Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH v3 1/2] usb: dwc2: alloc dma aligned buffer for isoc split
 in
To:     Doug Anderson <dianders@google.com>,
        William Wu <william.wu@rock-chips.com>
Cc:     hminas@synopsys.com, felipe.balbi@linux.intel.com,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>,
        =?UTF-8?Q?Heiko_St=c3=bcbner?= <heiko@sntech.de>,
        LKML <linux-kernel@vger.kernel.org>, linux-usb@vger.kernel.org,
        "open list:ARM/Rockchip SoC..." <linux-rockchip@lists.infradead.org>,
        Frank Wang <frank.wang@rock-chips.com>,
        =?UTF-8?B?6buE5rab?= <huangtao@rock-chips.com>,
        "daniel.meng" <daniel.meng@rock-chips.com>,
        John Youn <John.Youn@synopsys.com>,
        =?UTF-8?B?546L5b6B5aKe?= <wzz@rock-chips.com>,
        zsq@rock-chips.com,
        =?UTF-8?B?6Kix5ZiJ6YqY?= <Allen.Hsu@quantatw.com>,
        Stan Tsui <StanTsui@aopen.com>,
        =?UTF-8?B?U3BydWNlIFd1ICjlkLPlu7rli7Mp?= <Spruce.Wu@quantatw.com>,
        Martin.Tsai@quantatw.com, Kevin.Shai@quantatw.com,
        =?UTF-8?B?TW9uLUplciBXdSAo5ZCz5a2f5ZOyKQ==?= 
        <Mon-Jer.Wu@quantatw.com>,
        =?UTF-8?B?Q2xhdWQgQ2hhbmcgKOW8teaBreeviSk=?= 
        <Claud.Chang@quantatw.com>,
        =?UTF-8?B?U2FuIExpbiAo5p6X5bu66I+xKQ==?= <San.Lin@quantatw.com>,
        Ren.Kuo@quantatw.com, "David H.T. Wang" <davidhtwang@aopen.com>,
        Fong Lin <fonglin@aopen.com>,
        Steven Cheng <stevencheng@aopen.com>,
        Tom Chen <tomchen@aopen.com>, donchang@aopen.com,
        milesschofield@aopen.com
References: <1525748846-7767-1-git-send-email-william.wu@rock-chips.com>
 <1525748846-7767-2-git-send-email-william.wu@rock-chips.com>
 <CAD=FV=VKBVp++gPXzYN0N46NVarPNWqs1YZjfcd7twkU25WNyQ@mail.gmail.com>
From:   wlf <wulf@rock-chips.com>
Message-ID: <f2e0fba2-da0b-38ba-6e25-3da9ccf1e4ee@rock-chips.com>
Date:   Tue, 8 May 2018 15:43:55 +0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.7.0
MIME-Version: 1.0
In-Reply-To: <CAD=FV=VKBVp++gPXzYN0N46NVarPNWqs1YZjfcd7twkU25WNyQ@mail.gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Dear Doug,

在 2018年05月08日 13:11, Doug Anderson 写道:
> Hi,
>
> On Mon, May 7, 2018 at 8:07 PM, William Wu <william.wu@rock-chips.com> wrote:
>> +static int dwc2_alloc_split_dma_aligned_buf(struct dwc2_hsotg *hsotg,
>> +                                           struct dwc2_qh *qh,
>> +                                           struct dwc2_host_chan *chan)
>> +{
>> +       if (!hsotg->unaligned_cache)
>> +               return -ENOMEM;
>> +
>> +       if (!qh->dw_align_buf) {
>> +               qh->dw_align_buf = kmem_cache_alloc(hsotg->unaligned_cache,
>> +                                                   GFP_ATOMIC | GFP_DMA);
>> +               if (!qh->dw_align_buf)
>> +                       return -ENOMEM;
>> +
>> +               qh->dw_align_buf_size = min_t(u32, chan->max_packet,
>> +                                             DWC2_KMEM_UNALIGNED_BUF_SIZE);
> Rather than using min_t, wouldn't it be better to return -ENOMEM if
> "max_packet" > DWC2_KMEM_UNALIGNED_BUF_SIZE?  As it is, you might
> allocate less space than you need, right?  That seems like it would be
> bad (even though this is probably impossible).
Yes, good idea! So is it good to fix it like this?

if (!qh->dw_align_buf || chan->max_packet >
     DWC2_KMEM_UNALIGNED_BUF_SIZE)
	return -ENOMEM;	

qh->dw_align_buf_size = chan->max_packet;

>
>> @@ -2797,6 +2837,32 @@ static int dwc2_assign_and_init_hc(struct dwc2_hsotg *hsotg, struct dwc2_qh *qh)
>>          /* Set the transfer attributes */
>>          dwc2_hc_init_xfer(hsotg, chan, qtd);
>>
>> +       /* For non-dword aligned buffers */
>> +       if (hsotg->params.host_dma && qh->do_split &&
>> +           chan->ep_is_in && (chan->xfer_dma & 0x3)) {
>> +               dev_vdbg(hsotg->dev, "Non-aligned buffer\n");
>> +               if (dwc2_alloc_split_dma_aligned_buf(hsotg, qh, chan)) {
>> +                       dev_err(hsotg->dev,
>> +                               "Failed to allocate memory to handle non-aligned buffer\n");
>> +                       /* Add channel back to free list */
>> +                       chan->align_buf = 0;
>> +                       chan->multi_count = 0;
>> +                       list_add_tail(&chan->hc_list_entry,
>> +                                     &hsotg->free_hc_list);
>> +                       qtd->in_process = 0;
>> +                       qh->channel = NULL;
>> +                       return -ENOMEM;
>> +               }
>> +       } else {
>> +               /*
>> +                * We assume that DMA is always aligned in non-split
>> +                * case or split out case. Warn if not.
>> +                */
>> +               WARN_ON_ONCE(hsotg->params.host_dma &&
>> +                            (chan->xfer_dma & 0x3));
>> +               chan->align_buf = 0;
>> +       }
>> +
>>          if (chan->ep_type == USB_ENDPOINT_XFER_INT ||
>>              chan->ep_type == USB_ENDPOINT_XFER_ISOC)
>>                  /*
>> @@ -5241,6 +5307,17 @@ int dwc2_hcd_init(struct dwc2_hsotg *hsotg)
>>                          hsotg->params.dma_desc_enable = false;
>>                          hsotg->params.dma_desc_fs_enable = false;
>>                  }
>> +       } else if (hsotg->params.host_dma) {
> Are you sure this is "else if"?  Can't you have descriptor DMA enabled
> in the controller and still need to do a normal DMA transfer if you
> plug in a hub?  Seems like this should be just "if".
Sorry, I don't understand the case "have descriptor DMA enabled in the 
controller and still need to do a normal DMA transfer". But maybe it 
still has another problem if just use "if" here, because it will create 
kmem caches for Slave mode which actually doesn't need aligned DMA buf.
>
>
>> +               /*
>> +                * Create kmem caches to handle non-aligned buffer
>> +                * in Buffer DMA mode.
>> +                */
>> +               hsotg->unaligned_cache = kmem_cache_create("dwc2-unaligned-dma",
>> +                                               DWC2_KMEM_UNALIGNED_BUF_SIZE, 4,
> Worth using "DWC2_USB_DMA_ALIGN" rather than 4?
>
>
>> +                                               SLAB_CACHE_DMA, NULL);
>> +               if (!hsotg->unaligned_cache)
>> +                       dev_err(hsotg->dev,
>> +                               "unable to create dwc2 unaligned cache\n");
>>          }
>>
>>          hsotg->otg_port = 1;
>> @@ -5279,6 +5356,7 @@ int dwc2_hcd_init(struct dwc2_hsotg *hsotg)
>>   error4:
>>          kmem_cache_destroy(hsotg->desc_gen_cache);
>>          kmem_cache_destroy(hsotg->desc_hsisoc_cache);
>> +       kmem_cache_destroy(hsotg->unaligned_cache);
> nitty nit: freeing order should be opposite of allocation, so the new
> line should be above the other two.
Ah, I got it. But note that it's impossible to allocate the 
"unaligned_cache" and "desc *cache" at the same time. Should we still 
fix the free order? If yes, maybe the correct free order is:

kmem_cache_destroy(hsotg->unaligned_cache);
kmem_cache_destroy(hsotg->desc_hsisoc_cache);
kmem_cache_destroy(hsotg->desc_gen_cache);

Right?

And should we also need to fix the same free order in the "dwc2_hcd_remove"?
Best regards,
         wulf
>
>