RE: [PATCH] mei: me: do not bind the 700 series PCH-S xHCI controller

[Usyskin, Alexander]

> Subject: [PATCH] mei: me: do not bind the 700 series PCH-S xHCI controller
> 
> From: Joe Finzel <joe@xxxxxxxxx>
> 
> PCI ID 8086:7a60 is listed as the "Alder Lake Point LP" HECI device, but
> on the 700 series PCH-S (Raptor Lake-S) the same ID belongs to the USB 3.2
> xHCI controller (class 0x0c0330). The HECI device there is 8086:7a68.
> 
> With CONFIG_INTEL_MEI_ME=y, drivers/misc is linked before drivers/usb, so
> mei_me probes the xHCI controller before xhci_pci can claim it. The probe
> writes the HECI host CSR into the xHCI MMIO space, waits out three
> hw_start timeouts and gives up, blocking boot for about 5.5 s:
> 
>   mei mei1: H_RST is not set = 0x18000840
>   mei mei1: wait hw ready failed
>   mei mei1: hw_start failed ret = -62 fw status = 803401FD 00000000
> 00006032 00000000 00000000 00000000
>   ...
>   mei mei1: reset: reached maximal consecutive resets: disabling the device
>   mei_me 0000:00:14.0: init hw failure.
>   mei_me 0000:00:14.0: initialization failed.
>   xhci_hcd 0000:00:14.0: xHCI Host Controller
> 
> When mei_me is a module, xhci_pci is usually bound first, which is why
> this has gone unnoticed since the ID was added.
> 
> Match the entry on the communication controller class as well, so it can
> never bind a device of another class.
> 
> Seen on an ASUS ROG STRIX Z790-F GAMING WIFI (BIOS 3201), i7-13700K.
> 
> Fixes: 930c922a987a ("mei: me: add adler lake point LP DID")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Joe Finzel <joe@xxxxxxxxx>
> ---
> Seen on an ASUS ROG STRIX Z790-F with CONFIG_INTEL_MEI_ME=y (the log
> in
> the commit message is from that board). With mei_me built as a module,
> xhci_pci binds first and the stall is gone, which hid the problem so far.
> The patch is compile-tested on 7.2.9 and applies to char-misc-next; I
> can test it on the board with mei_me built in if that helps.
> ---
>  drivers/misc/mei/pci-me.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/misc/mei/pci-me.c b/drivers/misc/mei/pci-me.c
> index 790a46478..26e89140f 100644
> --- a/drivers/misc/mei/pci-me.c
> +++ b/drivers/misc/mei/pci-me.c
> @@ -113,7 +113,10 @@ static const struct pci_device_id mei_me_pci_tbl[] =
> {
>  	{ PCI_DEVICE_DATA(INTEL, MEI_EBG, MEI_ME_PCH15_SPS_CFG) },
> 
>  	{ PCI_DEVICE_DATA(INTEL, MEI_ADP_S, MEI_ME_PCH15_CFG) },
> -	{ PCI_DEVICE_DATA(INTEL, MEI_ADP_LP, MEI_ME_PCH15_CFG) },
> +	/* 8086:7a60 is also the xHCI controller of the 700 series PCH-S */
> +	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL,
> PCI_DEVICE_ID_INTEL_MEI_ADP_LP),
> +	  .class = PCI_CLASS_COMMUNICATION_OTHER << 8, .class_mask =
> 0xffff00,
> +	  .driver_data = (kernel_ulong_t)MEI_ME_PCH15_CFG },

This DeviceID is not a HECI one at all, added by mistake, it seems.
Need to revert patch that have added this ID, I'll send this revert.

- - 
Thanks,
Sasha


>  	{ PCI_DEVICE_DATA(INTEL, MEI_ADP_P, MEI_ME_PCH15_CFG) },
>  	{ PCI_DEVICE_DATA(INTEL, MEI_ADP_N, MEI_ME_PCH15_CFG) },
> 
> 
> ---
> base-commit: c26cc979e28f5193310e4379e14aa6417375c565
> change-id: 20261011-mei-xhci-02772d0506e3
> 
> Best regards,
> --
> Joe Finzel <joe@xxxxxxxxx>
>