The patch seems missed.
GFP_KERNEL isn't ok for runtime (cpu hotplug).
Signed-off-by: Shaohua Li<[email protected]>
---
linux-2.6.15-root/arch/i386/kernel/cpu/common.c | 2 +-
1 files changed, 1 insertion(+), 1 deletion(-)
diff -puN arch/i386/kernel/cpu/common.c~cpuhp arch/i386/kernel/cpu/common.c
--- linux-2.6.15/arch/i386/kernel/cpu/common.c~cpuhp 2006-03-14 12:13:43.000000000 +0800
+++ linux-2.6.15-root/arch/i386/kernel/cpu/common.c 2006-03-14 12:14:12.000000000 +0800
@@ -605,7 +605,7 @@ void __devinit 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 (;;)
_
Shaohua Li <[email protected]> wrote:
>
> The patch seems missed.
> GFP_KERNEL isn't ok for runtime (cpu hotplug).
>
> Signed-off-by: Shaohua Li<[email protected]>
> ---
>
> linux-2.6.15-root/arch/i386/kernel/cpu/common.c | 2 +-
> 1 files changed, 1 insertion(+), 1 deletion(-)
>
> diff -puN arch/i386/kernel/cpu/common.c~cpuhp arch/i386/kernel/cpu/common.c
> --- linux-2.6.15/arch/i386/kernel/cpu/common.c~cpuhp 2006-03-14 12:13:43.000000000 +0800
> +++ linux-2.6.15-root/arch/i386/kernel/cpu/common.c 2006-03-14 12:14:12.000000000 +0800
> @@ -605,7 +605,7 @@ void __devinit 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 (;;)
This isn't good. GFP_ATOMIC can fail, and if it does, we'll lose this CPU
and probably the entire machine. It's OK to do this during initial boot,
but not so OK to do it during CPU hotplug.
So can we please fix it better?
You don't describe _why_ the CPU is running atomically here - I wish you had.
One approach would be to allocate the page earlier, before we enter the
atomic region, and to pass that page down to cpu_init(), or to save a
pointer to it in an array of page*'s somewhere.
On Wed, 2006-03-22 at 21:32 -0800, Andrew Morton wrote:
> Shaohua Li <[email protected]> wrote:
> >
> > The patch seems missed.
> > GFP_KERNEL isn't ok for runtime (cpu hotplug).
> >
> > Signed-off-by: Shaohua Li<[email protected]>
> > ---
> >
> > linux-2.6.15-root/arch/i386/kernel/cpu/common.c | 2 +-
> > 1 files changed, 1 insertion(+), 1 deletion(-)
> >
> > diff -puN arch/i386/kernel/cpu/common.c~cpuhp arch/i386/kernel/cpu/common.c
> > --- linux-2.6.15/arch/i386/kernel/cpu/common.c~cpuhp 2006-03-14 12:13:43.000000000 +0800
> > +++ linux-2.6.15-root/arch/i386/kernel/cpu/common.c 2006-03-14 12:14:12.000000000 +0800
> > @@ -605,7 +605,7 @@ void __devinit 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 (;;)
>
>
> This isn't good. GFP_ATOMIC can fail, and if it does, we'll lose this CPU
> and probably the entire machine. It's OK to do this during initial boot,
> but not so OK to do it during CPU hotplug.
>
> So can we please fix it better?
>
> You don't describe _why_ the CPU is running atomically here - I wish you had.
>
> One approach would be to allocate the page earlier, before we enter the
> atomic region, and to pass that page down to cpu_init(), or to save a
> pointer to it in an array of page*'s somewhere.
Thanks for the suggestion. Here is the updated patch.
The patch fixes two issues:
1. cpu_init is called with interrupt disabled. Allocating gdt table
there isn't good at runtime.
2. gdt table page cause memory leak in CPU hotplug case.
Signed-off-by: Shaohua Li <[email protected]>
---
linux-2.6.16-root/arch/i386/kernel/cpu/common.c | 8 +++++++-
linux-2.6.16-root/arch/i386/kernel/smpboot.c | 13 +++++++++++++
2 files changed, 20 insertions(+), 1 deletion(-)
diff -puN arch/i386/kernel/cpu/common.c~gdt_table arch/i386/kernel/cpu/common.c
--- linux-2.6.16/arch/i386/kernel/cpu/common.c~gdt_table 2006-03-22 11:57:56.000000000 +0800
+++ linux-2.6.16-root/arch/i386/kernel/cpu/common.c 2006-03-22 12:10:04.000000000 +0800
@@ -594,6 +594,12 @@ void __devinit cpu_init(void)
set_in_cr4(X86_CR4_TSD);
}
+ /* The CPU hotplug case */
+ if (cpu_gdt_descr->address) {
+ gdt = (struct desc_struct *)cpu_gdt_descr->address;
+ memset(gdt, 0, PAGE_SIZE);
+ goto old_gdt;
+ }
/*
* This is a horrible hack to allocate the GDT. The problem
* is that cpu_init() is called really early for the boot CPU
@@ -612,7 +618,7 @@ void __devinit cpu_init(void)
local_irq_enable();
}
}
-
+old_gdt:
/*
* Initialize the per-CPU GDT with the boot GDT,
* and set up the GDT descriptor:
diff -puN arch/i386/kernel/smpboot.c~gdt_table arch/i386/kernel/smpboot.c
--- linux-2.6.16/arch/i386/kernel/smpboot.c~gdt_table 2006-03-22 12:01:41.000000000 +0800
+++ linux-2.6.16-root/arch/i386/kernel/smpboot.c 2006-03-22 12:18:28.000000000 +0800
@@ -1027,6 +1027,7 @@ int __devinit smp_prepare_cpu(int cpu)
struct warm_boot_cpu_info info;
struct work_struct task;
int apicid, ret;
+ struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
lock_cpu_hotplug();
@@ -1045,6 +1046,18 @@ int __devinit smp_prepare_cpu(int cpu)
goto exit;
}
+ /*
+ * the CPU isn't initialized at boot time, allocate gdt table here.
+ * cpu_init will initialize it
+ */
+ if (!cpu_gdt_descr->address) {
+ cpu_gdt_descr->address = get_zeroed_page(GFP_KERNEL);
+ if (!cpu_gdt_descr->address)
+ printk(KERN_CRIT "CPU%d failed to allocate GDT\n", cpu);
+ ret = -ENOMEM;
+ goto exit;
+ }
+
info.complete = &done;
info.apicid = apicid;
info.cpu = cpu;
_