linux-kernel - Re: efisubsys_init takes more than a few milliseconds

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <13ebec39-6656-ee60-58be-afe56adfde7f@molgen.mpg.de>
Date:   Wed, 22 Aug 2018 15:18:33 +0200
From:   Paul Menzel <pmenzel+linux-efi@...gen.mpg.de>
To:     Ard Biesheuvel <ard.biesheuvel@...aro.org>
Cc:     linux-efi@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: efisubsys_init takes more than a few milliseconds

Dear Ard,


On 03/25/18 17:21, Paul Menzel wrote:

> On 03/25/2018 09:41 AM, Paul Menzel wrote:
> 
>> On 03/24/2018 11:35 PM, Ard Biesheuvel wrote:
>>
>>> On 24 March 2018 at 22:10, Paul Menzel wrote:
> 
>>>> According to `initcall_debug`, `efisubsys_init` takes more than a few
>>>> milliseconds to execute on a Dell XPS 13 9370 (Intel(R) Core(TM) i7-8550U CPU @ 1.80GHz).
>>>>
>>>>> ```
>>>>> […]
>>>>> [ 0.144474] calling  efisubsys_init+0x0/0x2cf @ 1
>>>>> [ 0.144474] Registered efivars operations
>>>>> [ 0.173690] initcall efisubsys_init+0x0/0x2cf returned 0 after 27343 usecs
>>>>> […]
>>>>> ```
>>>>
>>>>
>>>> To get a vanilla Linux kernel to boot in well under one second, it’d be nice
>>>> if the time could be improved. Do you know, why it takes so long?
>>>>
>>>> According to `bootgraph.py` from pm-graph [1][2] it takes even a little
>>>> longer.
>>>>
>>>>> efisubsys_init: start=690.841, end=720.493, length(w/o overhead)=31.250
>>>>> ms, return=0
>>>>
>>>> There are several dozen calls to `virt_efi_get_next_variable()` all but one
>>>> taking around 0.335 ms. This path needs to be optimized. Is that possible?
>>>
>>> That depends. These are firmware calls, so to make these calls faster,
>>> you need to modify the firmware, not the kernel.
>>
>> Yeah, unfortunately, no free firmware runs on this laptop, and Dell doesn’t respond to these kind of reports, as they think, it’s not important.
>>
>>> We may be able to make more intrusive changes to get rid of this
>>> delay, e.g., spin up a special kernel thread, but I'd have to check in
>>> more detail.
>>
>> That’d be great.

On the MSI B350M MORTAR (MS-7A37) it takes even longer, almost 50 ms.

```
[    0.204009] calling  efisubsys_init+0x0/0x30a @ 1
[    0.204009] Registered efivars operations
[    0.252262] initcall efisubsys_init+0x0/0x30a returned 0 after 46875 usecs
```

If a solution could be found to move this out of the hotpath, that’d be great.
But I am not sure, if threads can be used that early.

>>> In the mean time, you can try passing 'efi=noruntime' to the kernel.

[…]

>>>> To reproduce this, clone the pm-graph repository [2], use `sudo
>>>> ./bootgraph.py -f -fstat -maxdepth 10 -manual` to see what to add to
>>>> `/boot/grub/grub.cfg`. Then reboot, and execute `sudo ./bootgraph.py -f
>>>> -fstat -maxdepth 10`.
>>>>
>>>> If your system is powerful enough, you can use a higher maximum depth. I
>>>> didn’t get around how `-cgfilter` works to get smaller HTML files.


Kind regards,

Paul


>>>> [1] https://01.org/suspendresume
>>>> [2] https://github.com/01org/pm-graph


Download attachment "smime.p7s" of type "application/pkcs7-signature" (5174 bytes)