powerpc/powernv: Improve kexec reliability

We add a machine_shutdown hook that frees the OPAL interrupts (so they get masked at the source and don't fire while kexec'ing) and which triggers an IODA reset on all the PCIe host bridges which will have the effect of blocking all DMAs and subsequent PCIs interrupts. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

powerpc/powernv: Improve kexec reliability
We add a machine_shutdown hook that frees the OPAL interrupts (so they get masked at the source and don't fire while kexec'ing) and which triggers an IODA reset on all the PCIe host bridges which will have the effect of blocking all DMAs and subsequent PCIs interrupts. Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Benjamin Herrenschmidt
1 parent 1de1455f33
Showing 7 changed files with 56 additions and 0 deletions Side-by-side Diff
arch/powerpc/include/asm/opal.h
arch/powerpc/platforms/powernv/opal.c
arch/powerpc/platforms/powernv/pci-ioda.c
arch/powerpc/platforms/powernv/pci.c
arch/powerpc/platforms/powernv/pci.h
arch/powerpc/platforms/powernv/powernv.h
arch/powerpc/platforms/powernv/setup.c
@@ -563,6 +563,8 @@
  
 extern int opal_machine_check(struct pt_regs *regs);
  
+extern void opal_shutdown(void);
+
 #endif /* __ASSEMBLY__ */
  
 #endif /* __OPAL_H */
@@ -15,6 +15,7 @@
 #include <linux/of.h>
 #include <linux/of_platform.h>
 #include <linux/interrupt.h>
+#include <linux/slab.h>
 #include <asm/opal.h>
 #include <asm/firmware.h>
  
@@ -28,6 +29,8 @@
 static struct device_node *opal_node;
 static DEFINE_SPINLOCK(opal_write_lock);
 extern u64 opal_mc_secondary_handler[];
+static unsigned int *opal_irqs;
+static unsigned int opal_irq_count;
  
 int __init early_init_dt_scan_opal(unsigned long node,
 				   const char *uname, int depth, void *data)
@@ -323,6 +326,8 @@
 	irqs = of_get_property(opal_node, "opal-interrupts", &irqlen);
 	pr_debug("opal: Found %d interrupts reserved for OPAL\n",
 		 irqs ? (irqlen / 4) : 0);
+	opal_irq_count = irqlen / 4;
+	opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL);
 	for (i = 0; irqs && i < (irqlen / 4); i++, irqs++) {
 		unsigned int hwirq = be32_to_cpup(irqs);
 		unsigned int irq = irq_create_mapping(NULL, hwirq);
  
@@ -334,8 +339,20 @@
 		if (rc)
 			pr_warning("opal: Error %d requesting irq %d"
 				   " (0x%x)\n", rc, irq, hwirq);
+		opal_irqs[i] = irq;
 	}
 	return 0;
 }
 subsys_initcall(opal_init);
+
+void opal_shutdown(void)
+{
+	unsigned int i;
+
+	for (i = 0; i < opal_irq_count; i++) {
+		if (opal_irqs[i])
+			free_irq(opal_irqs[i], 0);
+		opal_irqs[i] = 0;
+	}
+}
@@ -1048,6 +1048,12 @@
 	return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
 }
  
+static void pnv_pci_ioda_shutdown(struct pnv_phb *phb)
+{
+	opal_pci_reset(phb->opal_id, OPAL_PCI_IODA_TABLE_RESET,
+		       OPAL_ASSERT_RESET);
+}
+
 void __init pnv_pci_init_ioda_phb(struct device_node *np, int ioda_type)
 {
 	struct pci_controller *hose;
@@ -1177,6 +1183,9 @@
  
 	/* Setup TCEs */
 	phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
+
+	/* Setup shutdown function for kexec */
+	phb->shutdown = pnv_pci_ioda_shutdown;
  
 	/* Setup MSI support */
 	pnv_pci_init_ioda_msis(phb);
@@ -450,6 +450,18 @@
 		pnv_pci_dma_fallback_setup(hose, pdev);
 }
  
+void pnv_pci_shutdown(void)
+{
+	struct pci_controller *hose;
+
+	list_for_each_entry(hose, &hose_list, list_node) {
+		struct pnv_phb *phb = hose->private_data;
+
+		if (phb && phb->shutdown)
+			phb->shutdown(phb);
+	}
+}
+
 /* Fixup wrong class code in p7ioc and p8 root complex */
 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
 {
@@ -86,6 +86,7 @@
 	void (*dma_dev_setup)(struct pnv_phb *phb, struct pci_dev *pdev);
 	void (*fixup_phb)(struct pci_controller *hose);
 	u32 (*bdfn_to_pe)(struct pnv_phb *phb, struct pci_bus *bus, u32 devfn);
+	void (*shutdown)(struct pnv_phb *phb);
  
 	union {
 		struct {
@@ -158,5 +159,6 @@
 extern void pnv_pci_init_ioda2_phb(struct device_node *np);
 extern void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl,
 					u64 *startp, u64 *endp);
+
 #endif /* __POWERNV_PCI_H */
@@ -9,8 +9,10 @@
  
 #ifdef CONFIG_PCI
 extern void pnv_pci_init(void);
+extern void pnv_pci_shutdown(void);
 #else
 static inline void pnv_pci_init(void) { }
+static inline void pnv_pci_shutdown(void) { }
 #endif
  
 #endif /* _POWERNV_H */
@@ -126,6 +126,17 @@
 {
 }
  
+static void pnv_shutdown(void)
+{
+	/* Let the PCI code clear up IODA tables */
+	pnv_pci_shutdown();
+
+	/* And unregister all OPAL interrupts so they don't fire
+	 * up while we kexec
+	 */
+	opal_shutdown();
+}
+
 #ifdef CONFIG_KEXEC
 static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
 {
@@ -187,6 +198,7 @@
 	.init_IRQ		= pnv_init_IRQ,
 	.show_cpuinfo		= pnv_show_cpuinfo,
 	.progress		= pnv_progress,
+	.machine_shutdown	= pnv_shutdown,
 	.power_save             = power7_idle,
 	.calibrate_decr		= generic_calibrate_decr,
 #ifdef CONFIG_KEXEC
...	...	@@ -563,6 +563,8 @@
563	563
564	564	extern int opal_machine_check(struct pt_regs *regs);
565	565
	566	+extern void opal_shutdown(void);
	567	+
566	568	#endif /* __ASSEMBLY__ */
567	569
568	570	#endif /* __OPAL_H */
...	...	@@ -15,6 +15,7 @@
15	15	#include <linux/of.h>
16	16	#include <linux/of_platform.h>
17	17	#include <linux/interrupt.h>
	18	+#include <linux/slab.h>
18	19	#include <asm/opal.h>
19	20	#include <asm/firmware.h>
20	21
...	...	@@ -28,6 +29,8 @@
28	29	static struct device_node *opal_node;
29	30	static DEFINE_SPINLOCK(opal_write_lock);
30	31	extern u64 opal_mc_secondary_handler[];
	32	+static unsigned int *opal_irqs;
	33	+static unsigned int opal_irq_count;
31	34
32	35	int __init early_init_dt_scan_opal(unsigned long node,
33	36	const char uname, int depth, void data)
...	...	@@ -323,6 +326,8 @@
323	326	irqs = of_get_property(opal_node, "opal-interrupts", &irqlen);
324	327	pr_debug("opal: Found %d interrupts reserved for OPAL\n",
325	328	irqs ? (irqlen / 4) : 0);
	329	+ opal_irq_count = irqlen / 4;
	330	+ opal_irqs = kzalloc(opal_irq_count * sizeof(unsigned int), GFP_KERNEL);
326	331	for (i = 0; irqs && i < (irqlen / 4); i++, irqs++) {
327	332	unsigned int hwirq = be32_to_cpup(irqs);
328	333	unsigned int irq = irq_create_mapping(NULL, hwirq);
329	334
...	...	@@ -334,8 +339,20 @@
334	339	if (rc)
335	340	pr_warning("opal: Error %d requesting irq %d"
336	341	" (0x%x)\n", rc, irq, hwirq);
	342	+ opal_irqs[i] = irq;
337	343	}
338	344	return 0;
339	345	}
340	346	subsys_initcall(opal_init);
	347	+
	348	+void opal_shutdown(void)
	349	+{
	350	+ unsigned int i;
	351	+
	352	+ for (i = 0; i < opal_irq_count; i++) {
	353	+ if (opal_irqs[i])
	354	+ free_irq(opal_irqs[i], 0);
	355	+ opal_irqs[i] = 0;
	356	+ }
	357	+}
...	...	@@ -1048,6 +1048,12 @@
1048	1048	return phb->ioda.pe_rmap[(bus->number << 8) \| devfn];
1049	1049	}
1050	1050
	1051	+static void pnv_pci_ioda_shutdown(struct pnv_phb *phb)
	1052	+{
	1053	+ opal_pci_reset(phb->opal_id, OPAL_PCI_IODA_TABLE_RESET,
	1054	+ OPAL_ASSERT_RESET);
	1055	+}
	1056	+
1051	1057	void __init pnv_pci_init_ioda_phb(struct device_node *np, int ioda_type)
1052	1058	{
1053	1059	struct pci_controller *hose;
...	...	@@ -1177,6 +1183,9 @@
1177	1183
1178	1184	/* Setup TCEs */
1179	1185	phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
	1186	+
	1187	+ /* Setup shutdown function for kexec */
	1188	+ phb->shutdown = pnv_pci_ioda_shutdown;
1180	1189
1181	1190	/* Setup MSI support */
1182	1191	pnv_pci_init_ioda_msis(phb);
...	...	@@ -450,6 +450,18 @@
450	450	pnv_pci_dma_fallback_setup(hose, pdev);
451	451	}
452	452
	453	+void pnv_pci_shutdown(void)
	454	+{
	455	+ struct pci_controller *hose;
	456	+
	457	+ list_for_each_entry(hose, &hose_list, list_node) {
	458	+ struct pnv_phb *phb = hose->private_data;
	459	+
	460	+ if (phb && phb->shutdown)
	461	+ phb->shutdown(phb);
	462	+ }
	463	+}
	464	+
453	465	/* Fixup wrong class code in p7ioc and p8 root complex */
454	466	static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
455	467	{
...	...	@@ -86,6 +86,7 @@
86	86	void (dma_dev_setup)(struct pnv_phb phb, struct pci_dev *pdev);
87	87	void (fixup_phb)(struct pci_controller hose);
88	88	u32 (bdfn_to_pe)(struct pnv_phb phb, struct pci_bus *bus, u32 devfn);
	89	+ void (shutdown)(struct pnv_phb phb);
89	90
90	91	union {
91	92	struct {
...	...	@@ -158,5 +159,6 @@
158	159	extern void pnv_pci_init_ioda2_phb(struct device_node *np);
159	160	extern void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl,
160	161	u64 startp, u64 endp);
	162	+
161	163	#endif /* __POWERNV_PCI_H */
...	...	@@ -9,8 +9,10 @@
9	9
10	10	#ifdef CONFIG_PCI
11	11	extern void pnv_pci_init(void);
	12	+extern void pnv_pci_shutdown(void);
12	13	#else
13	14	static inline void pnv_pci_init(void) { }
	15	+static inline void pnv_pci_shutdown(void) { }
14	16	#endif
15	17
16	18	#endif /* _POWERNV_H */
...	...	@@ -126,6 +126,17 @@
126	126	{
127	127	}
128	128
	129	+static void pnv_shutdown(void)
	130	+{
	131	+ /* Let the PCI code clear up IODA tables */
	132	+ pnv_pci_shutdown();
	133	+
	134	+ /* And unregister all OPAL interrupts so they don't fire
	135	+ * up while we kexec
	136	+ */
	137	+ opal_shutdown();
	138	+}
	139	+
129	140	#ifdef CONFIG_KEXEC
130	141	static void pnv_kexec_cpu_down(int crash_shutdown, int secondary)
131	142	{
...	...	@@ -187,6 +198,7 @@
187	198	.init_IRQ = pnv_init_IRQ,
188	199	.show_cpuinfo = pnv_show_cpuinfo,
189	200	.progress = pnv_progress,
	201	+ .machine_shutdown = pnv_shutdown,
190	202	.power_save = power7_idle,
191	203	.calibrate_decr = generic_calibrate_decr,
192	204	#ifdef CONFIG_KEXEC