Reduce init time of `lp_init()` and `ppdev_init()`

Reduce init time of `lp_init()` and `ppdev_init()`

[Paul Menzel]

[-- Attachment #1: Type: text/plain, Size: 1646 bytes --]

Dear Linux folks,


I am trying to reduce the start-up time of the Linux kernel on an old 
Lenovo X60. Looking through the time stamps of Linux 4.16-rc7+, the 
modules `lp` and `ppdev` both take more than ten milliseconds to 
initialize according to `initcall_debug`.

```
[    8.337692] calling  parport_default_proc_register+0x0/0x1000 
[parport] @ 245
[    8.337735] initcall parport_default_proc_register+0x0/0x1000 
[parport] returned 0 after 30 usecs
[    8.359066] calling  lp_init_module+0x0/0x1000 [lp] @ 245
[    8.381317] lp: driver loaded but no devices found
[    8.385310] initcall lp_init_module+0x0/0x1000 [lp] returned 0 after 
25613 usecs
[    8.401727] calling  ppdev_init+0x0/0x1000 [ppdev] @ 245
[    8.411427] ppdev: user-space parallel port driver
[    8.415390] initcall ppdev_init+0x0/0x1000 [ppdev] returned 0 after 
13329 usecs
[    8.434018] calling  parport_pc_init+0x0/0xeed [parport_pc] @ 245
[    8.435657] initcall parport_pc_init+0x0/0xeed [parport_pc] returned 
0 after 1589 usecs
[    9.343691] EXT4-fs (dm-0): re-mounted. Opts: errors=remount-ro,discard
[    9.742084] systemd-journald[271]: Received request to flush runtime 
journal from PID 1
[   10.049050] calling  acpi_cpufreq_init+0x0/0x1000 [acpi_cpufreq] @ 297
[   10.049668] initcall acpi_cpufreq_init+0x0/0x1000 [acpi_cpufreq] 
returned 0 after 581 usecs
```

Furthermore, looking at the messages, it looks like it’s in the 
“hotpath”. If that is true, is there a way to load that in parallel of 
the other Linux kernel stuff?

Hints to get more insight are much appreciated.


Kind regards,

Paul


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 5174 bytes --]

Re: Reduce init time of `lp_init()` and `ppdev_init()`

[Greg Kroah-Hartman]

On Thu, Mar 29, 2018 at 05:57:20PM +0200, Paul Menzel wrote:
> Dear Linux folks,
> 
> 
> I am trying to reduce the start-up time of the Linux kernel on an old Lenovo
> X60. Looking through the time stamps of Linux 4.16-rc7+, the modules `lp`
> and `ppdev` both take more than ten milliseconds to initialize according to
> `initcall_debug`.
> 
> ```
> [    8.337692] calling  parport_default_proc_register+0x0/0x1000 [parport] @
> 245
> [    8.337735] initcall parport_default_proc_register+0x0/0x1000 [parport]
> returned 0 after 30 usecs
> [    8.359066] calling  lp_init_module+0x0/0x1000 [lp] @ 245
> [    8.381317] lp: driver loaded but no devices found
> [    8.385310] initcall lp_init_module+0x0/0x1000 [lp] returned 0 after
> 25613 usecs
> [    8.401727] calling  ppdev_init+0x0/0x1000 [ppdev] @ 245
> [    8.411427] ppdev: user-space parallel port driver
> [    8.415390] initcall ppdev_init+0x0/0x1000 [ppdev] returned 0 after 13329
> usecs
> [    8.434018] calling  parport_pc_init+0x0/0xeed [parport_pc] @ 245
> [    8.435657] initcall parport_pc_init+0x0/0xeed [parport_pc] returned 0
> after 1589 usecs
> [    9.343691] EXT4-fs (dm-0): re-mounted. Opts: errors=remount-ro,discard
> [    9.742084] systemd-journald[271]: Received request to flush runtime
> journal from PID 1
> [   10.049050] calling  acpi_cpufreq_init+0x0/0x1000 [acpi_cpufreq] @ 297
> [   10.049668] initcall acpi_cpufreq_init+0x0/0x1000 [acpi_cpufreq] returned
> 0 after 581 usecs
> ```
> 
> Furthermore, looking at the messages, it looks like it’s in the “hotpath”.
> If that is true, is there a way to load that in parallel of the other Linux
> kernel stuff?
> 
> Hints to get more insight are much appreciated.

Yes, either turn those into modules, which will get loaded later at boot
time, or turn on async probing for them and see if that works.

Or better yet, don't load them at all until after userspace fully starts
with an init script as part of an early bringup of cups.  Odds are you
don't need those modules at all to start the machine, right?

thanks,

greg k-h