BIOS region alignment on ivybridge platforms.

[persmule]

I have reported that a coreboot image with a cleansed and truncated ME region may cause the first boot costing more than 5 minutes. Now I confirm that in order to prevent this, the bios region should be aligned to 64KiB (0x10000) on ivybridges. Sandybridges do not have such limitation.

[corna]

Interesting.
Are you sure that this isn't an isolated case?

[dogroll]

This might be related to the issue I'm having.

I am running coreboot 4.6-86-g72f730e23c on my T520 (with i5-3320M). With me_cleaner enabled in the coreboot build, I get frequent boot times (Windows 10) of around 60 seconds---incredibly long. The unusually long boots tend to occur immediately after a restart/shutdown. However, sometimes when starting the machine from a long term power off state (e.g. I shutdown and leave the laptop for a few minutes before turning it back on again) I would get an incredibly fast ~3 second boot. I have tried narrowing down under which conditions the 60 and ~3 second boots occur, but they seem somewhat unpredictable. All I can really tell is that sometimes when I would cold boot the machine after it was off for a while I would get a quick boot, but restarting the machine would yield a slow 60 second boot 100% of the time.

With me_cleaner disabled, I get a standard ~8 second boot from my mSATA SSD.

[yhaenggi]

@dogroll im planing to flash my T520 aswell. Are there any other issues or everything working fine after applying me_cleaner? Excuse the offtopic and thanks in advance :)

[dogroll]

Hi, I couldn't discern any other issues, no 30 minute shutoff etc, so me_cleaner seems to work fine for the most part. But obviously the boot time issue is a bit of a turn off.

I'm not a linux user so I didn't test that, perhaps this issue only affects windows.

[persmule]

Are you sure that this isn't an isolated case?

[persmule]

Are you sure that this isn't an isolated case?

Confirmed both on x230 and ga-b75m-d3h.

[corna]

I finally had the time to check this.
On my Sapphire Pure Platinum H61 with an i7-3770 and coreboot I have this layout

00000000:00000fff fd
00018000:003fffff bios
00001000:00017fff me

but the PC boots fine without any delay. Do you have a timestamp log of those boots? If not build coreboot with the option CONFIG_COLLECT_TIMESTAMPS, flash it and then run cbmem -t on the first boot after the flash.

@dogroll having a timestamp log for each boot (slow and fast) could be particularly useful. I'm afraid you'll need Linux though.

[persmule]

I have completed the test on my thinkpad x230.
Below are two time stamp reports, each of which is concatenated after its corresponding ifd layout:

bios region aligned to 0x1000

00000000:00000fff fd
0001b000:00bfffff bios
00003000:0001afff me
00001000:00002fff gbe
23 entries total:

0:1st timestamp 2,294
1:start of rom stage 65,436 (63,141)
2:before ram initialization 74,347 (8,910)
3:after ram initialization 89,931,234 (89,856,887)
4:end of romstage 89,985,753 (54,519)
8:starting to load ramstage 89,988,974 (3,220)
15:starting LZMA decompress (ignore for x86) 89,989,162 (187)
16:finished LZMA decompress (ignore for x86) 90,014,342 (25,180)
9:finished loading ramstage 90,014,573 (231)
10:start of ramstage 90,014,708 (134)
30:device enumeration 90,014,719 (10)
40:device configuration 90,020,601 (5,882)
50:device enable 90,022,900 (2,298)
60:device initialization 90,023,040 (140)
70:device setup done 90,188,315 (165,274)
75:cbmem post 90,188,319 (3)
80:write tables 90,881,194 (692,875)
85:finalize chips 90,913,790 (32,595)
90:load payload 90,913,794 (4)
15:starting LZMA decompress (ignore for x86) 90,914,222 (427)
16:finished LZMA decompress (ignore for x86) 90,928,435 (14,213)
15:starting LZMA decompress (ignore for x86) 90,928,443 (8)
16:finished LZMA decompress (ignore for x86) 91,152,623 (224,179)
99:selfboot jump 91,152,651 (27)

Total Time: 91,150,344

bios region aligned to 0x10000:

00000000:00000fff fd
00020000:00bfffff bios
00003000:0001ffff me
00001000:00002fff gbe
23 entries total:

0:1st timestamp 2,312
1:start of rom stage 65,350 (63,037)
2:before ram initialization 67,009 (1,659)
3:after ram initialization 1,526,964 (1,459,955)
4:end of romstage 1,575,840 (48,875)
8:starting to load ramstage 1,579,432 (3,592)
15:starting LZMA decompress (ignore for x86) 1,579,768 (336)
16:finished LZMA decompress (ignore for x86) 1,609,570 (29,801)
9:finished loading ramstage 1,609,778 (208)
10:start of ramstage 1,609,892 (113)
30:device enumeration 1,609,903 (11)
40:device configuration 1,615,792 (5,889)
50:device enable 1,618,090 (2,297)
60:device initialization 1,618,231 (141)
70:device setup done 1,783,521 (165,289)
75:cbmem post 1,783,525 (4)
80:write tables 2,475,097 (691,571)
85:finalize chips 2,507,645 (32,548)
90:load payload 2,507,649 (4)
15:starting LZMA decompress (ignore for x86) 2,508,034 (384)
16:finished LZMA decompress (ignore for x86) 2,522,401 (14,367)
15:starting LZMA decompress (ignore for x86) 2,522,409 (8)
16:finished LZMA decompress (ignore for x86) 2,745,607 (223,197)
99:selfboot jump 2,745,635 (27)

Total Time: 2,743,313

[corna]

I've checked again on my Ivy Bridge platform with coreboot with full RAM training (MRC cache cleared) and with different RAM sticks combinations and the raminit is still fast, even if the BIOS region is not 64k-aligned. At this point it seems to me that your problem is specific for your board.

You should try to get a verbose raminit log and ask on the coreboot ML.

I had problems with RAM training on my board (with the ME intact) that required me to dump and restore the MRC cache at every coreboot update, but with recent builds the RAM training gets performed correctly with good performance: can you try again with the latest coreboot commits?

[persmule]

Timestamp for the first boot after flash on my x230 with the latest coreboot commits.
timestamp-x230.txt

[corna]

Do you have a verbose raminit log?

[persmule]

I have just collected one, along with the corresponding timestamp:
x230-log.txt
timestamp-x230.txt