Re: [PATCH] io_uring: fix bug of wrong BUILD_BUG_ON check

[Pavel Begunkov <asml.silence@gmail.com>]

On 9/6/21 6:03 PM, Hao Xu wrote:
> 在 2021/9/6 下午11:22, Pavel Begunkov 写道:
>> On 9/6/21 4:12 PM, Hao Xu wrote:
>>> Some check should be large than not equal or large than.
>>>
>>> Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
>>> ---
>>>   fs/io_uring.c | 6 +++---
>>>   1 file changed, 3 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>> index 2bde732a1183..3a833037af43 100644
>>> --- a/fs/io_uring.c
>>> +++ b/fs/io_uring.c
>>> @@ -10637,13 +10637,13 @@ static int __init io_uring_init(void)
>>>                sizeof(struct io_uring_rsrc_update2));
>>>         /* ->buf_index is u16 */
>>> -    BUILD_BUG_ON(IORING_MAX_REG_BUFFERS >= (1u << 16));
>>> +    BUILD_BUG_ON(IORING_MAX_REG_BUFFERS > (1u << 16));
>>>         /* should fit into one byte */
>>> -    BUILD_BUG_ON(SQE_VALID_FLAGS >= (1 << 8));
>>> +    BUILD_BUG_ON(SQE_VALID_FLAGS > (1 << 8));
>>
>> 0xff = 255 is the largest number fitting in u8,
>> 1<<8 = 256.
>>
>> let SQE_VALID_FLAGS = 256,
>> (256 > (1<<8)) == (256 > 256) == false,  even though it can't
>> be represented by u8.
> Isn't SQE_VALID_FLAGS = 256 a valid value for it?

SQE_VALID_FLAGS is a "bitwise OR" combination of valid flags, so
can't be go beyond 0xff

>>
>>
>>>       BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
>>> -    BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
>>> +    BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int));
>>>         req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
>>>                   SLAB_ACCOUNT);
>>>
>>
> 

-- 
Pavel Begunkov