Skip to contentSkip to conversation

cpu_init is called during resume

PATCH Pavel Machek 3 participants 6 messages Aug 31, 2006 – Sep 3, 2006

lore
Share & tools
lore
#0 lore raw

cpu_init() is called during resume, at time when GFP_KERNEL is not available. This silences warning, and adds few small cleanups.

---
commit 0c23b1939d38a6bbdf1fd22078bd4fd3ec932c68 tree 66e19bcca9940e3cab2df78aaef010e51260cb9f parent de9ee8a9c1e2cf18adee5d1569daa19deedc7c87 author <[email protected]> Thu, 31 Aug 2006 15:55:00 +0200 committer <[email protected]> Thu, 31 Aug 2006 15:55:00 +0200

 arch/i386/kernel/cpu/common.c |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)
arch/i386/kernel/cpu/common.c +3 -3
index 70c87de..d0565d3 100644
@@ -591,10 +591,10 @@ void __init early_cpu_init(void)
 void __cpuinit cpu_init(void)
 {
 	int cpu = smp_processor_id();
-	struct tss_struct * t = &per_cpu(init_tss, cpu);
+	struct tss_struct *t = &per_cpu(init_tss, cpu);
 	struct thread_struct *thread = &current->thread;
 	struct desc_struct *gdt;
-	__u32 stk16_off = (__u32)&per_cpu(cpu_16bit_stack, cpu);
+	u32 stk16_off = (u32)&per_cpu(cpu_16bit_stack, cpu);
 	struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
 
 	if (cpu_test_and_set(cpu, cpu_initialized)) {
@@ -629,7 +629,7 @@ void __cpuinit cpu_init(void)
 		/* alloc_bootmem_pages panics on failure, so no check */
 		memset(gdt, 0, PAGE_SIZE);
 	} else {
-		gdt = (struct desc_struct *)get_zeroed_page(GFP_KERNEL);
+		gdt = (struct desc_struct *)get_zeroed_page(GFP_ATOMIC);
 		if (unlikely(!gdt)) {
 			printk(KERN_CRIT "CPU%d failed to allocate GDT\n", cpu);
 			for (;;)
Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#1 lore raw
Quoted 3 lines · Pavel Machek

cpu_init() is called during resume, at time when GFP_KERNEL is not available. This silences warning, and adds few small cleanups.

I presume this is resume from disk. If you're doing resume from RAM, all the CPU-related stuff should already be allocated, unless you're bringing up a new CPU which wasn't previously there, right?

What's the call path for this on resume? In my i386-pda patches, I've rearranged this so that the secondary CPU's GDT (and PDA) are pre-allocated on the boot CPU. Does this help this case, or would they still need to be atomic allocations?

And wouldn't making these allocations atomic make real CPU hotplug (ie, on an active, running system) more likely to fail? This code doesn't deal with allocation failure very elegantly.

J

#2 lore raw

Hi!

Quoted 7 lines
Quoted 3 lines

cpu_init() is called during resume, at time when GFP_KERNEL is not available. This silences warning, and adds few small cleanups.

I presume this is resume from disk. If you're doing resume from RAM, all the CPU-related stuff should already be allocated, unless you're bringing up a new CPU which wasn't previously there, right?

We are doing virtual cpu hotplug/unplug... actually suspend to RAM *and* disk. Just try it :-).

Quoted 3 lines

And wouldn't making these allocations atomic make real CPU hotplug (ie, on an active, running system) more likely to fail? This code doesn't deal with allocation failure very elegantly.

Hmm... well hopefully free_pages_min will cover this case.
Pavel

Signature · 2 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
#3 lore raw
Quoted 3 lines · Pavel Machek

We are doing virtual cpu hotplug/unplug... actually suspend to RAM *and* disk. Just try it :-).

Yes, I know, but if you're replugging a CPU it should already have a GDT allocated, so the kmalloc() won't be called in the resume case. So if there's a problem, it can only be on the resume-from-disk path, on the assumption that that secondary CPU hasn't already been brought up.

J

#4 lore raw
Quoted 9 lines · Jeremy Fitzhardinge
Quoted 3 lines · Pavel Machek

We are doing virtual cpu hotplug/unplug... actually suspend to RAM *and* disk. Just try it :-).

Yes, I know, but if you're replugging a CPU it should already have a GDT allocated, so the kmalloc() won't be called in the resume case. So if there's a problem, it can only be on the resume-from-disk path, on the assumption that that secondary CPU hasn't already been brought up.

I think gdt is destroyed on cpu unplug, no? I guess I'll have to redo my testing...

Signature · 5 lines
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

-- 
VGER BF report: H 0.410571
#5 lore raw
Quoted 3 lines · Pavel Machek

I think gdt is destroyed on cpu unplug, no? I guess I'll have to redo my testing...

No, I don't think it is. arch/i386/cpu/common.c:cpu_uninit() doesn't free anything.

J

Signature · 1 line
VGER BF report: H 0