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;
Date:   Mon, 24 Jun 2019 23:58:53 +0900
From:   Suwan Kim <suwan.kim027@gmail.com>
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     shuah@kernel.org, gregkh@linuxfoundation.org,
        linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] usbip: Implement SG support to vhci
Message-ID: <20190624145852.GC7547@localhost.localdomain>
References: <20190621174553.28862-3-suwan.kim027@gmail.com>
 <Pine.LNX.4.44L0.1906211548560.1471-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Pine.LNX.4.44L0.1906211548560.1471-100000@iolanthe.rowland.org>
User-Agent: Mutt/1.11.3 (2019-02-01)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jun 21, 2019 at 04:05:24PM -0400, Alan Stern wrote:
> On Sat, 22 Jun 2019, Suwan Kim wrote:
> 
> > There are bugs on vhci with usb 3.0 storage device. Originally, vhci
> > doesn't supported SG. So, USB storage driver on vhci divides SG list
> > into multiple URBs and it causes buffer overflow on the xhci of the
> > server. So we need to add SG support to vhci
> 
> It doesn't cause buffer overflow.  The problem was that a transfer got
> terminated too early because the transfer length for one of the URBs
> was not divisible by the maxpacket size.

Oh.. I misunderstood the problem. I will rewrite the problem
situation.

> > In this patch, vhci basically support SG and it sends each SG list
> > entry to the stub driver. Then, the stub driver sees total length of
> > the buffer and allocates SG table and pages according to the total
> > buffer length calling sgl_alloc(). After the stub driver receives
> > completed URB, it again sends each SG list entry to the vhci.
> > 
> > If HCD of server doesn't support SG, the stub driver allocates
> > big buffer using kmalloc() instead of using sgl_alloc() which
> > allocates SG list and pages.
> 
> You might be better off not using kmalloc.  It's easier for the kernel 
> to allocate a bunch of small buffers than a single big one.  Then you 
> can create a separate URB for each buffer.

Ok. I will implement it as usb_sg_init() does and send v2 patch
including the logic of submitting separate URBs.

> > Alan fixed vhci bug with the USB 3.0 storage device by modifying
> > USB storage driver.
> > ("usb-storage: Set virt_boundary_mask to avoid SG overflows")
> > But the fundamental solution of it is to add SG support to vhci.
> > 
> > This patch works well with the USB 3.0 storage devices without Alan's
> > patch, and we can revert Alan's patch if it causes some troubles.
> 
> These last two paragraphs don't need to be in the patch description.

I will remove these paragraphs in v2 patch.

> > Signed-off-by: Suwan Kim <suwan.kim027@gmail.com>
> > ---
> 
> I'm not sufficiently familiar with the usbip drivers to review most of 
> this.  However...
> 
> > diff --git a/drivers/usb/usbip/vhci_hcd.c b/drivers/usb/usbip/vhci_hcd.c
> > index be87c8a63e24..cc93c1a87a3e 100644
> > --- a/drivers/usb/usbip/vhci_hcd.c
> > +++ b/drivers/usb/usbip/vhci_hcd.c
> > @@ -696,7 +696,8 @@ static int vhci_urb_enqueue(struct usb_hcd *hcd, struct urb *urb, gfp_t mem_flag
> >  	}
> >  	vdev = &vhci_hcd->vdev[portnum-1];
> >  
> > -	if (!urb->transfer_buffer && urb->transfer_buffer_length) {
> > +	if (!urb->transfer_buffer && !urb->num_sgs &&
> > +	     urb->transfer_buffer_length) {
> >  		dev_dbg(dev, "Null URB transfer buffer\n");
> >  		return -EINVAL;
> >  	}
> > @@ -1142,6 +1143,11 @@ static int vhci_setup(struct usb_hcd *hcd)
> >  		hcd->speed = HCD_USB3;
> >  		hcd->self.root_hub->speed = USB_SPEED_SUPER;
> >  	}
> > +
> > +	/* support sg */
> > +	hcd->self.sg_tablesize = ~0;
> > +	hcd->self.no_sg_constraint = 1;
> 
> You probably shouldn't do this, for two reasons.  First, sg_tablesize
> of the server's HCD may be smaller than ~0.  If the client's value is
> larger than the server's, a transfer could be accepted on the client
> but then fail on the server because the SG list was too big.
> 
> Also, you may want to restrict the size of SG transfers even further,
> so that you don't have to allocate a tremendous amount of memory all at
> once on the server.  An SG transfer can be quite large.  I don't know 
> what a reasonable limit would be -- 16 perhaps?

Is there any reason why you think that 16 is ok? Or Can I set this
value as the smallest value of all HC? I think that sg_tablesize
cannot be a variable value because vhci interacts with different
machines and all machines has different sg_tablesize value.

Regards

Suwan Kim