Re: [PATCH 1/3] x86/numa: Fix SRAT lookup for CFMWS ranges with numa_fill_memblks()

From: Robert Richter
Date: Mon Mar 18 2024 - 17:50:25 EST


Hi Dan,

thanks for the quick review.

Yes, this is the 'old' patch. But only the subject was corrected. I
will send a v2 anyway. See below.

On 18.03.24 14:26:41, Dan Williams wrote:
> Robert Richter wrote:
> > With kconfig option NUMA_KEEP_MEMINFO disabled the SRAT lookup done
> > with numa_fill_memblks() fails returning NUMA_NO_MEMBLK (-1). An
> > existing SRAT memory range cannot be found for a CFMWS address range.
> > This causes the addition of a duplicate numa_memblk with a different
> > node id and a subsequent page fault and kernel crash during boot.
> >
> > Note that the issue was initially introduced with [1]. But since
> > phys_to_target_node() was originally used that returned the valid node
> > 0, an additional numa_memblk was not added. Though, the node id was
> > wrong too.
> >
> > Fix this by enabling NUMA_KEEP_MEMINFO for x86 with ACPI and NUMA
> > enabled.
> >
> > [1] fd49f99c1809 ("ACPI: NUMA: Add a node and memblk for each CFMWS not in SRAT")
> >
> > Fixes: 8f1004679987 ("ACPI/NUMA: Apply SRAT proximity domain to entire CFMWS window")
> > Cc: Derick Marks <derick.w.marks@xxxxxxxxx>
> > Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
> > Cc: Alison Schofield <alison.schofield@xxxxxxxxx>
> > Signed-off-by: Robert Richter <rrichter@xxxxxxx>
> > ---
> > drivers/acpi/numa/Kconfig | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/drivers/acpi/numa/Kconfig b/drivers/acpi/numa/Kconfig
> > index 849c2bd820b9..2f4ac6ac6768 100644
> > --- a/drivers/acpi/numa/Kconfig
> > +++ b/drivers/acpi/numa/Kconfig
> > @@ -3,6 +3,7 @@ config ACPI_NUMA
> > bool "NUMA support"
> > depends on NUMA
> > depends on (X86 || ARM64 || LOONGARCH)
> > + select NUMA_KEEP_MEMINFO if X86
> > default y if ARM64
>
> A fix is needed, yes, but this is the wrong one. NUMA_KEEP_MEMINFO is
> only about marking numa_meminfo data as not "__init". Since
> numa_fill_memblks() *is* an __init function, it should have no
> dependency on NUMA_KEEP_MEMINFO.

Right, the option is about just keeping it in non-init mem, but the
parsing is durint __init. Will take a look.

>
> The fix here involves moving the definition of numa_fill_memblks() out
> of the "#ifdef CONFIG_NUMA_KEEP_MEMINFO" in
> arch/x86/include/asm/sparsemem.h so that it does not fallback to the
> default definition in include/linux/numa.h.
>
> It should also be the case that cxl_acpi needs this:
>
> diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
> index 67998dbd1d46..1bf25185c35b 100644
> --- a/drivers/cxl/Kconfig
> +++ b/drivers/cxl/Kconfig
> @@ -6,6 +6,7 @@ menuconfig CXL_BUS
> select FW_UPLOAD
> select PCI_DOE
> select FIRMWARE_TABLE
> + select NUMA_KEEP_MEMINFO if NUMA

Ok, will take a look here too.

Thanks,

-Robert

> help
> CXL is a bus that is electrically compatible with PCI Express, but
> layers three protocols on that signalling (CXL.io, CXL.cache, and