Threaded Mode | Linear Mode

pxe_curious · 2018-08-17, 23:22

Testing with a Dell PowerEdge R640 in UEFI mode and the X710 rNDC.

Loading iPXE and then WIMBOOT then WinPE based on Windows 10 1803 (and also tried 1607 and 1709).

Using the following iPXE script:

Code:

imgfree

imgselect ${ks-ipxe-url}/win/wimboot gui

imgfetch --name bcd                ${ks-ipxe-url}/win/bcd                       bcd

imgfetch --name boot.sdi           ${ks-ipxe-url}/win/boot.sdi                  boot.sdi

imgfetch --name segmono_boot.ttf   ${ks-ipxe-url}/win/fonts/segmono_boot.ttf    segmono_boot.ttf

imgfetch --name segoe_slboot.ttf   ${ks-ipxe-url}/win/fonts/segoe_slboot.ttf    segoe_slboot.ttf

imgfetch --name segoen_slboot.ttf  ${ks-ipxe-url}/win/fonts/segoen_slboot.ttf   segoen_slboot.ttf

imgfetch --name wgl4_boot.ttf      ${ks-ipxe-url}/win/fonts/wgl4_boot.ttf       wgl4_boot.ttf

imgfetch --name boot.wim           ${ks-installs-url}/windows/pxeboot/${winpe}  boot.wim

imgexec

The above script works completely fine for all the hardware I have and the WinPE WIM gets successfully loaded even on with the X710 card.

But with the X710 card, once it has been loaded, I am not able to get a link. It appears to be flapping constantly. Doing a netsh int show int does show the interface to be connected and then disconnected and it just keeps going like that. TCP/IP is trying to get a DHCP address but keeps falling back to an APIPA because the communications with DHCP server never completes. I have tried wpeinit and other means to reset the NICs within WinPE, but nothing changes.

I thought it might be a driver issue, but was able to confirm that the WIM I am loading works perfectly fine when loaded from a USB drive, network and all.

So the only pointer I have is that it doesn't work when loaded via iPXE/WIMBOOT.

Other network cards do not show this issue, including Intel X550s and so on.

The X710 has no problem with iPXE and other operating systems. I am able to install various Linux OSes, VMware, etc. I am using the snponly.efi binary.

So what i am wondering:
- is it something with my BCD? Are there options I am missing? Should I reserve some memory area or block it off? (I compared the BCD I use with the one that comes of the Windows install media, nothing jumps out.)
- Is snponly.efi putting the NIC into an unsupported mode or hanging onto memory when it shouldn't?
- Is WIMBOOT holding onto something? (I read the WIMBOOT page again and it does state that WIMBOOT releases all the memory.)

I saw a couple of threads about the X710, but I am not having an issue with iPXE loading or WinPE hanging on boot up. My issue specifically is that once WinPE has loaded and correctly identified the devices and loaded the correct drivers, I am not able to establish a network connection and the link appears to be flapping (but mostly showing as disconnected).

I am using the latest build of WIMBOOT, version 2.6.0. My snponly.efi is based on commit d2063b7693e0e35db97b2264aa987eb6341ae779.

***NiKiZe*** · 2018-08-17, 23:27

You have a driver issue, either something is missing in iPXE in regards to the driver, or the windows driver. This is assuming that you are using ipxe.efi.

There is nothing with BCD

If you are using snponly.efi than you have a firmware bug or windows driver bug.

pxe_curious · 2018-08-17, 23:34

(2018-08-17 23:27)NiKiZe Wrote: If you are using snponly.efi than you have a firmware bug or windows driver bug.

Even if the same WinPE, when loaded from a USB stick, does not exhibit this issue and everything works fine?

I checked firmware updates for everything on the Dell server and it has all the latest updates. Intel NIC firmware is at 18.5.17.

***NiKiZe*** · 2018-08-18, 09:21

(2018-08-17 23:34)pxe_curious Wrote: Even if the same WinPE, when loaded from a USB stick, does not exhibit this issue and everything works fine?

Yes because when the nic is used by anything internal states are changed.
If those states are not reset by the windows driver than you might get failures.

Did you try latest git master and the ipxe.efi build which have native drivers for the x710 nics?

pxe_curious · 2018-08-18, 16:12

(2018-08-18 09:21)NiKiZe Wrote: Did you try latest git master and the ipxe.efi build which have native drivers for the x710 nics?

Yes, tried it this morning. Did see ipxe.efi use the X710 native driver, but unfortunately the issue is still there.
I will try a couple more things with the WinPE, but if no-go, I will escalate to vendor.

pxe_curious · 2018-08-20, 21:25

If I use bootmgfw.efi (renamed to bootx64.efi) with boot.sdi, my boot.wim and a new BCD, skipping iPXE and WIMBOOT, the WinPE loads fine and networking is also working perfectly fine with the X710.

To me this seems to still indicate an issue with iPXE or WIMBOOT.

I am trying to isolate a scenario that I can use to show the vendor that it is an issue they need to resolve. But right now I am thinking it is an issue with iPXE.

***NiKiZe*** · 2018-08-20, 21:37

(2018-08-20 21:25)pxe_curious Wrote: To me this seems to still indicate an issue with iPXE or WIMBOOT.

I am trying to isolate a scenario that I can use to show the vendor that it is an issue they need to resolve. But right now I am thinking it is an issue with iPXE.

To me this (still) sounds like an issue with the Windows drivers not resetting the nic to proper state.
Even if this could be solved by changing anything in iPXE - the Windows driver should still handle this.
And in regards to Wimboot - it does nothing to the network in any way - so should not be an issue.

pxe_curious · 2018-08-22, 00:19

The fault is all mine.

Back in 2016 I was trying to solve a BSOD issue (http://forum.ipxe.org/showthread.php?tid=8115) and ended up solving it by setting the BCD to allow only one CPU.

Having just one CPU in WinPE with the Intel X710 NIC appears to interfere with the proper operation of the NIC.

I am still working on replicating it so I can escalate it to the vendor(s).

pxe_curious · 2018-08-22, 02:12

(2018-08-22 00:19)pxe_curious Wrote: I am still working on replicating it so I can escalate it to the vendor(s).

I installed Windows Server 2016 on a Dell PowerEdge R640 and made sure it works as intended. Networking was working perfectly fine.
I set the option in the BCD for the Windows Boot Loader of 'onecpu' to 'on' and rebooted.
Upon boot-up the system exhibited the problems I described, with the network link constantly switching between a connected and disconnected state.
Deleting the value 'onecpu' from the Windows Boot Loader in the BCD returned the system back to working order.
I now have something I can bug the vendor about.