> From: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> > Sent: Friday, April 22, 2022 4:43 PM > To: Linyu Yuan (QUIC) <quic_linyyuan@xxxxxxxxxxx> > Cc: Felipe Balbi <balbi@xxxxxxxxxx>; linux-usb@xxxxxxxxxxxxxxx; Jack Pham > (QUIC) <quic_jackp@xxxxxxxxxxx> > Subject: Re: [PATCH v6 0/5] usb: gadget: configfs: new trace events > > On Thu, Apr 07, 2022 at 09:27:40AM +0800, Linyu Yuan wrote: > > Last year I try to add trace event support for usb gadget configfs [1], > > this time the idea is change a lot, the purpose is trace all user space > > operation to gadget configuration, include gadget and it's function. > > But why? Who will use this, and what for? Thanks for review it. It is not used by user space, just for kernel gadget issue debugging. Like android, the user space is complicate to kernel developers, With this trace events, kernel development will understand What is simplified action happen from user space. > > > In usb gadget configfs, mainly user can do mkdir/rmdir a group, > > link/unlink a function, change gadget/function attributes, > > each operation will touch a struct config_item. > > As userspace is the thing doing this, why do you need to tell userspace > again that this happened? > > > It only have one trace event entry which store string, > > provide several API which represent user operation and generate string > > from struct config_item. > > > > example output, > > mkdir-80 [000] ..... 44.555106: gadget_configfs: mkdir dummy > > sh-76 [000] ..... 44.562609: gadget_configfs: echo dummy/idVendor > 0x05C6 > > > > mkdir-81 [000] ..... 44.569795: gadget_configfs: mkdir > dummy/functions/eem.0 > > sh-76 [000] ..... 44.600221: gadget_configfs: echo > dummy/functions/eem.0/dev_addr 1e:77:46:4b:1e:96 > > > > mkdir-82 [000] ..... 44.615542: gadget_configfs: mkdir > dummy/configs/dummy.1 > > ln-83 [000] ..... 44.628997: gadget_configfs: link > dummy/configs/dummy.1 dummy/functions/eem.0 > > sh-76 [000] ..... 44.634506: gadget_configfs: echo > dummy/configs/dummy.1/MaxPower 500 > > > > mkdir-84 [000] ..... 44.642265: gadget_configfs: mkdir > dummy/configs/dummy.1/strings/0x409 > > sh-76 [000] ..... 44.663886: gadget_configfs: echo > dummy/configs/dummy.1/strings/0x409/configuration dummy > > > > rmdir-85 [000] ..... 64.255507: gadget_configfs: rmdir > dummy/configs/dummy.1/strings/0x409 > > rm-86 [000] ..... 64.263926: gadget_configfs: unlink > dummy/configs/dummy.1 dummy/functions/eem.0 > > rmdir-87 [000] ..... 64.279768: gadget_configfs: rmdir > dummy/configs/dummy.1 > > rmdir-88 [000] ..... 64.328124: gadget_configfs: rmdir > dummy/functions/eem.0 > > rmdir-89 [000] ..... 64.992085: gadget_configfs: rmdir dummy > > As I said in other places, why not just add this to configfs directly > instead of all over the individual users of this one subsystem? > > And again, why? I consider it also, but I found most of fs have no such trace event support. I will try that. Thanks very much again. > > thanks, > > greg k-h
