[i.MX8 PCIe] PCIe link fails after kexec

[Bjorn Helgaas]

On Tue, Sep 23, 2025 at 03:35:56AM +0000, He, Guocai (CN) wrote:
> Hi all,
> 
> We are seeing a PCIe issue on i.MX8-based board when using kexec on
> Linux kernel 6.6,  we would appreciate your help to clarify whether
> this is a known problem or if there is a recommended fix.

Can you boot a recent kernel, e.g., v6.16 or v6.17-rc7, to make sure
you're not seeing an already-fixed problem?

If you can't try a recent kernel, please include your devicetree and
the complete dmesg log including both the initial boot and the kexec.

> ## Board & Setup
> Linux version 6.6.103-yocto-standard (oe-user at oe-host) (aarch64-wrs-linux-gcc (GCC) 13.4.0, GNU ld (GNU Binutils) 2
> .42.0.20240723) #1 SMP PREEMPT Wed Sep  3 20:13:35 UTC 2025
> 
> Machine model: Freescale i.MX8QM MEK
> PCIe device: Intel Corporation Wireless 7260
> 
> 
> ##Boot flow:
> #step 1: Initial boot: Linux kernel boot from tftp .
> setenv ipaddr 128.224.165.120
> setenv gatewayip 128.224.165.1
> setenv netmask 255.255.255.0
> setenv serverip 128.224.165.20
> 
> tftp 0x8a0000000 vlm-boards/29106/kernel
> tftp 0x8d0000000 vlm-boards/29106/dtb
> booti 0x8a0000000 - 0x8d0000000
> 
> 
> #step 2: Switch to another kernel on disk using kexec
> Reproduction steps
> root at nxp-imx8:~# mkdir /mnt/sdisk
> root at nxp-imx8:~# mount /dev/mmcblk0p1 /mnt/sdisk/
> root at nxp-imx8:~# scp ghe-cn at 128.224.153.151:/folk/ghe-cn/images/Image /mnt/sdisk/kernel     //the images is the same.
> 
> root at nxp-imx8:~# kexec -l /mnt/sdisk/kernel --append="console=ttyLP0,115200 video=HDMI-A-1:1920x1080-32 at 60 rw root=
> /dev/nfs nfsroot=128.224.165.20:/export/pxeboot/vlm-boards/29105/rootfs,tcp,v3 ip=128.224.165.120:128.224.165.20:12
> 8.224.165.1:255.255.255.0::eth0:off selinux=0 enforcing=0"
> 
> root at nxp-imx8:~# kexec -e
> 
> 
> #After the second kernel boots, the PCIe link does not come up.
> #Key log differences
> 
> #####  boot (boot_cold.log):
> imx6q-pcie 5f010000.pcie: host bridge /bus at 5f000000/pcie at 5f010000 ranges:
> imx-drm display-subsystem: bound imx-drm-dpu-bliteng.2 (ops dpu_bliteng_ops)
> imx6q-pcie 5f000000.pcie: host bridge /bus at 5f000000/pcie at 5f000000 ranges:
> imx6q-pcie 5f000000.pcie:       IO 0x006ff80000..0x006ff8ffff -> 0x0000000000
> imx6q-pcie 5f000000.pcie:      MEM 0x0060000000..0x006fefffff -> 0x0060000000
> imx6q-pcie 5f000000.pcie: iATU: unroll F, 6 ob, 6 ib, align 4K, limit 4G
> imx6q-pcie 5f000000.pcie: eDMA: unroll F, 1 wr, 1 rd
> imx6q-pcie 5f000000.pcie: PCIe Gen.1 x1 link up
> imx6q-pcie 5f000000.pcie: PCIe Gen.1 x1 link up
> imx6q-pcie 5f000000.pcie: Link up, Gen1
> imx6q-pcie 5f000000.pcie: PCIe Gen.1 x1 link up
> imx6q-pcie 5f000000.pcie: PCI host bridge to bus 0000:00
> 
> #→ Link up successfully
> 
> root at nxp-imx8:~# lspci
> 00:00.0 PCI bridge: Freescale Semiconductor Inc Device 0000 (rev 01)
> 01:00.0 Network controller: Intel Corporation Wireless 7260 (rev 6b)
> root at nxp-imx8:~#
> 
> 
> #####After kexec (boot_kexec.log):
> 
> imx6q-pcie 5f000000.pcie:       IO 0x006ff80000..0x006ff8ffff -> 0x0000000000
> imx6q-pcie 5f000000.pcie: iATU: unroll F, 6 ob, 6 ib, align 4K, limit 4G
> imx6q-pcie 5f000000.pcie: eDMA: unroll F, 1 wr, 1 rd
> 
> imx6q-pcie 5f000000.pcie: Phy link never came up
> imx6q-pcie 5f000000.pcie: Phy link never came up
> imx6q-pcie 5f000000.pcie: PCI host bridge to bus 0000:00
> 
> #→ Link never comes up, no PCIe devices detected.
> 
> root at nxp-imx8:~# lspci
> 00:00.0 PCI bridge: Freescale Semiconductor Inc Device 0000 (rev 01)
> root at nxp-imx8:~#
> 
> 
> #Questions
> Are there existing patches or recommended workarounds for this scenario?
> Thanks for your time and any guidance you can provide.
> 
> Best regards,
> 
> Guocai