Re: [PATCH 3/5] lib/bitmap: add test for bitmap_{from,to}_arr64

[Guenter Roeck <linux@roeck-us.net>]

On 2/27/23 11:24, Yury Norov wrote:
> On Mon, Feb 27, 2023 at 06:59:12AM -0800, Guenter Roeck wrote:
>> On 2/27/23 06:46, Alexander Lobakin wrote:
>>> From: Yury Norov <yury.norov@gmail.com>
>>> Date: Sat, 25 Feb 2023 16:06:45 -0800
>>>
>>>> On Sat, Feb 25, 2023 at 04:05:02PM -0800, Yury Norov wrote:
>>>>> On Sat, Feb 25, 2023 at 10:47:02AM -0800, Guenter Roeck wrote:
>>>>>> Hi,
>>>>>>
>>>>>> On Thu, Apr 28, 2022 at 01:51:14PM -0700, Yury Norov wrote:
>>>>>>> Test newly added bitmap_{from,to}_arr64() functions similarly to
>>>>>>> already existing bitmap_{from,to}_arr32() tests.
>>>>>>>
>>>>>>> Signed-off-by: Yury Norov <yury.norov@gmail.com>
>>>>>>
>>>>>> Ever since this test is in the tree, several of my boot tests show
>>>>>> lots of messages such as
>>>>>>
>>>>>> test_bitmap: bitmap_to_arr64(nbits == 1): tail is not safely cleared: 0xa5a5a5a500000001 (must be 0x0000000000000001)
>>>
>>> Hmmm, the whole 4 bytes weren't touched.
>>>
>>>>>> test_bitmap: bitmap_to_arr64(nbits == 2): tail is not safely cleared: 0xa5a5a5a500000001 (must be 0x0000000000000003)
>>>>>> test_bitmap: bitmap_to_arr64(nbits == 3): tail is not safely cleared: 0xa5a5a5a500000001 (must be 0x0000000000000007)
>>>
>>> This is where it gets worse...
>>>
>>>>>> ...
>>>>>> test_bitmap: bitmap_to_arr64(nbits == 927): tail is not safely cleared: 0xa5a5a5a500000000 (must be 0x000000007fffffff)
>>>>>> test_bitmap: bitmap_to_arr64(nbits == 928): tail is not safely cleared: 0xa5a5a5a580000000 (must be 0x00000000ffffffff)
>>>
>>> I don't see the pattern how the actual result gets generated. But the
>>> problem is in the bitmap code rather than in the subtest -- "must be"s
>>> are fully correct.
>>>
>>> Given that the 0xa5s are present in the upper 32 bits, it is Big Endian
>>> I guess? Maybe even 32-bit Big Endian? Otherwise I'd start concerning
>>> how comes it doesn't reproduce on x86_64s :D
>>>
>>
>> It does reproduce on 32-bit x86 builds, and as far as I can see
>> it is only seen with 32-bit little endian systems.
> 
> Hi Guenter, Alexander,
> 
> I think that the reason for the failures like this:
> 
>> test_bitmap: bitmap_to_arr64(nbits == 1): tail is not safely cleared: 0xa5a5a5a500000001 (must be 0x0000000000000001)
> 
> is that bitmap_to_arr64 is overly optimized for 32-bit LE architectures.
> 
> Regarding this:
> 
>> test_bitmap: bitmap_to_arr64(nbits == 927): tail is not safely cleared: 0xa5a5a5a500000000 (must be 0x000000007fffffff)
> 
> I am not sure what happens, but because this again happens on 32-bit
> LE only, I hope the following fix would help too.
> 
> Can you please check if the patch works for you? I don't have a 32-bit LE
> machine in hand, and all my 32-bit VMs (arm and i386) refuse to load the
> latest kernels for some weird reason, so it's only build-tested.
> 
> I'll give it a full-run when restore my 32-bit setups.
> 
> Thanks,
> Yury
> 
>>From 2881714db497aed103e310865da075e7b0ce7e1a Mon Sep 17 00:00:00 2001
> From: Yury Norov <yury.norov@gmail.com>
> Date: Mon, 27 Feb 2023 09:21:59 -0800
> Subject: [PATCH] lib/bitmap: drop optimization of bitmap_{from,to}_arr64
> 
> bitmap_{from,to}_arr64() optimization is overly optimistic on 32-bit LE
> architectures when it's wired to bitmap_copy_clear_tail().
> 
> bitmap_copy_clear_tail() takes care of unused bits in the bitmap up to
> the next word boundary. But on 32-bit machines when copying bits from
> bitmap to array of 64-bit words, it's expected that the unused part of
> a recipient array must be cleared up to 64-bit boundary, so the last 4
> bytes may stay untouched.
> 
> While the copying part of the optimization works correct, that clear-tail
> trick makes corresponding tests reasonably fail when nbits % 64 <= 32:
> 
> test_bitmap: bitmap_to_arr64(nbits == 1): tail is not safely cleared: 0xa5a5a5a500000001 (must be 0x0000000000000001)
> 
> Fix it by removing bitmap_{from,to}_arr64() optimization for 32-bit LE
> arches.
> 
> Reported-by: Guenter Roeck <linux@roeck-us.net>
> Fixes: 0a97953fd2210 ("lib: add bitmap_{from,to}_arr64")
> Signed-off-by: Yury Norov <yury.norov@gmail.com>

Tested with 32-bit i386 image. With this patch on top of
v6.2-12765-g982818426a0f, the log messages are gone. Without this patch,
they are still seen.

Tested-by: Guenter Roeck <linux@roeck-us.net>

Guenter

> ---
>   include/linux/bitmap.h | 8 +++-----
>   lib/bitmap.c           | 2 +-
>   2 files changed, 4 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/bitmap.h b/include/linux/bitmap.h
> index 40e53a2ecc0d..5abc993903fb 100644
> --- a/include/linux/bitmap.h
> +++ b/include/linux/bitmap.h
> @@ -302,12 +302,10 @@ void bitmap_to_arr32(u32 *buf, const unsigned long *bitmap,
>   #endif
>   
>   /*
> - * On 64-bit systems bitmaps are represented as u64 arrays internally. On LE32
> - * machines the order of hi and lo parts of numbers match the bitmap structure.
> - * In both cases conversion is not needed when copying data from/to arrays of
> - * u64.
> + * On 64-bit systems bitmaps are represented as u64 arrays internally. So,
> + * conversion is not needed when copying data from/to arrays of u64.
>    */
> -#if (BITS_PER_LONG == 32) && defined(__BIG_ENDIAN)
> +#if BITS_PER_LONG == 32
>   void bitmap_from_arr64(unsigned long *bitmap, const u64 *buf, unsigned int nbits);
>   void bitmap_to_arr64(u64 *buf, const unsigned long *bitmap, unsigned int nbits);
>   #else
> diff --git a/lib/bitmap.c b/lib/bitmap.c
> index 1c81413c51f8..ddb31015e38a 100644
> --- a/lib/bitmap.c
> +++ b/lib/bitmap.c
> @@ -1495,7 +1495,7 @@ void bitmap_to_arr32(u32 *buf, const unsigned long *bitmap, unsigned int nbits)
>   EXPORT_SYMBOL(bitmap_to_arr32);
>   #endif
>   
> -#if (BITS_PER_LONG == 32) && defined(__BIG_ENDIAN)
> +#if BITS_PER_LONG == 32
>   /**
>    * bitmap_from_arr64 - copy the contents of u64 array of bits to bitmap
>    *	@bitmap: array of unsigned longs, the destination bitmap