Re: [PATCH] mm/mremap_pages: Fix static key devmap_managed_key updates

[Ralph Campbell <rcampbell@nvidia.com>]

On 10/22/20 8:41 AM, Ira Weiny wrote:
> On Thu, Oct 22, 2020 at 11:37:53AM +0530, Aneesh Kumar K.V wrote:
>> commit 6f42193fd86e ("memremap: don't use a separate devm action for
>> devmap_managed_enable_get") changed the static key updates such that we
>> now call devmap_managed_enable_put() without doing the equivalent
>> devmap_managed_enable_get().
>>
>> devmap_managed_enable_get() is only called for MEMORY_DEVICE_PRIVATE and
>> MEMORY_DEVICE_FS_DAX, But memunmap_pages() get called for other pgmap
>> types too. This results in the below warning when switching between
>> system-ram and devdax mode for devdax namespace.
>>
>>   jump label: negative count!
>>   WARNING: CPU: 52 PID: 1335 at kernel/jump_label.c:235 static_key_slow_try_dec+0x88/0xa0
>>   Modules linked in:
>>   ....
>>
>>   NIP [c000000000433318] static_key_slow_try_dec+0x88/0xa0
>>   LR [c000000000433314] static_key_slow_try_dec+0x84/0xa0
>>   Call Trace:
>>   [c000000025c1f660] [c000000000433314] static_key_slow_try_dec+0x84/0xa0 (unreliable)
>>   [c000000025c1f6d0] [c000000000433664] __static_key_slow_dec_cpuslocked+0x34/0xd0
>>   [c000000025c1f700] [c0000000004337a4] static_key_slow_dec+0x54/0xf0
>>   [c000000025c1f770] [c00000000059c49c] memunmap_pages+0x36c/0x500
>>   [c000000025c1f820] [c000000000d91d10] devm_action_release+0x30/0x50
>>   [c000000025c1f840] [c000000000d92e34] release_nodes+0x2f4/0x3e0
>>   [c000000025c1f8f0] [c000000000d8b15c] device_release_driver_internal+0x17c/0x280
>>   [c000000025c1f930] [c000000000d883a4] bus_remove_device+0x124/0x210
>>   [c000000025c1f9b0] [c000000000d80ef4] device_del+0x1d4/0x530
>>   [c000000025c1fa70] [c000000000e341e8] unregister_dev_dax+0x48/0xe0
>>   [c000000025c1fae0] [c000000000d91d10] devm_action_release+0x30/0x50
>>   [c000000025c1fb00] [c000000000d92e34] release_nodes+0x2f4/0x3e0
>>   [c000000025c1fbb0] [c000000000d8b15c] device_release_driver_internal+0x17c/0x280
>>   [c000000025c1fbf0] [c000000000d87000] unbind_store+0x130/0x170
>>   [c000000025c1fc30] [c000000000d862a0] drv_attr_store+0x40/0x60
>>   [c000000025c1fc50] [c0000000006d316c] sysfs_kf_write+0x6c/0xb0
>>   [c000000025c1fc90] [c0000000006d2328] kernfs_fop_write+0x118/0x280
>>   [c000000025c1fce0] [c0000000005a79f8] vfs_write+0xe8/0x2a0
>>   [c000000025c1fd30] [c0000000005a7d94] ksys_write+0x84/0x140
>>   [c000000025c1fd80] [c00000000003a430] system_call_exception+0x120/0x270
>>   [c000000025c1fe20] [c00000000000c540] system_call_common+0xf0/0x27c
>>
>> Cc: Christoph Hellwig <hch@infradead.org>
>> Cc: Dan Williams <dan.j.williams@intel.com>
>> Cc: Sachin Sant <sachinp@linux.vnet.ibm.com>
>> Cc: linux-nvdimm@lists.01.org
>> Cc: Ira Weiny <ira.weiny@intel.com>
>> Cc: Jason Gunthorpe <jgg@mellanox.com>
>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
>> ---
>>   mm/memremap.c | 19 +++++++++++++++----
>>   1 file changed, 15 insertions(+), 4 deletions(-)
>>
>> diff --git a/mm/memremap.c b/mm/memremap.c
>> index 73a206d0f645..d4402ff3e467 100644
>> --- a/mm/memremap.c
>> +++ b/mm/memremap.c
>> @@ -158,6 +158,16 @@ void memunmap_pages(struct dev_pagemap *pgmap)
>>   {
>>   	unsigned long pfn;
>>   	int i;
>> +	bool need_devmap_managed = false;
>> +
>> +	switch (pgmap->type) {
>> +	case MEMORY_DEVICE_PRIVATE:
>> +	case MEMORY_DEVICE_FS_DAX:
>> +		need_devmap_managed = true;
>> +		break;
>> +	default:
>> +		break;
>> +	}
> 
> Is it overkill to avoid duplicating this switch logic in
> page_is_devmap_managed() by creating another call which can be used here?

Perhaps. I can imagine a helper defined in include/linux/mm.h which
page_is_devmap_managed() could also call but that would impact a lot of
places that include mm.h. Since memremap.c already has to have intimate
knowledge of the pgmap->type, I think limiting the change to just what
is needed is better for now. So the patch looks OK to me.

Looking at this some more, I would suggest changing devmap_managed_enable_get()
and devmap_managed_enable_put() to do the special case checking instead of
doing it in memremap_pages() and memunmap_pages().
Then devmap_managed_enable_get() doesn't need to return an error if
CONFIG_DEV_PAGEMAP_OPS isn't defined. I have only compile tested the
following.

Signed-off-by: Ralph Campbell <rcampbell@nvidia.com>
---
  mm/memremap.c | 39 ++++++++++++++++-----------------------
  1 file changed, 16 insertions(+), 23 deletions(-)

diff --git a/mm/memremap.c b/mm/memremap.c
index 73a206d0f645..16b2fb482da1 100644
--- a/mm/memremap.c
+++ b/mm/memremap.c
@@ -41,28 +41,24 @@ EXPORT_SYMBOL_GPL(memremap_compat_align);
  DEFINE_STATIC_KEY_FALSE(devmap_managed_key);
  EXPORT_SYMBOL(devmap_managed_key);
  
-static void devmap_managed_enable_put(void)
+static void devmap_managed_enable_put(struct dev_pagemap *pgmap)
  {
-	static_branch_dec(&devmap_managed_key);
+	if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
+	    pgmap->type == MEMORY_DEVICE_FS_DAX)
+		static_branch_dec(&devmap_managed_key);
  }
  
-static int devmap_managed_enable_get(struct dev_pagemap *pgmap)
+static void devmap_managed_enable_get(struct dev_pagemap *pgmap)
  {
-	if (pgmap->type == MEMORY_DEVICE_PRIVATE &&
-	    (!pgmap->ops || !pgmap->ops->page_free)) {
-		WARN(1, "Missing page_free method\n");
-		return -EINVAL;
-	}
-
-	static_branch_inc(&devmap_managed_key);
-	return 0;
+	if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
+	    pgmap->type == MEMORY_DEVICE_FS_DAX)
+		static_branch_inc(&devmap_managed_key);
  }
  #else
-static int devmap_managed_enable_get(struct dev_pagemap *pgmap)
+static void devmap_managed_enable_get(struct dev_pagemap *pgmap)
  {
-	return -EINVAL;
  }
-static void devmap_managed_enable_put(void)
+static void devmap_managed_enable_put(struct dev_pagemap *pgmap)
  {
  }
  #endif /* CONFIG_DEV_PAGEMAP_OPS */
@@ -169,7 +165,7 @@ void memunmap_pages(struct dev_pagemap *pgmap)
  		pageunmap_range(pgmap, i);
  
  	WARN_ONCE(pgmap->altmap.alloc, "failed to free all reserved pages\n");
-	devmap_managed_enable_put();
+	devmap_managed_enable_put(pgmap);
  }
  EXPORT_SYMBOL_GPL(memunmap_pages);
  
@@ -307,7 +303,6 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
  		.pgprot = PAGE_KERNEL,
  	};
  	const int nr_range = pgmap->nr_range;
-	bool need_devmap_managed = true;
  	int error, i;
  
  	if (WARN_ONCE(!nr_range, "nr_range must be specified\n"))
@@ -323,6 +318,10 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
  			WARN(1, "Missing migrate_to_ram method\n");
  			return ERR_PTR(-EINVAL);
  		}
+		if (!pgmap->ops->page_free) {
+			WARN(1, "Missing page_free method\n");
+			return ERR_PTR(-EINVAL);
+		}
  		if (!pgmap->owner) {
  			WARN(1, "Missing owner\n");
  			return ERR_PTR(-EINVAL);
@@ -336,11 +335,9 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
  		}
  		break;
  	case MEMORY_DEVICE_GENERIC:
-		need_devmap_managed = false;
  		break;
  	case MEMORY_DEVICE_PCI_P2PDMA:
  		params.pgprot = pgprot_noncached(params.pgprot);
-		need_devmap_managed = false;
  		break;
  	default:
  		WARN(1, "Invalid pgmap type %d\n", pgmap->type);
@@ -364,11 +361,7 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
  		}
  	}
  
-	if (need_devmap_managed) {
-		error = devmap_managed_enable_get(pgmap);
-		if (error)
-			return ERR_PTR(error);
-	}
+	devmap_managed_enable_get(pgmap);
  
  	/*
  	 * Clear the pgmap nr_range as it will be incremented for each
-- 
2.20.1

>>   
>>   	dev_pagemap_kill(pgmap);
>>   	for (i = 0; i < pgmap->nr_range; i++)
>> @@ -169,7 +179,8 @@ void memunmap_pages(struct dev_pagemap *pgmap)
>>   		pageunmap_range(pgmap, i);
>>   
>>   	WARN_ONCE(pgmap->altmap.alloc, "failed to free all reserved pages\n");
>> -	devmap_managed_enable_put();
>> +	if (need_devmap_managed)
>> +		devmap_managed_enable_put();
>>   }
>>   EXPORT_SYMBOL_GPL(memunmap_pages);
>>   
>> @@ -307,7 +318,7 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
>>   		.pgprot = PAGE_KERNEL,
>>   	};
>>   	const int nr_range = pgmap->nr_range;
>> -	bool need_devmap_managed = true;
>> +	bool need_devmap_managed = false;
> 
> I'm CC'ing Ralph Campbell because I think some of his work has proposed this
> same change.
> 
> Ira

This part of the patch isn't strictly needed, it just reverses the default value of
need_devmap_managed.

>>   	int error, i;
>>   
>>   	if (WARN_ONCE(!nr_range, "nr_range must be specified\n"))
>> @@ -327,6 +338,7 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
>>   			WARN(1, "Missing owner\n");
>>   			return ERR_PTR(-EINVAL);
>>   		}
>> +		need_devmap_managed = true;
>>   		break;
>>   	case MEMORY_DEVICE_FS_DAX:
>>   		if (!IS_ENABLED(CONFIG_ZONE_DEVICE) ||
>> @@ -334,13 +346,12 @@ void *memremap_pages(struct dev_pagemap *pgmap, int nid)
>>   			WARN(1, "File system DAX not supported\n");
>>   			return ERR_PTR(-EINVAL);
>>   		}
>> +		need_devmap_managed = true;
>>   		break;
>>   	case MEMORY_DEVICE_GENERIC:
>> -		need_devmap_managed = false;
>>   		break;
>>   	case MEMORY_DEVICE_PCI_P2PDMA:
>>   		params.pgprot = pgprot_noncached(params.pgprot);
>> -		need_devmap_managed = false;
>>   		break;
>>   	default:
>>   		WARN(1, "Invalid pgmap type %d\n", pgmap->type);
>> -- 
>> 2.26.2
>>