Bug 75721

Summary:	3.14x Kernels Boot Up Slower Than Previous Versions
Product:	Other	Reporter:	KernelBug (77777)
Component:	Other	Assignee:	other_other
Status:	NEW ---
Severity:	normal	CC:	alan
Priority:	P1
Hardware:	x86-64
OS:	Linux
Kernel Version:	3.14.3	Subsystem:
Regression:	No	Bisected commit-id:
Attachments:	cpuinfo dmesg lspci meminfo messages syslog dmesg-3.10.40 dmesg-3.14.4

Description KernelBug 2014-05-08 08:30:07 UTC

Hi,

These are some of my system specs;

Asus G75VW
Slackware 14.1 x86_64
Samsung 840 Pro SSD 128GB
Intel i7-3610QM
Nvidia GTX660M - 331.67 drivers

Since I'm running an SSD the bootup speeds are very fast and it's easy to notice this visually, seeing the text flying past the screen. I noticed right away with the 3.14x kernels that the text was moving much slower across the screen so I clocked the time with a stop watch and found out that the 3.14x kernels on my box boot up 3-4 seconds slower.

My box typically boots up on my ssd in 5 seconds, then using 3.14x 8-9 seconds, this is quite a slow down.

Another Slackware user said he literally went back to 3.10x because of the same issue of slow booting.

I'm attaching a few logs, please let me know what else you need.

I hope this issue gets fixed soon...

thanks

Comment 1 KernelBug 2014-05-08 08:30:29 UTC

Created attachment 135381 [details]
cpuinfo

Comment 2 KernelBug 2014-05-08 08:30:45 UTC

Created attachment 135391 [details]
dmesg

Comment 3 KernelBug 2014-05-08 08:31:04 UTC

Created attachment 135401 [details]
lspci

Comment 4 KernelBug 2014-05-08 08:31:19 UTC

Created attachment 135411 [details]
meminfo

Comment 5 KernelBug 2014-05-08 08:31:37 UTC

Created attachment 135421 [details]
messages

Comment 6 KernelBug 2014-05-08 08:31:51 UTC

Created attachment 135431 [details]
syslog

Comment 7 KernelBug 2014-05-14 20:09:56 UTC

I was hoping this was just some glitch/bug in 3.14.3, but I'm also experiencing the same thing in 3.14.4

I only run Openbox 3.5.2 and I don't use any display managers, only startx to log into X. I also noticed it's around 2 seconds slower when I type startx before getting into X, compared to 3.13x & 3.10x kernels.

I mean no one any disrespect, you all are doing a great job, but I want to make it clear I didn't buy an SSD to go slower, so PLEASE we really need to figure this out.

thanks

Comment 8 Alan 2014-05-19 11:03:57 UTC

As you are using the proprietary Nvidia drivers you'll to reproduce it without them, or take it up with Nvidia - they have our source code, we don't have theirs.

If you can reproduce it without them then capture the dmesg from boot with the timestamps for both a fast and a slow boot, that usually gives a good pointer as to what has caused the problem when compared side by side.

Also please understand the kernel bugzilla is used to track bugs across distributions and is not a bug fixing service. For that please contact your distribution.

Comment 9 KernelBug 2014-05-21 01:06:15 UTC

timestamps for both a fast and a slow boot, sorry I don't follow, how am I doing this?

thaks...

Comment 10 KernelBug 2014-05-22 03:46:07 UTC

Ok I added to lilo.conf; 

printk.time=1

I'm only getting timestamps up to this section below for 3.14.4;

USB Video Class driver (1.1.1)

I'm not seeing this as an accurate representation of what is fully happening, I'm noticing where the file system loads at this point;

EXT4-fs (sda1): re-mounted. Opts: (null)

This is where I'm seeing more of a slow down in 3.14x kernels over 3.10x

I'm attaching a dmesg for 3.10.40 and 3.14.4

Comment 11 KernelBug 2014-05-22 03:46:32 UTC

Created attachment 137061 [details]
dmesg-3.10.40

Comment 12 KernelBug 2014-05-22 03:46:53 UTC

Created attachment 137071 [details]
dmesg-3.14.4

Comment 13 Alan 2014-05-22 14:06:43 UTC

The kernel time stamping should be accurate, but it doesn't tell you what user space is doing. (although you can of course always double check that by recording it on a phone at a known frame rate and seeing if it tallies up). In this case it appears to slowly diverge.

I dont' see anything obvious in the two traces (usually you see a big delay in one driver or similar for such problems).

Do the two kernels have the same configuration options ?

Comment 14 KernelBug 2014-05-23 01:51:42 UTC

You don't see that 3.10 was faster?

3.10;

2.534971] USB Video Class driver (1.1.1)

3.14x

4.159329] USB Video Class driver (1.1.1)

Well at the userspace 3.14x does drag slower then 3.10x

I've recorded this off a stopwatch many times between the two versions and from start, to the prompt login I do get anywhere from a 2-4 second difference, from beginning to end between the two versions.

Both kernels have the exact configuration files being used to test this.