[tip:x86/apic] irqdomain: Correct early allocation of irq domains with IRQs off

From: tip-bot for Borislav Petkov
Date: Mon Dec 01 2014 - 04:47:08 EST


Commit-ID: eda7516e1d428caf3cfc88e4292f2a1357a2e569
Gitweb: http://git.kernel.org/tip/eda7516e1d428caf3cfc88e4292f2a1357a2e569
Author: Borislav Petkov <bp@xxxxxxx>
AuthorDate: Sun, 30 Nov 2014 13:37:43 +0100
Committer: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
CommitDate: Mon, 1 Dec 2014 10:38:30 +0100

irqdomain: Correct early allocation of irq domains with IRQs off

I'm seeing the following lockdep splat on an x2APIC machine with
interrupts remapping. The problem is that enable_IR_x2apic() disables
interrupts before doing any further initialization. However, after
having moved to irq domains, domain allocation cannot happen with
interrupts disabled (GFP_KERNEL).

A proper fix would be to move that initialization to the early irq setup
path, which is something for another day. It is more involved work too.

So do a temporary fix which should not encourage the behaviour of even
assuming the irq domains code should be called with interrupts disabled.
Make it x86-only too.

...
dmar: Host address width 36
dmar: DRHD base: 0x000000fed90000 flags: 0x0
dmar: IOMMU 0: reg_base_addr fed90000 ver 1:0 cap c0000020e60262 ecap f0101a
dmar: DRHD base: 0x000000fed91000 flags: 0x1
dmar: IOMMU 1: reg_base_addr fed91000 ver 1:0 cap c9008020660262 ecap f0105a
dmar: RMRR base: 0x000000da2ba000 end: 0x000000da2d0fff
dmar: RMRR base: 0x000000db800000 end: 0x000000df9fffff
IOAPIC id 2 under DRHD base 0xfed91000 IOMMU 1
HPET id 0 under DRHD base 0xfed91000
Queued invalidation will be enabled to support x2apic and Intr-remapping.
------------[ cut here ]------------
WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2744 lockdep_trace_alloc+0xd4/0xe0()
DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.18.0-rc6+ #1
Hardware name: LENOVO 2320CTO/2320CTO, BIOS G2ET86WW (2.06 ) 11/13/2012
0000000000000009 ffff880213d07b58 ffffffff816502d7 0000000000000000
ffff880213d07ba8 ffff880213d07b98 ffffffff81059790 0000000000000001
0000000000000092 0000000000000000 00000000000080d0 0000000000000000
Call Trace:
dump_stack
warn_slowpath_common
warn_slowpath_fmt
lockdep_trace_alloc
__alloc_pages_nodemask
? get_page_from_freelist
? trace_hardirqs_off_caller
alloc_kmem_pages_node
kmalloc_large_node
__kmalloc_node
? __dmar_enable_qi
__irq_domain_add
irq_domain_add_hierarchy
intel_setup_irq_remapping.part.4
intel_enable_irq_remapping
irq_remapping_enable
enable_IR
enable_IR_x2apic
default_setup_apic_routing
native_smp_prepare_cpus
kernel_init_freeable
? ret_from_fork
? rest_init
kernel_init
ret_from_fork
? rest_init
---[ end trace fac50e785fc22942 ]---
Enabled IRQ remapping in x2apic mode
Enabling x2apic
Enabled x2apic
Switched APIC routing to cluster x2apic.
..TIMER: vector=0x30 apic1=0 pin1=2 apic2=-1 pin2=-1
...

Signed-off-by: Borislav Petkov <bp@xxxxxxx>
Cc: Jiang Liu <jiang.liu@xxxxxxxxxxxxxxx>
Cc: Tony Luck <tony.luck@xxxxxxxxx>
Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
Cc: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
Cc: Grant Likely <grant.likely@xxxxxxxxxx>
Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
Cc: Yijing Wang <wangyijing@xxxxxxxxxx>
Cc: Yingjoe Chen <yingjoe.chen@xxxxxxxxxxxx>
Cc: Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx>
Cc: Matthias Brugger <matthias.bgg@xxxxxxxxx>
Cc: Joerg Roedel <joro@xxxxxxxxxx>
Link: http://lkml.kernel.org/r/20141129125319.GA6491@xxxxxxx
Link: http://lkml.kernel.org/r/1417351063-19039-1-git-send-email-bp@xxxxxxxxx
Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
---
kernel/irq/irqdomain.c | 21 +++++++++++++++++++--
1 file changed, 19 insertions(+), 2 deletions(-)

diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 7fac311..3395d89 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -46,14 +46,31 @@ struct irq_domain *__irq_domain_add(struct device_node *of_node, int size,
void *host_data)
{
struct irq_domain *domain;
+ gfp_t gfp_flags = GFP_KERNEL;
+
+#ifdef CONFIG_X86
+ /*
+ * BIG FAT COMMENT: Early initialization paths like enable_IR_x2apic(),
+ * for example, call into here with interrupts disabled but then we do
+ * allocate memory and can sleep so no-no. A proper fix would be to do
+ * x2APIC IR setup in the early irq setup path but it is too late for
+ * fixing it this way now, shortly before the merge window.
+ *
+ * So we do this little brown paper bag, which is temporary! Do not even
+ * think of calling irq domain setup code with IRQs disabled. You will
+ * get frozen-sharked!
+ */
+ if (irqs_disabled())
+ gfp_flags = GFP_NOFS;
+#endif

domain = kzalloc_node(sizeof(*domain) + (sizeof(unsigned int) * size),
- GFP_KERNEL, of_node_to_nid(of_node));
+ gfp_flags, of_node_to_nid(of_node));
if (WARN_ON(!domain))
return NULL;

/* Fill structure */
- INIT_RADIX_TREE(&domain->revmap_tree, GFP_KERNEL);
+ INIT_RADIX_TREE(&domain->revmap_tree, gfp_flags);
domain->ops = ops;
domain->host_data = host_data;
domain->of_node = of_node_get(of_node);
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/