s390/nmi: allocation of the extended save area

The machine check extended save area is needed to store the vector
registers and the guarded storage control block when a CPU is
interrupted by a machine check.

Move the slab cache allocation of the full save area to nmi.c,
for early boot use a static __initdata block.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/nmi.h

@@ -80,6 +80,8 @@ union mci {
 
 #define MCESA_ORIGIN_MASK	(~0x3ffUL)
 #define MCESA_LC_MASK		(0xfUL)
+#define MCESA_MIN_SIZE		(1024)
+#define MCESA_MAX_SIZE		(2048)
 
 struct mcesa {
 	u8 vector_save_area[1024];
@@ -88,8 +90,12 @@ struct mcesa {
 
 struct pt_regs;
 
-extern void s390_handle_mcck(void);
-extern void s390_do_machine_check(struct pt_regs *regs);
+void nmi_alloc_boot_cpu(struct lowcore *lc);
+int nmi_alloc_per_cpu(struct lowcore *lc);
+void nmi_free_per_cpu(struct lowcore *lc);
+
+void s390_handle_mcck(void);
+void s390_do_machine_check(struct pt_regs *regs);
 
 #endif /* __ASSEMBLY__ */
 #endif /* _ASM_S390_NMI_H */

arch/s390/kernel/nmi.c

@@ -12,7 +12,9 @@
 #include <linux/init.h>
 #include <linux/errno.h>
 #include <linux/hardirq.h>
+#include <linux/log2.h>
 #include <linux/kprobes.h>
+#include <linux/slab.h>
 #include <linux/time.h>
 #include <linux/module.h>
 #include <linux/sched/signal.h>
@@ -38,6 +40,86 @@ struct mcck_struct {
 };
 
 static DEFINE_PER_CPU(struct mcck_struct, cpu_mcck);
+static struct kmem_cache *mcesa_cache;
+static unsigned long mcesa_origin_lc;
+
+static inline int nmi_needs_mcesa(void)
+{
+	return MACHINE_HAS_VX || MACHINE_HAS_GS;
+}
+
+static inline unsigned long nmi_get_mcesa_size(void)
+{
+	if (MACHINE_HAS_GS)
+		return MCESA_MAX_SIZE;
+	return MCESA_MIN_SIZE;
+}
+
+/*
+ * The initial machine check extended save area for the boot CPU.
+ * It will be replaced by nmi_init() with an allocated structure.
+ * The structure is required for machine check happening early in
+ * the boot process.
+ */
+static struct mcesa boot_mcesa __initdata __aligned(MCESA_MAX_SIZE);
+
+void __init nmi_alloc_boot_cpu(struct lowcore *lc)
+{
+	if (!nmi_needs_mcesa())
+		return;
+	lc->mcesad = (unsigned long) &boot_mcesa;
+	if (MACHINE_HAS_GS)
+		lc->mcesad |= ilog2(MCESA_MAX_SIZE);
+}
+
+static int __init nmi_init(void)
+{
+	unsigned long origin, cr0, size;
+
+	if (!nmi_needs_mcesa())
+		return 0;
+	size = nmi_get_mcesa_size();
+	if (size > MCESA_MIN_SIZE)
+		mcesa_origin_lc = ilog2(size);
+	/* create slab cache for the machine-check-extended-save-areas */
+	mcesa_cache = kmem_cache_create("nmi_save_areas", size, size, 0, NULL);
+	if (!mcesa_cache)
+		panic("Couldn't create nmi save area cache");
+	origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
+	if (!origin)
+		panic("Couldn't allocate nmi save area");
+	/* The pointer is stored with mcesa_bits ORed in */
+	kmemleak_not_leak((void *) origin);
+	__ctl_store(cr0, 0, 0);
+	__ctl_clear_bit(0, 28); /* disable lowcore protection */
+	/* Replace boot_mcesa on the boot CPU */
+	S390_lowcore.mcesad = origin | mcesa_origin_lc;
+	__ctl_load(cr0, 0, 0);
+	return 0;
+}
+early_initcall(nmi_init);
+
+int nmi_alloc_per_cpu(struct lowcore *lc)
+{
+	unsigned long origin;
+
+	if (!nmi_needs_mcesa())
+		return 0;
+	origin = (unsigned long) kmem_cache_alloc(mcesa_cache, GFP_KERNEL);
+	if (!origin)
+		return -ENOMEM;
+	/* The pointer is stored with mcesa_bits ORed in */
+	kmemleak_not_leak((void *) origin);
+	lc->mcesad = origin | mcesa_origin_lc;
+	return 0;
+}
+
+void nmi_free_per_cpu(struct lowcore *lc)
+{
+	if (!nmi_needs_mcesa())
+		return;
+	kmem_cache_free(mcesa_cache, (void *)(lc->mcesad & MCESA_ORIGIN_MASK));
+}
 
 static notrace void s390_handle_damage(void)
 {

arch/s390/kernel/setup.c

@@ -55,6 +55,7 @@
 #include <asm/mmu_context.h>
 #include <asm/cpcmd.h>
 #include <asm/lowcore.h>
+#include <asm/nmi.h>
 #include <asm/irq.h>
 #include <asm/page.h>
 #include <asm/ptrace.h>
@@ -340,15 +341,7 @@ static void __init setup_lowcore(void)
 	lc->stfl_fac_list = S390_lowcore.stfl_fac_list;
 	memcpy(lc->stfle_fac_list, S390_lowcore.stfle_fac_list,
 	       MAX_FACILITY_BIT/8);
-	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
-		unsigned long bits, size;
-
-		bits = MACHINE_HAS_GS ? 11 : 10;
-		size = 1UL << bits;
-		lc->mcesad = (__u64) memblock_virt_alloc(size, size);
-		if (MACHINE_HAS_GS)
-			lc->mcesad |= bits;
-	}
+	nmi_alloc_boot_cpu(lc);
 	vdso_alloc_boot_cpu(lc);
 	lc->sync_enter_timer = S390_lowcore.sync_enter_timer;
 	lc->async_enter_timer = S390_lowcore.async_enter_timer;

arch/s390/kernel/smp.c

@@ -81,8 +81,6 @@ struct pcpu {
 static u8 boot_core_type;
 static struct pcpu pcpu_devices[NR_CPUS];
 
-static struct kmem_cache *pcpu_mcesa_cache;
-
 unsigned int smp_cpu_mt_shift;
 EXPORT_SYMBOL(smp_cpu_mt_shift);
 
@@ -193,10 +191,8 @@ static void pcpu_ec_call(struct pcpu *pcpu, int ec_bit)
 static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 {
 	unsigned long async_stack, panic_stack;
-	unsigned long mcesa_origin, mcesa_bits;
 	struct lowcore *lc;
 
-	mcesa_origin = mcesa_bits = 0;
 	if (pcpu != &pcpu_devices[0]) {
 		pcpu->lowcore =	(struct lowcore *)
 			__get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
@@ -204,40 +200,30 @@ static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 		panic_stack = __get_free_page(GFP_KERNEL);
 		if (!pcpu->lowcore || !panic_stack || !async_stack)
 			goto out;
-		if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
-			mcesa_origin = (unsigned long)
-				kmem_cache_alloc(pcpu_mcesa_cache, GFP_KERNEL);
-			if (!mcesa_origin)
-				goto out;
-			/* The pointer is stored with mcesa_bits ORed in */
-			kmemleak_not_leak((void *) mcesa_origin);
-			mcesa_bits = MACHINE_HAS_GS ? 11 : 0;
-		}
 	} else {
 		async_stack = pcpu->lowcore->async_stack - ASYNC_FRAME_OFFSET;
 		panic_stack = pcpu->lowcore->panic_stack - PANIC_FRAME_OFFSET;
-		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
-		mcesa_bits = pcpu->lowcore->mcesad & MCESA_LC_MASK;
 	}
 	lc = pcpu->lowcore;
 	memcpy(lc, &S390_lowcore, 512);
 	memset((char *) lc + 512, 0, sizeof(*lc) - 512);
 	lc->async_stack = async_stack + ASYNC_FRAME_OFFSET;
 	lc->panic_stack = panic_stack + PANIC_FRAME_OFFSET;
-	lc->mcesad = mcesa_origin | mcesa_bits;
 	lc->cpu_nr = cpu;
 	lc->spinlock_lockval = arch_spin_lockval(cpu);
 	lc->spinlock_index = 0;
-	if (vdso_alloc_per_cpu(lc))
+	if (nmi_alloc_per_cpu(lc))
 		goto out;
+	if (vdso_alloc_per_cpu(lc))
+		goto out_mcesa;
 	lowcore_ptr[cpu] = lc;
 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, (u32)(unsigned long) lc);
 	return 0;
+
+out_mcesa:
+	nmi_free_per_cpu(lc);
 out:
 	if (pcpu != &pcpu_devices[0]) {
-		if (mcesa_origin)
-			kmem_cache_free(pcpu_mcesa_cache,
-					(void *) mcesa_origin);
 		free_page(panic_stack);
 		free_pages(async_stack, ASYNC_ORDER);
 		free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
@@ -249,17 +235,12 @@ static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 
 static void pcpu_free_lowcore(struct pcpu *pcpu)
 {
-	unsigned long mcesa_origin;
-
 	pcpu_sigp_retry(pcpu, SIGP_SET_PREFIX, 0);
 	lowcore_ptr[pcpu - pcpu_devices] = NULL;
 	vdso_free_per_cpu(pcpu->lowcore);
+	nmi_free_per_cpu(pcpu->lowcore);
 	if (pcpu == &pcpu_devices[0])
 		return;
-	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
-		mcesa_origin = pcpu->lowcore->mcesad & MCESA_ORIGIN_MASK;
-		kmem_cache_free(pcpu_mcesa_cache, (void *) mcesa_origin);
-	}
 	free_page(pcpu->lowcore->panic_stack-PANIC_FRAME_OFFSET);
 	free_pages(pcpu->lowcore->async_stack-ASYNC_FRAME_OFFSET, ASYNC_ORDER);
 	free_pages((unsigned long) pcpu->lowcore, LC_ORDER);
@@ -936,22 +917,12 @@ void __init smp_fill_possible_mask(void)
 
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
-	unsigned long size;
-
 	/* request the 0x1201 emergency signal external interrupt */
 	if (register_external_irq(EXT_IRQ_EMERGENCY_SIG, do_ext_call_interrupt))
 		panic("Couldn't request external interrupt 0x1201");
 	/* request the 0x1202 external call external interrupt */
 	if (register_external_irq(EXT_IRQ_EXTERNAL_CALL, do_ext_call_interrupt))
 		panic("Couldn't request external interrupt 0x1202");
-	/* create slab cache for the machine-check-extended-save-areas */
-	if (MACHINE_HAS_VX || MACHINE_HAS_GS) {
-		size = 1UL << (MACHINE_HAS_GS ? 11 : 10);
-		pcpu_mcesa_cache = kmem_cache_create("nmi_save_areas",
-						     size, size, 0, NULL);
-		if (!pcpu_mcesa_cache)
-			panic("Couldn't create nmi save area cache");
-	}
 }
 
 void __init smp_prepare_boot_cpu(void)