[KVM][PATCH] smp_processor_id() and sleeping functions used in invalid context

[Luca Tettamanti <kronos.it@gmail.com>]

Hello,
I'm testing KVM on a Core2 CPU. I'm running kernel 2.6.20-git (pulled
few hours ago), configured with SMP and PREEMPT.

I'm hitting 2 different warnings:
BUG: using smp_processor_id() in preemptible [00000001] code: kvm/7726
caller is vmx_create_vcpu+0x9/0x2f [kvm_intel]
 [<b01ccbc8>] debug_smp_processor_id+0xa0/0xb4
 [<f1a35b34>] vmx_create_vcpu+0x9/0x2f [kvm_intel]
 [<f1b68d4f>] kvm_dev_ioctl+0x160/0xf97 [kvm]
 [<b0130a7d>] autoremove_wake_function+0x0/0x35
 [<b01481a9>] get_page_from_freelist+0xe9/0x353
 [<b02f0c46>] _write_unlock+0x25/0x3b
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b01364c4>] mark_held_locks+0x46/0x62
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b01366a5>] trace_hardirqs_on+0x11e/0x141
 [<b014f189>] __handle_mm_fault+0x463/0x864
 [<b014f189>] __handle_mm_fault+0x463/0x864
 [<b02f0c0b>] _spin_unlock+0x25/0x3b
 [<b014f55f>] __handle_mm_fault+0x839/0x864
 [<b0117398>] do_page_fault+0x15a/0x52a
 [<f1b68bef>] kvm_dev_ioctl+0x0/0xf97 [kvm]
 [<b0168ed7>] do_ioctl+0x1f/0x62
 [<b016915e>] vfs_ioctl+0x244/0x256
 [<b01366a5>] trace_hardirqs_on+0x11e/0x141
 [<b01691a3>] sys_ioctl+0x33/0x4c
 [<b0102f10>] syscall_call+0x7/0xb
 =======================

vmx_create_vcpu calls alloc_vmcs which uses smp_processor_id() in
preemptible context and pass the result to alloc_vmcs_cpu(); at a later
point the function may be running on a different CPU (hence the result
of cpu_to_node may be meaningless).

Second one:
BUG: sleeping function called from invalid context at
/home/kronos/src/linux-2.6.git/mm/slab.c:3034
in_atomic():1, irqs_disabled():0
1 lock held by kvm/12706:
 #0:  (&vcpu->mutex){--..}, at: [<f1b68d02>] kvm_dev_ioctl+0x113/0xf97
[kvm]
 [<b015c32a>] kmem_cache_alloc+0x1b/0x6f
 [<f1a360ab>] vmx_vcpu_setup+0x528/0x703 [kvm_intel]
 [<f1a35a7d>] vmx_vcpu_load+0xdb/0xe3 [kvm_intel]
 [<f1b68d8c>] kvm_dev_ioctl+0x19d/0xf97 [kvm]
 [<b0130a7d>] autoremove_wake_function+0x0/0x35
 [<b02d2ce3>] ip4_datagram_connect+0x293/0x2ec
 [<b02f0c46>] _write_unlock+0x25/0x3b
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b01364c4>] mark_held_locks+0x46/0x62
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b0148295>] get_page_from_freelist+0x1d5/0x353
 [<b01366a5>] trace_hardirqs_on+0x11e/0x141
 [<b0102235>] setup_sigcontext+0x105/0x189
 [<b014f189>] __handle_mm_fault+0x463/0x864
 [<b014f189>] __handle_mm_fault+0x463/0x864
 [<b02f0c0b>] _spin_unlock+0x25/0x3b
 [<b014f55f>] __handle_mm_fault+0x839/0x864
 [<b0117398>] do_page_fault+0x15a/0x52a
 [<f1b68bef>] kvm_dev_ioctl+0x0/0xf97 [kvm]
 [<b0168ed7>] do_ioctl+0x1f/0x62
 [<b016915e>] vfs_ioctl+0x244/0x256
 [<b01366a5>] trace_hardirqs_on+0x11e/0x141
 [<b01691a3>] sys_ioctl+0x33/0x4c
 [<b0102f10>] syscall_call+0x7/0xb
 =======================

I think it's caused the GFP_KERNEL allocation (vmx_vcpu_setup,
vmx.c:1103) done inside get_cpu().


This first warning can be fixed by moving the cpu_to_node conversion.
Not sure that it's correct though. Between vmx_create_vcpu() and
vmx_vcpu_load() the CPU may change and the 'vmcs' may be actually on a
differnt node, no?

Singed-Off-By: Luca Tettamanti <kronos.it@gmail.com>

---
 drivers/kvm/vmx.c |   11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/kvm/vmx.c b/drivers/kvm/vmx.c
index d0a2c2d..4d088a0 100644
--- a/drivers/kvm/vmx.c
+++ b/drivers/kvm/vmx.c
@@ -524,9 +524,8 @@ static __init void setup_vmcs_descriptor(void)
 	vmcs_descriptor.revision_id = vmx_msr_low;
 }
 
-static struct vmcs *alloc_vmcs_cpu(int cpu)
+static struct vmcs *alloc_vmcs_cpu(int node)
 {
-	int node = cpu_to_node(cpu);
 	struct page *pages;
 	struct vmcs *vmcs;
 
@@ -541,7 +540,13 @@ static struct vmcs *alloc_vmcs_cpu(int cpu)
 
 static struct vmcs *alloc_vmcs(void)
 {
-	return alloc_vmcs_cpu(smp_processor_id());
+	int cpu, node;
+
+	cpu = get_cpu();
+	node = cpu_to_node(cpu);
+	put_cpu();
+
+	return alloc_vmcs_cpu(node);
 }
 
 static void free_vmcs(struct vmcs *vmcs)


Luca
-- 
"L'amore consiste nell'essere cretini insieme." -- P. Valery
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/