LSI 9300-8i, flaky boot handshake

[tramblock]

I have an Inspur branded 9300-8i which has an unreliable handshake at boot with my Asus Z97 motherboard.

When it handshakes with no lane errors, I get full 8x speed and no issues even under heavy disk IO until rolling the dice at the next boot. It will run for weeks or months.

Most of the time, booting (warm or cold seems not to matter) results in a reduced speed handshake at 2.5x and or many lane errors. I don't trust it as once under heavy disk IO I had a kernel panic. Basically I just reboot, trying this or that until I get a proper handshake hich is not great for the disks.

I haven't worked out exactly the pattern of how I get it to the solid handshake, but it seems like entering the mobo bios settings and interrogating the HBA attached disks 1 by 1 and exiting without changing any BIOS settings might be the answer. Not too sure yet.

I've updated the card FW to 16.00.12.00. Updated the mobo BIOS to latest stable.

Not sure where to go next. Google searches don't really find anything and Gemini chats just lead to neurosis.

I would just buy a new motherboard but it would entail replacing the DDR3 RAM I already have. I may have to take that option.

Would you try a different card first?

Or can someone recommend what mobo settings I should be using here? Gemini seems to alternately recommend UEFI and Legacy booting of the card as the "gold standard" depending on what day it is.

 

[TrevorH]

Did you try reseating it in the slot? Or cleaning the contacts on the pcie connector with isopropyl alcohol & a cotton bud? And when screwing the bracket down at the back, does it make the card lift out of the slot?

 

[tramblock]

Did you try reseating it in the slot? Or cleaning the contacts on the pcie connector with isopropyl alcohol & a cotton bud? And when screwing the bracket down at the back, does it make the card lift out of the slot?

Thanks for the reply.

Yes, I've reseated it many times. Also tried the card in the other PCI-E 3.0 slot. I cleaned the contacts, but with an eraser rather than isopropyl. When screwing the bracket down It does affect the card seating, but it's barely perceptible and I've gotten stable connections both with it screwed down and not screwed down.

 

[BLinux]

@tramblock when it does PCIe link negotiation, does it always link at x8 lanes? or does it go down to x4, x2, x1? Is it just the link speed that is unstable but all 8 lanes are stable?

also, check your PCIe power management settings and disable it and see if that helps.

 

[tramblock]

@tramblock when it does PCIe link negotiation, does it always link at x8 lanes? or does it go down to x4, x2, x1? Is it just the link speed that is unstable but all 8 lanes are stable?

also, check your PCIe power management settings and disable it and see if that helps.

@BLinux Thanks for the reply.

I think it said x4 lanes, but I'll have to brave a restart to know for sure.
For a good handshake, pciconf outputs:
link x8(x8) speed 8.0(8.0)

Most of the time when the handshake is bad, I guess this would appear as:
link x8(x4) speed 8.0(2.5)

That said, sometimes this line looks OK - full lanes and speed, but if lane errors are present I will eventually see problems.
PCIe Sec 1 lane errors 0xff <---anything other than 0 here and eventually I see problems

The key to a stable handshake is seeing:
PCIe Sec 1 lane errors 0

PCie power management is disabled in the BIOS.

 

[BLinux]

@BLinux Thanks for the reply.

I think it said x4 lanes, but I'll have to brave a restart to know for sure.
For a good handshake, pciconf outputs:
link x8(x8) speed 8.0(8.0)

Most of the time when the handshake is bad, I guess this would appear as:
link x8(x4) speed 8.0(2.5)

Definitely seems like you have some bad PCIe lanes...

That said, sometimes this line looks OK - full lanes and speed, but if lane errors are present I will eventually see problems.
PCIe Sec 1 lane errors 0xff <---anything other than 0 here and eventually I see problems

The key to a stable handshake is seeing:
PCIe Sec 1 lane errors 0

PCie power management is disabled in the BIOS.

A couple of things that have helped me in the past:

1. If this is a used card, check the PCIe gold fingers and see if they are dirty. You can use a pencil eraser to clean it up, or mild abrasive polishing compound.

2. Each PCIe lane has a pair of capacitors on the PCIe connector. Check them out closely and see if any of them are loose. Gently rub your fingers over them left to right and vice versa (with the gold fingers pointing down) and see if it moves them. Sometimes these get damaged and if they are somewhat loose, they can work intermittently or cause signal downgrade.