Re: [PATCH v4 2/5] gpio: Add support for Intel LJCA USB GPIO driver

[Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>]

On Fri, Mar 10, 2023 at 03:39:07PM +0800, Ye, Xiang wrote:
> On Fri, Mar 10, 2023 at 08:11:04AM +0100, Greg Kroah-Hartman wrote:
> > On Fri, Mar 10, 2023 at 01:01:11PM +0800, Ye, Xiang wrote:
> > > On Thu, Mar 09, 2023 at 02:40:10PM +0100, Oliver Neukum wrote:
> > > > 
> > > > 
> > > > On 09.03.23 08:10, Ye Xiang wrote:
> > > > 
> > > > > +#define LJCA_GPIO_BUF_SIZE 60
> > > > > +struct ljca_gpio_dev {
> > > > > +	struct platform_device *pdev;
> > > > > +	struct gpio_chip gc;
> > > > > +	struct ljca_gpio_info *gpio_info;
> > > > > +	DECLARE_BITMAP(unmasked_irqs, LJCA_MAX_GPIO_NUM);
> > > > > +	DECLARE_BITMAP(enabled_irqs, LJCA_MAX_GPIO_NUM);
> > > > > +	DECLARE_BITMAP(reenable_irqs, LJCA_MAX_GPIO_NUM);
> > > > > +	u8 *connect_mode;
> > > > > +	/* mutex to protect irq bus */
> > > > > +	struct mutex irq_lock;
> > > > > +	struct work_struct work;
> > > > > +	/* lock to protect package transfer to Hardware */
> > > > > +	struct mutex trans_lock;
> > > > > +
> > > > > +	u8 obuf[LJCA_GPIO_BUF_SIZE];
> > > > > +	u8 ibuf[LJCA_GPIO_BUF_SIZE];
> > > > 
> > > > And here we have a violation of DMA coherency rules.
> > > > Basically you cannot embed buffers into other data structures
> > > > if they can be subject to DMA.
> > > But obuf and ibuf does not used to do DMA transfer here.
> > > It is actually copied from or to ljca buffer to do URB transfer.
> > 
> > urb transfers _ARE_ DMA transfers.
> > 
> > > Should it still need to follow the DMA coherency rules?
> > 
> > Yes, all buffers for USB urbs are required to follow those rules.
> But these two buffers are not used to do USB urb transfer directly.
> For the "u8 obuf[LJCA_GPIO_BUF_SIZE]",  it will be copied to ljca buffer
> ("header->data" as below code [1] showed) first. Then the "header" is used
> to do the actual urb transfer.
> 
> And the "header" is allocated by using kmalloc. It should has met the DMA
> coherency rules.
> 
> [1] """
> struct ljca_msg *header;
> ...
> header = kmalloc(msg_len, GFP_KERNEL);
> if (!header)
> 	return -ENOMEM;

Ok, that's good, but why have 2 buffers for this then?

thanks,

greg k-h