Merge tag 'stable/for-linus-3.12-rc0-tag-two' of git://git.kernel.org/pub/scm/li…

…nux/kernel/git/xen/tip Pull Xen bug-fixes from Konrad Rzeszutek Wilk: "This pull I usually do after rc1 is out but because we have a nice amount of fixes, some bootup related fixes for ARM, and it is early in the cycle we figured to do it now to help with tracking of potential regressions. The simple ones are the ARM ones - one of the patches fell through the cracks, other fixes a bootup issue (unconditionally using Xen functions). Then a fix for a regression causing preempt count being off (patch causing this went in v3.12). Lastly are the fixes to make Xen PVHVM guests use PV ticketlocks (Xen PV already does). The enablement of that was supposed to be part of the x86 spinlock merge in commit 816434ec4a67 ("The biggest change here are paravirtualized ticket spinlocks (PV spinlocks), which bring a nice speedup on various benchmarks...") but unfortunatly it would cause hang when booting Xen PVHVM guests. Yours truly got all of the bugs fixed last week and they (six of them) are included in this pull. Bug-fixes: - Boot on ARM without using Xen unconditionally - On Xen ARM don't run cpuidle/cpufreq - Fix regression in balloon driver, preempt count warnings - Fixes to make PVHVM able to use pv ticketlock. - Revert Xen PVHVM disabling pv ticketlock (aka, re-enable pv ticketlocks)" * tag 'stable/for-linus-3.12-rc0-tag-two' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip: xen/spinlock: Don't use __initdate for xen_pv_spin Revert "xen/spinlock: Disable IRQ spinlock (PV) allocation on PVHVM" xen/spinlock: Don't setup xen spinlock IPI kicker if disabled. xen/smp: Update pv_lock_ops functions before alternative code starts under PVHVM xen/spinlock: We don't need the old structure anymore xen/spinlock: Fix locking path engaging too soon under PVHVM. xen/arm: disable cpuidle and cpufreq when linux is running as dom0 xen/p2m: Don't call get_balloon_scratch_page() twice, keep interrupts disabled for multicalls ARM: xen: only set pm function ptrs for Xen guests

Merge tag 'stable/for-linus-3.12-rc0-tag-two' of git://git.kernel.org/pub/scm/li…
…nux/kernel/git/xen/tip Pull Xen bug-fixes from Konrad Rzeszutek Wilk: "This pull I usually do after rc1 is out but because we have a nice amount of fixes, some bootup related fixes for ARM, and it is early in the cycle we figured to do it now to help with tracking of potential regressions. The simple ones are the ARM ones - one of the patches fell through the cracks, other fixes a bootup issue (unconditionally using Xen functions). Then a fix for a regression causing preempt count being off (patch causing this went in v3.12). Lastly are the fixes to make Xen PVHVM guests use PV ticketlocks (Xen PV already does). The enablement of that was supposed to be part of the x86 spinlock merge in commit 816434ec4a67 ("The biggest change here are paravirtualized ticket spinlocks (PV spinlocks), which bring a nice speedup on various benchmarks...") but unfortunatly it would cause hang when booting Xen PVHVM guests. Yours truly got all of the bugs fixed last week and they (six of them) are included in this pull. Bug-fixes: - Boot on ARM without using Xen unconditionally - On Xen ARM don't run cpuidle/cpufreq - Fix regression in balloon driver, preempt count warnings - Fixes to make PVHVM able to use pv ticketlock. - Revert Xen PVHVM disabling pv ticketlock (aka, re-enable pv ticketlocks)" * tag 'stable/for-linus-3.12-rc0-tag-two' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip: xen/spinlock: Don't use __initdate for xen_pv_spin Revert "xen/spinlock: Disable IRQ spinlock (PV) allocation on PVHVM" xen/spinlock: Don't setup xen spinlock IPI kicker if disabled. xen/smp: Update pv_lock_ops functions before alternative code starts under PVHVM xen/spinlock: We don't need the old structure anymore xen/spinlock: Fix locking path engaging too soon under PVHVM. xen/arm: disable cpuidle and cpufreq when linux is running as dom0 xen/p2m: Don't call get_balloon_scratch_page() twice, keep interrupts disabled for multicalls ARM: xen: only set pm function ptrs for Xen guests
Linus Torvalds
2 parents fa1586a7e4 c3b7cb1fd8
Showing 5 changed files Side-by-side Diff
arch/arm/xen/enlighten.c
arch/x86/xen/enlighten.c
arch/x86/xen/p2m.c
arch/x86/xen/smp.c
arch/x86/xen/spinlock.c
@@ -21,6 +21,8 @@
 #include <linux/of.h>
 #include <linux/of_irq.h>
 #include <linux/of_address.h>
+#include <linux/cpuidle.h>
+#include <linux/cpufreq.h>
  
 #include <linux/mm.h>
  
  
  
@@ -267,18 +269,28 @@
 	if (!xen_initial_domain())
 		xenbus_probe(NULL);
  
+	/*
+	 * Making sure board specific code will not set up ops for
+	 * cpu idle and cpu freq.
+	 */
+	disable_cpuidle();
+	disable_cpufreq();
+
 	return 0;
 }
 core_initcall(xen_guest_init);
  
 static int __init xen_pm_init(void)
 {
+	if (!xen_domain())
+		return -ENODEV;
+
 	pm_power_off = xen_power_off;
 	arm_pm_restart = xen_restart;
  
 	return 0;
 }
-subsys_initcall(xen_pm_init);
+late_initcall(xen_pm_init);
  
 static irqreturn_t xen_arm_callback(int irq, void *arg)
 {
@@ -1692,7 +1692,6 @@
 	case CPU_UP_PREPARE:
 		xen_vcpu_setup(cpu);
 		if (xen_have_vector_callback) {
-			xen_init_lock_cpu(cpu);
 			if (xen_feature(XENFEAT_hvm_safe_pvclock))
 				xen_setup_timer(cpu);
 		}
@@ -990,10 +990,13 @@
 				printk(KERN_WARNING "m2p_remove_override: "
 						"pfn %lx mfn %lx, failed to modify kernel mappings",
 						pfn, mfn);
+				put_balloon_scratch_page();
 				return -1;
 			}
  
-			mcs = xen_mc_entry(
+			xen_mc_batch();
+
+			mcs = __xen_mc_entry(
 					sizeof(struct gnttab_unmap_and_replace));
 			unmap_op = mcs.args;
 			unmap_op->host_addr = kmap_op->host_addr;
  
  
@@ -1003,12 +1006,11 @@
 			MULTI_grant_table_op(mcs.mc,
 					GNTTABOP_unmap_and_replace, unmap_op, 1);
  
-			xen_mc_issue(PARAVIRT_LAZY_MMU);
-
 			mcs = __xen_mc_entry(0);
 			MULTI_update_va_mapping(mcs.mc, scratch_page_address,
-					pfn_pte(page_to_pfn(get_balloon_scratch_page()),
+					pfn_pte(page_to_pfn(scratch_page),
 					PAGE_KERNEL_RO), 0);
+
 			xen_mc_issue(PARAVIRT_LAZY_MMU);
  
 			kmap_op->host_addr = 0;
@@ -273,12 +273,20 @@
 	BUG_ON(smp_processor_id() != 0);
 	native_smp_prepare_boot_cpu();
  
-	/* We've switched to the "real" per-cpu gdt, so make sure the
-	   old memory can be recycled */
-	make_lowmem_page_readwrite(xen_initial_gdt);
+	if (xen_pv_domain()) {
+		/* We've switched to the "real" per-cpu gdt, so make sure the
+		   old memory can be recycled */
+		make_lowmem_page_readwrite(xen_initial_gdt);
  
-	xen_filter_cpu_maps();
-	xen_setup_vcpu_info_placement();
+		xen_filter_cpu_maps();
+		xen_setup_vcpu_info_placement();
+	}
+	/*
+	 * The alternative logic (which patches the unlock/lock) runs before
+	 * the smp bootup up code is activated. Hence we need to set this up
+	 * the core kernel is being patched. Otherwise we will have only
+	 * modules patched but not core code.
+	 */
 	xen_init_spinlocks();
 }
  
@@ -709,6 +717,15 @@
 	WARN_ON(rc);
 	if (!rc)
 		rc =  native_cpu_up(cpu, tidle);
+
+	/*
+	 * We must initialize the slowpath CPU kicker _after_ the native
+	 * path has executed. If we initialized it before none of the
+	 * unlocker IPI kicks would reach the booting CPU as the booting
+	 * CPU had not set itself 'online' in cpu_online_mask. That mask
+	 * is checked when IPIs are sent (on HVM at least).
+	 */
+	xen_init_lock_cpu(cpu);
 	return rc;
 }
  
@@ -728,5 +745,6 @@
 	smp_ops.cpu_die = xen_hvm_cpu_die;
 	smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
 	smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
+	smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu;
 }
@@ -81,7 +81,6 @@
 	spinlock_stats.time_blocked += delta;
 }
 #else  /* !CONFIG_XEN_DEBUG_FS */
-#define TIMEOUT			(1 << 10)
 static inline void add_stats(enum xen_contention_stat var, u32 val)
 {
 }
@@ -96,23 +95,6 @@
 }
 #endif  /* CONFIG_XEN_DEBUG_FS */
  
-/*
- * Size struct xen_spinlock so it's the same as arch_spinlock_t.
- */
-#if NR_CPUS < 256
-typedef u8 xen_spinners_t;
-# define inc_spinners(xl) \
-	asm(LOCK_PREFIX " incb %0" : "+m" ((xl)->spinners) : : "memory");
-# define dec_spinners(xl) \
-	asm(LOCK_PREFIX " decb %0" : "+m" ((xl)->spinners) : : "memory");
-#else
-typedef u16 xen_spinners_t;
-# define inc_spinners(xl) \
-	asm(LOCK_PREFIX " incw %0" : "+m" ((xl)->spinners) : : "memory");
-# define dec_spinners(xl) \
-	asm(LOCK_PREFIX " decw %0" : "+m" ((xl)->spinners) : : "memory");
-#endif
-
 struct xen_lock_waiting {
 	struct arch_spinlock *lock;
 	__ticket_t want;
@@ -123,6 +105,7 @@
 static DEFINE_PER_CPU(struct xen_lock_waiting, lock_waiting);
 static cpumask_t waiting_cpus;
  
+static bool xen_pvspin = true;
 static void xen_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
 {
 	int irq = __this_cpu_read(lock_kicker_irq);
  
@@ -241,16 +224,12 @@
 	int irq;
 	char *name;
  
+	if (!xen_pvspin)
+		return;
+
 	WARN(per_cpu(lock_kicker_irq, cpu) >= 0, "spinlock on CPU%d exists on IRQ%d!\n",
 	     cpu, per_cpu(lock_kicker_irq, cpu));
  
-	/*
-	 * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
-	 * (xen: disable PV spinlocks on HVM)
-	 */
-	if (xen_hvm_domain())
-		return;
-
 	name = kasprintf(GFP_KERNEL, "spinlock%d", cpu);
 	irq = bind_ipi_to_irqhandler(XEN_SPIN_UNLOCK_VECTOR,
 				     cpu,
@@ -270,11 +249,7 @@
  
 void xen_uninit_lock_cpu(int cpu)
 {
-	/*
-	 * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
-	 * (xen: disable PV spinlocks on HVM)
-	 */
-	if (xen_hvm_domain())
+	if (!xen_pvspin)
 		return;
  
 	unbind_from_irqhandler(per_cpu(lock_kicker_irq, cpu), NULL);
  
@@ -283,16 +258,9 @@
 	per_cpu(irq_name, cpu) = NULL;
 }
  
-static bool xen_pvspin __initdata = true;
  
 void __init xen_init_spinlocks(void)
 {
-	/*
-	 * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
-	 * (xen: disable PV spinlocks on HVM)
-	 */
-	if (xen_hvm_domain())
-		return;
  
 	if (!xen_pvspin) {
 		printk(KERN_DEBUG "xen: PV spinlocks disabled\n");
@@ -322,6 +290,9 @@
  
 	if (d_xen == NULL)
 		return -ENOMEM;
+
+	if (!xen_pvspin)
+		return 0;
  
 	d_spin_debug = debugfs_create_dir("spinlocks", d_xen);
...	...	@@ -21,6 +21,8 @@
21	21	#include <linux/of.h>
22	22	#include <linux/of_irq.h>
23	23	#include <linux/of_address.h>
	24	+#include <linux/cpuidle.h>
	25	+#include <linux/cpufreq.h>
24	26
25	27	#include <linux/mm.h>
26	28
27	29
28	30
...	...	@@ -267,18 +269,28 @@
267	269	if (!xen_initial_domain())
268	270	xenbus_probe(NULL);
269	271
	272	+ /*
	273	+ * Making sure board specific code will not set up ops for
	274	+ * cpu idle and cpu freq.
	275	+ */
	276	+ disable_cpuidle();
	277	+ disable_cpufreq();
	278	+
270	279	return 0;
271	280	}
272	281	core_initcall(xen_guest_init);
273	282
274	283	static int __init xen_pm_init(void)
275	284	{
	285	+ if (!xen_domain())
	286	+ return -ENODEV;
	287	+
276	288	pm_power_off = xen_power_off;
277	289	arm_pm_restart = xen_restart;
278	290
279	291	return 0;
280	292	}
281		-subsys_initcall(xen_pm_init);
	293	+late_initcall(xen_pm_init);
282	294
283	295	static irqreturn_t xen_arm_callback(int irq, void *arg)
284	296	{
...	...	@@ -1692,7 +1692,6 @@
1692	1692	case CPU_UP_PREPARE:
1693	1693	xen_vcpu_setup(cpu);
1694	1694	if (xen_have_vector_callback) {
1695		- xen_init_lock_cpu(cpu);
1696	1695	if (xen_feature(XENFEAT_hvm_safe_pvclock))
1697	1696	xen_setup_timer(cpu);
1698	1697	}
...	...	@@ -990,10 +990,13 @@
990	990	printk(KERN_WARNING "m2p_remove_override: "
991	991	"pfn %lx mfn %lx, failed to modify kernel mappings",
992	992	pfn, mfn);
	993	+ put_balloon_scratch_page();
993	994	return -1;
994	995	}
995	996
996		- mcs = xen_mc_entry(
	997	+ xen_mc_batch();
	998	+
	999	+ mcs = __xen_mc_entry(
997	1000	sizeof(struct gnttab_unmap_and_replace));
998	1001	unmap_op = mcs.args;
999	1002	unmap_op->host_addr = kmap_op->host_addr;
1000	1003
1001	1004
...	...	@@ -1003,12 +1006,11 @@
1003	1006	MULTI_grant_table_op(mcs.mc,
1004	1007	GNTTABOP_unmap_and_replace, unmap_op, 1);
1005	1008
1006		- xen_mc_issue(PARAVIRT_LAZY_MMU);
1007		-
1008	1009	mcs = __xen_mc_entry(0);
1009	1010	MULTI_update_va_mapping(mcs.mc, scratch_page_address,
1010		- pfn_pte(page_to_pfn(get_balloon_scratch_page()),
	1011	+ pfn_pte(page_to_pfn(scratch_page),
1011	1012	PAGE_KERNEL_RO), 0);
	1013	+
1012	1014	xen_mc_issue(PARAVIRT_LAZY_MMU);
1013	1015
1014	1016	kmap_op->host_addr = 0;
...	...	@@ -273,12 +273,20 @@
273	273	BUG_ON(smp_processor_id() != 0);
274	274	native_smp_prepare_boot_cpu();
275	275
276		- /* We've switched to the "real" per-cpu gdt, so make sure the
277		- old memory can be recycled */
278		- make_lowmem_page_readwrite(xen_initial_gdt);
	276	+ if (xen_pv_domain()) {
	277	+ /* We've switched to the "real" per-cpu gdt, so make sure the
	278	+ old memory can be recycled */
	279	+ make_lowmem_page_readwrite(xen_initial_gdt);
279	280
280		- xen_filter_cpu_maps();
281		- xen_setup_vcpu_info_placement();
	281	+ xen_filter_cpu_maps();
	282	+ xen_setup_vcpu_info_placement();
	283	+ }
	284	+ /*
	285	+ * The alternative logic (which patches the unlock/lock) runs before
	286	+ * the smp bootup up code is activated. Hence we need to set this up
	287	+ * the core kernel is being patched. Otherwise we will have only
	288	+ * modules patched but not core code.
	289	+ */
282	290	xen_init_spinlocks();
283	291	}
284	292
...	...	@@ -709,6 +717,15 @@
709	717	WARN_ON(rc);
710	718	if (!rc)
711	719	rc = native_cpu_up(cpu, tidle);
	720	+
	721	+ /*
	722	+ * We must initialize the slowpath CPU kicker _after_ the native
	723	+ * path has executed. If we initialized it before none of the
	724	+ * unlocker IPI kicks would reach the booting CPU as the booting
	725	+ * CPU had not set itself 'online' in cpu_online_mask. That mask
	726	+ * is checked when IPIs are sent (on HVM at least).
	727	+ */
	728	+ xen_init_lock_cpu(cpu);
712	729	return rc;
713	730	}
714	731
...	...	@@ -728,5 +745,6 @@
728	745	smp_ops.cpu_die = xen_hvm_cpu_die;
729	746	smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
730	747	smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
	748	+ smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu;
731	749	}
...	...	@@ -81,7 +81,6 @@
81	81	spinlock_stats.time_blocked += delta;
82	82	}
83	83	#else /* !CONFIG_XEN_DEBUG_FS */
84		-#define TIMEOUT (1 << 10)
85	84	static inline void add_stats(enum xen_contention_stat var, u32 val)
86	85	{
87	86	}
...	...	@@ -96,23 +95,6 @@
96	95	}
97	96	#endif /* CONFIG_XEN_DEBUG_FS */
98	97
99		-/*
100		- * Size struct xen_spinlock so it's the same as arch_spinlock_t.
101		- */
102		-#if NR_CPUS < 256
103		-typedef u8 xen_spinners_t;
104		-# define inc_spinners(xl) \
105		- asm(LOCK_PREFIX " incb %0" : "+m" ((xl)->spinners) : : "memory");
106		-# define dec_spinners(xl) \
107		- asm(LOCK_PREFIX " decb %0" : "+m" ((xl)->spinners) : : "memory");
108		-#else
109		-typedef u16 xen_spinners_t;
110		-# define inc_spinners(xl) \
111		- asm(LOCK_PREFIX " incw %0" : "+m" ((xl)->spinners) : : "memory");
112		-# define dec_spinners(xl) \
113		- asm(LOCK_PREFIX " decw %0" : "+m" ((xl)->spinners) : : "memory");
114		-#endif
115		-
116	98	struct xen_lock_waiting {
117	99	struct arch_spinlock *lock;
118	100	__ticket_t want;
...	...	@@ -123,6 +105,7 @@
123	105	static DEFINE_PER_CPU(struct xen_lock_waiting, lock_waiting);
124	106	static cpumask_t waiting_cpus;
125	107
	108	+static bool xen_pvspin = true;
126	109	static void xen_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
127	110	{
128	111	int irq = __this_cpu_read(lock_kicker_irq);
129	112
...	...	@@ -241,16 +224,12 @@
241	224	int irq;
242	225	char *name;
243	226
	227	+ if (!xen_pvspin)
	228	+ return;
	229	+
244	230	WARN(per_cpu(lock_kicker_irq, cpu) >= 0, "spinlock on CPU%d exists on IRQ%d!\n",
245	231	cpu, per_cpu(lock_kicker_irq, cpu));
246	232
247		- /*
248		- * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
249		- * (xen: disable PV spinlocks on HVM)
250		- */
251		- if (xen_hvm_domain())
252		- return;
253		-
254	233	name = kasprintf(GFP_KERNEL, "spinlock%d", cpu);
255	234	irq = bind_ipi_to_irqhandler(XEN_SPIN_UNLOCK_VECTOR,
256	235	cpu,
...	...	@@ -270,11 +249,7 @@
270	249
271	250	void xen_uninit_lock_cpu(int cpu)
272	251	{
273		- /*
274		- * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
275		- * (xen: disable PV spinlocks on HVM)
276		- */
277		- if (xen_hvm_domain())
	252	+ if (!xen_pvspin)
278	253	return;
279	254
280	255	unbind_from_irqhandler(per_cpu(lock_kicker_irq, cpu), NULL);
281	256
...	...	@@ -283,16 +258,9 @@
283	258	per_cpu(irq_name, cpu) = NULL;
284	259	}
285	260
286		-static bool xen_pvspin __initdata = true;
287	261
288	262	void __init xen_init_spinlocks(void)
289	263	{
290		- /*
291		- * See git commit f10cd522c5fbfec9ae3cc01967868c9c2401ed23
292		- * (xen: disable PV spinlocks on HVM)
293		- */
294		- if (xen_hvm_domain())
295		- return;
296	264
297	265	if (!xen_pvspin) {
298	266	printk(KERN_DEBUG "xen: PV spinlocks disabled\n");
...	...	@@ -322,6 +290,9 @@
322	290
323	291	if (d_xen == NULL)
324	292	return -ENOMEM;
	293	+
	294	+ if (!xen_pvspin)
	295	+ return 0;
325	296
326	297	d_spin_debug = debugfs_create_dir("spinlocks", d_xen);
327	298