Re: [PATCH] mm/vmalloc: fix align value calculation error

[zijun_hu <zijun_hu@zoho.com>]

On 08/10/2016 05:28 AM, Andrew Morton wrote:
> On Fri, 5 Aug 2016 23:48:21 +0800 zijun_hu <zijun_hu@zoho.com> wrote:
> 
>> From: zijun_hu <zijun_hu@htc.com>
>> Date: Fri, 5 Aug 2016 22:10:07 +0800
>> Subject: [PATCH 1/1] mm/vmalloc: fix align value calculation error
>>
>> it causes double align requirement for __get_vm_area_node() if parameter
>> size is power of 2 and VM_IOREMAP is set in parameter flags
>>
>> get_order_long() is implemented and used instead of fls_long() for
>> fixing the bug
> 
> Makes sense.  I think.
> 
>> --- a/include/linux/bitops.h
>> +++ b/include/linux/bitops.h
>> @@ -192,6 +192,23 @@ static inline unsigned fls_long(unsigned long l)
>>  }
>>  
>>  /**
>> + * get_order_long - get order after rounding @l up to power of 2
>> + * @l: parameter
>> + *
>> + * it is same as get_count_order() but long type parameter
>> + * or 0 is returned if @l == 0UL
>> + */
>> +static inline int get_order_long(unsigned long l)
>> +{
>> +	if (l == 0UL)
>> +		return 0;
>> +	else if (l & (l - 1UL))
>> +		return fls_long(l);
>> +	else
>> +		return fls_long(l) - 1;
>> +}
>> +
>> +/**
>>   * __ffs64 - find first set bit in a 64 bit word
>>   * @word: The 64 bit word
>>   *
>> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
>> index 91f44e7..7d717f3 100644
>> --- a/mm/vmalloc.c
>> +++ b/mm/vmalloc.c
>> @@ -1360,7 +1360,7 @@ static struct vm_struct *__get_vm_area_node(unsigned long size,
>>  
>>  	BUG_ON(in_interrupt());
>>  	if (flags & VM_IOREMAP)
>> -		align = 1ul << clamp_t(int, fls_long(size),
>> +		align = 1ul << clamp_t(int, get_order_long(size),
>>  				       PAGE_SHIFT, IOREMAP_MAX_ORDER);
>>  
>>  	size = PAGE_ALIGN(size);
> 
> It would be better to call this get_count_order_long(), I think?  To
> match get_count_order()?
> 
yes, i agree with you to correct function name

i provide another patch called v2 based on your suggestion as shown below
it have following correction against original patch v1
1) use name get_count_order_long() instead of get_order_long()
2) return -1 if @l == 0 to consist with get_order_long()
3) cast type to int before returning from get_count_order_long()
4) move up function parameter checking for __get_vm_area_node()
5) more commit message is offered to make issue and approach clear
any comments about new patch is welcome

this new patch called patch v2 is shown below

>From 868d3c100f41e16136eed50e47bbfb03d4a16d25 Mon Sep 17 00:00:00 2001
From: zijun_hu <zijun_hu@htc.com>
Date: Wed, 10 Aug 2016 12:13:41 +0800
Subject: [PATCH v2 1/1] mm/vmalloc: fix align value calculation error

it causes double align requirement for __get_vm_area_node() if parameter
size is power of 2 and VM_IOREMAP is set in parameter flags, for example
size=0x10000 -> fls_long(0x10000)=17 -> align=0x20000

get_count_order_long() is implemented and used instead of fls_long() for
fixing the bug, for example
size=0x10000 -> get_count_order_long(0x10000)=16 -> align=0x10000

Signed-off-by: zijun_hu <zijun_hu@htc.com>
---
 include/linux/bitops.h | 16 ++++++++++++++++
 mm/vmalloc.c           |  8 ++++----
 2 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/include/linux/bitops.h b/include/linux/bitops.h
index 299e76b..93a07d1 100644
--- a/include/linux/bitops.h
+++ b/include/linux/bitops.h
@@ -192,6 +192,22 @@ static inline unsigned fls_long(unsigned long l)
 }
 
 /**
+ * get_count_order_long - get order after rounding @l up to power of 2
+ * @l: parameter
+ *
+ * it is same as get_count_order() but with long type parameter
+ */
+static inline int get_count_order_long(unsigned long l)
+{
+	if (l == 0UL)
+		return -1;
+	else if (l & (l - 1UL))
+		return (int)fls_long(l);
+	else
+		return (int)fls_long(l) - 1;
+}
+
+/**
  * __ffs64 - find first set bit in a 64 bit word
  * @word: The 64 bit word
  *
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 91f44e7..80660a0 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -1359,14 +1359,14 @@ static struct vm_struct *__get_vm_area_node(unsigned long size,
 	struct vm_struct *area;
 
 	BUG_ON(in_interrupt());
-	if (flags & VM_IOREMAP)
-		align = 1ul << clamp_t(int, fls_long(size),
-				       PAGE_SHIFT, IOREMAP_MAX_ORDER);
-
 	size = PAGE_ALIGN(size);
 	if (unlikely(!size))
 		return NULL;
 
+	if (flags & VM_IOREMAP)
+		align = 1ul << clamp_t(int, get_count_order_long(size),
+				       PAGE_SHIFT, IOREMAP_MAX_ORDER);
+
 	area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
 	if (unlikely(!area))
 		return NULL;
-- 
1.9.1

> get_count_order() is a weird name and perhaps both of these should be
> renamed to things which actually make sense.  That's a separate issue.
> 
okay, perhaps, another patch is applied to correct this weird name issue
in the future due to it is a separate issue now
we use get_count_order_long() to consist with get_order_long() now