Re: [PATCH] usbip: Fix the error limitation on max_hw_sectors for usbip device

[Zongmin Zhou <min_halo@163.com>]

On 2025/3/29 05:14, Shuah Khan wrote:
> On 3/13/25 04:02, Zongmin Zhou wrote:
>>
>> On 2025/3/11 00:49, Shuah Khan wrote:
>>> On 3/5/25 03:03, Zongmin Zhou wrote:
>>>> At 2025-03-05 03:45:28, "Shuah Khan" <skhan@linuxfoundation.org> 
>>>> wrote:
>>>>
>>>>> On 3/2/25 05:37, Zongmin Zhou wrote:
>>>>>> Dear shuah,
>>>>>>
>>>>>>
>>>>>> Yes, I agree with you.It would be better if there have a more 
>>>>>> simpler fixes than This patch.
>>>>>>
>>>>>> I can just think of the two possible solutions that mentioned 
>>>>>> before.
>>>>>
>>>>  >What are the two possible solutions?
>>>> 1. The patch we are discussing now,have to change the API between 
>>>> the kernel and user-space.
>>>
>>> 2. Simply set vhci-hcd dma mask to 64 by default,just modify the 
>>> vhci-hcd driver. Then dma_max_mapping_size() will always return 
>>> SIZE_MAX.
>>>
>>> I prefer option #2 - What are the downsides if any with this option?
>>>
>> If set vhci-hcd dma mask to 64 by default,I can't predict what will 
>> happen when the real USB controller support less than 64bit?
>>
>> After all, the data flows from vhci-hcd to usbip-host and finally to 
>> the USB controller to which the device is actually connected.
>>
>> the data is ultimately processed through the real USB controller?
>
> Sorry for the delay.
>
> That is the case. I have to check the code to see what the host
> would do if it receives larger buffers from the client (vhci)
>>
>> However, the default setting to 64-bit is equivalent to eliminating 
>> the impact of
>>
>> the patch(commit d74ffae8b8dd) on usbip protocol devices, sounds 
>> feasible?
>>
>> I am not very professional in this field, waiting for your evaluation.
>
> We can give this a try. Send me the patch with default testing the
> following cases:
>
> Host - swiotlb enabled and disabled in your environment to see what
> happens when there is a mismatch swiotlb enabled case and client
> side doesn't limit the size.

If you want to test swiotlb disabled mode, you can modify the kernel cmd 
to force disable swiotlb:

modify the grub.cfg, add the swiotlb=noforce parameter to kernel command 
line,and reboot.

cat /proc/cmdline to check whether modified successfully.


The patch set vhci-hcd dma mask to 64 by default like below:

---
  drivers/usb/usbip/vhci_hcd.c | 3 +++
  1 file changed, 3 insertions(+)

diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
index e70fba9f55d6..fca3a4a6e94d 100644
--- a/drivers/usb/usbip/vhci_hcd.c
+++ b/drivers/usb/usbip/vhci_hcd.c
@@ -1345,6 +1345,9 @@ static int vhci_hcd_probe(struct platform_device 
*pdev)

      usbip_dbg_vhci_hc("name %s id %d\n", pdev->name, pdev->id);

+    /* Set the dma mask to support 64bit for vhci-hcd driver. */
+    dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
+
      /*
       * Allocate and initialize hcd.
       * Our private data is also allocated automatically.
-- 
2.34.1

>
> thanks,
> -- Shuah