On 9/15/2023 9:51 AM, Alan Stern wrote:

On Fri, Sep 15, 2023 at 09:02:48AM +0800, Linyu Yuan wrote:

quoted

On 9/14/2023 10:54 PM, Alan Stern wrote:

quoted

You didn't include the version number in the Subject: line.  Undoubtedly
Greg's automatic error checker will warn you about this.  Unless the
version number is clearly marked for each patch, it's difficult for his
programs to tell which email message contains the most recent version.

On Thu, Sep 14, 2023 at 06:02:56PM +0800, Linyu Yuan wrote:

quoted

Some UDC trace event will save usb udc information, but it use one int
size buffer to save one bit information of usb udc, it is wast trace
buffer.

Add anonymous union which have one u32 member can be used by trace event
during fast assign stage to save more entries with same trace ring buffer
size.

Signed-off-by: Linyu Yuan <redacted>
---

And you didn't include the version change information here, below the
"---" line.

Apart from that, this is a _lot_ better than before!  I don't know if
Greg will think this change is worth merging, but at least now it's
possible to read the code and understand what's going on.

according Steven's comment, maybe will always save data in little endian at
trace event

fast assign stage.

it will add definition of bit field back.

Yes, that would be even better because you wouldn't have to change the
definition of struct usb_gadget or struct usb_endpoint at all.  The fast
assign stage can simply do:

	__entry->dw1 = (g->sg_supported << 0) |
			(g->is_otg << 1) |
			...

and then you can easily access the individual bits in __entry.  It
wouldn't be as fast but it would still save a lot of space.

how about __entry->dw1 = cpu_to_le32(g->dw1) ?

Alan Stern