Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [PATCH v8 2/3] tty: hvc: pass DMA capable memory to put_chars()
From:   Xianting Tian <xianting.tian@linux.alibaba.com>
To:     Daniel Axtens <dja@axtens.net>, gregkh@linuxfoundation.org,
        jirislaby@kernel.org, amit@kernel.org, arnd@arndb.de,
        osandov@fb.com
Cc:     shile.zhang@linux.alibaba.com, linuxppc-dev@lists.ozlabs.org,
        linux-kernel@vger.kernel.org,
        virtualization@lists.linux-foundation.org
References: <20210818082122.166881-1-xianting.tian@linux.alibaba.com>
 <20210818082122.166881-3-xianting.tian@linux.alibaba.com>
 <87pmu8ehkk.fsf@linkitivity.dja.id.au>
 <6e36640d-b55f-ece4-4cab-437ecec0964a@linux.alibaba.com>
Message-ID: <614b778b-8486-c20f-d5ed-37cc3b92ada1@linux.alibaba.com>
Date:   Sat, 18 Sep 2021 20:32:01 +0800
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:78.0)
 Gecko/20100101 Thunderbird/78.10.1
MIME-Version: 1.0
In-Reply-To: <6e36640d-b55f-ece4-4cab-437ecec0964a@linux.alibaba.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

hi

Will you consider to continue the disscussion of this patch? thanks

在 2021/8/20 下午4:43, Xianting TIan 写道:
>
> 在 2021/8/20 下午2:49, Daniel Axtens 写道:
>> Xianting Tian <xianting.tian@linux.alibaba.com> writes:
>>
>>> As well known, hvc backend driver(eg, virtio-console) can register its
>>> operations to hvc framework. The operations can contain put_chars(),
>>> get_chars() and so on.
>>>
>>> Some hvc backend may do dma in its operations. eg, put_chars() of
>>> virtio-console. But in the code of hvc framework, it may pass DMA
>>> incapable memory to put_chars() under a specific configuration, which
>>> is explained in commit c4baad5029(virtio-console: avoid DMA from 
>>> stack):
>> We could also run into issues on powerpc where Andrew is working on
>> adding vmap-stack but the opal hvc driver assumes that it is passed a
>> buffer which is not in vmalloc space but in the linear mapping. So it
>> would be good to fix this (or more clearly document what drivers can
>> expect).
>>
>>> 1, c[] is on stack,
>>>     hvc_console_print():
>>>     char c[N_OUTBUF] __ALIGNED__;
>>>     cons_ops[index]->put_chars(vtermnos[index], c, i);
>>> 2, ch is on stack,
>>>     static void hvc_poll_put_char(,,char ch)
>>>     {
>>>     struct tty_struct *tty = driver->ttys[0];
>>>     struct hvc_struct *hp = tty->driver_data;
>>>     int n;
>>>
>>>     do {
>>>         n = hp->ops->put_chars(hp->vtermno, &ch, 1);
>>>     } while (n <= 0);
>>>     }
>>>
>>> Commit c4baad5029 is just the fix to avoid DMA from stack memory, which
>>> is passed to virtio-console by hvc framework in above code. But I think
>>> the fix is aggressive, it directly uses kmemdup() to alloc new buffer
>>> from kmalloc area and do memcpy no matter the memory is in kmalloc area
>>> or not. But most importantly, it should better be fixed in the hvc
>>> framework, by changing it to never pass stack memory to the put_chars()
>>> function in the first place. Otherwise, we still face the same issue if
>>> a new hvc backend using dma added in the future.
>>>
>>> In this patch, we make 'char out_buf[N_OUTBUF]' and 'chat out_ch' part
>>> of 'struct hvc_struct', so both two buf are no longer the stack memory.
>>> we can use it in above two cases separately.
>>>
>>> Introduce another array(cons_outbufs[]) for buffer pointers next to
>>> the cons_ops[] and vtermnos[] arrays. With the array, we can easily 
>>> find
>>> the buffer, instead of traversing hp list.
>>>
>>> With the patch, we can remove the fix c4baad5029.
>>>
>>> Signed-off-by: Xianting Tian <xianting.tian@linux.alibaba.com>
>>> Reviewed-by: Shile Zhang <shile.zhang@linux.alibaba.com>
>>>   struct hvc_struct {
>>>       struct tty_port port;
>>>       spinlock_t lock;
>>>       int index;
>>>       int do_wakeup;
>>> -    char *outbuf;
>>> -    int outbuf_size;
>>>       int n_outbuf;
>>>       uint32_t vtermno;
>>>       const struct hv_ops *ops;
>>> @@ -48,6 +56,10 @@ struct hvc_struct {
>>>       struct work_struct tty_resize;
>>>       struct list_head next;
>>>       unsigned long flags;
>>> +    char out_ch;
>>> +    char out_buf[N_OUTBUF] __ALIGNED__;
>>> +    int outbuf_size;
>>> +    char outbuf[0] __ALIGNED__;
>> I'm trying to understand this patch but I am finding it very difficult
>> to understand what the difference between `out_buf` and `outbuf`
>> (without the underscore) is supposed to be. `out_buf` is statically
>> sized and the size of `outbuf` is supposed to depend on the arguments to
>> hvc_alloc(), but I can't quite figure out what the roles of each one are
>> and their names are confusingly similiar!
>>
>> I looked briefly at the older revisions of the series but it didn't make
>> things much clearer.
>>
>> Could you give them clearer names?
>
> thanks for the comments,
>
> It is indeed not easy to understand by the name. I will change it to a 
> proper name if we have next version patch.
>
> Jiri Slaby is worring about the performance, because we need add two 
> locks to protect 'out_ch' and 'out_buf' separately, the origin 
> on-stack buffer is lockless.
>
> I don't know whether this solution can be accepted, just waiting for 
> Jiri's further commtents.
>
>>
>> Also, looking at Documentation/process/deprecated.rst, it looks like
>> maybe we want to use a 'flexible array member' instead:
>>
>> .. note:: If you are using struct_size() on a structure containing a 
>> zero-length
>>          or a one-element array as a trailing array member, please 
>> refactor such
>>          array usage and switch to a `flexible array member
>>          <#zero-length-and-one-element-arrays>`_ instead.
>>
>> I think we want:
> thanks, we should use [], not [0].
>>
>>> +    char outbuf[] __ALIGNED__;
>> Kind regards,
>> Daniel