On 10/23/24 02:43, Dan Williams wrote:
When the CXL subsystem is built-in the module init order is determined
by Makefile order. That order violates expectations. The expectation is
that cxl_acpi and cxl_mem can race to attach and that if cxl_acpi wins
the race cxl_mem will find the enabled CXL root ports it needs and if
cxl_acpi loses the race it will retrigger cxl_mem to attach via
cxl_bus_rescan(). That only works if cxl_acpi can assume ports are
enabled immediately upon cxl_acpi_probe() return. That in turn can only
happen in the CONFIG_CXL_ACPI=y case if the cxl_port object appears
before the cxl_acpi object in the Makefile.
I'm having problems with understanding this. The acpi module is
initialised following the initcall levels, as defined by the code with
the subsys_initcall(cxl_acpi_init), and the cxl_mem module is not, so
AFAIK, there should not be any race there with the acpi module always
being initialised first. It I'm right, the problem should be another one
we do not know yet ...
Fix up the order to prevent initialization failures, and make sure that
cxl_port is built-in if cxl_acpi is also built-in.
... or forcing cxl_port to be built-in is enough. I wonder how, without
it, the cxl root ports can be there for cxl_mem ...
As for what contributed to this not being found earlier, the CXL
regression environment, cxl_test, builds all CXL functionality as a
module to allow to symbol mocking and other dynamic reload tests. As a
result there is no regression coverage for the built-in case.
Reported-by: Gregory Price <gourry@xxxxxxxxxx>
Closes: http://lore.kernel.org/20241004212504.1246-1-gourry@xxxxxxxxxx
Tested-by: Gregory Price <gourry@xxxxxxxxxx>
Fixes: 8dd2bc0f8e02 ("cxl/mem: Add the cxl_mem driver")
Cc: <stable@xxxxxxxxxxxxxxx>
Cc: Davidlohr Bueso <dave@xxxxxxxxxxxx>
Cc: Jonathan Cameron <jonathan.cameron@xxxxxxxxxx>
Cc: Dave Jiang <dave.jiang@xxxxxxxxx>
Cc: Alison Schofield <alison.schofield@xxxxxxxxx>
Cc: Vishal Verma <vishal.l.verma@xxxxxxxxx>
Cc: Ira Weiny <ira.weiny@xxxxxxxxx>
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
drivers/cxl/Kconfig | 1 +
drivers/cxl/Makefile | 20 ++++++++++++++------
2 files changed, 15 insertions(+), 6 deletions(-)
diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig
index 29c192f20082..876469e23f7a 100644
--- a/drivers/cxl/Kconfig
+++ b/drivers/cxl/Kconfig
@@ -60,6 +60,7 @@ config CXL_ACPI
default CXL_BUS
select ACPI_TABLE_LIB
select ACPI_HMAT
+ select CXL_PORT
help
Enable support for host managed device memory (HDM) resources
published by a platform's ACPI CXL memory layout description. See
diff --git a/drivers/cxl/Makefile b/drivers/cxl/Makefile
index db321f48ba52..2caa90fa4bf2 100644
--- a/drivers/cxl/Makefile
+++ b/drivers/cxl/Makefile
@@ -1,13 +1,21 @@
# SPDX-License-Identifier: GPL-2.0
+
+# Order is important here for the built-in case:
+# - 'core' first for fundamental init
+# - 'port' before platform root drivers like 'acpi' so that CXL-root ports
+# are immediately enabled
+# - 'mem' and 'pmem' before endpoint drivers so that memdevs are
+# immediately enabled
+# - 'pci' last, also mirrors the hardware enumeration hierarchy
obj-y += core/
-obj-$(CONFIG_CXL_PCI) += cxl_pci.o
-obj-$(CONFIG_CXL_MEM) += cxl_mem.o
+obj-$(CONFIG_CXL_PORT) += cxl_port.o
obj-$(CONFIG_CXL_ACPI) += cxl_acpi.o
obj-$(CONFIG_CXL_PMEM) += cxl_pmem.o
-obj-$(CONFIG_CXL_PORT) += cxl_port.o
+obj-$(CONFIG_CXL_MEM) += cxl_mem.o
+obj-$(CONFIG_CXL_PCI) += cxl_pci.o
-cxl_mem-y := mem.o
-cxl_pci-y := pci.o
+cxl_port-y := port.o
cxl_acpi-y := acpi.o
cxl_pmem-y := pmem.o security.o
-cxl_port-y := port.o
+cxl_mem-y := mem.o
+cxl_pci-y := pci.o
