Skip to contentSkip to conversation

i386: fix suspend/resume with dynamically allocated irq stacks

PATCH In review Jeremy Fitzhardinge 2 participants 10 messages May 3, 2007

lore
Share & tools
lore

Jeremy Fitzhardinge

author
#0 lore raw

PATCH 1/1 i386: fix suspend/resume with dynamically allocated irq stacks

This fixes two bugs:
- the stack allocation must be marked __cpuinit, since it gets called on resume as well.
- presumably the interrupt stack should be freed on unplug if its going to get reallocated on every plug.

[ Only non-vmalloced stacks tested. ]

Signed-off-by Jeremy Fitzhardinge <[email protected]>
---
 arch/i386/kernel/irq.c |   42 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)

===================================================================

arch/i386/kernel/irq.c +37 -5
@@ -195,10 +195,13 @@ static int __init irq_guard_cpu0(void)
 }
 core_initcall(irq_guard_cpu0);
 
-static void * __init __alloc_irqstack(int cpu)
+static DEFINE_PER_CPU(struct page *, irqstack_pages[THREAD_SIZE/PAGE_SIZE]);
+static DEFINE_PER_CPU(struct vm_struct *, irqstack_area);
+
+static void * __cpuinit __alloc_irqstack(int cpu)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
+	struct page **pages = per_cpu(irqstack_pages, cpu), **tmp = pages;
 	struct vm_struct *area;
 
 	if (!cpu)

@@ -207,13 +210,27 @@ static void * __init __alloc_irqstack(in
 
 	/* failures here are unrecoverable anyway */
 	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	for (i = 0; i < ARRAY_SIZE(pages); ++i)
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
 		pages[i] = alloc_page(GFP_HIGHUSER);
 	map_vm_area(area, PAGE_KERNEL, &tmp);
+	per_cpu(irqstack_area, cpu) = area;
 	return area->addr;
 }
+
+static void __cpuinit __free_irqstack(int cpu, void *stk)
+{
+	int i;
+
+	if (!cpu)
+		return;
+
+	unmap_vm_area(per_cpu(irqstack_area, cpu));
+
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(per_cpu(irqstack_pages, cpu)[i]);
+}
 #else /* !CONFIG_VMALLOC_STACK */
-static void * __init __alloc_irqstack(int cpu)
+static void * __cpuinit __alloc_irqstack(int cpu)
 {
 	if (!cpu)
 		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,

@@ -222,12 +239,26 @@ static void * __init __alloc_irqstack(in
 	return (void *)__get_free_pages(GFP_KERNEL,
 					ilog2(THREAD_SIZE/PAGE_SIZE));
 }
+
+static void __cpuinit __free_irqstack(int cpu, void *stk)
+{
+	if (!cpu)
+		return;
+
+	free_pages((unsigned long)stk, ilog2(THREAD_SIZE/PAGE_SIZE));
+}
 #endif /* !CONFIG_VMALLOC_STACK */
 
-static void __init alloc_irqstacks(int cpu)
+static void __cpuinit alloc_irqstacks(int cpu)
 {
 	per_cpu(softirq_stack, cpu) = __alloc_irqstack(cpu);
 	per_cpu(hardirq_stack, cpu) = __alloc_irqstack(cpu);
+}
+
+static void __cpuinit free_irqstacks(int cpu)
+{
+	__free_irqstack(cpu, per_cpu(softirq_stack, cpu));
+	__free_irqstack(cpu, per_cpu(hardirq_stack, cpu));
 }
 
 /*

@@ -266,6 +297,7 @@ void irq_ctx_init(int cpu)
 
 void irq_ctx_exit(int cpu)
 {
+	free_irqstacks(cpu);
 	per_cpu(hardirq_ctx, cpu) = NULL;
 }
#1 lore raw

arch/i386/kernel/irq.c static void * __init __alloc_irqstack(in · line 231

Quoted 12 lines · Jeremy Fitzhardinge
arch/i386/kernel/irq.c +12 -0
+static void __cpuinit __free_irqstack(int cpu, void *stk)
+{
+	int i;
+
+	if (!cpu)
+		return;
+
+	unmap_vm_area(per_cpu(irqstack_area, cpu));
+
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(per_cpu(irqstack_pages, cpu)[i]);
+}

This will leak the vm_struct and also leave it in the vmlist. work_free_thread_info() needs to be redone too. It should be

	remove_vm_area( /* unmap and remove the vm_struct from vmlist */ );
	kfree( /* free the vm_struct */ );
	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
		__free_page( /* dredge up the appropriate page to free */ );

-- wli

#2 lore raw
Quoted 4 lines · Jeremy Fitzhardinge
Quoted 4 lines
arch/i386/kernel/irq.c +4 -0
+static void __cpuinit __free_irqstack(int cpu, void *stk)
+{
+	int i;
+
Quoted 8 lines

arch/i386/kernel/irq.c static void * __init __alloc_irqstack(in · line 231

Quoted 8 lines
arch/i386/kernel/irq.c +8 -0
+	if (!cpu)
+		return;
+
+	unmap_vm_area(per_cpu(irqstack_area, cpu));
+
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(per_cpu(irqstack_pages, cpu)[i]);
+}

On Wed, May 02, 2007 at 07:25:34PM -0700, Bill Irwin wrote:
[...]

Not sure if cpu 0 can ever be offlined, but it is remapped and so on. So its virtual mapping should be undone if it can be offlined, but we probably shouldn't attempt to free its underlying bootmem allocation.

-- wli

#3 lore raw
In reply to Jeremy Fitzhardinge
Quoted 10 lines · Jeremy Fitzhardinge

This fixes two bugs:
- the stack allocation must be marked __cpuinit, since it gets called on resume as well.
- presumably the interrupt stack should be freed on unplug if its going to get reallocated on every plug.

[ Only non-vmalloced stacks tested. ]
Signed-off-by Jeremy Fitzhardinge <[email protected]>
---
 arch/i386/kernel/irq.c |   42 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)

Updated patch follows. Please add your Signed-off-by: if it meets your approval; I am operating on the assumption I should never do so myself. I'm a bit unsure of how to handle cpu 0 vs. potential freeing of per_cpu areas and error returns from __cpuinit affairs, but anyhow:

This fixes three bugs:
- the stack allocation must be marked __cpuinit, since it gets called on resume as well.
- presumably the interrupt stack should be freed on unplug if its going to get reallocated on every plug.
- the vm_struct got leaked by thread info freeing callbacks.

Signed-off-by William Irwin <[email protected]>
arch/i386/kernel/irq.c +87 -34
Index: stack-paranoia/arch/i386/kernel/irq.c
===================================================================
@@ -142,18 +142,19 @@
  * These should really be __section__(".bss.page_aligned") as well, but
  * gcc's 3.0 and earlier don't handle that correctly.
  */
-static DEFINE_PER_CPU(char *, softirq_stack);
-static DEFINE_PER_CPU(char *, hardirq_stack);
-
+struct irq_stack_info {
+	char *stack;
 #ifdef CONFIG_DEBUG_STACK
-static void * __init irq_remap_stack(void *stack)
-{
-	int i;
 	struct page *pages[THREAD_SIZE/PAGE_SIZE];
+#endif /* CONFIG_DEBUG_STACK */
+};
+static DEFINE_PER_CPU(struct irq_stack_info, softirq_stack_info);
+static DEFINE_PER_CPU(struct irq_stack_info, hardirq_stack_info);
 
-	for (i = 0; i < ARRAY_SIZE(pages); ++i)
-		pages[i] =  virt_to_page(stack + PAGE_SIZE*i);
-	return vmap(pages, THREAD_SIZE/PAGE_SIZE, VM_IOREMAP, PAGE_KERNEL);
+#ifdef CONFIG_DEBUG_STACK
+static void * __init irq_remap_stack(struct irq_stack_info *info)
+{
+	return vmap(info->pages, ARRAY_SIZE(info->pages), VM_IOREMAP, PAGE_KERNEL);
 }
 
 static int __init irq_guard_cpu0(void)

@@ -161,59 +162,110 @@
 	unsigned long flags;
 	void *tmp;
 
-	tmp = irq_remap_stack(per_cpu(softirq_stack, 0));
+	tmp = irq_remap_stack(&per_cpu(softirq_stack_info, 0));
 	if (!tmp)
 		return -ENOMEM;
 	else {
 		local_irq_save(flags);
-		per_cpu(softirq_stack, 0) = tmp;
+		per_cpu(softirq_stack_info, 0).stack = tmp;
 		local_irq_restore(flags);
 	}
-	tmp = irq_remap_stack(per_cpu(hardirq_stack, 0));
+	tmp = irq_remap_stack(&per_cpu(hardirq_stack_info, 0));
 	if (!tmp)
 		return -ENOMEM;
 	else {
 		local_irq_save(flags);
-		per_cpu(hardirq_stack, 0) = tmp;
+		per_cpu(hardirq_stack_info, 0).stack = tmp;
 		local_irq_restore(flags);
 	}
 	return 0;
 }
 core_initcall(irq_guard_cpu0);
 
-static void * __init __alloc_irqstack(int cpu)
+static int __cpuinit __alloc_irqstack(int cpu, struct irq_stack_info *info)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
-	struct vm_struct *area;
 
-	if (!slab_is_available())
-		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
+	if (!slab_is_available()) {
+		WARN_ON(cpu != 0);
+		info->stack = __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
 						__pa(MAX_DMA_ADDRESS));
+		info->pages[0] = virt_to_page(info->stack);
+		for (i = 1; i < ARRAY_SIZE(info->pages); ++i)
+			info->pages[i] = info->pages[0] + i;
+		return 0;
+	}
+	for (i = 0; i < ARRAY_SIZE(info->pages); ++i) {
+		if (!cpu)
+			WARN_ON(!info->pages[i]);
+		else {
+			info->pages[i] = alloc_page(GFP_HIGHUSER);
+			if (!info->pages[i])
+				goto out;
+		}
+	}
+	info->stack = irq_remap_stack(info);
+	if (info->stack)
+		return 0;
+out:
+	if (cpu) {
+		for (--i; i >= 0; --i) {
+			__free_page(info->pages[i]);
+			info->pages[i] = NULL;
+		}
+	}
+	return -1;
+}
+
+static void __cpuinit __free_irqstack(int cpu, struct irq_stack_info *info)
+{
+	int i;
 
-	/* failures here are unrecoverable anyway */
-	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	for (i = 0; i < ARRAY_SIZE(pages); ++i)
-		pages[i] = alloc_page(GFP_HIGHUSER);
-	map_vm_area(area, PAGE_KERNEL, &tmp);
-	return area->addr;
+	vunmap(info->stack);
+	/* cpu 0 bootmem allocates its underlying pages */
+	if (!cpu)
+		return;
+	for (i = 0; i < ARRAY_SIZE(info->pages); ++i) {
+		__free_page(info->pages[i]);
+		info->pages[i] = NULL;
+	}
 }
 #else /* !CONFIG_DEBUG_STACK */
-static void * __init __alloc_irqstack(int cpu)
+static int __cpuinit __alloc_irqstack(int cpu, struct irq_stack_info *info)
 {
-	if (!slab_is_available())
-		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
+	if (cpu)
+		info->stack = (void *)__get_free_pages(GFP_KERNEL,
+					ilog2(THREAD_SIZE/PAGE_SIZE));
+	else if (!slab_is_available())
+		info->stack = __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
 						__pa(MAX_DMA_ADDRESS));
+	else
+		BUG_ON(!info->stack);
+	return 0;
+}
 
-	return (void *)__get_free_pages(GFP_KERNEL,
-					ilog2(THREAD_SIZE/PAGE_SIZE));
+static void __cpuinit __free_irqstack(int cpu, struct irq_stack_info *info)
+{
+	if (cpu)
+		free_pages((unsigned long)info->stack, ilog2(THREAD_SIZE/PAGE_SIZE));
 }
 #endif /* !CONFIG_DEBUG_STACK */
 
-static void __init alloc_irqstacks(int cpu)
+static int __cpuinit alloc_irqstacks(int cpu)
+{
+	if (__alloc_irqstack(cpu, &per_cpu(softirq_stack_info, cpu)))
+		return -1;
+	if (__alloc_irqstack(cpu, &per_cpu(hardirq_stack_info, cpu))) {
+		__free_irqstack(cpu, &per_cpu(softirq_stack_info, cpu));
+		return -1;
+	}
+	return 0;
+}
+
+static void __cpuinit free_irqstacks(int cpu)
 {
-	per_cpu(softirq_stack, cpu) = __alloc_irqstack(cpu);
-	per_cpu(hardirq_stack, cpu) = __alloc_irqstack(cpu);
+	__free_irqstack(cpu, &per_cpu(softirq_stack_info, cpu));
+	__free_irqstack(cpu, &per_cpu(hardirq_stack_info, cpu));
 }
 
 /*

@@ -228,7 +280,7 @@
 
 	alloc_irqstacks(cpu);
 
-	irqctx = (union irq_ctx*)per_cpu(hardirq_stack, cpu);
+	irqctx = (union irq_ctx*)per_cpu(hardirq_stack_info, cpu).stack;
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;

@@ -237,7 +289,7 @@
 
 	per_cpu(hardirq_ctx, cpu) = irqctx;
 
-	irqctx = (union irq_ctx*)per_cpu(softirq_stack, cpu);
+	irqctx = (union irq_ctx*)per_cpu(softirq_stack_info, cpu).stack;
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;

@@ -252,6 +304,7 @@
 
 void irq_ctx_exit(int cpu)
 {
+	free_irqstacks(cpu);
 	per_cpu(hardirq_ctx, cpu) = NULL;
 }
arch/i386/kernel/process.c +12 -17
Index: stack-paranoia/arch/i386/kernel/process.c
===================================================================
@@ -327,43 +327,38 @@
 struct thread_info *alloc_thread_info(struct task_struct *unused)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
-	struct vm_struct *area;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
+	struct thread_info *info;
 
 	/*
 	 * passing VM_IOREMAP for the sake of alignment is why
 	 * all this is done by hand.
 	 */
-	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	if (!area)
-		return NULL;
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i) {
 		pages[i] = alloc_page(GFP_HIGHUSER);
 		if (!pages[i])
 			goto out_free_pages;
 	}
-	/* implicitly transfer page refcounts to the vm_struct */
-	if (map_vm_area(area, PAGE_KERNEL, &tmp))
-		goto out_remove_area;
-	/* it may be worth poisoning, save thread_info proper */
-	return (struct thread_info *)area->addr;
-out_remove_area:
-	remove_vm_area(area);
+	info = vmap(pages, THREAD_SIZE/PAGE_SIZE, VM_IOREMAP, PAGE_KERNEL);
+	if (info)
+		return info;
 out_free_pages:
-	do {
-		__free_page(pages[--i]);
-	} while (i >= 0);
+	for (--i; i >= 0; --i)
+		__free_page(pages[i]);
 	return NULL;
 }
 
 static void work_free_thread_info(struct work_struct *work)
 {
 	int i;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
 	void *p = work;
 
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
-		__free_page(vmalloc_to_page(p + PAGE_SIZE*i));
-	vfree(p);
+		pages[i] = vmalloc_to_page(p + PAGE_SIZE*i);
+	vunmap(work);
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(pages[i]);
 }
 
 void free_thread_info(struct thread_info *info)
author
#4 lore raw
Quoted 14 lines · Bill Irwin
Quoted 11 lines · Jeremy Fitzhardinge

This fixes two bugs:
- the stack allocation must be marked __cpuinit, since it gets called on resume as well.
- presumably the interrupt stack should be freed on unplug if its going to get reallocated on every plug.

[ Only non-vmalloced stacks tested. ]
Signed-off-by Jeremy Fitzhardinge <[email protected]>
---
 arch/i386/kernel/irq.c |   42 +++++++++++++++++++++++++++++++++++++-----
 1 file changed, 37 insertions(+), 5 deletions(-)
Quoted 2 lines

Updated patch follows. Please add your Signed-off-by: if it meets your approval;

What does it apply to? I'm getting conflicts if I replace my patch with this. Or does it replace one of your patches?

J

#5 lore raw
Quoted 2 lines · Bill Irwin
Quoted 2 lines

Updated patch follows. Please add your Signed-off-by: if it meets your approval;

Quoted 2 lines · Jeremy Fitzhardinge

What does it apply to? I'm getting conflicts if I replace my patch with this. Or does it replace one of your patches?

I had the same question about yours and just brute-force merged. Not a big deal for me to rediff against whatever everyone's working off of.

-- wli

author
#6 lore raw
Quoted 3 lines · Bill Irwin

I had the same question about yours and just brute-force merged. Not a big deal for me to rediff against whatever everyone's working off of.

I picked up one version of your patches you posted a couple of days ago, but I guess you've posted the series multiple times, presumably with differences.

How about posting the series again and I can test it out?

J

#7 lore raw
Quoted 2 lines · Bill Irwin
Quoted 2 lines

I had the same question about yours and just brute-force merged. Not a big deal for me to rediff against whatever everyone's working off of.

Quoted 6 lines · Jeremy Fitzhardinge

I picked up one version of your patches you posted a couple of days ago, but I guess you've posted the series multiple times, presumably with differences.
How about posting the series again and I can test it out?

From the looks of it, I'm totally out-of-synch. Here are the 4 (the

fourth should be partially folded into the third but also partially folded into the first), included as MIME attachments.

As an aside, it looks like failures here need to eventually propagate to __cpu_up(). irq_ctx_init() needs to return a status, and its callers need to check it. irq_ctx_init() probably also needs to be __cpuinit.

-- wli

Subject: dynamically allocate IRQ stacks

Dynamically allocate IRQ stacks in order to conserve memory when using IRQ stacks. cpu_possible_map is not now initialized in such a manner as to provide a meaningful indication of how many CPU's might be in the system, and features to appear in the sequel also require indirection, so they themselves are not allocatable as per_cpu variables, but rather only pointers to them.

Signed-off-by William Irwin <[email protected]>
arch/i386/kernel/irq.c +32 -15
Index: stack-paranoia/arch/i386/kernel/irq.c
===================================================================
@@ -17,9 +17,11 @@
 #include <linux/notifier.h>
 #include <linux/cpu.h>
 #include <linux/delay.h>
+#include <linux/bootmem.h>
 
 #include <asm/apic.h>
 #include <asm/uaccess.h>
+#include <asm/pgtable.h>
 
 DEFINE_PER_CPU(irq_cpustat_t, irq_stat) ____cacheline_internodealigned_in_smp;
 EXPORT_PER_CPU_SYMBOL(irq_stat);

@@ -56,8 +58,8 @@
 	u32                     stack[THREAD_SIZE/sizeof(u32)];
 };
 
-static union irq_ctx *hardirq_ctx[NR_CPUS] __read_mostly;
-static union irq_ctx *softirq_ctx[NR_CPUS] __read_mostly;
+static DEFINE_PER_CPU(union irq_ctx *, hardirq_ctx);
+static DEFINE_PER_CPU(union irq_ctx *, softirq_ctx);
 #endif
 
 /*

@@ -102,7 +104,7 @@
 #ifdef CONFIG_4KSTACKS
 
 	curctx = (union irq_ctx *) current_thread_info();
-	irqctx = hardirq_ctx[smp_processor_id()];
+	irqctx = per_cpu(hardirq_ctx, smp_processor_id());
 
 	/*
 	 * this is where we switch to the IRQ stack. However, if we are

@@ -150,11 +152,24 @@
  * These should really be __section__(".bss.page_aligned") as well, but
  * gcc's 3.0 and earlier don't handle that correctly.
  */
-static char softirq_stack[NR_CPUS * THREAD_SIZE]
-		__attribute__((__aligned__(THREAD_SIZE)));
+static DEFINE_PER_CPU(char *, softirq_stack);
+static DEFINE_PER_CPU(char *, hardirq_stack);
 
-static char hardirq_stack[NR_CPUS * THREAD_SIZE]
-		__attribute__((__aligned__(THREAD_SIZE)));
+static void * __init __alloc_irqstack(int cpu)
+{
+	if (!slab_is_available())
+		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
+						__pa(MAX_DMA_ADDRESS));
+
+	return (void *)__get_free_pages(GFP_KERNEL,
+					ilog2(THREAD_SIZE/PAGE_SIZE));
+}
+
+static void __init alloc_irqstacks(int cpu)
+{
+	per_cpu(softirq_stack, cpu) = __alloc_irqstack(cpu);
+	per_cpu(hardirq_stack, cpu) = __alloc_irqstack(cpu);
+}
 
 /*
  * allocate per-cpu stacks for hardirq and for softirq processing

@@ -163,34 +178,36 @@
 {
 	union irq_ctx *irqctx;
 
-	if (hardirq_ctx[cpu])
+	if (per_cpu(hardirq_ctx, cpu))
 		return;
 
-	irqctx = (union irq_ctx*) &hardirq_stack[cpu*THREAD_SIZE];
+	alloc_irqstacks(cpu);
+
+	irqctx = (union irq_ctx*)per_cpu(hardirq_stack, cpu);
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;
 	irqctx->tinfo.preempt_count     = HARDIRQ_OFFSET;
 	irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
 
-	hardirq_ctx[cpu] = irqctx;
+	per_cpu(hardirq_ctx, cpu) = irqctx;
 
-	irqctx = (union irq_ctx*) &softirq_stack[cpu*THREAD_SIZE];
+	irqctx = (union irq_ctx*)per_cpu(softirq_stack, cpu);
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;
 	irqctx->tinfo.preempt_count     = 0;
 	irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
 
-	softirq_ctx[cpu] = irqctx;
+	per_cpu(softirq_ctx, cpu) = irqctx;
 
 	printk("CPU %u irqstacks, hard=%p soft=%p\n",
-		cpu,hardirq_ctx[cpu],softirq_ctx[cpu]);
+		cpu, per_cpu(hardirq_ctx, cpu), per_cpu(softirq_ctx, cpu));
 }
 
 void irq_ctx_exit(int cpu)
 {
-	hardirq_ctx[cpu] = NULL;
+	per_cpu(hardirq_ctx, cpu) = NULL;
 }
 
 extern asmlinkage void __do_softirq(void);

@@ -209,7 +226,7 @@
 
 	if (local_softirq_pending()) {
 		curctx = current_thread_info();
-		irqctx = softirq_ctx[smp_processor_id()];
+		irqctx = per_cpu(softirq_ctx, smp_processor_id());
 		irqctx->tinfo.task = curctx->task;
 		irqctx->tinfo.previous_esp = current_stack_pointer;

IRQ stacks are a valuable stability feature. This patch makes them unconditional, as there is no circumstance under which they do not improve stability and they have no meaningful performance impact.

Signed-off-by William Irwin <[email protected]>
include/asm-i386/irq.h +3 -8
Index: stack-paranoia/include/asm-i386/irq.h
===================================================================
@@ -24,14 +24,9 @@
 # define ARCH_HAS_NMI_WATCHDOG		/* See include/linux/nmi.h */
 #endif
 
-#ifdef CONFIG_4KSTACKS
-  extern void irq_ctx_init(int cpu);
-  extern void irq_ctx_exit(int cpu);
-# define __ARCH_HAS_DO_SOFTIRQ
-#else
-# define irq_ctx_init(cpu) do { } while (0)
-# define irq_ctx_exit(cpu) do { } while (0)
-#endif
+void irq_ctx_init(int);
+void irq_ctx_exit(int);
+#define __ARCH_HAS_DO_SOFTIRQ
 
 #ifdef CONFIG_IRQBALANCE
 extern int irqbalance_disable(char *str);
arch/i386/kernel/irq.c +0 -11
Index: stack-paranoia/arch/i386/kernel/irq.c
===================================================================
@@ -49,7 +49,6 @@
 #endif
 }
 
-#ifdef CONFIG_4KSTACKS
 /*
  * per-CPU IRQ handling contexts (thread information and stack)
  */

@@ -60,7 +59,6 @@
 
 static DEFINE_PER_CPU(union irq_ctx *, hardirq_ctx);
 static DEFINE_PER_CPU(union irq_ctx *, softirq_ctx);
-#endif
 
 /*
  * do_IRQ handles all normal device IRQ's (the special

@@ -73,10 +71,8 @@
 	/* high bit used in ret_from_ code */
 	int irq = ~regs->orig_eax;
 	struct irq_desc *desc = irq_desc + irq;
-#ifdef CONFIG_4KSTACKS
 	union irq_ctx *curctx, *irqctx;
 	u32 *isp;
-#endif
 
 	if (unlikely((unsigned)irq >= NR_IRQS)) {
 		printk(KERN_EMERG "%s: cannot handle IRQ %d\n",

@@ -100,9 +96,6 @@
 		}
 	}
 #endif
-
-#ifdef CONFIG_4KSTACKS
-
 	curctx = (union irq_ctx *) current_thread_info();
 	irqctx = per_cpu(hardirq_ctx, smp_processor_id());
 

@@ -138,7 +131,6 @@
 			: "memory", "cc"
 		);
 	} else
-#endif
 		desc->handle_irq(irq, desc);
 
 	irq_exit();

@@ -146,8 +138,6 @@
 	return 1;
 }
 
-#ifdef CONFIG_4KSTACKS
-
 /*
  * These should really be __section__(".bss.page_aligned") as well, but
  * gcc's 3.0 and earlier don't handle that correctly.

@@ -251,7 +241,6 @@
 }
 
 EXPORT_SYMBOL(do_softirq);
-#endif
 
 /*
  * Interrupt statistics:

This patch introduces CONFIG_DEBUG_STACK, which vmalloc()'s task and IRQ stacks in order to establish guard pages. In such a manner any stack overflow that references pages immediately adjacent to the stack is immediately trapped with a fault, which precludes silent memory corruption or difficult-to-decipher failure modes resulting from stack corruption.

It furthermore adds a check to __pa() to catch drivers trying to DMA off the stack, which more generally flags incorrect attempts to use __pa() on vmallocspace addresses.

Signed-off-by William Irwin <[email protected]>
arch/i386/Kconfig.debug +10 -0
Index: stack-paranoia/arch/i386/Kconfig.debug
===================================================================
@@ -35,6 +35,16 @@
 
 	  This option will slow down process creation somewhat.
 
+config DEBUG_STACK
+	bool "Debug stack overflows"
+	depends on DEBUG_KERNEL
+	help
+	  Allocates the stack physically discontiguously and from high
+	  memory. Furthermore an unmapped guard page follows the stack,
+	  which results in immediately trapping stack overflows instead
+	  of silent corruption. This is not for end-users. It's intended
+	  to trigger fatal system errors under various forms of stack abuse.
+
 comment "Page alloc debug is incompatible with Software Suspend on i386"
 	depends on DEBUG_KERNEL && SOFTWARE_SUSPEND
arch/i386/kernel/process.c +53 -0

65 lines not on this page · raw message

include/asm-i386/module.h +8 -1
Index: stack-paranoia/include/asm-i386/module.h
===================================================================
@@ -68,6 +68,13 @@
 #define MODULE_STACKSIZE ""
 #endif
 
-#define MODULE_ARCH_VERMAGIC MODULE_PROC_FAMILY MODULE_STACKSIZE
+#ifdef CONFIG_DEBUG_STACK
+#define MODULE_DEBUG_STACK "DEBUG_STACKS "
+#else
+#define MODULE_DEBUG_STACK ""
+#endif
+
+#define MODULE_ARCH_VERMAGIC MODULE_PROC_FAMILY MODULE_STACKSIZE \
+		MODULE_DEBUG_STACK
 
 #endif /* _ASM_I386_MODULE_H */
include/asm-i386/thread_info.h +6 -0
Index: stack-paranoia/include/asm-i386/thread_info.h
===================================================================
@@ -94,6 +94,11 @@
 }
 
 /* thread information allocation */
+#ifdef CONFIG_DEBUG_STACK
+struct task_struct;
+struct thread_info *alloc_thread_info(struct task_struct *);
+void free_thread_info(struct thread_info *);
+#else /* !CONFIG_DEBUG_STACK */
 #ifdef CONFIG_DEBUG_STACK_USAGE
 #define alloc_thread_info(tsk) kzalloc(THREAD_SIZE, GFP_KERNEL)
 #else

@@ -101,6 +106,7 @@
 #endif
 
 #define free_thread_info(info)	kfree(info)
+#endif /* !CONFIG_DEBUG_STACK */
 
 #else /* !__ASSEMBLY__ */
arch/i386/kernel/doublefault.c +1 -1
Index: stack-paranoia/arch/i386/kernel/doublefault.c
===================================================================
@@ -62,5 +62,5 @@
 	.ss		= __KERNEL_DS,
 	.ds		= __USER_DS,
 
-	.__cr3		= __pa(swapper_pg_dir)
+	.__cr3		= (unsigned long)swapper_pg_dir - PAGE_OFFSET,
 };
arch/i386/kernel/irq.c +56 -1

75 lines not on this page · raw message

arch/i386/mm/pgtable.c +12 -0
Index: stack-paranoia/arch/i386/mm/pgtable.c
===================================================================
@@ -181,6 +181,18 @@
 #endif
 }
 
+#ifdef CONFIG_DEBUG_STACK
+unsigned long __kvaddr_to_paddr(unsigned long kvaddr)
+{
+	if (high_memory)
+		BUG_ON(kvaddr >= VMALLOC_START);
+	else
+		BUG_ON(kvaddr >= (unsigned long)__va(MAXMEM));
+	return kvaddr - PAGE_OFFSET;
+}
+EXPORT_SYMBOL(__kvaddr_to_paddr);
+#endif /* CONFIG_DEBUG_STACK */
+
 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 {
 	return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
include/asm-i386/page.h +8 -2
Index: stack-paranoia/include/asm-i386/page.h
===================================================================
@@ -118,11 +118,17 @@
 #define __PAGE_OFFSET		((unsigned long)CONFIG_PAGE_OFFSET)
 #endif
 
-
 #define PAGE_OFFSET		((unsigned long)__PAGE_OFFSET)
+
+#if defined(CONFIG_DEBUG_STACK) && !defined(__ASSEMBLY__)
+unsigned long __kvaddr_to_paddr(unsigned long);
+#define __pa(x)			__kvaddr_to_paddr((unsigned long)(x))
+#else /* !CONFIG_DEBUG_STACK */
+#define __pa(x)			((unsigned long)(x)-PAGE_OFFSET)
+#endif /* !CONFIG_DEBUG_STACK */
+
 #define VMALLOC_RESERVE		((unsigned long)__VMALLOC_RESERVE)
 #define MAXMEM			(-__PAGE_OFFSET-__VMALLOC_RESERVE)
-#define __pa(x)			((unsigned long)(x)-PAGE_OFFSET)
 /* __pa_symbol should be used for C visible symbols.
    This seems to be the official gcc blessed way to do such arithmetic. */
 #define __pa_symbol(x)          __pa(RELOC_HIDE((unsigned long)(x),0))
arch/i386/kernel/irq.c +103 -42

201 lines not on this page · raw message

arch/i386/kernel/process.c +12 -17
Index: stack-paranoia/arch/i386/kernel/process.c
===================================================================
@@ -327,43 +327,38 @@
 struct thread_info *alloc_thread_info(struct task_struct *unused)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
-	struct vm_struct *area;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
+	struct thread_info *info;
 
 	/*
 	 * passing VM_IOREMAP for the sake of alignment is why
 	 * all this is done by hand.
 	 */
-	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	if (!area)
-		return NULL;
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i) {
 		pages[i] = alloc_page(GFP_HIGHUSER);
 		if (!pages[i])
 			goto out_free_pages;
 	}
-	/* implicitly transfer page refcounts to the vm_struct */
-	if (map_vm_area(area, PAGE_KERNEL, &tmp))
-		goto out_remove_area;
-	/* it may be worth poisoning, save thread_info proper */
-	return (struct thread_info *)area->addr;
-out_remove_area:
-	remove_vm_area(area);
+	info = vmap(pages, THREAD_SIZE/PAGE_SIZE, VM_IOREMAP, PAGE_KERNEL);
+	if (info)
+		return info;
 out_free_pages:
-	do {
-		__free_page(pages[--i]);
-	} while (i >= 0);
+	for (--i; i >= 0; --i)
+		__free_page(pages[i]);
 	return NULL;
 }
 
 static void work_free_thread_info(struct work_struct *work)
 {
 	int i;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
 	void *p = work;
 
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
-		__free_page(vmalloc_to_page(p + PAGE_SIZE*i));
-	vfree(p);
+		pages[i] = vmalloc_to_page(p + PAGE_SIZE*i);
+	vunmap(work);
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(pages[i]);
 }
 
 void free_thread_info(struct thread_info *info)
#8 lore raw
Quoted 3 lines · Bill Irwin

As an aside, it looks like failures here need to eventually propagate to __cpu_up(). irq_ctx_init() needs to return a status, and its callers need to check it. irq_ctx_init() probably also needs to be __cpuinit.

Ignoring the general irq_ctx_init() cleanup this takes care of it.

arch/i386/kernel/irq.c +20 -24
Index: stack-paranoia/arch/i386/kernel/irq.c
===================================================================
@@ -279,35 +279,31 @@
 /*
  * allocate per-cpu stacks for hardirq and for softirq processing
  */
-void irq_ctx_init(int cpu)
+static void __cpuinit
+__irq_ctx_init(union irq_ctx **irqctx, struct irq_stack_info *info,
+	       int cpu, int preempt_count)
 {
-	union irq_ctx *irqctx;
+	*irqctx = (union irq_ctx*)info->stack;
+	(*irqctx)->tinfo.task              = NULL;
+	(*irqctx)->tinfo.exec_domain       = NULL;
+	(*irqctx)->tinfo.cpu               = cpu;
+	(*irqctx)->tinfo.preempt_count     = preempt_count;
+	(*irqctx)->tinfo.addr_limit        = MAKE_MM_SEG(0);
+}
 
+int irq_ctx_init(int cpu)
+{
 	if (per_cpu(hardirq_ctx, cpu))
-		return;
-
-	alloc_irqstacks(cpu);
-
-	irqctx = (union irq_ctx*)per_cpu(hardirq_stack_info, cpu).stack;
-	irqctx->tinfo.task              = NULL;
-	irqctx->tinfo.exec_domain       = NULL;
-	irqctx->tinfo.cpu               = cpu;
-	irqctx->tinfo.preempt_count     = HARDIRQ_OFFSET;
-	irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
-
-	per_cpu(hardirq_ctx, cpu) = irqctx;
-
-	irqctx = (union irq_ctx*)per_cpu(softirq_stack_info, cpu).stack;
-	irqctx->tinfo.task              = NULL;
-	irqctx->tinfo.exec_domain       = NULL;
-	irqctx->tinfo.cpu               = cpu;
-	irqctx->tinfo.preempt_count     = 0;
-	irqctx->tinfo.addr_limit        = MAKE_MM_SEG(0);
-
-	per_cpu(softirq_ctx, cpu) = irqctx;
-
+		return 0;
+	if (alloc_irqstacks(cpu))
+		return -1;
+	__irq_ctx_init(&per_cpu(hardirq_ctx, cpu),
+		       &per_cpu(hardirq_stack_info, cpu), cpu, HARDIRQ_OFFSET);
+	__irq_ctx_init(&per_cpu(softirq_ctx, cpu),
+		       &per_cpu(softirq_stack_info, cpu), cpu, 0);
 	printk("CPU %u irqstacks, hard=%p soft=%p\n",
 		cpu, per_cpu(hardirq_ctx, cpu), per_cpu(softirq_ctx, cpu));
+	return 0;
 }
 
 void irq_ctx_exit(int cpu)
include/asm-i386/irq.h +1 -1
Index: stack-paranoia/include/asm-i386/irq.h
===================================================================
@@ -24,7 +24,7 @@
 # define ARCH_HAS_NMI_WATCHDOG		/* See include/linux/nmi.h */
 #endif
 
-void irq_ctx_init(int);
+int irq_ctx_init(int);
 void irq_ctx_exit(int);
 #define __ARCH_HAS_DO_SOFTIRQ
arch/i386/kernel/i8259.c +4 -1
Index: stack-paranoia/arch/i386/kernel/i8259.c
===================================================================
@@ -417,5 +417,8 @@
 	if (boot_cpu_data.hard_math && !cpu_has_fpu)
 		setup_irq(FPU_IRQ, &fpu_irq);
 
-	irq_ctx_init(smp_processor_id());
+	if (irq_ctx_init(smp_processor_id()))
+		printk(KERN_INFO
+		       "Couldn't allocate IRQ context for CPU %d\n",
+		       smp_processor_id());
 }
arch/i386/kernel/smpboot.c +5 -3
Index: stack-paranoia/arch/i386/kernel/smpboot.c
===================================================================
@@ -828,9 +828,11 @@
 	printk("Booting processor %d/%d eip %lx\n", cpu, apicid, start_eip);
 	/* Stack for startup_32 can be just as for start_secondary onwards */
 	stack_start.esp = (void *) idle->thread.esp;
-
-	irq_ctx_init(cpu);
-
+	if (irq_ctx_init(cpu)) {
+		printk(KERN_INFO
+		       "Couldn't allocate IRQ contexts for CPU %d\n", cpu);
+		return -1;
+	}
 	x86_cpu_to_apicid[cpu] = apicid;
 	/*
 	 * This grunge runs the startup process for
arch/i386/mach-voyager/voyager_smp.c +6 -2
Index: stack-paranoia/arch/i386/mach-voyager/voyager_smp.c
===================================================================
@@ -589,8 +589,12 @@
 		return;
 	}
 
-	irq_ctx_init(cpu);
-
+	if (irq_ctx_init(cpu)) {
+		printk(KERN_INFO
+		       "Couldn't allocate IRQ context for CPU %d\n", cpu);
+		cpucount--;
+		return;
+	}
 	/* Note: Don't modify initial ss override */
 	VDEBUG(("VOYAGER SMP: Booting CPU%d at 0x%lx[%x:%x], stack %p\n", cpu, 
 		(unsigned long)hijack_source.val, hijack_source.idt.Segment,
#9 lore raw
In reply to Bill Irwin
Quoted 5 lines · Bill Irwin

Updated patch follows. Please add your Signed-off-by: if it meets your approval; I am operating on the assumption I should never do so myself. I'm a bit unsure of how to handle cpu 0 vs. potential freeing of per_cpu areas and error returns from __cpuinit affairs, but anyhow:

Quoted 7 lines

This fixes three bugs:
- the stack allocation must be marked __cpuinit, since it gets called on resume as well.
- presumably the interrupt stack should be freed on unplug if its going to get reallocated on every plug.
- the vm_struct got leaked by thread info freeing callbacks.

Signed-off-by William Irwin <[email protected]>

I think just normalizing cpu 0's stack suffices here. Still no idea what to do about backpropagating errors from __cpuinit bits just yet.

arch/i386/kernel/irq.c +103 -42
Index: stack-paranoia/arch/i386/kernel/irq.c
===================================================================
@@ -142,78 +142,138 @@
  * These should really be __section__(".bss.page_aligned") as well, but
  * gcc's 3.0 and earlier don't handle that correctly.
  */
-static DEFINE_PER_CPU(char *, softirq_stack);
-static DEFINE_PER_CPU(char *, hardirq_stack);
-
+struct irq_stack_info {
+	char *stack;
 #ifdef CONFIG_DEBUG_STACK
-static void * __init irq_remap_stack(void *stack)
-{
-	int i;
 	struct page *pages[THREAD_SIZE/PAGE_SIZE];
+#endif /* CONFIG_DEBUG_STACK */
+};
+static DEFINE_PER_CPU(struct irq_stack_info, softirq_stack_info);
+static DEFINE_PER_CPU(struct irq_stack_info, hardirq_stack_info);
 
-	for (i = 0; i < ARRAY_SIZE(pages); ++i)
-		pages[i] =  virt_to_page(stack + PAGE_SIZE*i);
-	return vmap(pages, THREAD_SIZE/PAGE_SIZE, VM_IOREMAP, PAGE_KERNEL);
-}
-
-static int __init irq_guard_cpu0(void)
+#ifdef CONFIG_DEBUG_STACK
+static int __init irq_remap_stack(struct irq_stack_info *info)
 {
+	struct page *page, *pages[THREAD_SIZE/PAGE_SIZE];
 	unsigned long flags;
+	int i;
 	void *tmp;
 
-	tmp = irq_remap_stack(per_cpu(softirq_stack, 0));
-	if (!tmp)
-		return -ENOMEM;
-	else {
-		local_irq_save(flags);
-		per_cpu(softirq_stack, 0) = tmp;
-		local_irq_restore(flags);
+	for (i = 0; i < ARRAY_SIZE(pages); ++i) {
+		pages[i] = alloc_page(GFP_HIGHUSER);
+		if (!pages[i])
+			goto out_free_pages;
 	}
-	tmp = irq_remap_stack(per_cpu(hardirq_stack, 0));
+	tmp = vmap(pages, ARRAY_SIZE(info->pages), VM_IOREMAP, PAGE_KERNEL);
 	if (!tmp)
-		return -ENOMEM;
+		goto out_free_pages;
 	else {
 		local_irq_save(flags);
-		per_cpu(hardirq_stack, 0) = tmp;
+		memcpy(info->pages, pages, sizeof(pages));
+		page = virt_to_page(info->stack);
+		for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i) {
+			ClearPageReserved(&page[i]);
+			init_page_count(&page[i]);
+			__free_page(&page[i]);
+		}
+		info->stack = tmp;
 		local_irq_restore(flags);
 	}
 	return 0;
+out_free_pages:
+	for (--i; i >= 0; --i)
+		__free_page(pages[i]);
+	return -1;
+}
+
+static int __init irq_guard_cpu0(void)
+{
+	if (irq_remap_stack(&per_cpu(softirq_stack_info, 0)))
+		return -ENOMEM;
+	if (irq_remap_stack(&per_cpu(hardirq_stack_info, 0)))
+		return -ENOMEM;
+	return 0;
 }
 core_initcall(irq_guard_cpu0);
 
-static void * __init __alloc_irqstack(int cpu)
+static int __cpuinit __alloc_irqstack(int cpu, struct irq_stack_info *info)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
-	struct vm_struct *area;
 
-	if (!slab_is_available())
-		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
+	if (!slab_is_available()) {
+		info->stack = __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
 						__pa(MAX_DMA_ADDRESS));
+		info->pages[0] = virt_to_page(info->stack);
+		for (i = 1; i < ARRAY_SIZE(info->pages); ++i)
+			info->pages[i] = info->pages[0] + i;
+		return 0;
+	}
+	for (i = 0; i < ARRAY_SIZE(info->pages); ++i) {
+		info->pages[i] = alloc_page(GFP_HIGHUSER);
+		if (!info->pages[i])
+			goto out;
+	}
+	info->stack = vmap(info->pages, ARRAY_SIZE(info->pages), VM_IOREMAP,
+								 PAGE_KERNEL);
+	if (info->stack)
+		return 0;
+out:
+	for (--i; i >= 0; --i) {
+		__free_page(info->pages[i]);
+		info->pages[i] = NULL;
+	}
+	return -1;
+}
 
-	/* failures here are unrecoverable anyway */
-	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	for (i = 0; i < ARRAY_SIZE(pages); ++i)
-		pages[i] = alloc_page(GFP_HIGHUSER);
-	map_vm_area(area, PAGE_KERNEL, &tmp);
-	return area->addr;
+static void __cpuinit __free_irqstack(int cpu, struct irq_stack_info *info)
+{
+	int i;
+
+	vunmap(info->stack);
+	for (i = 0; i < ARRAY_SIZE(info->pages); ++i) {
+		if (!PageReserved(info->pages[i]))
+			__free_page(info->pages[i]);
+		info->pages[i] = NULL;
+	}
+	info->stack = NULL;
 }
 #else /* !CONFIG_DEBUG_STACK */
-static void * __init __alloc_irqstack(int cpu)
+static int __cpuinit __alloc_irqstack(int cpu, struct irq_stack_info *info)
 {
 	if (!slab_is_available())
-		return __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
+		info->stack = __alloc_bootmem(THREAD_SIZE, THREAD_SIZE,
 						__pa(MAX_DMA_ADDRESS));
-
-	return (void *)__get_free_pages(GFP_KERNEL,
+	else
+		info->stack = (void *)__get_free_pages(GFP_KERNEL,
 					ilog2(THREAD_SIZE/PAGE_SIZE));
+	return info->stack ? 0 : -1;
+}
+
+static void __cpuinit __free_irqstack(int cpu, struct irq_stack_info *info)
+{
+	struct page *page = virt_to_page(info->stack);
+
+	if (!PageReserved(page))
+		__free_pages(page, ilog2(THREAD_SIZE/PAGE_SIZE));
+	info->stack = NULL;
 }
 #endif /* !CONFIG_DEBUG_STACK */
 
-static void __init alloc_irqstacks(int cpu)
+static int __cpuinit alloc_irqstacks(int cpu)
+{
+	if (__alloc_irqstack(cpu, &per_cpu(softirq_stack_info, cpu)))
+		return -1;
+	if (__alloc_irqstack(cpu, &per_cpu(hardirq_stack_info, cpu))) {
+		__free_irqstack(cpu, &per_cpu(softirq_stack_info, cpu));
+		return -1;
+	}
+	return 0;
+}
+
+static void __cpuinit free_irqstacks(int cpu)
 {
-	per_cpu(softirq_stack, cpu) = __alloc_irqstack(cpu);
-	per_cpu(hardirq_stack, cpu) = __alloc_irqstack(cpu);
+	__free_irqstack(cpu, &per_cpu(softirq_stack_info, cpu));
+	__free_irqstack(cpu, &per_cpu(hardirq_stack_info, cpu));
 }
 
 /*

@@ -228,7 +288,7 @@
 
 	alloc_irqstacks(cpu);
 
-	irqctx = (union irq_ctx*)per_cpu(hardirq_stack, cpu);
+	irqctx = (union irq_ctx*)per_cpu(hardirq_stack_info, cpu).stack;
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;

@@ -237,7 +297,7 @@
 
 	per_cpu(hardirq_ctx, cpu) = irqctx;
 
-	irqctx = (union irq_ctx*)per_cpu(softirq_stack, cpu);
+	irqctx = (union irq_ctx*)per_cpu(softirq_stack_info, cpu).stack;
 	irqctx->tinfo.task              = NULL;
 	irqctx->tinfo.exec_domain       = NULL;
 	irqctx->tinfo.cpu               = cpu;

@@ -252,6 +312,7 @@
 
 void irq_ctx_exit(int cpu)
 {
+	free_irqstacks(cpu);
 	per_cpu(hardirq_ctx, cpu) = NULL;
 }
arch/i386/kernel/process.c +12 -17
Index: stack-paranoia/arch/i386/kernel/process.c
===================================================================
@@ -327,43 +327,38 @@
 struct thread_info *alloc_thread_info(struct task_struct *unused)
 {
 	int i;
-	struct page *pages[THREAD_SIZE/PAGE_SIZE], **tmp = pages;
-	struct vm_struct *area;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
+	struct thread_info *info;
 
 	/*
 	 * passing VM_IOREMAP for the sake of alignment is why
 	 * all this is done by hand.
 	 */
-	area = get_vm_area(THREAD_SIZE, VM_IOREMAP);
-	if (!area)
-		return NULL;
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i) {
 		pages[i] = alloc_page(GFP_HIGHUSER);
 		if (!pages[i])
 			goto out_free_pages;
 	}
-	/* implicitly transfer page refcounts to the vm_struct */
-	if (map_vm_area(area, PAGE_KERNEL, &tmp))
-		goto out_remove_area;
-	/* it may be worth poisoning, save thread_info proper */
-	return (struct thread_info *)area->addr;
-out_remove_area:
-	remove_vm_area(area);
+	info = vmap(pages, THREAD_SIZE/PAGE_SIZE, VM_IOREMAP, PAGE_KERNEL);
+	if (info)
+		return info;
 out_free_pages:
-	do {
-		__free_page(pages[--i]);
-	} while (i >= 0);
+	for (--i; i >= 0; --i)
+		__free_page(pages[i]);
 	return NULL;
 }
 
 static void work_free_thread_info(struct work_struct *work)
 {
 	int i;
+	struct page *pages[THREAD_SIZE/PAGE_SIZE];
 	void *p = work;
 
 	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
-		__free_page(vmalloc_to_page(p + PAGE_SIZE*i));
-	vfree(p);
+		pages[i] = vmalloc_to_page(p + PAGE_SIZE*i);
+	vunmap(work);
+	for (i = 0; i < THREAD_SIZE/PAGE_SIZE; ++i)
+		__free_page(pages[i]);
 }
 
 void free_thread_info(struct thread_info *info)