[RFC PATCH] x86, espfix: use spin_lock rather than mutex

[RFC PATCH] x86, espfix: use spin_lock rather than mutex

[Gu Zheng]

The following lockdep warning occurrs when running with latest kernel:
[    3.178000] ------------[ cut here ]------------
[    3.183000] WARNING: CPU: 128 PID: 0 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0xdd/0xe0()
[    3.193000] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))
[    3.199000] Modules linked in:

[    3.203000] CPU: 128 PID: 0 Comm: swapper/128 Not tainted 4.1.0-rc3 #70
[    3.221000]  0000000000000000 2d6601fb3e6d4e4c ffff88086fd5fc38 ffffffff81773f0a
[    3.230000]  0000000000000000 ffff88086fd5fc90 ffff88086fd5fc78 ffffffff8108c85a
[    3.238000]  ffff88086fd60000 0000000000000092 ffff88086fd60000 00000000000000d0
[    3.246000] Call Trace:
[    3.249000]  [<ffffffff81773f0a>] dump_stack+0x4c/0x65
[    3.255000]  [<ffffffff8108c85a>] warn_slowpath_common+0x8a/0xc0
[    3.261000]  [<ffffffff8108c8e5>] warn_slowpath_fmt+0x55/0x70
[    3.268000]  [<ffffffff810ee24d>] lockdep_trace_alloc+0xdd/0xe0
[    3.274000]  [<ffffffff811cda0d>] __alloc_pages_nodemask+0xad/0xca0
[    3.281000]  [<ffffffff810ec7ad>] ? __lock_acquire+0xf6d/0x1560
[    3.288000]  [<ffffffff81219c8a>] alloc_page_interleave+0x3a/0x90
[    3.295000]  [<ffffffff8121b32d>] alloc_pages_current+0x17d/0x1a0
[    3.301000]  [<ffffffff811c869e>] ? __get_free_pages+0xe/0x50
[    3.308000]  [<ffffffff811c869e>] __get_free_pages+0xe/0x50
[    3.314000]  [<ffffffff8102640b>] init_espfix_ap+0x17b/0x320
[    3.320000]  [<ffffffff8105c691>] start_secondary+0xf1/0x1f0
[    3.327000] ---[ end trace 1b3327d9d6a1d62c ]---

This seems a mis-warning by lockdep, as we alloc pages with GFP_KERNEL in
init_espfix_ap() which is called before enabled local irq, and the lockdep
sub-system considers this behaviour as allocating memory with GFP_FS with
local irq disabled, then trigger the warning as mentioned about.
Though here we use GFP_NOFS rather GFP_KERNEL to avoid the warning, but
you know, init_espfix_ap is called with preempt and local irq disabled,
it is not a good idea to use mutex (might sleep) here.
So we convert the initialization lock to spin_lock here to avoid the noise.

Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
Cc: Stable <stable@vger.kernel.org>
---
 arch/x86/kernel/espfix_64.c |   13 +++++++------
 1 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c
index f5d0730..ceb35a3 100644
--- a/arch/x86/kernel/espfix_64.c
+++ b/arch/x86/kernel/espfix_64.c
@@ -57,14 +57,14 @@
 # error "Need more than one PGD for the ESPFIX hack"
 #endif
 
-#define PGALLOC_GFP (GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO)
+#define PGALLOC_GFP (GFP_ATOMIC | __GFP_NOTRACK | __GFP_ZERO)
 
 /* This contains the *bottom* address of the espfix stack */
 DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_stack);
 DEFINE_PER_CPU_READ_MOSTLY(unsigned long, espfix_waddr);
 
-/* Initialization mutex - should this be a spinlock? */
-static DEFINE_MUTEX(espfix_init_mutex);
+/* Initialization lock */
+static DEFINE_SPINLOCK(espfix_init_lock);
 
 /* Page allocation bitmap - each page serves ESPFIX_STACKS_PER_PAGE CPUs */
 #define ESPFIX_MAX_PAGES  DIV_ROUND_UP(CONFIG_NR_CPUS, ESPFIX_STACKS_PER_PAGE)
@@ -144,6 +144,7 @@ void init_espfix_ap(void)
 	int n;
 	void *stack_page;
 	pteval_t ptemask;
+	unsigned long flags;
 
 	/* We only have to do this once... */
 	if (likely(this_cpu_read(espfix_stack)))
@@ -158,7 +159,7 @@ void init_espfix_ap(void)
 	if (likely(stack_page))
 		goto done;
 
-	mutex_lock(&espfix_init_mutex);
+	spin_lock_irqsave(&espfix_init_lock, flags);
 
 	/* Did we race on the lock? */
 	stack_page = ACCESS_ONCE(espfix_pages[page]);
@@ -188,7 +189,7 @@ void init_espfix_ap(void)
 	}
 
 	pte_p = pte_offset_kernel(&pmd, addr);
-	stack_page = (void *)__get_free_page(GFP_KERNEL);
+	stack_page = (void *)__get_free_page(PGALLOC_GFP);
 	pte = __pte(__pa(stack_page) | (__PAGE_KERNEL_RO & ptemask));
 	for (n = 0; n < ESPFIX_PTE_CLONES; n++)
 		set_pte(&pte_p[n*PTE_STRIDE], pte);
@@ -197,7 +198,7 @@ void init_espfix_ap(void)
 	ACCESS_ONCE(espfix_pages[page]) = stack_page;
 
 unlock_done:
-	mutex_unlock(&espfix_init_mutex);
+	spin_unlock_irqrestore(&espfix_init_lock, flags);
 done:
 	this_cpu_write(espfix_stack, addr);
 	this_cpu_write(espfix_waddr, (unsigned long)stack_page
-- 
1.7.7

Re: [RFC PATCH] x86, espfix: use spin_lock rather than mutex

[Ingo Molnar]

* Borislav Petkov <bp@alien8.de> wrote:

> On Thu, May 14, 2015 at 07:37:45PM +0800, Gu Zheng wrote:
> > The following lockdep warning occurrs when running with latest kernel:
> > [    3.178000] ------------[ cut here ]------------
> > [    3.183000] WARNING: CPU: 128 PID: 0 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0xdd/0xe0()
> > [    3.193000] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))
> > [    3.199000] Modules linked in:
> > 
> > [    3.203000] CPU: 128 PID: 0 Comm: swapper/128 Not tainted 4.1.0-rc3 #70
> > [    3.221000]  0000000000000000 2d6601fb3e6d4e4c ffff88086fd5fc38 ffffffff81773f0a
> > [    3.230000]  0000000000000000 ffff88086fd5fc90 ffff88086fd5fc78 ffffffff8108c85a
> > [    3.238000]  ffff88086fd60000 0000000000000092 ffff88086fd60000 00000000000000d0
> > [    3.246000] Call Trace:
> > [    3.249000]  [<ffffffff81773f0a>] dump_stack+0x4c/0x65
> > [    3.255000]  [<ffffffff8108c85a>] warn_slowpath_common+0x8a/0xc0
> > [    3.261000]  [<ffffffff8108c8e5>] warn_slowpath_fmt+0x55/0x70
> > [    3.268000]  [<ffffffff810ee24d>] lockdep_trace_alloc+0xdd/0xe0
> > [    3.274000]  [<ffffffff811cda0d>] __alloc_pages_nodemask+0xad/0xca0
> > [    3.281000]  [<ffffffff810ec7ad>] ? __lock_acquire+0xf6d/0x1560
> > [    3.288000]  [<ffffffff81219c8a>] alloc_page_interleave+0x3a/0x90
> > [    3.295000]  [<ffffffff8121b32d>] alloc_pages_current+0x17d/0x1a0
> > [    3.301000]  [<ffffffff811c869e>] ? __get_free_pages+0xe/0x50
> > [    3.308000]  [<ffffffff811c869e>] __get_free_pages+0xe/0x50
> > [    3.314000]  [<ffffffff8102640b>] init_espfix_ap+0x17b/0x320
> > [    3.320000]  [<ffffffff8105c691>] start_secondary+0xf1/0x1f0
> > [    3.327000] ---[ end trace 1b3327d9d6a1d62c ]---
> >
> > This seems a mis-warning by lockdep, as we alloc pages with GFP_KERNEL in
> > init_espfix_ap() which is called before enabled local irq, and the lockdep
> > sub-system considers this behaviour as allocating memory with GFP_FS with
> > local irq disabled, then trigger the warning as mentioned about.
> > Though here we use GFP_NOFS rather GFP_KERNEL to avoid the warning, but
> > you know, init_espfix_ap is called with preempt and local irq disabled,
> > it is not a good idea to use mutex (might sleep) here.
> > So we convert the initialization lock to spin_lock here to avoid the noise.
> > 
> > Signed-off-by: Gu Zheng <guz.fnst@cn.fujitsu.com>
> > Cc: Stable <stable@vger.kernel.org>
> > ---
> >  arch/x86/kernel/espfix_64.c |   13 +++++++------
> >  1 files changed, 7 insertions(+), 6 deletions(-)
> > 
> > diff --git a/arch/x86/kernel/espfix_64.c b/arch/x86/kernel/espfix_64.c
> > index f5d0730..ceb35a3 100644
> > --- a/arch/x86/kernel/espfix_64.c
> > +++ b/arch/x86/kernel/espfix_64.c
> > @@ -57,14 +57,14 @@
> >  # error "Need more than one PGD for the ESPFIX hack"
> >  #endif
> >  
> > -#define PGALLOC_GFP (GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO)
> > +#define PGALLOC_GFP (GFP_ATOMIC | __GFP_NOTRACK | __GFP_ZERO)
> 
> IINM, that's ESPFIX_MAX_PAGES with GFP_ATOMIC which for 8K CPUs are 128
> pages.
> 
> That's a lotta waste in my book for espfix stack pages.
> 
> Enabling interrupts earlier in start_secondary() is probably out of 
> the question, maybe we should prealloc all those pages...

We could allocate them on the boot CPU side and hand them over to the 
secondary CPU.

Thanks,

	Ingo