On Tue, 25 Nov 2025, Daniele Ceraolo Spurio <[email protected]> wrote: > On 11/25/2025 2:26 AM, Jani Nikula wrote: >> On Fri, 14 Nov 2025, Daniele Ceraolo Spurio >> <[email protected]> wrote: >>> The intel GFX drivers (i915/xe) interface with the ME device for some of >>> their features (e.g. PXP, HDCP) via the component interface. Given that >>> the ME device can be hidden by BIOS/Coreboot, the GFX drivers need a >>> way to check if the device is available before attempting to bind the >>> component, otherwise they'll go ahead and initialize features that will >>> never work. >>> The simplest way to check if the ME device is available is to check the >>> available devices against the PCI ID list of the mei_me driver. To avoid >>> duplication, this patch exports the list, so that it can be used directly >>> from the GFX drivers. >>> >>> Signed-off-by: Daniele Ceraolo Spurio <[email protected]> >>> Cc: Alexander Usyskin <[email protected]> >>> Cc: Greg Kroah-Hartman <[email protected]> >>> --- >>> drivers/misc/mei/pci-me.c | 12 +++++++++++- >>> include/linux/mei_me.h | 15 +++++++++++++++ >>> 2 files changed, 26 insertions(+), 1 deletion(-) >>> create mode 100644 include/linux/mei_me.h >>> >>> diff --git a/drivers/misc/mei/pci-me.c b/drivers/misc/mei/pci-me.c >>> index b017ff29dbd1..632756f9da66 100644 >>> --- a/drivers/misc/mei/pci-me.c >>> +++ b/drivers/misc/mei/pci-me.c >>> @@ -18,6 +18,7 @@ >>> #include <linux/pm_runtime.h> >>> >>> #include <linux/mei.h> >>> +#include <linux/mei_me.h> >>> >>> #include "mei_dev.h" >>> #include "client.h" >>> @@ -25,7 +26,7 @@ >>> #include "hw-me.h" >>> >>> /* mei_pci_tbl - PCI Device ID Table */ >>> -static const struct pci_device_id mei_me_pci_tbl[] = { >>> +const struct pci_device_id mei_me_pci_tbl[] = { >>> {MEI_PCI_DEVICE(MEI_DEV_ID_82946GZ, MEI_ME_ICH_CFG)}, >>> {MEI_PCI_DEVICE(MEI_DEV_ID_82G35, MEI_ME_ICH_CFG)}, >>> {MEI_PCI_DEVICE(MEI_DEV_ID_82Q965, MEI_ME_ICH_CFG)}, >>> @@ -135,6 +136,15 @@ static const struct pci_device_id mei_me_pci_tbl[] = { >>> >>> MODULE_DEVICE_TABLE(pci, mei_me_pci_tbl); >>> >>> +/* >>> + * Other drivers (e.g., i915, xe) interface with the ME device for some of >>> their >>> + * features (e.g., PXP, HDCP). However, the ME device can be unplugged via >>> the >>> + * pci subsystem or hidden by BIOS/coreboot, so those drivers might want to >>> + * check if the device is available before initializing those features. To >>> + * allow them to perform such a check, we export the list of ME device IDs. >>> + */ >>> +EXPORT_SYMBOL_GPL(mei_me_pci_tbl); >> Data is not an interface. >> >> Please add an exported helper function (with a sensible stub for >> CONFIG_INTEL_MEI_ME=n) and everything becomes much cleaner both mei and >> i915 side. > > That is actually what I had in v2 [1], but Greg suggested to export the > table directly instead. I am ok either way.
Well I'm not asking to add a function to return the device id table, which certainly has the worst of both approaches. I'm asking to add a function that does the required check, which I think is overall simplest. BR, Jani. > > [1] https://patchwork.freedesktop.org/patch/674368/?series=151677&rev=2 > > Daniele > >> >> BR, >> Jani. >> >>> + >>> #ifdef CONFIG_PM >>> static inline void mei_me_set_pm_domain(struct mei_device *dev); >>> static inline void mei_me_unset_pm_domain(struct mei_device *dev); >>> diff --git a/include/linux/mei_me.h b/include/linux/mei_me.h >>> new file mode 100644 >>> index 000000000000..48fd913a3d95 >>> --- /dev/null >>> +++ b/include/linux/mei_me.h >>> @@ -0,0 +1,15 @@ >>> +/* SPDX-License-Identifier: GPL-2.0 */ >>> +/* >>> + * Copyright (c) 2025, Intel Corporation. All rights reserved. >>> + */ >>> + >>> +#ifndef _LINUX_MEI_ME_H >>> +#define _LINUX_MEI_ME_H >>> + >>> +#include <linux/pci.h> >>> + >>> +#if IS_ENABLED(CONFIG_INTEL_MEI_ME) >>> +extern const struct pci_device_id mei_me_pci_tbl[]; >>> +#endif >>> + >>> +#endif /* _LINUX_MEI_ME_H */ > -- Jani Nikula, Intel
