Re: Bug report: the extended PCI config space is missed with 6.2-rc2

From: Dan Williams
Date: Thu Jan 05 2023 - 18:41:18 EST


Bjorn Helgaas wrote:
> [+cc Tony, Dan]
>
> On Wed, Jan 04, 2023 at 09:39:56AM -0500, Liang, Kan wrote:
> > Hi Bjorn,
> >
> > Happy new year!
> >
> > We found some PCI issues with the latest 6.2-rc2.
> >
> > - Using the lspci -xxxx, the extended PCI config space of all PCI
> > devices are missed with the latest 6.2-rc2. The system we used had 932
> > PCI devices, at least 800 which have extended space as seen when booted
> > into a 5.15 kernel. But none of them appeared in 6.2-rc2.
> > - The drivers which rely on the information in the extended PCI config
> > space don't work anymore. We have confirmed that the perf uncore driver
> > (uncore performance monitoring) and Intel VSEC driver (telemetry) don't
> > work in 6.2-rc2. There could be more drivers which are impacted.
> >
> > After a bisect, we found the regression is caused by the below commit
> > 07eab0901ede ("efi/x86: Remove EfiMemoryMappedIO from E820 map").
> > After reverting the commit, the issues are gone.
>
> Can you try this patch (based on v6.2-rc1):

Looks good to me, one question below, but either way:

Reviewed-by: Dan Williams <dan.j.williams@xxxxxxxxx>

...and Dave, who reported that CXL enumeration was busted in -rc2, says
this patch fixes that. So you can also add:

Tested-by: Dave Jiang <dave.jiang@xxxxxxxxx>

>
>
> commit 89a0067217b0 ("x86/pci: Treat EfiMemoryMappedIO as reservation of ECAM space")
> parent 1b929c02afd3
> Author: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> Date: Thu Jan 5 16:02:58 2023 -0600
>
> x86/pci: Treat EfiMemoryMappedIO as reservation of ECAM space
>
> Normally we reject ECAM space unless it is reported as reserved in the E820
> table or via a PNP0C02 _CRS method (PCI Firmware, r3.3, sec 4.1.2). This
> means extended config space (offsets 0x100-0xfff) may not be accessible.
>
> Some firmware doesn't report ECAM space via PNP0C02 _CRS methods, but does
> mention it as an EfiMemoryMappedIO region via EFI GetMemoryMap(), which is
> normally converted to an E820 entry by a bootloader or EFI stub.
>
> 07eab0901ede ("efi/x86: Remove EfiMemoryMappedIO from E820 map"), removes
> E820 entries that correspond to EfiMemoryMappedIO regions because some
> other firmware uses EfiMemoryMappedIO for PCI host bridge windows, and the
> E820 entries prevent Linux from allocating BAR space for hot-added devices.
>
> Allow use of ECAM for extended config space when the region is covered by
> an EfiMemoryMappedIO region, even if it's not included in E820 or PNP0C02
> _CRS.
>
> Fixes: 07eab0901ede ("efi/x86: Remove EfiMemoryMappedIO from E820 map")
> Link: https://lore.kernel.org/r/ac2693d8-8ba3-72e0-5b66-b3ae008d539d@xxxxxxxxxxxxxxx
>
> diff --git a/arch/x86/pci/mmconfig-shared.c b/arch/x86/pci/mmconfig-shared.c
> index 758cbfe55daa..4adc587a4c94 100644
> --- a/arch/x86/pci/mmconfig-shared.c
> +++ b/arch/x86/pci/mmconfig-shared.c
> @@ -12,6 +12,7 @@
> */
>
> #include <linux/acpi.h>
> +#include <linux/efi.h>
> #include <linux/pci.h>
> #include <linux/init.h>
> #include <linux/bitmap.h>
> @@ -442,6 +443,25 @@ static bool is_acpi_reserved(u64 start, u64 end, enum e820_type not_used)
> return mcfg_res.flags;
> }
>
> +static bool is_efi_reserved(u64 start, u64 end, enum e820_type not_used)
> +{
> + efi_memory_desc_t *md;
> + u64 size, mmio_start, mmio_end;
> +
> + for_each_efi_memory_desc(md) {
> + if (md->type == EFI_MEMORY_MAPPED_IO) {

Should this also consider EFI_RESERVED_TYPE? Not that any known BIOS
needs that accommodation. This is more a question than a suggestion.