so that when the domain later shuts down it
  * reports this code to the control tools.
- * @arg == as for SCHEDOP_shutdown.
+ * @arg == sched_shutdown, as for SCHEDOP_shutdown.
  */
 #define SCHEDOP_shutdown_code 5
 
@@ -92,10 +117,47 @@ struct sched_remote_shutdown {
  * With id != 0 and timeout != 0, poke watchdog timer and set new timeout.
  */
 #define SCHEDOP_watchdog    6
+
+/*
+ * Override the current vcpu affinity by pinning it to one physical cpu or
+ * undo this override restoring the previous affinity.
+ * @arg == pointer to sched_pin_override structure.
+ *
+ * A negative pcpu value will undo a previous pin override and restore the
+ * previous cpu affinity.
+ * This call is allowed for the hardware domain only and requires the cpu
+ * to be part of the domain's cpupool.
+ */
+#define SCHEDOP_pin_override 7
+
+struct sched_shutdown {
+    unsigned int reason; /* SHUTDOWN_* => shutdown reason */
+};
+DEFINE_GUEST_HANDLE_STRUCT(sched_shutdown);
+
+struct sched_poll {
+    GUEST_HANDLE(evtchn_port_t) ports;
+    unsigned int nr_ports;
+    uint64_t timeout;
+};
+DEFINE_GUEST_HANDLE_STRUCT(sched_poll);
+
+struct sched_remote_shutdown {
+    domid_t domain_id;         /* Remote domain ID */
+    unsigned int reason;       /* SHUTDOWN_* => shutdown reason */
+};
+DEFINE_GUEST_HANDLE_STRUCT(sched_remote_shutdown);
+
 struct sched_watchdog {
     uint32_t id;                /* watchdog ID */
     uint32_t timeout;           /* timeout */
 };
+DEFINE_GUEST_HANDLE_STRUCT(sched_watchdog);
+
+struct sched_pin_override {
+    int32_t pcpu;
+};
+DEFINE_GUEST_HANDLE_STRUCT(sched_pin_override);
 
 /*
  * Reason codes for SCHEDOP_shutdown. These may be interpreted by control
@@ -107,6 +169,7 @@ struct sched_watchdog {
 #define SHUTDOWN_suspend    2  /* Clean up, save suspend info, kill.         */
 #define SHUTDOWN_crash      3  /* Tell controller we've crashed.             */
 #define SHUTDOWN_watchdog   4  /* Restart because watchdog time expired.     */
+
 /*
  * Domain asked to perform 'soft reset' for it. The expected behavior is to
  * reset internal Xen state for the domain returning it to the point where it
@@ -115,5 +178,6 @@ struct sched_watchdog {
  * interfaces again.
  */
 #define SHUTDOWN_soft_reset 5
+#define SHUTDOWN_MAX        5  /* Maximum valid shutdown reason.             */
 
 #endif /* __XEN_PUBLIC_SCHED_H__ */
-- 
2.6.2

[PATCH v3 2/6] smp: add function to execute a function synchronously on a physical cpu

[Juergen Gross]

On some hardware models (e.g. Dell Studio 1555 laptop) some hardware
related functions (e.g. SMIs) are to be executed on physical cpu 0
only. Instead of open coding such a functionality multiple times in
the kernel add a service function for this purpose. This will enable
the possibility to take special measures in virtualized environments
like Xen, too.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
V2: instead of manipulating the allowed set of cpus use cpu specific
    workqueue as requested by Peter Zijlstra
---
 include/linux/smp.h |  2 ++
 kernel/smp.c        | 44 ++++++++++++++++++++++++++++++++++++++++++++
 kernel/up.c         |  9 +++++++++
 3 files changed, 55 insertions(+)

diff --git a/include/linux/smp.h b/include/linux/smp.h
index c441407..fc9d21b 100644
--- a/include/linux/smp.h
+++ b/include/linux/smp.h
@@ -196,4 +196,6 @@ extern void arch_enable_nonboot_cpus_end(void);
 
 void smp_setup_processor_id(void);
 
+int smp_call_sync_on_phys_cpu(unsigned int cpu, int (*func)(void *), void *par);
+
 #endif /* __LINUX_SMP_H */
diff --git a/kernel/smp.c b/kernel/smp.c
index 7416544..62da74b 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -739,3 +739,47 @@ void wake_up_all_idle_cpus(void)
 	preempt_enable();
 }
 EXPORT_SYMBOL_GPL(wake_up_all_idle_cpus);
+
+/**
+ * smp_call_sync_on_phys_cpu - Call a function on a specific physical cpu
+ *
+ * Used to call a function on a specific physical cpu. Even if the specified
+ * cpu isn't the current one, return only after the called function has
+ * returned.
+ */
+struct smp_sync_call_struct {
+	struct work_struct	work;
+	struct completion	done;
+	int			(*func)(void *);
+	void			*data;
+	int			ret;
+};
+
+static void smp_call_sync_callback(struct work_struct *work)
+{
+	struct smp_sync_call_struct *sscs;
+
+	sscs = container_of(work, struct smp_sync_call_struct, work);
+	sscs->ret = sscs->func(sscs->data);
+
+	complete(&sscs->done);
+}
+
+int smp_call_sync_on_phys_cpu(unsigned int cpu, int (*func)(void *), void *par)
+{
+	struct smp_sync_call_struct sscs = {
+		.work = __WORK_INITIALIZER(sscs.work, smp_call_sync_callback),
+		.done = COMPLETION_INITIALIZER_ONSTACK(sscs.done),
+		.func = func,
+		.data = par,
+	};
+
+	if (cpu >= nr_cpu_ids)
+		return -EINVAL;
+
+	queue_work_on(cpu, system_wq, &sscs.work);
+	wait_for_completion(&sscs.done);
+
+	return sscs.ret;
+}
+EXPORT_SYMBOL_GPL(smp_call_sync_on_phys_cpu);
diff --git a/kernel/up.c b/kernel/up.c
index 1760bf3..afd395c 100644
--- a/kernel/up.c
+++ b/kernel/up.c
@@ -82,3 +82,12 @@ void on_each_cpu_cond(bool (*cond_func)(int cpu, void *info),
 	preempt_enable();
 }
 EXPORT_SYMBOL(on_each_cpu_cond);
+
+int smp_call_sync_on_phys_cpu(unsigned int cpu, int (*func)(void *), void *par)
+{
+	if (cpu != 0)
+		return -EINVAL;
+
+	return func(par);
+}
+EXPORT_SYMBOL_GPL(smp_call_sync_on_phys_cpu);
-- 
2.6.2

Re: [PATCH v3 2/6] smp: add function to execute a function synchronously on a physical cpu

[Peter Zijlstra]

On Fri, Apr 01, 2016 at 09:14:30AM +0200, Juergen Gross wrote:
> +	if (cpu >= nr_cpu_ids)
> +		return -EINVAL;

> +	if (cpu != 0)
> +		return -EINVAL;

The other functions return -ENXIO for this.

Re: [PATCH v3 2/6] smp: add function to execute a function synchronously on a physical cpu

[Juergen Gross]

On 01/04/16 09:37, Peter Zijlstra wrote:
> On Fri, Apr 01, 2016 at 09:14:30AM +0200, Juergen Gross wrote:
>> +	if (cpu >= nr_cpu_ids)
>> +		return -EINVAL;
> 
>> +	if (cpu != 0)
>> +		return -EINVAL;
> 
> The other functions return -ENXIO for this.

Aah, okay. Will change.


Juergen

[PATCH v3 3/6] dcdbas: make use of smp_call_sync_on_phys_cpu()

[Juergen Gross]

Use smp_call_sync_on_phys_cpu() to raise SMI on cpu 0.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
 drivers/firmware/dcdbas.c | 46 ++++++++++++++++++++--------------------------
 1 file changed, 20 insertions(+), 26 deletions(-)

diff --git a/drivers/firmware/dcdbas.c b/drivers/firmware/dcdbas.c
index 829eec8..d15ad0b 100644
--- a/drivers/firmware/dcdbas.c
+++ b/drivers/firmware/dcdbas.c
@@ -238,33 +238,14 @@ static ssize_t host_control_on_shutdown_store(struct device *dev,
 	return count;
 }
 
-/**
- * dcdbas_smi_request: generate SMI request
- *
- * Called with smi_data_lock.
- */
-int dcdbas_smi_request(struct smi_cmd *smi_cmd)
+static int raise_smi(void *par)
 {
-	cpumask_var_t old_mask;
-	int ret = 0;
+	struct smi_cmd *smi_cmd = par;
 
-	if (smi_cmd->magic != SMI_CMD_MAGIC) {
-		dev_info(&dcdbas_pdev->dev, "%s: invalid magic value\n",
-			 __func__);
-		return -EBADR;
-	}
-
-	/* SMI requires CPU 0 */
-	if (!alloc_cpumask_var(&old_mask, GFP_KERNEL))
-		return -ENOMEM;
-
-	cpumask_copy(old_mask, &current->cpus_allowed);
-	set_cpus_allowed_ptr(current, cpumask_of(0));
 	if (smp_processor_id() != 0) {
 		dev_dbg(&dcdbas_pdev->dev, "%s: failed to get CPU 0\n",
 			__func__);
-		ret = -EBUSY;
-		goto out;
+		return -EBUSY;
 	}
 
 	/* generate SMI */
@@ -280,10 +261,23 @@ int dcdbas_smi_request(struct smi_cmd *smi_cmd)
 		: "memory"
 	);
 
-out:
-	set_cpus_allowed_ptr(current, old_mask);
-	free_cpumask_var(old_mask);
-	return ret;
+	return 0;
+}
+/**
+ * dcdbas_smi_request: generate SMI request
+ *
+ * Called with smi_data_lock.
+ */
+int dcdbas_smi_request(struct smi_cmd *smi_cmd)
+{
+	if (smi_cmd->magic != SMI_CMD_MAGIC) {
+		dev_info(&dcdbas_pdev->dev, "%s: invalid magic value\n",
+			 __func__);
+		return -EBADR;
+	}
+
+	/* SMI requires CPU 0 */
+	return smp_call_sync_on_phys_cpu(0, raise_smi, smi_cmd);
 }
 
 /**
-- 
2.6.2

[PATCH v3 4/6] hwmon: use smp_call_sync_on_phys_cpu() for dell-smm i8k

[Juergen Gross]

Use the smp_call_sync_on_phys_cpu() function to call system management
mode on cpu 0.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
 drivers/hwmon/dell-smm-hwmon.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)

diff --git a/drivers/hwmon/dell-smm-hwmon.c b/drivers/hwmon/dell-smm-hwmon.c
index c43318d..4875462 100644
--- a/drivers/hwmon/dell-smm-hwmon.c
+++ b/drivers/hwmon/dell-smm-hwmon.c
@@ -130,23 +130,15 @@ static inline const char *i8k_get_dmi_data(int field)
 /*
  * Call the System Management Mode BIOS. Code provided by Jonathan Buzzard.
  */
-static int i8k_smm(struct smm_regs *regs)
+static int i8k_smm_func(void *par)
 {
 	int rc;
+	struct smm_regs *regs = par;
 	int eax = regs->eax;
-	cpumask_var_t old_mask;
 
 	/* SMM requires CPU 0 */
-	if (!alloc_cpumask_var(&old_mask, GFP_KERNEL))
-		return -ENOMEM;
-	cpumask_copy(old_mask, &current->cpus_allowed);
-	rc = set_cpus_allowed_ptr(current, cpumask_of(0));
-	if (rc)
-		goto out;
-	if (smp_processor_id() != 0) {
-		rc = -EBUSY;
-		goto out;
-	}
+	if (smp_processor_id() != 0)
+		return -EBUSY;
 
 #if defined(CONFIG_X86_64)
 	asm volatile("pushq %%rax\n\t"
@@ -204,13 +196,18 @@ static int i8k_smm(struct smm_regs *regs)
 	if (rc != 0 || (regs->eax & 0xffff) == 0xffff || regs->eax == eax)
 		rc = -EINVAL;
 
-out:
-	set_cpus_allowed_ptr(current, old_mask);
-	free_cpumask_var(old_mask);
 	return rc;
 }
 
 /*
+ * Call the System Management Mode BIOS.
+ */
+static int i8k_smm(struct smm_regs *regs)
+{
+	return smp_call_sync_on_phys_cpu(0, i8k_smm_func, regs);
+}
+
+/*
  * Read the fan status.
  */
 static int i8k_get_fan_status(int fan)
-- 
2.6.2

[PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Juergen Gross]

Add generic virtualization support for pinning the current vcpu to a
specified physical cpu. As this operation isn't performance critical
(a very limited set of operations like BIOS calls and SMIs is expected
to need this) just add a hypervisor specific indirection.

Such a pinning should last as short as possible as it might block
sensible vcpu scheduling and maybe other hypervisor functions like
suspending the system which rely on scheduling. To ensure this don't
let the current thread be preempted while the vcpu is pinned in
smp_call_sync_on_phys_cpu().

Signed-off-by: Juergen Gross <jgross@suse.com>
---
V3: use getc_cpu()/put_cpu() as suggested by David Vrabel

V2: adapt to using workqueues
    add include/linux/hypervisor.h to hide architecture specific stuff
    from generic kernel code

In case paravirt maintainers don't want to be responsible for
include/linux/hypervisor.h I could take it.
---
 MAINTAINERS                       |  1 +
 arch/x86/include/asm/hypervisor.h |  9 +++++++++
 include/linux/hypervisor.h        | 17 +++++++++++++++++
 kernel/smp.c                      |  6 ++++++
 kernel/up.c                       | 11 ++++++++++-
 5 files changed, 43 insertions(+), 1 deletion(-)
 create mode 100644 include/linux/hypervisor.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 378ebff..3571dff 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -8329,6 +8329,7 @@ S:	Supported
 F:	Documentation/virtual/paravirt_ops.txt
 F:	arch/*/kernel/paravirt*
 F:	arch/*/include/asm/paravirt.h
+F:	include/linux/hypervisor.h
 
 PARIDE DRIVERS FOR PARALLEL PORT IDE DEVICES
 M:	Tim Waugh <tim@cyberelk.net>
diff --git a/arch/x86/include/asm/hypervisor.h b/arch/x86/include/asm/hypervisor.h
index 055ea99..13f80a2 100644
--- a/arch/x86/include/asm/hypervisor.h
+++ b/arch/x86/include/asm/hypervisor.h
@@ -43,6 +43,9 @@ struct hypervisor_x86 {
 
 	/* X2APIC detection (run once per boot) */
 	bool		(*x2apic_available)(void);
+
+	/* pin current vcpu to specified physical cpu (run rarely) */
+	void		(*pin_vcpu)(int);
 };
 
 extern const struct hypervisor_x86 *x86_hyper;
@@ -56,6 +59,12 @@ extern const struct hypervisor_x86 x86_hyper_kvm;
 extern void init_hypervisor(struct cpuinfo_x86 *c);
 extern void init_hypervisor_platform(void);
 extern bool hypervisor_x2apic_available(void);
+
+static inline void hypervisor_pin_vcpu(int cpu)
+{
+	if (x86_hyper->pin_vcpu)
+		x86_hyper->pin_vcpu(cpu);
+}
 #else
 static inline void init_hypervisor(struct cpuinfo_x86 *c) { }
 static inline void init_hypervisor_platform(void) { }
diff --git a/include/linux/hypervisor.h b/include/linux/hypervisor.h
new file mode 100644
index 0000000..3fa5ef2
--- /dev/null
+++ b/include/linux/hypervisor.h
@@ -0,0 +1,17 @@
+#ifndef __LINUX_HYPEVISOR_H
+#define __LINUX_HYPEVISOR_H
+
+/*
+ *	Generic Hypervisor support
+ *		Juergen Gross <jgross@suse.com>
+ */
+
+#ifdef CONFIG_HYPERVISOR_GUEST
+#include <asm/hypervisor.h>
+#else
+static inline void hypervisor_pin_vcpu(int cpu)
+{
+}
+#endif
+
+#endif /* __LINUX_HYPEVISOR_H */
diff --git a/kernel/smp.c b/kernel/smp.c
index 62da74b..f9cfb7d 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -14,6 +14,7 @@
 #include <linux/smp.h>
 #include <linux/cpu.h>
 #include <linux/sched.h>
+#include <linux/hypervisor.h>
 
 #include "smpboot.h"
 
@@ -758,9 +759,14 @@ struct smp_sync_call_struct {
 static void smp_call_sync_callback(struct work_struct *work)
 {
 	struct smp_sync_call_struct *sscs;
+	unsigned int cpu;
 
 	sscs = container_of(work, struct smp_sync_call_struct, work);
+	cpu = get_cpu();
+	hypervisor_pin_vcpu(cpu);
 	sscs->ret = sscs->func(sscs->data);
+	hypervisor_pin_vcpu(-1);
+	put_cpu();
 
 	complete(&sscs->done);
 }
diff --git a/kernel/up.c b/kernel/up.c
index afd395c..725ec44 100644
--- a/kernel/up.c
+++ b/kernel/up.c
@@ -6,6 +6,7 @@
 #include <linux/kernel.h>
 #include <linux/export.h>
 #include <linux/smp.h>
+#include <linux/hypervisor.h>
 
 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 				int wait)
@@ -85,9 +86,17 @@ EXPORT_SYMBOL(on_each_cpu_cond);
 
 int smp_call_sync_on_phys_cpu(unsigned int cpu, int (*func)(void *), void *par)
 {
+	int ret;
+
 	if (cpu != 0)
 		return -EINVAL;
 
-	return func(par);
+	preempt_disable();
+	hypervisor_pin_vcpu(0);
+	ret = func(par);
+	hypervisor_pin_vcpu(-1);
+	preempt_enable();
+
+	return ret;
 }
 EXPORT_SYMBOL_GPL(smp_call_sync_on_phys_cpu);
-- 
2.6.2

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Peter Zijlstra]

On Fri, Apr 01, 2016 at 09:14:33AM +0200, Juergen Gross wrote:
> --- a/kernel/smp.c
> +++ b/kernel/smp.c
> @@ -14,6 +14,7 @@
>  #include <linux/smp.h>
>  #include <linux/cpu.h>
>  #include <linux/sched.h>
> +#include <linux/hypervisor.h>
>  
>  #include "smpboot.h"
>  
> @@ -758,9 +759,14 @@ struct smp_sync_call_struct {
>  static void smp_call_sync_callback(struct work_struct *work)
>  {
>  	struct smp_sync_call_struct *sscs;
> +	unsigned int cpu;
>  
>  	sscs = container_of(work, struct smp_sync_call_struct, work);
> +	cpu = get_cpu();
> +	hypervisor_pin_vcpu(cpu);
>  	sscs->ret = sscs->func(sscs->data);
> +	hypervisor_pin_vcpu(-1);
> +	put_cpu();
>  
>  	complete(&sscs->done);
>  }

So I don't really like this; it adds the requirement that the function
cannot schedule, which greatly limits the utility of the construct. At
this point you might as well use the regular IPI stuff.

You can easily avoid this constraint by using:

	hypervisor_pin_vcpu(smp_processor_id());

Also, for the vpinning stuff, the UP version below is sufficient, even
on SMP systems (with the current !preempt constraint). Which seems to
suggest we're not having the right interface for this.

So I would propose you add:

	smp_call_on_cpu()

As per patch 2. No promises about physical or anything. This means it
can be used freely by anyone that wants to run a function on another
cpu -- a much more useful thing.

And then build a phys variant on top.


> diff --git a/kernel/up.c b/kernel/up.c
> index afd395c..725ec44 100644
> --- a/kernel/up.c
> +++ b/kernel/up.c
> @@ -6,6 +6,7 @@
>  #include <linux/kernel.h>
>  #include <linux/export.h>
>  #include <linux/smp.h>
> +#include <linux/hypervisor.h>
>  
>  int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
>  				int wait)
> @@ -85,9 +86,17 @@ EXPORT_SYMBOL(on_each_cpu_cond);
>  
>  int smp_call_sync_on_phys_cpu(unsigned int cpu, int (*func)(void *), void *par)
>  {
> +	int ret;
> +
>  	if (cpu != 0)
>  		return -EINVAL;
>  
> -	return func(par);
> +	preempt_disable();
> +	hypervisor_pin_vcpu(0);
> +	ret = func(par);
> +	hypervisor_pin_vcpu(-1);
> +	preempt_enable();
> +
> +	return ret;
>  }
>  EXPORT_SYMBOL_GPL(smp_call_sync_on_phys_cpu);
> -- 
> 2.6.2
> 

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Juergen Gross]

On 01/04/16 09:43, Peter Zijlstra wrote:
> On Fri, Apr 01, 2016 at 09:14:33AM +0200, Juergen Gross wrote:
>> --- a/kernel/smp.c
>> +++ b/kernel/smp.c
>> @@ -14,6 +14,7 @@
>>  #include <linux/smp.h>
>>  #include <linux/cpu.h>
>>  #include <linux/sched.h>
>> +#include <linux/hypervisor.h>
>>  
>>  #include "smpboot.h"
>>  
>> @@ -758,9 +759,14 @@ struct smp_sync_call_struct {
>>  static void smp_call_sync_callback(struct work_struct *work)
>>  {
>>  	struct smp_sync_call_struct *sscs;
>> +	unsigned int cpu;
>>  
>>  	sscs = container_of(work, struct smp_sync_call_struct, work);
>> +	cpu = get_cpu();
>> +	hypervisor_pin_vcpu(cpu);
>>  	sscs->ret = sscs->func(sscs->data);
>> +	hypervisor_pin_vcpu(-1);
>> +	put_cpu();
>>  
>>  	complete(&sscs->done);
>>  }
> 
> So I don't really like this; it adds the requirement that the function
> cannot schedule, which greatly limits the utility of the construct. At
> this point you might as well use the regular IPI stuff.

Main reason for disabling preemption was to avoid any suspend/resume
cycles while vcpu pinning is active.

With the switch to workqueues this might not be necessary, if I've read
try_to_freeze_tasks() correctly. Can you confirm, please?

> You can easily avoid this constraint by using:
> 
> 	hypervisor_pin_vcpu(smp_processor_id());
> 
> Also, for the vpinning stuff, the UP version below is sufficient, even
> on SMP systems (with the current !preempt constraint). Which seems to
> suggest we're not having the right interface for this.
> 
> So I would propose you add:
> 
> 	smp_call_on_cpu()
> 
> As per patch 2. No promises about physical or anything. This means it
> can be used freely by anyone that wants to run a function on another
> cpu -- a much more useful thing.

Okay.

> And then build a phys variant on top.

Hmm, I'm not sure I understand what you are suggesting here.

Should this phys variant make use of smp_call_on_cpu() via an
intermediate function called on the dedicated cpu which is doing the
pinning and calling the user function then?

Or do you want the phys variant to either use smp_call_on_cpu() or to
do the pinning and call the user function by itself depending on the
environment (pinning supported)?


Juergen

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Peter Zijlstra]

On Fri, Apr 01, 2016 at 10:28:46AM +0200, Juergen Gross wrote:
> On 01/04/16 09:43, Peter Zijlstra wrote:
> > On Fri, Apr 01, 2016 at 09:14:33AM +0200, Juergen Gross wrote:
> >> --- a/kernel/smp.c
> >> +++ b/kernel/smp.c
> >> @@ -14,6 +14,7 @@
> >>  #include <linux/smp.h>
> >>  #include <linux/cpu.h>
> >>  #include <linux/sched.h>
> >> +#include <linux/hypervisor.h>
> >>  
> >>  #include "smpboot.h"
> >>  
> >> @@ -758,9 +759,14 @@ struct smp_sync_call_struct {
> >>  static void smp_call_sync_callback(struct work_struct *work)
> >>  {
> >>  	struct smp_sync_call_struct *sscs;
> >> +	unsigned int cpu;
> >>  
> >>  	sscs = container_of(work, struct smp_sync_call_struct, work);
> >> +	cpu = get_cpu();
> >> +	hypervisor_pin_vcpu(cpu);
> >>  	sscs->ret = sscs->func(sscs->data);
> >> +	hypervisor_pin_vcpu(-1);
> >> +	put_cpu();
> >>  
> >>  	complete(&sscs->done);
> >>  }
> > 
> > So I don't really like this; it adds the requirement that the function
> > cannot schedule, which greatly limits the utility of the construct. At
> > this point you might as well use the regular IPI stuff.
> 
> Main reason for disabling preemption was to avoid any suspend/resume
> cycles while vcpu pinning is active.
> 
> With the switch to workqueues this might not be necessary, if I've read
> try_to_freeze_tasks() correctly. Can you confirm, please?

This is not something we should worry about; the caller should ensure
the CPU stays valid; typically I would expect a caller to do
get_online_cpus() before 'computing' what CPU to send the function to.

> > So I would propose you add:
> > 
> > 	smp_call_on_cpu()
> > 
> > As per patch 2. No promises about physical or anything. This means it
> > can be used freely by anyone that wants to run a function on another
> > cpu -- a much more useful thing.
> 
> Okay.
> 
> > And then build a phys variant on top.
> 
> Hmm, I'm not sure I understand what you are suggesting here.
> 
> Should this phys variant make use of smp_call_on_cpu() via an
> intermediate function called on the dedicated cpu which is doing the
> pinning and calling the user function then?
> 
> Or do you want the phys variant to either use smp_call_on_cpu() or to
> do the pinning and call the user function by itself depending on the
> environment (pinning supported)?

Yeah, uhmm.. not sure on the details; my brain is having a hard time
engaging this morning.

Maybe just make the vpin thing an option like:

	smp_call_on_cpu(int (*func)(void *), int phys_cpu);

Also; is something like the vpin thing possible on KVM? because if we're
going to expose it to generic code like this we had maybe look at wider
support.

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Juergen Gross]

On 01/04/16 10:44, Peter Zijlstra wrote:
> On Fri, Apr 01, 2016 at 10:28:46AM +0200, Juergen Gross wrote:
>> On 01/04/16 09:43, Peter Zijlstra wrote:
>>> On Fri, Apr 01, 2016 at 09:14:33AM +0200, Juergen Gross wrote:
>>>> --- a/kernel/smp.c
>>>> +++ b/kernel/smp.c
>>>> @@ -14,6 +14,7 @@
>>>>  #include <linux/smp.h>
>>>>  #include <linux/cpu.h>
>>>>  #include <linux/sched.h>
>>>> +#include <linux/hypervisor.h>
>>>>  
>>>>  #include "smpboot.h"
>>>>  
>>>> @@ -758,9 +759,14 @@ struct smp_sync_call_struct {
>>>>  static void smp_call_sync_callback(struct work_struct *work)
>>>>  {
>>>>  	struct smp_sync_call_struct *sscs;
>>>> +	unsigned int cpu;
>>>>  
>>>>  	sscs = container_of(work, struct smp_sync_call_struct, work);
>>>> +	cpu = get_cpu();
>>>> +	hypervisor_pin_vcpu(cpu);
>>>>  	sscs->ret = sscs->func(sscs->data);
>>>> +	hypervisor_pin_vcpu(-1);
>>>> +	put_cpu();
>>>>  
>>>>  	complete(&sscs->done);
>>>>  }
>>>
>>> So I don't really like this; it adds the requirement that the function
>>> cannot schedule, which greatly limits the utility of the construct. At
>>> this point you might as well use the regular IPI stuff.
>>
>> Main reason for disabling preemption was to avoid any suspend/resume
>> cycles while vcpu pinning is active.
>>
>> With the switch to workqueues this might not be necessary, if I've read
>> try_to_freeze_tasks() correctly. Can you confirm, please?
> 
> This is not something we should worry about; the caller should ensure
> the CPU stays valid; typically I would expect a caller to do
> get_online_cpus() before 'computing' what CPU to send the function to.

Okay.

> 
>>> So I would propose you add:
>>>
>>> 	smp_call_on_cpu()
>>>
>>> As per patch 2. No promises about physical or anything. This means it
>>> can be used freely by anyone that wants to run a function on another
>>> cpu -- a much more useful thing.
>>
>> Okay.
>>
>>> And then build a phys variant on top.
>>
>> Hmm, I'm not sure I understand what you are suggesting here.
>>
>> Should this phys variant make use of smp_call_on_cpu() via an
>> intermediate function called on the dedicated cpu which is doing the
>> pinning and calling the user function then?
>>
>> Or do you want the phys variant to either use smp_call_on_cpu() or to
>> do the pinning and call the user function by itself depending on the
>> environment (pinning supported)?
> 
> Yeah, uhmm.. not sure on the details; my brain is having a hard time
> engaging this morning.
> 
> Maybe just make the vpin thing an option like:
> 
> 	smp_call_on_cpu(int (*func)(void *), int phys_cpu);

Okay.

> Also; is something like the vpin thing possible on KVM? because if we're
> going to expose it to generic code like this we had maybe look at wider
> support.

It is necessary for dom0 under Xen. I don't think there is a need to do
this on KVM as a guest has no direct access to e.g. BIOS functions of
the real hardware and the host system needs no vcpu pinning. I'm not
sure about VMWare.

Juergen

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Peter Zijlstra]

On Fri, Apr 01, 2016 at 11:03:21AM +0200, Juergen Gross wrote:
> > Maybe just make the vpin thing an option like:
> > 
> > 	smp_call_on_cpu(int (*func)(void *), int phys_cpu);

> > Also; is something like the vpin thing possible on KVM? because if we're
> > going to expose it to generic code like this we had maybe look at wider
> > support.
> 
> It is necessary for dom0 under Xen. I don't think there is a need to do
> this on KVM as a guest has no direct access to e.g. BIOS functions of
> the real hardware and the host system needs no vcpu pinning. I'm not
> sure about VMWare.

OK, then can we WARN if .phys=1 and the platform doesn't support it?

Re: [PATCH v3 5/6] virt, sched: add cpu pinning to smp_call_sync_on_phys_cpu()

[Juergen Gross]

On 01/04/16 11:15, Peter Zijlstra wrote:
> On Fri, Apr 01, 2016 at 11:03:21AM +0200, Juergen Gross wrote:
>>> Maybe just make the vpin thing an option like:
>>>
>>> 	smp_call_on_cpu(int (*func)(void *), int phys_cpu);
> 
>>> Also; is something like the vpin thing possible on KVM? because if we're
>>> going to expose it to generic code like this we had maybe look at wider
>>> support.
>>
>> It is necessary for dom0 under Xen. I don't think there is a need to do
>> this on KVM as a guest has no direct access to e.g. BIOS functions of
>> the real hardware and the host system needs no vcpu pinning. I'm not
>> sure about VMWare.
> 
> OK, then can we WARN if .phys=1 and the platform doesn't support it?
> 

Yes, good idea.


Juergen

[PATCH v3 6/6] xen: add xen_pin_vcpu() to support calling functions on a dedicated pcpu

[Juergen Gross]

Some hardware models (e.g. Dell Studio 1555 laptops) require calls to
the firmware to be issued on cpu 0 only. As Dom0 might have to use
these calls, add xen_pin_vcpu() to achieve this functionality.

In case either the domain doesn't have the privilege to make the
related hypercall or the hypervisor isn't supporting it, issue a
warning once and disable further pinning attempts.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
 arch/x86/xen/enlighten.c | 40 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 40 insertions(+)

diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 880862c..7907bcf8 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -1885,6 +1885,45 @@ static void xen_set_cpu_features(struct cpuinfo_x86 *c)
 	}
 }
 
+static void xen_pin_vcpu(int cpu)
+{
+	static bool disable_pinning;
+	struct sched_pin_override pin_override;
+	int ret;
+
+	if (disable_pinning)
+		return;
+
+	pin_override.pcpu = cpu;
+	ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
+	if (cpu < 0)
+		return;
+
+	switch (ret) {
+	case -ENOSYS:
+		pr_warn("The kernel tried to call a function on physical cpu %d, but Xen isn't\n"
+			"supporting this. In case of problems you might consider vcpu pinning.\n",
+			cpu);
+		disable_pinning = true;
+		break;
+	case -EPERM:
+		WARN(1, "Trying to pin vcpu without having privilege to do so\n");
+		disable_pinning = true;
+		break;
+	case -EINVAL:
+	case -EBUSY:
+		pr_warn("The kernel tried to call a function on physical cpu %d, but this cpu\n"
+			"seems not to be available. Please check your Xen cpu configuration.\n",
+			cpu);
+		break;
+	case 0:
+		break;
+	default:
+		WARN(1, "rc %d while trying to pin vcpu\n", ret);
+		disable_pinning = true;
+	}
+}
+
 const struct hypervisor_x86 x86_hyper_xen = {
 	.name			= "Xen",
 	.detect			= xen_platform,
@@ -1893,6 +1932,7 @@ const struct hypervisor_x86 x86_hyper_xen = {
 #endif
 	.x2apic_available	= xen_x2apic_para_available,
 	.set_cpu_features       = xen_set_cpu_features,
+	.pin_vcpu               = xen_pin_vcpu,
 };
 EXPORT_SYMBOL(x86_hyper_xen);
 
-- 
2.6.2