[PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[Dan Williams]

Changes since v3 [1]:
- Rework / simplify CXL to LIBNVDIMM coordination to remove a
  flush_work() locking dependency from underneath the root device lock.
- Move the root device rescan to a workqueue
- Connect RCDs directly as endpoints reachable through a CXL host bridge
  as a dport, i.e. drop the extra dport indirection from v3
- Add unit test infrastructure for an RCD configuration

[1]: http://lore.kernel.org/r/20221109104059.766720-1-rrichter@amd.com/

---

>From [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

Unlike a CXL memory expander in a VH topology that has at least one
intervening 'struct cxl_port' instance between itself and the CXL root
device, an RCD attaches one-level higher. For example:

               VH
          ┌──────────┐
          │ ACPI0017 │
          │  root0   │
          └─────┬────┘
                │
          ┌─────┴────┐
          │  dport0  │
    ┌─────┤ ACPI0016 ├─────┐
    │     │  port1   │     │
    │     └────┬─────┘     │
    │          │           │
 ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
 │dport0│   │dport1│   │dport2│
 │ RP0  │   │ RP1  │   │ RP2  │
 └──────┘   └──┬───┘   └──────┘
               │
           ┌───┴─────┐
           │endpoint0│
           │  port2  │
           └─────────┘

...vs:

              RCH
          ┌──────────┐
          │ ACPI0017 │
          │  root0   │
          └────┬─────┘
               │
           ┌───┴────┐
           │ dport0 │
           │ACPI0016│
           └───┬────┘
               │
          ┌────┴─────┐
          │endpoint0 │
          │  port1   │
          └──────────┘

So arrange for endpoint port in the RCH/RCD case to appear directly
connected to the host-bridge in its singular role as a dport. Compare
that to the VH case where the host-bridge serves a dual role as a
'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
the Root Ports in the Root Complex that are modeled as 'cxl_dport'
instances in the CXL topology.

Another deviation from the VH case is that RCDs may need to look up
their component registers from the Root Complex Register Block (RCRB).
That platform firmware specified RCRB area is cached by the cxl_acpi
driver and conveyed via the host-bridge dport to the cxl_mem driver to
perform the cxl_rcrb_to_component() lookup for the endpoint port
(See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
upstream port component registers).

---

Dan Williams (9):
      cxl/acpi: Simplify cxl_nvdimm_bridge probing
      cxl/region: Drop redundant pmem region release handling
      cxl/pmem: Refactor nvdimm device registration, delete the workqueue
      cxl/pmem: Remove the cxl_pmem_wq and related infrastructure
      cxl/acpi: Move rescan to the workqueue
      tools/testing/cxl: Make mock CEDT parsing more robust
      cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem
      cxl/port: Add RCD endpoint port enumeration
      tools/testing/cxl: Add an RCH topology

Robert Richter (2):
      cxl/ACPI: Register CXL host ports by bridge device
      cxl/acpi: Extract component registers of restricted hosts from RCRB

Terry Bowman (1):
      cxl/acpi: Set ACPI's CXL _OSC to indicate CXL1.1 support


 drivers/acpi/pci_root.c       |    1 
 drivers/cxl/acpi.c            |  105 +++++++++---
 drivers/cxl/core/core.h       |    8 -
 drivers/cxl/core/pmem.c       |   94 +++++++----
 drivers/cxl/core/port.c       |  111 +++++++------
 drivers/cxl/core/region.c     |   54 ++++++
 drivers/cxl/core/regs.c       |   56 +++++++
 drivers/cxl/cxl.h             |   46 +++--
 drivers/cxl/cxlmem.h          |   15 ++
 drivers/cxl/mem.c             |   72 ++++++++
 drivers/cxl/pci.c             |   13 +-
 drivers/cxl/pmem.c            |  351 +++++------------------------------------
 tools/testing/cxl/Kbuild      |    1 
 tools/testing/cxl/test/cxl.c  |  241 ++++++++++++++++++++++------
 tools/testing/cxl/test/mem.c  |   40 ++++-
 tools/testing/cxl/test/mock.c |   19 ++
 tools/testing/cxl/test/mock.h |    3 
 17 files changed, 712 insertions(+), 518 deletions(-)

base-commit: 3b39fd6cf12ceda2a2582dcb9b9ee9f4d197b857

[PATCH v4 01/12] cxl/acpi: Simplify cxl_nvdimm_bridge probing

[Dan Williams]

The 'struct cxl_nvdimm_bridge' object advertises platform CXL PMEM
resources. It coordinates with libnvdimm to attach nvdimm devices and
regions for each corresponding CXL object. That coordination is
complicated, i.e. difficult to reason about, and it turns out redundant.
It is already the case that the CXL core knows how to tear down a
cxl_region when a cxl_memdev goes through ->remove(), so that pathway
can be extended to directly cleanup cxl_nvdimm and cxl_pmem_region
objects.

Towards the goal of ripping out the cxl_nvdimm_bridge state machine,
arrange for cxl_acpi to optionally pre-load the cxl_pmem driver so that
the nvdimm bridge is active synchronously with
devm_cxl_add_nvdimm_bridge(), and remove all the bind attributes for the
cxl_nvdimm* objects since the cxl root device and cxl_memdev bind
attributes are sufficient.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/acpi.c |    1 +
 drivers/cxl/pmem.c |    9 +++++++++
 2 files changed, 10 insertions(+)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index fb9f72813067..c540da0cbf1e 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -539,3 +539,4 @@ module_platform_driver(cxl_acpi_driver);
 MODULE_LICENSE("GPL v2");
 MODULE_IMPORT_NS(CXL);
 MODULE_IMPORT_NS(ACPI);
+MODULE_SOFTDEP("pre: cxl_pmem");
diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
index 4c627d67281a..946e171e7d4a 100644
--- a/drivers/cxl/pmem.c
+++ b/drivers/cxl/pmem.c
@@ -99,6 +99,9 @@ static struct cxl_driver cxl_nvdimm_driver = {
 	.name = "cxl_nvdimm",
 	.probe = cxl_nvdimm_probe,
 	.id = CXL_DEVICE_NVDIMM,
+	.drv = {
+		.suppress_bind_attrs = true,
+	},
 };
 
 static int cxl_pmem_get_config_size(struct cxl_dev_state *cxlds,
@@ -360,6 +363,9 @@ static struct cxl_driver cxl_nvdimm_bridge_driver = {
 	.probe = cxl_nvdimm_bridge_probe,
 	.remove = cxl_nvdimm_bridge_remove,
 	.id = CXL_DEVICE_NVDIMM_BRIDGE,
+	.drv = {
+		.suppress_bind_attrs = true,
+	},
 };
 
 static int match_cxl_nvdimm(struct device *dev, void *data)
@@ -583,6 +589,9 @@ static struct cxl_driver cxl_pmem_region_driver = {
 	.name = "cxl_pmem_region",
 	.probe = cxl_pmem_region_probe,
 	.id = CXL_DEVICE_PMEM_REGION,
+	.drv = {
+		.suppress_bind_attrs = true,
+	},
 };
 
 /*

[PATCH v4 02/12] cxl/region: Drop redundant pmem region release handling

[Dan Williams]

Now that a cxl_nvdimm object can only experience ->remove() via an
unregistration event (because the cxl_nvdimm bind attributes are
suppressed), additional cleanups are possible.

It is already the case that the removal of a cxl_memdev object triggers
->remove() on any associated region. With that mechanism in place there
is no need for the cxl_nvdimm removal to trigger the same. Just rely on
cxl_region_detach() to tear down the whole cxl_pmem_region.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/core/pmem.c |    2 -
 drivers/cxl/cxl.h       |    1 -
 drivers/cxl/pmem.c      |   90 -----------------------------------------------
 3 files changed, 93 deletions(-)

diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
index 36aa5070d902..1d12a8206444 100644
--- a/drivers/cxl/core/pmem.c
+++ b/drivers/cxl/core/pmem.c
@@ -188,7 +188,6 @@ static void cxl_nvdimm_release(struct device *dev)
 {
 	struct cxl_nvdimm *cxl_nvd = to_cxl_nvdimm(dev);
 
-	xa_destroy(&cxl_nvd->pmem_regions);
 	kfree(cxl_nvd);
 }
 
@@ -231,7 +230,6 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
 
 	dev = &cxl_nvd->dev;
 	cxl_nvd->cxlmd = cxlmd;
-	xa_init(&cxl_nvd->pmem_regions);
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_nvdimm_key);
 	device_set_pm_not_required(dev);
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 7d07127eade3..4ac7938eaf6c 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -424,7 +424,6 @@ struct cxl_nvdimm {
 	struct device dev;
 	struct cxl_memdev *cxlmd;
 	struct cxl_nvdimm_bridge *bridge;
-	struct xarray pmem_regions;
 };
 
 struct cxl_pmem_region_mapping {
diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
index 946e171e7d4a..652f00fc68ca 100644
--- a/drivers/cxl/pmem.c
+++ b/drivers/cxl/pmem.c
@@ -27,26 +27,7 @@ static void clear_exclusive(void *cxlds)
 
 static void unregister_nvdimm(void *nvdimm)
 {
-	struct cxl_nvdimm *cxl_nvd = nvdimm_provider_data(nvdimm);
-	struct cxl_nvdimm_bridge *cxl_nvb = cxl_nvd->bridge;
-	struct cxl_pmem_region *cxlr_pmem;
-	unsigned long index;
-
-	device_lock(&cxl_nvb->dev);
-	dev_set_drvdata(&cxl_nvd->dev, NULL);
-	xa_for_each(&cxl_nvd->pmem_regions, index, cxlr_pmem) {
-		get_device(&cxlr_pmem->dev);
-		device_unlock(&cxl_nvb->dev);
-
-		device_release_driver(&cxlr_pmem->dev);
-		put_device(&cxlr_pmem->dev);
-
-		device_lock(&cxl_nvb->dev);
-	}
-	device_unlock(&cxl_nvb->dev);
-
 	nvdimm_delete(nvdimm);
-	cxl_nvd->bridge = NULL;
 }
 
 static int cxl_nvdimm_probe(struct device *dev)
@@ -243,21 +224,6 @@ static int cxl_nvdimm_release_driver(struct device *dev, void *cxl_nvb)
 	return 0;
 }
 
-static int cxl_pmem_region_release_driver(struct device *dev, void *cxl_nvb)
-{
-	struct cxl_pmem_region *cxlr_pmem;
-
-	if (!is_cxl_pmem_region(dev))
-		return 0;
-
-	cxlr_pmem = to_cxl_pmem_region(dev);
-	if (cxlr_pmem->bridge != cxl_nvb)
-		return 0;
-
-	device_release_driver(dev);
-	return 0;
-}
-
 static void offline_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb,
 			       struct nvdimm_bus *nvdimm_bus)
 {
@@ -269,8 +235,6 @@ static void offline_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb,
 	 * nvdimm_bus_unregister() rips the nvdimm objects out from
 	 * underneath them.
 	 */
-	bus_for_each_dev(&cxl_bus_type, NULL, cxl_nvb,
-			 cxl_pmem_region_release_driver);
 	bus_for_each_dev(&cxl_bus_type, NULL, cxl_nvb,
 			 cxl_nvdimm_release_driver);
 	nvdimm_bus_unregister(nvdimm_bus);
@@ -378,48 +342,6 @@ static void unregister_nvdimm_region(void *nd_region)
 	nvdimm_region_delete(nd_region);
 }
 
-static int cxl_nvdimm_add_region(struct cxl_nvdimm *cxl_nvd,
-				 struct cxl_pmem_region *cxlr_pmem)
-{
-	int rc;
-
-	rc = xa_insert(&cxl_nvd->pmem_regions, (unsigned long)cxlr_pmem,
-		       cxlr_pmem, GFP_KERNEL);
-	if (rc)
-		return rc;
-
-	get_device(&cxlr_pmem->dev);
-	return 0;
-}
-
-static void cxl_nvdimm_del_region(struct cxl_nvdimm *cxl_nvd,
-				  struct cxl_pmem_region *cxlr_pmem)
-{
-	/*
-	 * It is possible this is called without a corresponding
-	 * cxl_nvdimm_add_region for @cxlr_pmem
-	 */
-	cxlr_pmem = xa_erase(&cxl_nvd->pmem_regions, (unsigned long)cxlr_pmem);
-	if (cxlr_pmem)
-		put_device(&cxlr_pmem->dev);
-}
-
-static void release_mappings(void *data)
-{
-	int i;
-	struct cxl_pmem_region *cxlr_pmem = data;
-	struct cxl_nvdimm_bridge *cxl_nvb = cxlr_pmem->bridge;
-
-	device_lock(&cxl_nvb->dev);
-	for (i = 0; i < cxlr_pmem->nr_mappings; i++) {
-		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
-		struct cxl_nvdimm *cxl_nvd = m->cxl_nvd;
-
-		cxl_nvdimm_del_region(cxl_nvd, cxlr_pmem);
-	}
-	device_unlock(&cxl_nvb->dev);
-}
-
 static void cxlr_pmem_remove_resource(void *res)
 {
 	remove_resource(res);
@@ -508,10 +430,6 @@ static int cxl_pmem_region_probe(struct device *dev)
 		goto out_nvb;
 	}
 
-	rc = devm_add_action_or_reset(dev, release_mappings, cxlr_pmem);
-	if (rc)
-		goto out_nvd;
-
 	for (i = 0; i < cxlr_pmem->nr_mappings; i++) {
 		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
 		struct cxl_memdev *cxlmd = m->cxlmd;
@@ -538,14 +456,6 @@ static int cxl_pmem_region_probe(struct device *dev)
 			goto out_nvd;
 		}
 
-		/*
-		 * Pin the region per nvdimm device as those may be released
-		 * out-of-order with respect to the region, and a single nvdimm
-		 * maybe associated with multiple regions
-		 */
-		rc = cxl_nvdimm_add_region(cxl_nvd, cxlr_pmem);
-		if (rc)
-			goto out_nvd;
 		m->cxl_nvd = cxl_nvd;
 		mappings[i] = (struct nd_mapping_desc) {
 			.nvdimm = nvdimm,

[PATCH v4 03/12] cxl/pmem: Refactor nvdimm device registration, delete the workqueue

[Dan Williams]

The three objects 'struct cxl_nvdimm_bridge', 'struct cxl_nvdimm', and
'struct cxl_pmem_region' manage CXL persistent memory resources. The
bridge represents base platform resources, the nvdimm represents one or
more endpoints, and the region is a collection of nvdimms that
contribute to an assembled address range.

Their relationship is such that a region is torn down if any component
endpoints are removed. All regions and endpoints are torn down if the
foundational bridge device goes down.

A workqueue was deployed to manage these interdependencies, but it is
difficult to reason about, and fragile. A recent attempt to take the CXL
root device lock in the cxl_mem driver was reported by lockdep as
colliding with the flush_work() in the cxl_pmem flows.

Instead of the workqueue, arrange for all pmem/nvdimm devices to be torn
down immediately and hierarchically. A similar change is made to both
the 'cxl_nvdimm' and 'cxl_pmem_region' objects. For bisect-ability both
changes are made in the same patch which unfortunately makes the patch
bigger than desired.

Arrange for cxl_memdev and cxl_region to register a cxl_nvdimm and
cxl_pmem_region as a devres release action of the bridge device.
Additionally, include a devres release action of the cxl_memdev or
cxl_region device that triggers the bridge's release action if an endpoint
exits before the bridge. I.e. this allows either unplugging the bridge,
or unplugging and endpoint to result in the same cleanup actions.

To keep the patch smaller the cleanup of the now defunct workqueue
infrastructure is saved for a follow-on patch.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/core/pmem.c      |   70 ++++++++++++++++++++----
 drivers/cxl/core/region.c    |   54 ++++++++++++++++++-
 drivers/cxl/cxl.h            |    7 ++
 drivers/cxl/cxlmem.h         |    4 +
 drivers/cxl/mem.c            |    9 +++
 drivers/cxl/pci.c            |    3 -
 drivers/cxl/pmem.c           |  122 ++++++++++++------------------------------
 tools/testing/cxl/test/mem.c |    3 -
 8 files changed, 164 insertions(+), 108 deletions(-)

diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
index 1d12a8206444..647b3a30638e 100644
--- a/drivers/cxl/core/pmem.c
+++ b/drivers/cxl/core/pmem.c
@@ -219,7 +219,8 @@ EXPORT_SYMBOL_NS_GPL(to_cxl_nvdimm, CXL);
 
 static struct lock_class_key cxl_nvdimm_key;
 
-static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
+static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_nvdimm_bridge *cxl_nvb,
+					   struct cxl_memdev *cxlmd)
 {
 	struct cxl_nvdimm *cxl_nvd;
 	struct device *dev;
@@ -230,6 +231,7 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
 
 	dev = &cxl_nvd->dev;
 	cxl_nvd->cxlmd = cxlmd;
+	cxlmd->cxl_nvd = cxl_nvd;
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_nvdimm_key);
 	device_set_pm_not_required(dev);
@@ -240,27 +242,52 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
 	return cxl_nvd;
 }
 
-static void cxl_nvd_unregister(void *dev)
+static void cxl_nvd_unregister(void *_cxl_nvd)
 {
-	device_unregister(dev);
+	struct cxl_nvdimm *cxl_nvd = _cxl_nvd;
+	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
+
+	device_lock_assert(&cxlmd->cxl_nvb->dev);
+	cxl_nvd->cxlmd = NULL;
+	cxlmd->cxl_nvd = NULL;
+	device_unregister(&cxl_nvd->dev);
+}
+
+static void cxlmd_release_nvdimm(void *_cxlmd)
+{
+	struct cxl_memdev *cxlmd = _cxlmd;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
+
+	device_lock(&cxl_nvb->dev);
+	if (cxlmd->cxl_nvd)
+		devm_release_action(&cxl_nvb->dev, cxl_nvd_unregister,
+				    cxlmd->cxl_nvd);
+	device_unlock(&cxl_nvb->dev);
+	put_device(&cxl_nvb->dev);
 }
 
 /**
  * devm_cxl_add_nvdimm() - add a bridge between a cxl_memdev and an nvdimm
- * @host: same host as @cxlmd
  * @cxlmd: cxl_memdev instance that will perform LIBNVDIMM operations
  *
  * Return: 0 on success negative error code on failure.
  */
-int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
+int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd)
 {
+	struct cxl_nvdimm_bridge *cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
 	struct cxl_nvdimm *cxl_nvd;
 	struct device *dev;
 	int rc;
 
-	cxl_nvd = cxl_nvdimm_alloc(cxlmd);
-	if (IS_ERR(cxl_nvd))
-		return PTR_ERR(cxl_nvd);
+	if (!cxl_nvb)
+		return -ENODEV;
+
+	cxl_nvd = cxl_nvdimm_alloc(cxl_nvb, cxlmd);
+	if (IS_ERR(cxl_nvd)) {
+		rc = PTR_ERR(cxl_nvd);
+		goto err_alloc;
+	}
+	cxlmd->cxl_nvb = cxl_nvb;
 
 	dev = &cxl_nvd->dev;
 	rc = dev_set_name(dev, "pmem%d", cxlmd->id);
@@ -271,13 +298,34 @@ int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
 	if (rc)
 		goto err;
 
-	dev_dbg(host, "%s: register %s\n", dev_name(dev->parent),
-		dev_name(dev));
+	dev_dbg(&cxlmd->dev, "register %s\n", dev_name(dev));
 
-	return devm_add_action_or_reset(host, cxl_nvd_unregister, dev);
+	/*
+	 * Remove this nvdimm connection if either the top-level PMEM
+	 * bridge goes down, or the endpoint device goes through
+	 * ->remove().
+	 */
+	device_lock(&cxl_nvb->dev);
+	if (cxl_nvb->dev.driver)
+		rc = devm_add_action_or_reset(&cxl_nvb->dev, cxl_nvd_unregister,
+					      cxl_nvd);
+	else
+		rc = -ENXIO;
+	device_unlock(&cxl_nvb->dev);
+
+	if (rc)
+		goto err_alloc;
+
+	/* @cxlmd carries a reference on @cxl_nvb until cxlmd_release_nvdimm */
+	return devm_add_action_or_reset(&cxlmd->dev, cxlmd_release_nvdimm, cxlmd);
 
 err:
 	put_device(dev);
+err_alloc:
+	put_device(&cxl_nvb->dev);
+	cxlmd->cxl_nvb = NULL;
+	cxlmd->cxl_nvd = NULL;
+
 	return rc;
 }
 EXPORT_SYMBOL_NS_GPL(devm_cxl_add_nvdimm, CXL);
diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
index f9ae5ad284ff..e73bec828032 100644
--- a/drivers/cxl/core/region.c
+++ b/drivers/cxl/core/region.c
@@ -1812,6 +1812,7 @@ static struct lock_class_key cxl_pmem_region_key;
 static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 {
 	struct cxl_region_params *p = &cxlr->params;
+	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct cxl_pmem_region *cxlr_pmem;
 	struct device *dev;
 	int i;
@@ -1839,6 +1840,14 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 		struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
 		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
 
+		if (i == 0) {
+			cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
+			if (!cxl_nvb) {
+				cxlr_pmem = ERR_PTR(-ENODEV);
+				goto out;
+			}
+			cxlr->cxl_nvb = cxl_nvb;
+		}
 		m->cxlmd = cxlmd;
 		get_device(&cxlmd->dev);
 		m->start = cxled->dpa_res->start;
@@ -1848,6 +1857,7 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 
 	dev = &cxlr_pmem->dev;
 	cxlr_pmem->cxlr = cxlr;
+	cxlr->cxlr_pmem = cxlr_pmem;
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_pmem_region_key);
 	device_set_pm_not_required(dev);
@@ -1860,9 +1870,30 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 	return cxlr_pmem;
 }
 
-static void cxlr_pmem_unregister(void *dev)
+static void cxlr_pmem_unregister(void *_cxlr_pmem)
+{
+	struct cxl_pmem_region *cxlr_pmem = _cxlr_pmem;
+	struct cxl_region *cxlr = cxlr_pmem->cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
+
+	device_lock_assert(&cxl_nvb->dev);
+	cxlr->cxlr_pmem = NULL;
+	cxlr_pmem->cxlr = NULL;
+	device_unregister(&cxlr_pmem->dev);
+}
+
+static void cxlr_release_nvdimm(void *_cxlr)
 {
-	device_unregister(dev);
+	struct cxl_region *cxlr = _cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
+
+	device_lock(&cxl_nvb->dev);
+	if (cxlr->cxlr_pmem)
+		devm_release_action(&cxl_nvb->dev, cxlr_pmem_unregister,
+				    cxlr->cxlr_pmem);
+	device_unlock(&cxl_nvb->dev);
+	cxlr->cxl_nvb = NULL;
+	put_device(&cxl_nvb->dev);
 }
 
 /**
@@ -1874,12 +1905,14 @@ static void cxlr_pmem_unregister(void *dev)
 static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
 {
 	struct cxl_pmem_region *cxlr_pmem;
+	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct device *dev;
 	int rc;
 
 	cxlr_pmem = cxl_pmem_region_alloc(cxlr);
 	if (IS_ERR(cxlr_pmem))
 		return PTR_ERR(cxlr_pmem);
+	cxl_nvb = cxlr->cxl_nvb;
 
 	dev = &cxlr_pmem->dev;
 	rc = dev_set_name(dev, "pmem_region%d", cxlr->id);
@@ -1893,10 +1926,25 @@ static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
 	dev_dbg(&cxlr->dev, "%s: register %s\n", dev_name(dev->parent),
 		dev_name(dev));
 
-	return devm_add_action_or_reset(&cxlr->dev, cxlr_pmem_unregister, dev);
+	device_lock(&cxl_nvb->dev);
+	if (cxl_nvb->dev.driver)
+		rc = devm_add_action_or_reset(&cxl_nvb->dev,
+					      cxlr_pmem_unregister, cxlr_pmem);
+	else
+		rc = -ENXIO;
+	device_unlock(&cxl_nvb->dev);
+
+	if (rc)
+		goto err_bridge;
+
+	/* @cxlr carries a reference on @cxl_nvb until cxlr_release_nvdimm */
+	return devm_add_action_or_reset(&cxlr->dev, cxlr_release_nvdimm, cxlr);
 
 err:
 	put_device(dev);
+err_bridge:
+	put_device(&cxl_nvb->dev);
+	cxlr->cxl_nvb = NULL;
 	return rc;
 }
 
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 4ac7938eaf6c..9b5ba9626636 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -386,6 +386,8 @@ struct cxl_region_params {
  * @id: This region's id. Id is globally unique across all regions
  * @mode: Endpoint decoder allocation / access mode
  * @type: Endpoint decoder target type
+ * @cxl_nvb: nvdimm bridge for coordinating @cxlr_pmem shutdown
+ * @cxlr_pmem: (for pmem regions) cached copy of the nvdimm bridge
  * @params: active + config params for the region
  */
 struct cxl_region {
@@ -393,6 +395,8 @@ struct cxl_region {
 	int id;
 	enum cxl_decoder_mode mode;
 	enum cxl_decoder_type type;
+	struct cxl_nvdimm_bridge *cxl_nvb;
+	struct cxl_pmem_region *cxlr_pmem;
 	struct cxl_region_params params;
 };
 
@@ -438,7 +442,6 @@ struct cxl_pmem_region {
 	struct device dev;
 	struct cxl_region *cxlr;
 	struct nd_region *nd_region;
-	struct cxl_nvdimm_bridge *bridge;
 	struct range hpa_range;
 	int nr_mappings;
 	struct cxl_pmem_region_mapping mapping[];
@@ -637,7 +640,7 @@ struct cxl_nvdimm_bridge *devm_cxl_add_nvdimm_bridge(struct device *host,
 struct cxl_nvdimm *to_cxl_nvdimm(struct device *dev);
 bool is_cxl_nvdimm(struct device *dev);
 bool is_cxl_nvdimm_bridge(struct device *dev);
-int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd);
+int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd);
 struct cxl_nvdimm_bridge *cxl_find_nvdimm_bridge(struct device *dev);
 
 #ifdef CONFIG_CXL_REGION
diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
index 88e3a8e54b6a..c1c9960ab05f 100644
--- a/drivers/cxl/cxlmem.h
+++ b/drivers/cxl/cxlmem.h
@@ -35,6 +35,8 @@
  * @cdev: char dev core object for ioctl operations
  * @cxlds: The device state backing this device
  * @detach_work: active memdev lost a port in its ancestry
+ * @cxl_nvb: coordinate removal of @cxl_nvd if present
+ * @cxl_nvd: optional bridge to an nvdimm if the device supports pmem
  * @id: id number of this memdev instance.
  */
 struct cxl_memdev {
@@ -42,6 +44,8 @@ struct cxl_memdev {
 	struct cdev cdev;
 	struct cxl_dev_state *cxlds;
 	struct work_struct detach_work;
+	struct cxl_nvdimm_bridge *cxl_nvb;
+	struct cxl_nvdimm *cxl_nvd;
 	int id;
 };
 
diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
index 64ccf053d32c..549b6b499bae 100644
--- a/drivers/cxl/mem.c
+++ b/drivers/cxl/mem.c
@@ -48,6 +48,7 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
 static int cxl_mem_probe(struct device *dev)
 {
 	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
+	struct cxl_dev_state *cxlds = cxlmd->cxlds;
 	struct cxl_port *parent_port;
 	struct cxl_dport *dport;
 	struct dentry *dentry;
@@ -95,6 +96,14 @@ static int cxl_mem_probe(struct device *dev)
 	if (rc)
 		return rc;
 
+	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM)) {
+		rc = devm_cxl_add_nvdimm(cxlmd);
+		if (rc == -ENODEV)
+			dev_info(dev, "PMEM disabled by platform\n");
+		else
+			return rc;
+	}
+
 	/*
 	 * The kernel may be operating out of CXL memory on this device,
 	 * there is no spec defined way to determine whether this device
diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
index 621a0522b554..e15da405b948 100644
--- a/drivers/cxl/pci.c
+++ b/drivers/cxl/pci.c
@@ -503,9 +503,6 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (IS_ERR(cxlmd))
 		return PTR_ERR(cxlmd);
 
-	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM))
-		rc = devm_cxl_add_nvdimm(&pdev->dev, cxlmd);
-
 	return rc;
 }
 
diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
index 652f00fc68ca..73357d0c3f25 100644
--- a/drivers/cxl/pmem.c
+++ b/drivers/cxl/pmem.c
@@ -34,26 +34,16 @@ static int cxl_nvdimm_probe(struct device *dev)
 {
 	struct cxl_nvdimm *cxl_nvd = to_cxl_nvdimm(dev);
 	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
 	unsigned long flags = 0, cmd_mask = 0;
 	struct cxl_dev_state *cxlds = cxlmd->cxlds;
-	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct nvdimm *nvdimm;
 	int rc;
 
-	cxl_nvb = cxl_find_nvdimm_bridge(dev);
-	if (!cxl_nvb)
-		return -ENXIO;
-
-	device_lock(&cxl_nvb->dev);
-	if (!cxl_nvb->nvdimm_bus) {
-		rc = -ENXIO;
-		goto out;
-	}
-
 	set_exclusive_cxl_commands(cxlds, exclusive_cmds);
 	rc = devm_add_action_or_reset(dev, clear_exclusive, cxlds);
 	if (rc)
-		goto out;
+		return rc;
 
 	set_bit(NDD_LABELING, &flags);
 	set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
@@ -61,19 +51,11 @@ static int cxl_nvdimm_probe(struct device *dev)
 	set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
 	nvdimm = nvdimm_create(cxl_nvb->nvdimm_bus, cxl_nvd, NULL, flags,
 			       cmd_mask, 0, NULL);
-	if (!nvdimm) {
-		rc = -ENOMEM;
-		goto out;
-	}
+	if (!nvdimm)
+		return -ENOMEM;
 
 	dev_set_drvdata(dev, nvdimm);
-	cxl_nvd->bridge = cxl_nvb;
-	rc = devm_add_action_or_reset(dev, unregister_nvdimm, nvdimm);
-out:
-	device_unlock(&cxl_nvb->dev);
-	put_device(&cxl_nvb->dev);
-
-	return rc;
+	return devm_add_action_or_reset(dev, unregister_nvdimm, nvdimm);
 }
 
 static struct cxl_driver cxl_nvdimm_driver = {
@@ -200,6 +182,16 @@ static int cxl_pmem_ctl(struct nvdimm_bus_descriptor *nd_desc,
 	return cxl_pmem_nvdimm_ctl(nvdimm, cmd, buf, buf_len);
 }
 
+static void unregister_nvdimm_bus(void *_cxl_nvb)
+{
+	struct cxl_nvdimm_bridge *cxl_nvb = _cxl_nvb;
+	struct nvdimm_bus *nvdimm_bus = cxl_nvb->nvdimm_bus;
+
+	cxl_nvb->nvdimm_bus = NULL;
+	nvdimm_bus_unregister(nvdimm_bus);
+}
+
+
 static bool online_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb)
 {
 	if (cxl_nvb->nvdimm_bus)
@@ -303,23 +295,21 @@ static int cxl_nvdimm_bridge_probe(struct device *dev)
 {
 	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
 
-	if (cxl_nvb->state == CXL_NVB_DEAD)
-		return -ENXIO;
+	cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor){
+		.provider_name = "CXL",
+		.module = THIS_MODULE,
+		.ndctl = cxl_pmem_ctl,
+	};
 
-	if (cxl_nvb->state == CXL_NVB_NEW) {
-		cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor) {
-			.provider_name = "CXL",
-			.module = THIS_MODULE,
-			.ndctl = cxl_pmem_ctl,
-		};
+	cxl_nvb->nvdimm_bus =
+		nvdimm_bus_register(&cxl_nvb->dev, &cxl_nvb->nd_desc);
 
-		INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
-	}
+	if (!cxl_nvb->nvdimm_bus)
+		return -ENOMEM;
 
-	cxl_nvb->state = CXL_NVB_ONLINE;
-	cxl_nvdimm_bridge_state_work(cxl_nvb);
+	INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
 
-	return 0;
+	return devm_add_action_or_reset(dev, unregister_nvdimm_bus, cxl_nvb);
 }
 
 static struct cxl_driver cxl_nvdimm_bridge_driver = {
@@ -332,11 +322,6 @@ static struct cxl_driver cxl_nvdimm_bridge_driver = {
 	},
 };
 
-static int match_cxl_nvdimm(struct device *dev, void *data)
-{
-	return is_cxl_nvdimm(dev);
-}
-
 static void unregister_nvdimm_region(void *nd_region)
 {
 	nvdimm_region_delete(nd_region);
@@ -357,8 +342,8 @@ static int cxl_pmem_region_probe(struct device *dev)
 	struct nd_mapping_desc mappings[CXL_DECODER_MAX_INTERLEAVE];
 	struct cxl_pmem_region *cxlr_pmem = to_cxl_pmem_region(dev);
 	struct cxl_region *cxlr = cxlr_pmem->cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
 	struct cxl_pmem_region_info *info = NULL;
-	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct nd_interleave_set *nd_set;
 	struct nd_region_desc ndr_desc;
 	struct cxl_nvdimm *cxl_nvd;
@@ -366,28 +351,12 @@ static int cxl_pmem_region_probe(struct device *dev)
 	struct resource *res;
 	int rc, i = 0;
 
-	cxl_nvb = cxl_find_nvdimm_bridge(&cxlr_pmem->mapping[0].cxlmd->dev);
-	if (!cxl_nvb) {
-		dev_dbg(dev, "bridge not found\n");
-		return -ENXIO;
-	}
-	cxlr_pmem->bridge = cxl_nvb;
-
-	device_lock(&cxl_nvb->dev);
-	if (!cxl_nvb->nvdimm_bus) {
-		dev_dbg(dev, "nvdimm bus not found\n");
-		rc = -ENXIO;
-		goto out_nvb;
-	}
-
 	memset(&mappings, 0, sizeof(mappings));
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 
 	res = devm_kzalloc(dev, sizeof(*res), GFP_KERNEL);
-	if (!res) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!res)
+		return -ENOMEM;
 
 	res->name = "Persistent Memory";
 	res->start = cxlr_pmem->hpa_range.start;
@@ -397,11 +366,11 @@ static int cxl_pmem_region_probe(struct device *dev)
 
 	rc = insert_resource(&iomem_resource, res);
 	if (rc)
-		goto out_nvb;
+		return rc;
 
 	rc = devm_add_action_or_reset(dev, cxlr_pmem_remove_resource, res);
 	if (rc)
-		goto out_nvb;
+		return rc;
 
 	ndr_desc.res = res;
 	ndr_desc.provider_data = cxlr_pmem;
@@ -415,39 +384,23 @@ static int cxl_pmem_region_probe(struct device *dev)
 	}
 
 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
-	if (!nd_set) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!nd_set)
+		return -ENOMEM;
 
 	ndr_desc.memregion = cxlr->id;
 	set_bit(ND_REGION_CXL, &ndr_desc.flags);
 	set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
 
 	info = kmalloc_array(cxlr_pmem->nr_mappings, sizeof(*info), GFP_KERNEL);
-	if (!info) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!info)
+		return -ENOMEM;
 
 	for (i = 0; i < cxlr_pmem->nr_mappings; i++) {
 		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
 		struct cxl_memdev *cxlmd = m->cxlmd;
 		struct cxl_dev_state *cxlds = cxlmd->cxlds;
-		struct device *d;
-
-		d = device_find_child(&cxlmd->dev, NULL, match_cxl_nvdimm);
-		if (!d) {
-			dev_dbg(dev, "[%d]: %s: no cxl_nvdimm found\n", i,
-				dev_name(&cxlmd->dev));
-			rc = -ENODEV;
-			goto out_nvd;
-		}
 
-		/* safe to drop ref now with bridge lock held */
-		put_device(d);
-
-		cxl_nvd = to_cxl_nvdimm(d);
+		cxl_nvd = cxlmd->cxl_nvd;
 		nvdimm = dev_get_drvdata(&cxl_nvd->dev);
 		if (!nvdimm) {
 			dev_dbg(dev, "[%d]: %s: no nvdimm found\n", i,
@@ -488,9 +441,6 @@ static int cxl_pmem_region_probe(struct device *dev)
 				      cxlr_pmem->nd_region);
 out_nvd:
 	kfree(info);
-out_nvb:
-	device_unlock(&cxl_nvb->dev);
-	put_device(&cxl_nvb->dev);
 
 	return rc;
 }
diff --git a/tools/testing/cxl/test/mem.c b/tools/testing/cxl/test/mem.c
index aa2df3a15051..a4ee8e61dd60 100644
--- a/tools/testing/cxl/test/mem.c
+++ b/tools/testing/cxl/test/mem.c
@@ -285,9 +285,6 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
 	if (IS_ERR(cxlmd))
 		return PTR_ERR(cxlmd);
 
-	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM))
-		rc = devm_cxl_add_nvdimm(dev, cxlmd);
-
 	return 0;
 }
 

[PATCH v4 04/12] cxl/pmem: Remove the cxl_pmem_wq and related infrastructure

[Dan Williams]

Now that cxl_nvdimm and cxl_pmem_region objects are torn down
sychronously with the removal of either the bridge, or an endpoint, the
cxl_pmem_wq infrastructure can be jettisoned.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/core/pmem.c |   22 -------
 drivers/cxl/cxl.h       |   17 ------
 drivers/cxl/pmem.c      |  144 -----------------------------------------------
 3 files changed, 1 insertion(+), 182 deletions(-)

diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
index 647b3a30638e..7171f8434a28 100644
--- a/drivers/cxl/core/pmem.c
+++ b/drivers/cxl/core/pmem.c
@@ -99,7 +99,6 @@ static struct cxl_nvdimm_bridge *cxl_nvdimm_bridge_alloc(struct cxl_port *port)
 
 	dev = &cxl_nvb->dev;
 	cxl_nvb->port = port;
-	cxl_nvb->state = CXL_NVB_NEW;
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_nvdimm_bridge_key);
 	device_set_pm_not_required(dev);
@@ -117,28 +116,7 @@ static struct cxl_nvdimm_bridge *cxl_nvdimm_bridge_alloc(struct cxl_port *port)
 static void unregister_nvb(void *_cxl_nvb)
 {
 	struct cxl_nvdimm_bridge *cxl_nvb = _cxl_nvb;
-	bool flush;
 
-	/*
-	 * If the bridge was ever activated then there might be in-flight state
-	 * work to flush. Once the state has been changed to 'dead' then no new
-	 * work can be queued by user-triggered bind.
-	 */
-	device_lock(&cxl_nvb->dev);
-	flush = cxl_nvb->state != CXL_NVB_NEW;
-	cxl_nvb->state = CXL_NVB_DEAD;
-	device_unlock(&cxl_nvb->dev);
-
-	/*
-	 * Even though the device core will trigger device_release_driver()
-	 * before the unregister, it does not know about the fact that
-	 * cxl_nvdimm_bridge_driver defers ->remove() work. So, do the driver
-	 * release not and flush it before tearing down the nvdimm device
-	 * hierarchy.
-	 */
-	device_release_driver(&cxl_nvb->dev);
-	if (flush)
-		flush_work(&cxl_nvb->state_work);
 	device_unregister(&cxl_nvb->dev);
 }
 
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 9b5ba9626636..a26efa195a3a 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -400,34 +400,17 @@ struct cxl_region {
 	struct cxl_region_params params;
 };
 
-/**
- * enum cxl_nvdimm_brige_state - state machine for managing bus rescans
- * @CXL_NVB_NEW: Set at bridge create and after cxl_pmem_wq is destroyed
- * @CXL_NVB_DEAD: Set at brige unregistration to preclude async probing
- * @CXL_NVB_ONLINE: Target state after successful ->probe()
- * @CXL_NVB_OFFLINE: Target state after ->remove() or failed ->probe()
- */
-enum cxl_nvdimm_brige_state {
-	CXL_NVB_NEW,
-	CXL_NVB_DEAD,
-	CXL_NVB_ONLINE,
-	CXL_NVB_OFFLINE,
-};
-
 struct cxl_nvdimm_bridge {
 	int id;
 	struct device dev;
 	struct cxl_port *port;
 	struct nvdimm_bus *nvdimm_bus;
 	struct nvdimm_bus_descriptor nd_desc;
-	struct work_struct state_work;
-	enum cxl_nvdimm_brige_state state;
 };
 
 struct cxl_nvdimm {
 	struct device dev;
 	struct cxl_memdev *cxlmd;
-	struct cxl_nvdimm_bridge *bridge;
 };
 
 struct cxl_pmem_region_mapping {
diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
index 73357d0c3f25..100a853983af 100644
--- a/drivers/cxl/pmem.c
+++ b/drivers/cxl/pmem.c
@@ -11,13 +11,6 @@
 #include "cxlmem.h"
 #include "cxl.h"
 
-/*
- * Ordered workqueue for cxl nvdimm device arrival and departure
- * to coordinate bus rescans when a bridge arrives and trigger remove
- * operations when the bridge is removed.
- */
-static struct workqueue_struct *cxl_pmem_wq;
-
 static __read_mostly DECLARE_BITMAP(exclusive_cmds, CXL_MEM_COMMAND_ID_MAX);
 
 static void clear_exclusive(void *cxlds)
@@ -191,106 +184,6 @@ static void unregister_nvdimm_bus(void *_cxl_nvb)
 	nvdimm_bus_unregister(nvdimm_bus);
 }
 
-
-static bool online_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb)
-{
-	if (cxl_nvb->nvdimm_bus)
-		return true;
-	cxl_nvb->nvdimm_bus =
-		nvdimm_bus_register(&cxl_nvb->dev, &cxl_nvb->nd_desc);
-	return cxl_nvb->nvdimm_bus != NULL;
-}
-
-static int cxl_nvdimm_release_driver(struct device *dev, void *cxl_nvb)
-{
-	struct cxl_nvdimm *cxl_nvd;
-
-	if (!is_cxl_nvdimm(dev))
-		return 0;
-
-	cxl_nvd = to_cxl_nvdimm(dev);
-	if (cxl_nvd->bridge != cxl_nvb)
-		return 0;
-
-	device_release_driver(dev);
-	return 0;
-}
-
-static void offline_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb,
-			       struct nvdimm_bus *nvdimm_bus)
-{
-	if (!nvdimm_bus)
-		return;
-
-	/*
-	 * Set the state of cxl_nvdimm devices to unbound / idle before
-	 * nvdimm_bus_unregister() rips the nvdimm objects out from
-	 * underneath them.
-	 */
-	bus_for_each_dev(&cxl_bus_type, NULL, cxl_nvb,
-			 cxl_nvdimm_release_driver);
-	nvdimm_bus_unregister(nvdimm_bus);
-}
-
-static void cxl_nvb_update_state(struct work_struct *work)
-{
-	struct cxl_nvdimm_bridge *cxl_nvb =
-		container_of(work, typeof(*cxl_nvb), state_work);
-	struct nvdimm_bus *victim_bus = NULL;
-	bool release = false, rescan = false;
-
-	device_lock(&cxl_nvb->dev);
-	switch (cxl_nvb->state) {
-	case CXL_NVB_ONLINE:
-		if (!online_nvdimm_bus(cxl_nvb)) {
-			dev_err(&cxl_nvb->dev,
-				"failed to establish nvdimm bus\n");
-			release = true;
-		} else
-			rescan = true;
-		break;
-	case CXL_NVB_OFFLINE:
-	case CXL_NVB_DEAD:
-		victim_bus = cxl_nvb->nvdimm_bus;
-		cxl_nvb->nvdimm_bus = NULL;
-		break;
-	default:
-		break;
-	}
-	device_unlock(&cxl_nvb->dev);
-
-	if (release)
-		device_release_driver(&cxl_nvb->dev);
-	if (rescan) {
-		int rc = bus_rescan_devices(&cxl_bus_type);
-
-		dev_dbg(&cxl_nvb->dev, "rescan: %d\n", rc);
-	}
-	offline_nvdimm_bus(cxl_nvb, victim_bus);
-
-	put_device(&cxl_nvb->dev);
-}
-
-static void cxl_nvdimm_bridge_state_work(struct cxl_nvdimm_bridge *cxl_nvb)
-{
-	/*
-	 * Take a reference that the workqueue will drop if new work
-	 * gets queued.
-	 */
-	get_device(&cxl_nvb->dev);
-	if (!queue_work(cxl_pmem_wq, &cxl_nvb->state_work))
-		put_device(&cxl_nvb->dev);
-}
-
-static void cxl_nvdimm_bridge_remove(struct device *dev)
-{
-	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
-
-	if (cxl_nvb->state == CXL_NVB_ONLINE)
-		cxl_nvb->state = CXL_NVB_OFFLINE;
-	cxl_nvdimm_bridge_state_work(cxl_nvb);
-}
-
 static int cxl_nvdimm_bridge_probe(struct device *dev)
 {
 	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
@@ -307,15 +200,12 @@ static int cxl_nvdimm_bridge_probe(struct device *dev)
 	if (!cxl_nvb->nvdimm_bus)
 		return -ENOMEM;
 
-	INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
-
 	return devm_add_action_or_reset(dev, unregister_nvdimm_bus, cxl_nvb);
 }
 
 static struct cxl_driver cxl_nvdimm_bridge_driver = {
 	.name = "cxl_nvdimm_bridge",
 	.probe = cxl_nvdimm_bridge_probe,
-	.remove = cxl_nvdimm_bridge_remove,
 	.id = CXL_DEVICE_NVDIMM_BRIDGE,
 	.drv = {
 		.suppress_bind_attrs = true,
@@ -454,31 +344,6 @@ static struct cxl_driver cxl_pmem_region_driver = {
 	},
 };
 
-/*
- * Return all bridges to the CXL_NVB_NEW state to invalidate any
- * ->state_work referring to the now destroyed cxl_pmem_wq.
- */
-static int cxl_nvdimm_bridge_reset(struct device *dev, void *data)
-{
-	struct cxl_nvdimm_bridge *cxl_nvb;
-
-	if (!is_cxl_nvdimm_bridge(dev))
-		return 0;
-
-	cxl_nvb = to_cxl_nvdimm_bridge(dev);
-	device_lock(dev);
-	cxl_nvb->state = CXL_NVB_NEW;
-	device_unlock(dev);
-
-	return 0;
-}
-
-static void destroy_cxl_pmem_wq(void)
-{
-	destroy_workqueue(cxl_pmem_wq);
-	bus_for_each_dev(&cxl_bus_type, NULL, NULL, cxl_nvdimm_bridge_reset);
-}
-
 static __init int cxl_pmem_init(void)
 {
 	int rc;
@@ -486,13 +351,9 @@ static __init int cxl_pmem_init(void)
 	set_bit(CXL_MEM_COMMAND_ID_SET_SHUTDOWN_STATE, exclusive_cmds);
 	set_bit(CXL_MEM_COMMAND_ID_SET_LSA, exclusive_cmds);
 
-	cxl_pmem_wq = alloc_ordered_workqueue("cxl_pmem", 0);
-	if (!cxl_pmem_wq)
-		return -ENXIO;
-
 	rc = cxl_driver_register(&cxl_nvdimm_bridge_driver);
 	if (rc)
-		goto err_bridge;
+		return rc;
 
 	rc = cxl_driver_register(&cxl_nvdimm_driver);
 	if (rc)
@@ -508,8 +369,6 @@ static __init int cxl_pmem_init(void)
 	cxl_driver_unregister(&cxl_nvdimm_driver);
 err_nvdimm:
 	cxl_driver_unregister(&cxl_nvdimm_bridge_driver);
-err_bridge:
-	destroy_cxl_pmem_wq();
 	return rc;
 }
 
@@ -518,7 +377,6 @@ static __exit void cxl_pmem_exit(void)
 	cxl_driver_unregister(&cxl_pmem_region_driver);
 	cxl_driver_unregister(&cxl_nvdimm_driver);
 	cxl_driver_unregister(&cxl_nvdimm_bridge_driver);
-	destroy_cxl_pmem_wq();
 }
 
 MODULE_LICENSE("GPL v2");

[PATCH v4 05/12] cxl/acpi: Move rescan to the workqueue

[Dan Williams]

Now that the cxl_mem driver has a need to take the root device lock, the
cxl_bus_rescan() needs to run outside of the root lock context.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/acpi.c      |   17 +++++++++++++++--
 drivers/cxl/core/port.c |   19 +++++++++++++++++--
 drivers/cxl/cxl.h       |    3 ++-
 3 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index c540da0cbf1e..b8407b77aff6 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -511,7 +511,8 @@ static int cxl_acpi_probe(struct platform_device *pdev)
 		return rc;
 
 	/* In case PCI is scanned before ACPI re-trigger memdev attach */
-	return cxl_bus_rescan();
+	cxl_bus_rescan();
+	return 0;
 }
 
 static const struct acpi_device_id cxl_acpi_ids[] = {
@@ -535,7 +536,19 @@ static struct platform_driver cxl_acpi_driver = {
 	.id_table = cxl_test_ids,
 };
 
-module_platform_driver(cxl_acpi_driver);
+static int __init cxl_acpi_init(void)
+{
+	return platform_driver_register(&cxl_acpi_driver);
+}
+
+static void __exit cxl_acpi_exit(void)
+{
+	platform_driver_unregister(&cxl_acpi_driver);
+	cxl_bus_drain();
+}
+
+module_init(cxl_acpi_init);
+module_exit(cxl_acpi_exit);
 MODULE_LICENSE("GPL v2");
 MODULE_IMPORT_NS(CXL);
 MODULE_IMPORT_NS(ACPI);
diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
index 0d2f5eaaca7d..d225267c69bb 100644
--- a/drivers/cxl/core/port.c
+++ b/drivers/cxl/core/port.c
@@ -1844,12 +1844,27 @@ static void cxl_bus_remove(struct device *dev)
 
 static struct workqueue_struct *cxl_bus_wq;
 
-int cxl_bus_rescan(void)
+static void cxl_bus_rescan_queue(struct work_struct *w)
 {
-	return bus_rescan_devices(&cxl_bus_type);
+	int rc = bus_rescan_devices(&cxl_bus_type);
+
+	pr_debug("CXL bus rescan result: %d\n", rc);
+}
+
+void cxl_bus_rescan(void)
+{
+	static DECLARE_WORK(rescan_work, cxl_bus_rescan_queue);
+
+	queue_work(cxl_bus_wq, &rescan_work);
 }
 EXPORT_SYMBOL_NS_GPL(cxl_bus_rescan, CXL);
 
+void cxl_bus_drain(void)
+{
+	drain_workqueue(cxl_bus_wq);
+}
+EXPORT_SYMBOL_NS_GPL(cxl_bus_drain, CXL);
+
 bool schedule_cxl_memdev_detach(struct cxl_memdev *cxlmd)
 {
 	return queue_work(cxl_bus_wq, &cxlmd->detach_work);
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index a26efa195a3a..9b33ae4b2aec 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -552,7 +552,8 @@ int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
 			  struct cxl_dport *parent_dport);
 struct cxl_port *find_cxl_root(struct device *dev);
 int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd);
-int cxl_bus_rescan(void);
+void cxl_bus_rescan(void);
+void cxl_bus_drain(void);
 struct cxl_port *cxl_mem_find_port(struct cxl_memdev *cxlmd,
 				   struct cxl_dport **dport);
 bool schedule_cxl_memdev_detach(struct cxl_memdev *cxlmd);

[PATCH v4 06/12] tools/testing/cxl: Make mock CEDT parsing more robust

[Dan Williams]

Accept any cxl_test topology device as the first argument in
cxl_chbs_context. This is in preparation for reworking the detection of
the component registers across VH and RCH topologies. Move
mock_acpi_table_parse_cedt() beneath the definition of is_mock_port()
and use is_mock_port() instead of the explicit mock cxl_acpi device
check.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 tools/testing/cxl/test/cxl.c |   80 +++++++++++++++++++++---------------------
 1 file changed, 40 insertions(+), 40 deletions(-)

diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
index facfcd11cb67..42a34650dd2f 100644
--- a/tools/testing/cxl/test/cxl.c
+++ b/tools/testing/cxl/test/cxl.c
@@ -320,46 +320,6 @@ static int populate_cedt(void)
 	return 0;
 }
 
-/*
- * WARNING, this hack assumes the format of 'struct
- * cxl_cfmws_context' and 'struct cxl_chbs_context' share the property that
- * the first struct member is the device being probed by the cxl_acpi
- * driver.
- */
-struct cxl_cedt_context {
-	struct device *dev;
-};
-
-static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id,
-				      acpi_tbl_entry_handler_arg handler_arg,
-				      void *arg)
-{
-	struct cxl_cedt_context *ctx = arg;
-	struct device *dev = ctx->dev;
-	union acpi_subtable_headers *h;
-	unsigned long end;
-	int i;
-
-	if (dev != &cxl_acpi->dev)
-		return acpi_table_parse_cedt(id, handler_arg, arg);
-
-	if (id == ACPI_CEDT_TYPE_CHBS)
-		for (i = 0; i < ARRAY_SIZE(mock_cedt.chbs); i++) {
-			h = (union acpi_subtable_headers *)&mock_cedt.chbs[i];
-			end = (unsigned long)&mock_cedt.chbs[i + 1];
-			handler_arg(h, arg, end);
-		}
-
-	if (id == ACPI_CEDT_TYPE_CFMWS)
-		for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
-			h = (union acpi_subtable_headers *) mock_cfmws[i];
-			end = (unsigned long) h + mock_cfmws[i]->header.length;
-			handler_arg(h, arg, end);
-		}
-
-	return 0;
-}
-
 static bool is_mock_bridge(struct device *dev)
 {
 	int i;
@@ -410,6 +370,46 @@ static bool is_mock_port(struct device *dev)
 	return false;
 }
 
+/*
+ * WARNING, this hack assumes the format of 'struct cxl_cfmws_context'
+ * and 'struct cxl_chbs_context' share the property that the first
+ * struct member is cxl_test device being probed by the cxl_acpi
+ * driver.
+ */
+struct cxl_cedt_context {
+	struct device *dev;
+};
+
+static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id,
+				      acpi_tbl_entry_handler_arg handler_arg,
+				      void *arg)
+{
+	struct cxl_cedt_context *ctx = arg;
+	struct device *dev = ctx->dev;
+	union acpi_subtable_headers *h;
+	unsigned long end;
+	int i;
+
+	if (!is_mock_port(dev) && !is_mock_dev(dev))
+		return acpi_table_parse_cedt(id, handler_arg, arg);
+
+	if (id == ACPI_CEDT_TYPE_CHBS)
+		for (i = 0; i < ARRAY_SIZE(mock_cedt.chbs); i++) {
+			h = (union acpi_subtable_headers *)&mock_cedt.chbs[i];
+			end = (unsigned long)&mock_cedt.chbs[i + 1];
+			handler_arg(h, arg, end);
+		}
+
+	if (id == ACPI_CEDT_TYPE_CFMWS)
+		for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
+			h = (union acpi_subtable_headers *) mock_cfmws[i];
+			end = (unsigned long) h + mock_cfmws[i]->header.length;
+			handler_arg(h, arg, end);
+		}
+
+	return 0;
+}
+
 static int host_bridge_index(struct acpi_device *adev)
 {
 	return adev - host_bridge;

[PATCH v4 07/12] cxl/ACPI: Register CXL host ports by bridge device

[Dan Williams]

From: Robert Richter <rrichter@amd.com>

A port of a CXL host bridge links to the bridge's ACPI device
(&adev->dev) with its corresponding uport/dport device (uport_dev and
dport_dev respectively). The device is not a direct parent device in
the PCI topology as pdev->dev.parent points to a PCI bridge's (struct
pci_host_bridge) device. The following CXL memory device hierarchy
would be valid for an endpoint once an RCD EP would be enabled (note
this will be done in a later patch):

VH mode:

 cxlmd->dev.parent->parent
        ^^^\^^^^^^\ ^^^^^^\
            \      \       pci_dev (Type 1, Downstream Port)
             \      pci_dev (Type 0, PCI Express Endpoint)
              cxl mem device

RCD mode:

 cxlmd->dev.parent->parent
        ^^^\^^^^^^\ ^^^^^^\
            \      \       pci_host_bridge
             \      pci_dev (Type 0, RCiEP)
              cxl mem device

In VH mode a downstream port is created by port enumeration and thus
always exists.

Now, in RCD mode the host bridge also already exists but it references
to an ACPI device. A port lookup by the PCI device's parent device
will fail as a direct link to the registered port is missing. The ACPI
device of the bridge must be determined first.

To prevent this, change port registration of a CXL host to use the
bridge device instead. Do this also for the VH case as port topology
will better reflect the PCI topology then.

Signed-off-by: Robert Richter <rrichter@amd.com>
[djbw: rebase on brige mocking]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/acpi.c |   35 +++++++++++++++++++----------------
 1 file changed, 19 insertions(+), 16 deletions(-)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index b8407b77aff6..50d82376097c 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -193,35 +193,34 @@ static int add_host_bridge_uport(struct device *match, void *arg)
 {
 	struct cxl_port *root_port = arg;
 	struct device *host = root_port->dev.parent;
-	struct acpi_device *bridge = to_cxl_host_bridge(host, match);
+	struct acpi_device *hb = to_cxl_host_bridge(host, match);
 	struct acpi_pci_root *pci_root;
 	struct cxl_dport *dport;
 	struct cxl_port *port;
+	struct device *bridge;
 	int rc;
 
-	if (!bridge)
+	if (!hb)
 		return 0;
 
-	dport = cxl_find_dport_by_dev(root_port, match);
+	pci_root = acpi_pci_find_root(hb->handle);
+	bridge = pci_root->bus->bridge;
+	dport = cxl_find_dport_by_dev(root_port, bridge);
 	if (!dport) {
 		dev_dbg(host, "host bridge expected and not found\n");
 		return 0;
 	}
 
-	/*
-	 * Note that this lookup already succeeded in
-	 * to_cxl_host_bridge(), so no need to check for failure here
-	 */
-	pci_root = acpi_pci_find_root(bridge->handle);
-	rc = devm_cxl_register_pci_bus(host, match, pci_root->bus);
+	rc = devm_cxl_register_pci_bus(host, bridge, pci_root->bus);
 	if (rc)
 		return rc;
 
-	port = devm_cxl_add_port(host, match, dport->component_reg_phys, dport);
+	port = devm_cxl_add_port(host, bridge, dport->component_reg_phys,
+				 dport);
 	if (IS_ERR(port))
 		return PTR_ERR(port);
 
-	dev_info(pci_root->bus->bridge, "host supports CXL\n");
+	dev_info(bridge, "host supports CXL\n");
 
 	return 0;
 }
@@ -253,18 +252,20 @@ static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
 static int add_host_bridge_dport(struct device *match, void *arg)
 {
 	acpi_status status;
+	struct device *bridge;
 	unsigned long long uid;
 	struct cxl_dport *dport;
 	struct cxl_chbs_context ctx;
+	struct acpi_pci_root *pci_root;
 	struct cxl_port *root_port = arg;
 	struct device *host = root_port->dev.parent;
-	struct acpi_device *bridge = to_cxl_host_bridge(host, match);
+	struct acpi_device *hb = to_cxl_host_bridge(host, match);
 
-	if (!bridge)
+	if (!hb)
 		return 0;
 
-	status = acpi_evaluate_integer(bridge->handle, METHOD_NAME__UID, NULL,
-				       &uid);
+	status =
+		acpi_evaluate_integer(hb->handle, METHOD_NAME__UID, NULL, &uid);
 	if (status != AE_OK) {
 		dev_err(match, "unable to retrieve _UID\n");
 		return -ENODEV;
@@ -285,7 +286,9 @@ static int add_host_bridge_dport(struct device *match, void *arg)
 
 	dev_dbg(match, "CHBCR found: 0x%08llx\n", (u64)ctx.chbcr);
 
-	dport = devm_cxl_add_dport(root_port, match, uid, ctx.chbcr);
+	pci_root = acpi_pci_find_root(hb->handle);
+	bridge = pci_root->bus->bridge;
+	dport = devm_cxl_add_dport(root_port, bridge, uid, ctx.chbcr);
 	if (IS_ERR(dport))
 		return PTR_ERR(dport);
 

[PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Dan Williams]

From: Robert Richter <rrichter@amd.com>

A downstream port must be connected to a component register block.
For restricted hosts the base address is determined from the RCRB. The
RCRB is provided by the host's CEDT CHBS entry. Rework CEDT parser to
get the RCRB and add code to extract the component register block from
it.

RCRB's BAR[0..1] point to the component block containing CXL subsystem
component registers. MEMBAR extraction follows the PCI base spec here,
esp. 64 bit extraction and memory range alignment (6.0, 7.5.1.2.1). The
RCRB base address is cached in the cxl_dport per-host bridge so that the
upstream port component registers can be retrieved later by an RCD
(RCIEP) associated with the host bridge.

Note: Right now the component register block is used for HDM decoder
capability only which is optional for RCDs. If unsupported by the RCD,
the HDM init will fail. It is future work to bypass it in this case.

Co-developed-by: Terry Bowman <terry.bowman@amd.com>
Signed-off-by: Terry Bowman <terry.bowman@amd.com>
Signed-off-by: Robert Richter <rrichter@amd.com>
[djbw: introduce devm_cxl_add_rch_dport()]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/acpi.c            |   54 ++++++++++++++++++++++++++++++++--------
 drivers/cxl/core/port.c       |   42 +++++++++++++++++++++++++++----
 drivers/cxl/core/regs.c       |   56 +++++++++++++++++++++++++++++++++++++++++
 drivers/cxl/cxl.h             |   16 ++++++++++++
 tools/testing/cxl/Kbuild      |    1 +
 tools/testing/cxl/test/cxl.c  |   10 +++++++
 tools/testing/cxl/test/mock.c |   19 ++++++++++++++
 tools/testing/cxl/test/mock.h |    3 ++
 8 files changed, 186 insertions(+), 15 deletions(-)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index 50d82376097c..1224add13529 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -9,6 +9,8 @@
 #include "cxlpci.h"
 #include "cxl.h"
 
+#define CXL_RCRB_SIZE	SZ_8K
+
 static unsigned long cfmws_to_decoder_flags(int restrictions)
 {
 	unsigned long flags = CXL_DECODER_F_ENABLE;
@@ -215,6 +217,11 @@ static int add_host_bridge_uport(struct device *match, void *arg)
 	if (rc)
 		return rc;
 
+	if (dport->rch) {
+		dev_info(bridge, "host supports CXL (restricted)\n");
+		return 0;
+	}
+
 	port = devm_cxl_add_port(host, bridge, dport->component_reg_phys,
 				 dport);
 	if (IS_ERR(port))
@@ -228,27 +235,46 @@ static int add_host_bridge_uport(struct device *match, void *arg)
 struct cxl_chbs_context {
 	struct device *dev;
 	unsigned long long uid;
-	resource_size_t chbcr;
+	struct acpi_cedt_chbs chbs;
 };
 
-static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
-			 const unsigned long end)
+static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
+			const unsigned long end)
 {
 	struct cxl_chbs_context *ctx = arg;
 	struct acpi_cedt_chbs *chbs;
 
-	if (ctx->chbcr)
+	if (ctx->chbs.base)
 		return 0;
 
 	chbs = (struct acpi_cedt_chbs *) header;
 
 	if (ctx->uid != chbs->uid)
 		return 0;
-	ctx->chbcr = chbs->base;
+	ctx->chbs = *chbs;
 
 	return 0;
 }
 
+static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
+{
+	struct acpi_cedt_chbs *chbs = &ctx->chbs;
+
+	if (!chbs->base)
+		return CXL_RESOURCE_NONE;
+
+	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
+		return chbs->base;
+
+	if (chbs->length != CXL_RCRB_SIZE)
+		return CXL_RESOURCE_NONE;
+
+	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
+		&chbs->base);
+
+	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
+}
+
 static int add_host_bridge_dport(struct device *match, void *arg)
 {
 	acpi_status status;
@@ -258,6 +284,7 @@ static int add_host_bridge_dport(struct device *match, void *arg)
 	struct cxl_chbs_context ctx;
 	struct acpi_pci_root *pci_root;
 	struct cxl_port *root_port = arg;
+	resource_size_t component_reg_phys;
 	struct device *host = root_port->dev.parent;
 	struct acpi_device *hb = to_cxl_host_bridge(host, match);
 
@@ -274,21 +301,28 @@ static int add_host_bridge_dport(struct device *match, void *arg)
 	dev_dbg(match, "UID found: %lld\n", uid);
 
 	ctx = (struct cxl_chbs_context) {
-		.dev = host,
+		.dev = match,
 		.uid = uid,
 	};
-	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbcr, &ctx);
+	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbs, &ctx);
 
-	if (ctx.chbcr == 0) {
+	component_reg_phys = cxl_get_chbcr(&ctx);
+	if (component_reg_phys == CXL_RESOURCE_NONE) {
 		dev_warn(match, "No CHBS found for Host Bridge (UID %lld)\n", uid);
 		return 0;
 	}
 
-	dev_dbg(match, "CHBCR found: 0x%08llx\n", (u64)ctx.chbcr);
+	dev_dbg(match, "CHBCR found: %pa\n", &component_reg_phys);
 
 	pci_root = acpi_pci_find_root(hb->handle);
 	bridge = pci_root->bus->bridge;
-	dport = devm_cxl_add_dport(root_port, bridge, uid, ctx.chbcr);
+	if (ctx.chbs.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
+		dport = devm_cxl_add_rch_dport(root_port, bridge, uid,
+					       component_reg_phys,
+					       ctx.chbs.base);
+	else
+		dport = devm_cxl_add_dport(root_port, bridge, uid,
+					   component_reg_phys);
 	if (IS_ERR(dport))
 		return PTR_ERR(dport);
 
diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
index d225267c69bb..d9fe06e1462f 100644
--- a/drivers/cxl/core/port.c
+++ b/drivers/cxl/core/port.c
@@ -628,6 +628,8 @@ static struct cxl_port *cxl_port_alloc(struct device *uport,
 			iter = to_cxl_port(iter->dev.parent);
 		if (iter->host_bridge)
 			port->host_bridge = iter->host_bridge;
+		else if (parent_dport->rch)
+			port->host_bridge = parent_dport->dport;
 		else
 			port->host_bridge = iter->uport;
 		dev_dbg(uport, "host-bridge: %s\n", dev_name(port->host_bridge));
@@ -899,10 +901,15 @@ static void cxl_dport_unlink(void *data)
 	sysfs_remove_link(&port->dev.kobj, link_name);
 }
 
-static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
-					      struct device *dport_dev,
-					      int port_id,
-					      resource_size_t component_reg_phys)
+enum cxl_dport_mode {
+	CXL_DPORT_VH,
+	CXL_DPORT_RCH,
+};
+
+static struct cxl_dport *
+__devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
+		     int port_id, resource_size_t component_reg_phys,
+		     enum cxl_dport_mode mode, resource_size_t rcrb)
 {
 	char link_name[CXL_TARGET_STRLEN];
 	struct cxl_dport *dport;
@@ -932,6 +939,9 @@ static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
 	dport->port_id = port_id;
 	dport->component_reg_phys = component_reg_phys;
 	dport->port = port;
+	if (mode == CXL_DPORT_RCH)
+		dport->rch = true;
+	dport->rcrb = rcrb;
 
 	cond_cxl_root_lock(port);
 	rc = add_dport(port, dport);
@@ -973,7 +983,8 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
 	struct cxl_dport *dport;
 
 	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
-				     component_reg_phys);
+				     component_reg_phys, CXL_DPORT_VH,
+				     CXL_RESOURCE_NONE);
 	if (IS_ERR(dport)) {
 		dev_dbg(dport_dev, "failed to add dport to %s: %ld\n",
 			dev_name(&port->dev), PTR_ERR(dport));
@@ -986,6 +997,27 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
 }
 EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
 
+struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
+					 struct device *dport_dev, int port_id,
+					 resource_size_t component_reg_phys,
+					 resource_size_t rcrb)
+{
+	struct cxl_dport *dport;
+
+	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
+				     component_reg_phys, CXL_DPORT_RCH, rcrb);
+	if (IS_ERR(dport)) {
+		dev_dbg(dport_dev, "failed to add RCH dport to %s: %ld\n",
+			dev_name(&port->dev), PTR_ERR(dport));
+	} else {
+		dev_dbg(dport_dev, "RCH dport added to %s\n",
+			dev_name(&port->dev));
+	}
+
+	return dport;
+}
+EXPORT_SYMBOL_NS_GPL(devm_cxl_add_rch_dport, CXL);
+
 static int add_ep(struct cxl_ep *new)
 {
 	struct cxl_port *port = new->dport->port;
diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c
index ec178e69b18f..7c2a85dc4125 100644
--- a/drivers/cxl/core/regs.c
+++ b/drivers/cxl/core/regs.c
@@ -307,3 +307,59 @@ int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
 	return -ENODEV;
 }
 EXPORT_SYMBOL_NS_GPL(cxl_find_regblock, CXL);
+
+resource_size_t cxl_rcrb_to_component(struct device *dev,
+				      resource_size_t rcrb,
+				      enum cxl_rcrb which)
+{
+	resource_size_t component_reg_phys;
+	u32 bar0, bar1;
+	void *addr;
+	u16 cmd;
+
+	if (which == CXL_RCRB_UPSTREAM)
+		rcrb += SZ_4K;
+
+	/*
+	 * RCRB's BAR[0..1] point to component block containing CXL
+	 * subsystem component registers. MEMBAR extraction follows
+	 * the PCI Base spec here, esp. 64 bit extraction and memory
+	 * ranges alignment (6.0, 7.5.1.2.1).
+	 */
+	if (!request_mem_region(rcrb, SZ_4K, "CXL RCRB"))
+		return CXL_RESOURCE_NONE;
+	addr = ioremap(rcrb, SZ_4K);
+	if (!addr) {
+		dev_err(dev, "Failed to map region %pr\n", addr);
+		release_mem_region(rcrb, SZ_4K);
+		return CXL_RESOURCE_NONE;
+	}
+
+	cmd = readw(addr + PCI_COMMAND);
+	bar0 = readl(addr + PCI_BASE_ADDRESS_0);
+	bar1 = readl(addr + PCI_BASE_ADDRESS_1);
+	iounmap(addr);
+	release_mem_region(rcrb, SZ_4K);
+
+	/* sanity check */
+	if (cmd == 0xffff)
+		return CXL_RESOURCE_NONE;
+	if ((cmd & PCI_COMMAND_MEMORY) == 0)
+		return CXL_RESOURCE_NONE;
+	if (bar0 & (PCI_BASE_ADDRESS_MEM_TYPE_1M | PCI_BASE_ADDRESS_SPACE_IO))
+		return CXL_RESOURCE_NONE;
+
+	component_reg_phys = bar0 & PCI_BASE_ADDRESS_MEM_MASK;
+	if (bar0 & PCI_BASE_ADDRESS_MEM_TYPE_64)
+		component_reg_phys |= ((u64)bar1) << 32;
+
+	if (!component_reg_phys)
+		return CXL_RESOURCE_NONE;
+
+	/* MEMBAR is block size (64k) aligned. */
+	if (!IS_ALIGNED(component_reg_phys, CXL_COMPONENT_REG_BLOCK_SIZE))
+		return CXL_RESOURCE_NONE;
+
+	return component_reg_phys;
+}
+EXPORT_SYMBOL_NS_GPL(cxl_rcrb_to_component, CXL);
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 9b33ae4b2aec..43c43d1ec069 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -223,6 +223,14 @@ enum cxl_regloc_type;
 int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
 		      struct cxl_register_map *map);
 
+enum cxl_rcrb {
+	CXL_RCRB_DOWNSTREAM,
+	CXL_RCRB_UPSTREAM,
+};
+resource_size_t cxl_rcrb_to_component(struct device *dev,
+				      resource_size_t rcrb,
+				      enum cxl_rcrb which);
+
 #define CXL_RESOURCE_NONE ((resource_size_t) -1)
 #define CXL_TARGET_STRLEN 20
 
@@ -486,12 +494,16 @@ cxl_find_dport_by_dev(struct cxl_port *port, const struct device *dport_dev)
  * @dport: PCI bridge or firmware device representing the downstream link
  * @port_id: unique hardware identifier for dport in decoder target list
  * @component_reg_phys: downstream port component registers
+ * @rcrb: base address for the Root Complex Register Block
+ * @rch: Indicate whether this dport was enumerated in RCH or VH mode
  * @port: reference to cxl_port that contains this downstream port
  */
 struct cxl_dport {
 	struct device *dport;
 	int port_id;
 	resource_size_t component_reg_phys;
+	resource_size_t rcrb;
+	bool rch;
 	struct cxl_port *port;
 };
 
@@ -561,6 +573,10 @@ bool schedule_cxl_memdev_detach(struct cxl_memdev *cxlmd);
 struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
 				     struct device *dport, int port_id,
 				     resource_size_t component_reg_phys);
+struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
+					 struct device *dport_dev, int port_id,
+					 resource_size_t component_reg_phys,
+					 resource_size_t rcrb);
 
 struct cxl_decoder *to_cxl_decoder(struct device *dev);
 struct cxl_root_decoder *to_cxl_root_decoder(struct device *dev);
diff --git a/tools/testing/cxl/Kbuild b/tools/testing/cxl/Kbuild
index 500be85729cc..9e4d94e81723 100644
--- a/tools/testing/cxl/Kbuild
+++ b/tools/testing/cxl/Kbuild
@@ -10,6 +10,7 @@ ldflags-y += --wrap=devm_cxl_add_passthrough_decoder
 ldflags-y += --wrap=devm_cxl_enumerate_decoders
 ldflags-y += --wrap=cxl_await_media_ready
 ldflags-y += --wrap=cxl_hdm_decode_init
+ldflags-y += --wrap=cxl_rcrb_to_component
 
 DRIVERS := ../../../drivers
 CXL_SRC := $(DRIVERS)/cxl
diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
index 42a34650dd2f..1823c61d7ba3 100644
--- a/tools/testing/cxl/test/cxl.c
+++ b/tools/testing/cxl/test/cxl.c
@@ -694,6 +694,15 @@ static int mock_cxl_port_enumerate_dports(struct cxl_port *port)
 	return 0;
 }
 
+resource_size_t mock_cxl_rcrb_to_component(struct device *dev,
+					   resource_size_t rcrb,
+					   enum cxl_rcrb which)
+{
+	dev_dbg(dev, "rcrb: %pa which: %d\n", &rcrb, which);
+
+	return 0;
+}
+
 static struct cxl_mock_ops cxl_mock_ops = {
 	.is_mock_adev = is_mock_adev,
 	.is_mock_bridge = is_mock_bridge,
@@ -702,6 +711,7 @@ static struct cxl_mock_ops cxl_mock_ops = {
 	.is_mock_dev = is_mock_dev,
 	.acpi_table_parse_cedt = mock_acpi_table_parse_cedt,
 	.acpi_evaluate_integer = mock_acpi_evaluate_integer,
+	.cxl_rcrb_to_component = mock_cxl_rcrb_to_component,
 	.acpi_pci_find_root = mock_acpi_pci_find_root,
 	.devm_cxl_port_enumerate_dports = mock_cxl_port_enumerate_dports,
 	.devm_cxl_setup_hdm = mock_cxl_setup_hdm,
diff --git a/tools/testing/cxl/test/mock.c b/tools/testing/cxl/test/mock.c
index bce6a21df0d5..5dface08e0de 100644
--- a/tools/testing/cxl/test/mock.c
+++ b/tools/testing/cxl/test/mock.c
@@ -224,6 +224,25 @@ int __wrap_cxl_hdm_decode_init(struct cxl_dev_state *cxlds,
 }
 EXPORT_SYMBOL_NS_GPL(__wrap_cxl_hdm_decode_init, CXL);
 
+resource_size_t __wrap_cxl_rcrb_to_component(struct device *dev,
+					     resource_size_t rcrb,
+					     enum cxl_rcrb which)
+{
+	int index;
+	resource_size_t component_reg_phys;
+	struct cxl_mock_ops *ops = get_cxl_mock_ops(&index);
+
+	if (ops && ops->is_mock_port(dev))
+		component_reg_phys =
+			ops->cxl_rcrb_to_component(dev, rcrb, which);
+	else
+		component_reg_phys = cxl_rcrb_to_component(dev, rcrb, which);
+	put_cxl_mock_ops(index);
+
+	return component_reg_phys;
+}
+EXPORT_SYMBOL_NS_GPL(__wrap_cxl_rcrb_to_component, CXL);
+
 MODULE_LICENSE("GPL v2");
 MODULE_IMPORT_NS(ACPI);
 MODULE_IMPORT_NS(CXL);
diff --git a/tools/testing/cxl/test/mock.h b/tools/testing/cxl/test/mock.h
index 738f24e3988a..ef33f159375e 100644
--- a/tools/testing/cxl/test/mock.h
+++ b/tools/testing/cxl/test/mock.h
@@ -15,6 +15,9 @@ struct cxl_mock_ops {
 					     acpi_string pathname,
 					     struct acpi_object_list *arguments,
 					     unsigned long long *data);
+	resource_size_t (*cxl_rcrb_to_component)(struct device *dev,
+						 resource_size_t rcrb,
+						 enum cxl_rcrb which);
 	struct acpi_pci_root *(*acpi_pci_find_root)(acpi_handle handle);
 	bool (*is_mock_bus)(struct pci_bus *bus);
 	bool (*is_mock_port)(struct device *dev);

[PATCH v4 09/12] cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem

[Dan Williams]

In preparation for devm_cxl_add_endpoint() to call out to
cxl_rcrb_to_component() to map the upstream port component registers,
move devm_cxl_add_endpoint() from the cxl_core to the cxl_mem driver.
This is due to the organization of cxl_test that mandates that the
cxl_core not call out to any mocked symbols. It also cleans up the
export of devm_cxl_add_endpoint() which is just a wrapper around
devm_cxl_add_port().

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/core/core.h |    8 --------
 drivers/cxl/core/port.c |   39 ---------------------------------------
 drivers/cxl/cxl.h       |    2 --
 drivers/cxl/cxlmem.h    |    9 +++++++++
 drivers/cxl/mem.c       |   38 ++++++++++++++++++++++++++++++++++++++
 5 files changed, 47 insertions(+), 49 deletions(-)

diff --git a/drivers/cxl/core/core.h b/drivers/cxl/core/core.h
index 1d8f87be283f..8c04672dca56 100644
--- a/drivers/cxl/core/core.h
+++ b/drivers/cxl/core/core.h
@@ -58,14 +58,6 @@ extern struct rw_semaphore cxl_dpa_rwsem;
 
 bool is_switch_decoder(struct device *dev);
 struct cxl_switch_decoder *to_cxl_switch_decoder(struct device *dev);
-static inline struct cxl_ep *cxl_ep_load(struct cxl_port *port,
-					 struct cxl_memdev *cxlmd)
-{
-	if (!port)
-		return NULL;
-
-	return xa_load(&port->endpoints, (unsigned long)&cxlmd->dev);
-}
 
 int cxl_memdev_init(void);
 void cxl_memdev_exit(void);
diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
index d9fe06e1462f..c7f58282b2c1 100644
--- a/drivers/cxl/core/port.c
+++ b/drivers/cxl/core/port.c
@@ -1203,45 +1203,6 @@ static void reap_dports(struct cxl_port *port)
 	}
 }
 
-int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
-			  struct cxl_dport *parent_dport)
-{
-	struct cxl_port *parent_port = parent_dport->port;
-	struct cxl_dev_state *cxlds = cxlmd->cxlds;
-	struct cxl_port *endpoint, *iter, *down;
-	int rc;
-
-	/*
-	 * Now that the path to the root is established record all the
-	 * intervening ports in the chain.
-	 */
-	for (iter = parent_port, down = NULL; !is_cxl_root(iter);
-	     down = iter, iter = to_cxl_port(iter->dev.parent)) {
-		struct cxl_ep *ep;
-
-		ep = cxl_ep_load(iter, cxlmd);
-		ep->next = down;
-	}
-
-	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
-				     cxlds->component_reg_phys, parent_dport);
-	if (IS_ERR(endpoint))
-		return PTR_ERR(endpoint);
-
-	rc = cxl_endpoint_autoremove(cxlmd, endpoint);
-	if (rc)
-		return rc;
-
-	if (!endpoint->dev.driver) {
-		dev_err(&cxlmd->dev, "%s failed probe\n",
-			dev_name(&endpoint->dev));
-		return -ENXIO;
-	}
-
-	return 0;
-}
-EXPORT_SYMBOL_NS_GPL(devm_cxl_add_endpoint, CXL);
-
 static void cxl_detach_ep(void *data)
 {
 	struct cxl_memdev *cxlmd = data;
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 43c43d1ec069..d94635e43a50 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -560,8 +560,6 @@ struct pci_bus *cxl_port_to_pci_bus(struct cxl_port *port);
 struct cxl_port *devm_cxl_add_port(struct device *host, struct device *uport,
 				   resource_size_t component_reg_phys,
 				   struct cxl_dport *parent_dport);
-int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
-			  struct cxl_dport *parent_dport);
 struct cxl_port *find_cxl_root(struct device *dev);
 int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd);
 void cxl_bus_rescan(void);
diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
index c1c9960ab05f..e082991bc58c 100644
--- a/drivers/cxl/cxlmem.h
+++ b/drivers/cxl/cxlmem.h
@@ -80,6 +80,15 @@ static inline bool is_cxl_endpoint(struct cxl_port *port)
 
 struct cxl_memdev *devm_cxl_add_memdev(struct cxl_dev_state *cxlds);
 
+static inline struct cxl_ep *cxl_ep_load(struct cxl_port *port,
+					 struct cxl_memdev *cxlmd)
+{
+	if (!port)
+		return NULL;
+
+	return xa_load(&port->endpoints, (unsigned long)&cxlmd->dev);
+}
+
 /**
  * struct cxl_mbox_cmd - A command to be submitted to hardware.
  * @opcode: (input) The command set and command submitted to hardware.
diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
index 549b6b499bae..aa63ce8c7ca6 100644
--- a/drivers/cxl/mem.c
+++ b/drivers/cxl/mem.c
@@ -45,6 +45,44 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
 	return 0;
 }
 
+static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
+				 struct cxl_dport *parent_dport)
+{
+	struct cxl_port *parent_port = parent_dport->port;
+	struct cxl_dev_state *cxlds = cxlmd->cxlds;
+	struct cxl_port *endpoint, *iter, *down;
+	int rc;
+
+	/*
+	 * Now that the path to the root is established record all the
+	 * intervening ports in the chain.
+	 */
+	for (iter = parent_port, down = NULL; !is_cxl_root(iter);
+	     down = iter, iter = to_cxl_port(iter->dev.parent)) {
+		struct cxl_ep *ep;
+
+		ep = cxl_ep_load(iter, cxlmd);
+		ep->next = down;
+	}
+
+	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
+				     cxlds->component_reg_phys, parent_dport);
+	if (IS_ERR(endpoint))
+		return PTR_ERR(endpoint);
+
+	rc = cxl_endpoint_autoremove(cxlmd, endpoint);
+	if (rc)
+		return rc;
+
+	if (!endpoint->dev.driver) {
+		dev_err(&cxlmd->dev, "%s failed probe\n",
+			dev_name(&endpoint->dev));
+		return -ENXIO;
+	}
+
+	return 0;
+}
+
 static int cxl_mem_probe(struct device *dev)
 {
 	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);

[PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dan Williams]

Unlike a CXL memory expander in a VH topology that has at least one
intervening 'struct cxl_port' instance between itself and the CXL root
device, an RCD attaches one-level higher. For example:

               VH
          ┌──────────┐
          │ ACPI0017 │
          │  root0   │
          └─────┬────┘
                │
          ┌─────┴────┐
          │  dport0  │
    ┌─────┤ ACPI0016 ├─────┐
    │     │  port1   │     │
    │     └────┬─────┘     │
    │          │           │
 ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
 │dport0│   │dport1│   │dport2│
 │ RP0  │   │ RP1  │   │ RP2  │
 └──────┘   └──┬───┘   └──────┘
               │
           ┌───┴─────┐
           │endpoint0│
           │  port2  │
           └─────────┘

...vs:

              RCH
          ┌──────────┐
          │ ACPI0017 │
          │  root0   │
          └────┬─────┘
               │
           ┌───┴────┐
           │ dport0 │
           │ACPI0016│
           └───┬────┘
               │
          ┌────┴─────┐
          │endpoint0 │
          │  port1   │
          └──────────┘

So arrange for endpoint port in the RCH/RCD case to appear directly
connected to the host-bridge in its singular role as a dport. Compare
that to the VH case where the host-bridge serves a dual role as a
'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
the Root Ports in the Root Complex that are modeled as 'cxl_dport'
instances in the CXL topology.

Another deviation from the VH case is that RCDs may need to look up
their component registers from the Root Complex Register Block (RCRB).
That platform firmware specified RCRB area is cached by the cxl_acpi
driver and conveyed via the host-bridge dport to the cxl_mem driver to
perform the cxl_rcrb_to_component() lookup for the endpoint port
(See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
upstream port component registers).

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/cxl/core/port.c |   11 +++++++++--
 drivers/cxl/cxlmem.h    |    2 ++
 drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
 drivers/cxl/pci.c       |   10 ++++++++++
 4 files changed, 45 insertions(+), 9 deletions(-)

diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
index c7f58282b2c1..2385ee00eb9a 100644
--- a/drivers/cxl/core/port.c
+++ b/drivers/cxl/core/port.c
@@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
 {
 	struct device *dev = &cxlmd->dev;
 	struct device *iter;
+	struct cxl_dport *dport;
+	struct cxl_port *port;
 	int rc;
 
+	/*
+	 * Skip intermediate port enumeration in the RCH case, there
+	 * are no ports in between a host bridge and an endpoint.
+	 */
+	if (cxlmd->cxlds->rcd)
+		return 0;
+
 	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
 	if (rc)
 		return rc;
@@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
 	for (iter = dev; iter; iter = grandparent(iter)) {
 		struct device *dport_dev = grandparent(iter);
 		struct device *uport_dev;
-		struct cxl_dport *dport;
-		struct cxl_port *port;
 
 		if (!dport_dev)
 			return 0;
diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
index e082991bc58c..35d485d041f0 100644
--- a/drivers/cxl/cxlmem.h
+++ b/drivers/cxl/cxlmem.h
@@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
  * @dev: The device associated with this CXL state
  * @regs: Parsed register blocks
  * @cxl_dvsec: Offset to the PCIe device DVSEC
+ * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
  * @payload_size: Size of space for payload
  *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
  * @lsa_size: Size of Label Storage Area
@@ -235,6 +236,7 @@ struct cxl_dev_state {
 	struct cxl_regs regs;
 	int cxl_dvsec;
 
+	bool rcd;
 	size_t payload_size;
 	size_t lsa_size;
 	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
index aa63ce8c7ca6..9a655b4b5e52 100644
--- a/drivers/cxl/mem.c
+++ b/drivers/cxl/mem.c
@@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
 	return 0;
 }
 
-static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
+static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
 				 struct cxl_dport *parent_dport)
 {
 	struct cxl_port *parent_port = parent_dport->port;
 	struct cxl_dev_state *cxlds = cxlmd->cxlds;
 	struct cxl_port *endpoint, *iter, *down;
+	resource_size_t component_reg_phys;
 	int rc;
 
 	/*
@@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
 		ep->next = down;
 	}
 
-	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
-				     cxlds->component_reg_phys, parent_dport);
+	/*
+	 * The component registers for an RCD might come from the
+	 * host-bridge RCRB if they are not already mapped via the
+	 * typical register locator mechanism.
+	 */
+	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
+		component_reg_phys = cxl_rcrb_to_component(
+			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
+	else
+		component_reg_phys = cxlds->component_reg_phys;
+	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
+				     parent_dport);
 	if (IS_ERR(endpoint))
 		return PTR_ERR(endpoint);
 
@@ -87,6 +98,7 @@ static int cxl_mem_probe(struct device *dev)
 {
 	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
 	struct cxl_dev_state *cxlds = cxlmd->cxlds;
+	struct device *endpoint_parent;
 	struct cxl_port *parent_port;
 	struct cxl_dport *dport;
 	struct dentry *dentry;
@@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
 		return -ENXIO;
 	}
 
-	device_lock(&parent_port->dev);
-	if (!parent_port->dev.driver) {
+	if (dport->rch)
+		endpoint_parent = parent_port->uport;
+	else
+		endpoint_parent = &parent_port->dev;
+
+	device_lock(endpoint_parent);
+	if (!endpoint_parent->driver) {
 		dev_err(dev, "CXL port topology %s not enabled\n",
 			dev_name(&parent_port->dev));
 		rc = -ENXIO;
 		goto unlock;
 	}
 
-	rc = devm_cxl_add_endpoint(cxlmd, dport);
+	rc = devm_cxl_add_endpoint(endpoint_parent, cxlmd, dport);
 unlock:
-	device_unlock(&parent_port->dev);
+	device_unlock(endpoint_parent);
 	put_device(&parent_port->dev);
 	if (rc)
 		return rc;
diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
index e15da405b948..73ff6c33a0c0 100644
--- a/drivers/cxl/pci.c
+++ b/drivers/cxl/pci.c
@@ -433,6 +433,15 @@ static void devm_cxl_pci_create_doe(struct cxl_dev_state *cxlds)
 	}
 }
 
+/*
+ * Assume that any RCIEP that emits the CXL memory expander class code
+ * is an RCD
+ */
+static bool is_cxl_restricted(struct pci_dev *pdev)
+{
+	return pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END;
+}
+
 static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	struct cxl_register_map map;
@@ -455,6 +464,7 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (IS_ERR(cxlds))
 		return PTR_ERR(cxlds);
 
+	cxlds->rcd = is_cxl_restricted(pdev);
 	cxlds->serial = pci_get_dsn(pdev);
 	cxlds->cxl_dvsec = pci_find_dvsec_capability(
 		pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_PCIE_DEVICE);

[PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Dan Williams]

In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
the represents the memory expander. Unlike a VH topology there is no
CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
as the CXL root object (ACPI0017 on ACPI based systems) targeting the
host-bridge as a dport, per usual, but then that dport directly hosts
the endpoint port.

Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
device instance as its immediate child.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 tools/testing/cxl/test/cxl.c |  151 +++++++++++++++++++++++++++++++++++++++---
 tools/testing/cxl/test/mem.c |   37 ++++++++++
 2 files changed, 176 insertions(+), 12 deletions(-)

diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
index 1823c61d7ba3..b1c362090b92 100644
--- a/tools/testing/cxl/test/cxl.c
+++ b/tools/testing/cxl/test/cxl.c
@@ -13,9 +13,11 @@
 
 #define NR_CXL_HOST_BRIDGES 2
 #define NR_CXL_SINGLE_HOST 1
+#define NR_CXL_RCH 1
 #define NR_CXL_ROOT_PORTS 2
 #define NR_CXL_SWITCH_PORTS 2
 #define NR_CXL_PORT_DECODERS 8
+#define NR_BRIDGES (NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + NR_CXL_RCH)
 
 static struct platform_device *cxl_acpi;
 static struct platform_device *cxl_host_bridge[NR_CXL_HOST_BRIDGES];
@@ -35,6 +37,8 @@ static struct platform_device *cxl_swd_single[NR_MEM_SINGLE];
 struct platform_device *cxl_mem[NR_MEM_MULTI];
 struct platform_device *cxl_mem_single[NR_MEM_SINGLE];
 
+static struct platform_device *cxl_rch[NR_CXL_RCH];
+static struct platform_device *cxl_rcd[NR_CXL_RCH];
 
 static inline bool is_multi_bridge(struct device *dev)
 {
@@ -57,7 +61,7 @@ static inline bool is_single_bridge(struct device *dev)
 }
 
 static struct acpi_device acpi0017_mock;
-static struct acpi_device host_bridge[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST] = {
+static struct acpi_device host_bridge[NR_BRIDGES] = {
 	[0] = {
 		.handle = &host_bridge[0],
 	},
@@ -67,7 +71,9 @@ static struct acpi_device host_bridge[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST]
 	[2] = {
 		.handle = &host_bridge[2],
 	},
-
+	[3] = {
+		.handle = &host_bridge[3],
+	},
 };
 
 static bool is_mock_dev(struct device *dev)
@@ -80,6 +86,9 @@ static bool is_mock_dev(struct device *dev)
 	for (i = 0; i < ARRAY_SIZE(cxl_mem_single); i++)
 		if (dev == &cxl_mem_single[i]->dev)
 			return true;
+	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++)
+		if (dev == &cxl_rcd[i]->dev)
+			return true;
 	if (dev == &cxl_acpi->dev)
 		return true;
 	return false;
@@ -101,7 +110,7 @@ static bool is_mock_adev(struct acpi_device *adev)
 
 static struct {
 	struct acpi_table_cedt cedt;
-	struct acpi_cedt_chbs chbs[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST];
+	struct acpi_cedt_chbs chbs[NR_BRIDGES];
 	struct {
 		struct acpi_cedt_cfmws cfmws;
 		u32 target[1];
@@ -122,6 +131,10 @@ static struct {
 		struct acpi_cedt_cfmws cfmws;
 		u32 target[1];
 	} cfmws4;
+	struct {
+		struct acpi_cedt_cfmws cfmws;
+		u32 target[1];
+	} cfmws5;
 } __packed mock_cedt = {
 	.cedt = {
 		.header = {
@@ -154,6 +167,14 @@ static struct {
 		.uid = 2,
 		.cxl_version = ACPI_CEDT_CHBS_VERSION_CXL20,
 	},
+	.chbs[3] = {
+		.header = {
+			.type = ACPI_CEDT_TYPE_CHBS,
+			.length = sizeof(mock_cedt.chbs[0]),
+		},
+		.uid = 3,
+		.cxl_version = ACPI_CEDT_CHBS_VERSION_CXL11,
+	},
 	.cfmws0 = {
 		.cfmws = {
 			.header = {
@@ -229,6 +250,21 @@ static struct {
 		},
 		.target = { 2 },
 	},
+	.cfmws5 = {
+		.cfmws = {
+			.header = {
+				.type = ACPI_CEDT_TYPE_CFMWS,
+				.length = sizeof(mock_cedt.cfmws5),
+			},
+			.interleave_ways = 0,
+			.granularity = 4,
+			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_TYPE3 |
+					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
+			.qtg_id = 5,
+			.window_size = SZ_256M,
+		},
+		.target = { 3 },
+	},
 };
 
 struct acpi_cedt_cfmws *mock_cfmws[] = {
@@ -237,6 +273,7 @@ struct acpi_cedt_cfmws *mock_cfmws[] = {
 	[2] = &mock_cedt.cfmws2.cfmws,
 	[3] = &mock_cedt.cfmws3.cfmws,
 	[4] = &mock_cedt.cfmws4.cfmws,
+	[5] = &mock_cedt.cfmws5.cfmws,
 };
 
 struct cxl_mock_res {
@@ -262,11 +299,11 @@ static void depopulate_all_mock_resources(void)
 	mutex_unlock(&mock_res_lock);
 }
 
-static struct cxl_mock_res *alloc_mock_res(resource_size_t size)
+static struct cxl_mock_res *alloc_mock_res(resource_size_t size, int align)
 {
 	struct cxl_mock_res *res = kzalloc(sizeof(*res), GFP_KERNEL);
 	struct genpool_data_align data = {
-		.align = SZ_256M,
+		.align = align,
 	};
 	unsigned long phys;
 
@@ -301,7 +338,7 @@ static int populate_cedt(void)
 		else
 			size = ACPI_CEDT_CHBS_LENGTH_CXL11;
 
-		res = alloc_mock_res(size);
+		res = alloc_mock_res(size, size);
 		if (!res)
 			return -ENOMEM;
 		chbs->base = res->range.start;
@@ -311,7 +348,7 @@ static int populate_cedt(void)
 	for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
 		struct acpi_cedt_cfmws *window = mock_cfmws[i];
 
-		res = alloc_mock_res(window->window_size);
+		res = alloc_mock_res(window->window_size, SZ_256M);
 		if (!res)
 			return -ENOMEM;
 		window->base_hpa = res->range.start;
@@ -330,6 +367,10 @@ static bool is_mock_bridge(struct device *dev)
 	for (i = 0; i < ARRAY_SIZE(cxl_hb_single); i++)
 		if (dev == &cxl_hb_single[i]->dev)
 			return true;
+	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++)
+		if (dev == &cxl_rch[i]->dev)
+			return true;
+
 	return false;
 }
 
@@ -439,7 +480,7 @@ mock_acpi_evaluate_integer(acpi_handle handle, acpi_string pathname,
 	return AE_OK;
 }
 
-static struct pci_bus mock_pci_bus[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST];
+static struct pci_bus mock_pci_bus[NR_BRIDGES];
 static struct acpi_pci_root mock_pci_root[ARRAY_SIZE(mock_pci_bus)] = {
 	[0] = {
 		.bus = &mock_pci_bus[0],
@@ -450,7 +491,9 @@ static struct acpi_pci_root mock_pci_root[ARRAY_SIZE(mock_pci_bus)] = {
 	[2] = {
 		.bus = &mock_pci_bus[2],
 	},
-
+	[3] = {
+		.bus = &mock_pci_bus[3],
+	},
 };
 
 static bool is_mock_bus(struct pci_bus *bus)
@@ -736,6 +779,87 @@ static void mock_companion(struct acpi_device *adev, struct device *dev)
 #define SZ_512G (SZ_64G * 8)
 #endif
 
+static __init int cxl_rch_init(void)
+{
+	int rc, i;
+
+	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++) {
+		int idx = NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + i;
+		struct acpi_device *adev = &host_bridge[idx];
+		struct platform_device *pdev;
+
+		pdev = platform_device_alloc("cxl_host_bridge", idx);
+		if (!pdev)
+			goto err_bridge;
+
+		mock_companion(adev, &pdev->dev);
+		rc = platform_device_add(pdev);
+		if (rc) {
+			platform_device_put(pdev);
+			goto err_bridge;
+		}
+
+		cxl_rch[i] = pdev;
+		mock_pci_bus[idx].bridge = &pdev->dev;
+		rc = sysfs_create_link(&pdev->dev.kobj, &pdev->dev.kobj,
+				       "firmware_node");
+		if (rc)
+			goto err_bridge;
+	}
+
+	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++) {
+		int idx = NR_MEM_MULTI + NR_MEM_SINGLE + i;
+		struct platform_device *rch = cxl_rch[i];
+		struct platform_device *pdev;
+
+		pdev = platform_device_alloc("cxl_rcd", idx);
+		if (!pdev)
+			goto err_mem;
+		pdev->dev.parent = &rch->dev;
+		set_dev_node(&pdev->dev, i % 2);
+
+		rc = platform_device_add(pdev);
+		if (rc) {
+			platform_device_put(pdev);
+			goto err_mem;
+		}
+		cxl_rcd[i] = pdev;
+	}
+
+	return 0;
+
+err_mem:
+	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
+		platform_device_unregister(cxl_rcd[i]);
+err_bridge:
+	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
+		struct platform_device *pdev = cxl_rch[i];
+
+		if (!pdev)
+			continue;
+		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
+		platform_device_unregister(cxl_rch[i]);
+	}
+
+	return rc;
+}
+
+static void cxl_rch_exit(void)
+{
+	int i;
+
+	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
+		platform_device_unregister(cxl_rcd[i]);
+	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
+		struct platform_device *pdev = cxl_rch[i];
+
+		if (!pdev)
+			continue;
+		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
+		platform_device_unregister(cxl_rch[i]);
+	}
+}
+
 static __init int cxl_single_init(void)
 {
 	int i, rc;
@@ -1008,9 +1132,13 @@ static __init int cxl_test_init(void)
 	if (rc)
 		goto err_mem;
 
+	rc = cxl_rch_init();
+	if (rc)
+		goto err_single;
+
 	cxl_acpi = platform_device_alloc("cxl_acpi", 0);
 	if (!cxl_acpi)
-		goto err_single;
+		goto err_rch;
 
 	mock_companion(&acpi0017_mock, &cxl_acpi->dev);
 	acpi0017_mock.dev.bus = &platform_bus_type;
@@ -1023,6 +1151,8 @@ static __init int cxl_test_init(void)
 
 err_add:
 	platform_device_put(cxl_acpi);
+err_rch:
+	cxl_rch_exit();
 err_single:
 	cxl_single_exit();
 err_mem:
@@ -1060,6 +1190,7 @@ static __exit void cxl_test_exit(void)
 	int i;
 
 	platform_device_unregister(cxl_acpi);
+	cxl_rch_exit();
 	cxl_single_exit();
 	for (i = ARRAY_SIZE(cxl_mem) - 1; i >= 0; i--)
 		platform_device_unregister(cxl_mem[i]);
diff --git a/tools/testing/cxl/test/mem.c b/tools/testing/cxl/test/mem.c
index a4ee8e61dd60..b59c5976b2d9 100644
--- a/tools/testing/cxl/test/mem.c
+++ b/tools/testing/cxl/test/mem.c
@@ -100,6 +100,24 @@ static int mock_get_log(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
 	return 0;
 }
 
+static int mock_rcd_id(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
+{
+	struct cxl_mbox_identify id = {
+		.fw_revision = { "mock fw v1 " },
+		.total_capacity =
+			cpu_to_le64(DEV_SIZE / CXL_CAPACITY_MULTIPLIER),
+		.volatile_capacity =
+			cpu_to_le64(DEV_SIZE / CXL_CAPACITY_MULTIPLIER),
+	};
+
+	if (cmd->size_out < sizeof(id))
+		return -EINVAL;
+
+	memcpy(cmd->payload_out, &id, sizeof(id));
+
+	return 0;
+}
+
 static int mock_id(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
 {
 	struct cxl_mbox_identify id = {
@@ -216,7 +234,10 @@ static int cxl_mock_mbox_send(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *
 		rc = mock_get_log(cxlds, cmd);
 		break;
 	case CXL_MBOX_OP_IDENTIFY:
-		rc = mock_id(cxlds, cmd);
+		if (cxlds->rcd)
+			rc = mock_rcd_id(cxlds, cmd);
+		else
+			rc = mock_id(cxlds, cmd);
 		break;
 	case CXL_MBOX_OP_GET_LSA:
 		rc = mock_get_lsa(cxlds, cmd);
@@ -245,6 +266,13 @@ static void label_area_release(void *lsa)
 	vfree(lsa);
 }
 
+static bool is_rcd(struct platform_device *pdev)
+{
+	const struct platform_device_id *id = platform_get_device_id(pdev);
+
+	return !!id->driver_data;
+}
+
 static int cxl_mock_mem_probe(struct platform_device *pdev)
 {
 	struct device *dev = &pdev->dev;
@@ -268,6 +296,10 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
 	cxlds->serial = pdev->id;
 	cxlds->mbox_send = cxl_mock_mbox_send;
 	cxlds->payload_size = SZ_4K;
+	if (is_rcd(pdev)) {
+		cxlds->rcd = true;
+		cxlds->component_reg_phys = CXL_RESOURCE_NONE;
+	}
 
 	rc = cxl_enumerate_cmds(cxlds);
 	if (rc)
@@ -289,7 +321,8 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
 }
 
 static const struct platform_device_id cxl_mock_mem_ids[] = {
-	{ .name = "cxl_mem", },
+	{ .name = "cxl_mem", 0 },
+	{ .name = "cxl_rcd", 1 },
 	{ },
 };
 MODULE_DEVICE_TABLE(platform, cxl_mock_mem_ids);

[PATCH v4 12/12] cxl/acpi: Set ACPI's CXL _OSC to indicate CXL1.1 support

[Dan Williams]

From: Terry Bowman <terry.bowman@amd.com>

ACPI includes a CXL _OSC for the OS to communicate what it knows of CXL
device topologies. To date Linux has added support for CXL 2.0 (VH) port
topologies, hotplug, and error handling. Now that the driver also know
how to enumerate CXL 1.1 (RCH) port topologies, indicate that capability
via CXL _OSC. See CXL3.0 Table 9-26 'Interpretation of CXL _OSC Support
Field'

Signed-off-by: Terry Bowman <terry.bowman@amd.com>
Signed-off-by: Robert Richter <rrichter@amd.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
[djbw: wordsmith changelog]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
 drivers/acpi/pci_root.c |    1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/acpi/pci_root.c b/drivers/acpi/pci_root.c
index 4e3db20e9cbb..b3c202d2a433 100644
--- a/drivers/acpi/pci_root.c
+++ b/drivers/acpi/pci_root.c
@@ -493,6 +493,7 @@ static u32 calculate_cxl_support(void)
 	u32 support;
 
 	support = OSC_CXL_2_0_PORT_DEV_REG_ACCESS_SUPPORT;
+	support |= OSC_CXL_1_1_PORT_REG_ACCESS_SUPPORT;
 	if (pci_aer_available())
 		support |= OSC_CXL_PROTOCOL_ERR_REPORTING_SUPPORT;
 	if (IS_ENABLED(CONFIG_HOTPLUG_PCI_PCIE))

Re: [PATCH v4 03/12] cxl/pmem: Refactor nvdimm device registration, delete the workqueue

[Jonathan Cameron]

On Thu, 24 Nov 2022 10:34:52 -0800
Dan Williams <dan.j.williams@intel.com> wrote:

> The three objects 'struct cxl_nvdimm_bridge', 'struct cxl_nvdimm', and
> 'struct cxl_pmem_region' manage CXL persistent memory resources. The
> bridge represents base platform resources, the nvdimm represents one or
> more endpoints, and the region is a collection of nvdimms that
> contribute to an assembled address range.
> 
> Their relationship is such that a region is torn down if any component
> endpoints are removed. All regions and endpoints are torn down if the
> foundational bridge device goes down.
> 
> A workqueue was deployed to manage these interdependencies, but it is
> difficult to reason about, and fragile. A recent attempt to take the CXL
> root device lock in the cxl_mem driver was reported by lockdep as
> colliding with the flush_work() in the cxl_pmem flows.
> 
> Instead of the workqueue, arrange for all pmem/nvdimm devices to be torn
> down immediately and hierarchically. A similar change is made to both
> the 'cxl_nvdimm' and 'cxl_pmem_region' objects. For bisect-ability both
> changes are made in the same patch which unfortunately makes the patch
> bigger than desired.
> 
> Arrange for cxl_memdev and cxl_region to register a cxl_nvdimm and
> cxl_pmem_region as a devres release action of the bridge device.
> Additionally, include a devres release action of the cxl_memdev or
> cxl_region device that triggers the bridge's release action if an endpoint
> exits before the bridge. I.e. this allows either unplugging the bridge,
> or unplugging and endpoint to result in the same cleanup actions.
> 
> To keep the patch smaller the cleanup of the now defunct workqueue
> infrastructure is saved for a follow-on patch.
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Hi Dan,

This is fiddly to follow, but then so was the original. A few minor comments inline.

Jonathan

> ---
>  drivers/cxl/core/pmem.c      |   70 ++++++++++++++++++++----
>  drivers/cxl/core/region.c    |   54 ++++++++++++++++++-
>  drivers/cxl/cxl.h            |    7 ++
>  drivers/cxl/cxlmem.h         |    4 +
>  drivers/cxl/mem.c            |    9 +++
>  drivers/cxl/pci.c            |    3 -
>  drivers/cxl/pmem.c           |  122 ++++++++++++------------------------------
>  tools/testing/cxl/test/mem.c |    3 -
>  8 files changed, 164 insertions(+), 108 deletions(-)
> 
> diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
> index 1d12a8206444..647b3a30638e 100644
> --- a/drivers/cxl/core/pmem.c
> +++ b/drivers/cxl/core/pmem.c
> @@ -219,7 +219,8 @@ EXPORT_SYMBOL_NS_GPL(to_cxl_nvdimm, CXL);
>  
>  static struct lock_class_key cxl_nvdimm_key;
>  
> -static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
> +static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_nvdimm_bridge *cxl_nvb,
> +					   struct cxl_memdev *cxlmd)
>  {
>  	struct cxl_nvdimm *cxl_nvd;
>  	struct device *dev;
> @@ -230,6 +231,7 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
>  
>  	dev = &cxl_nvd->dev;
>  	cxl_nvd->cxlmd = cxlmd;
> +	cxlmd->cxl_nvd = cxl_nvd;
>  	device_initialize(dev);
>  	lockdep_set_class(&dev->mutex, &cxl_nvdimm_key);
>  	device_set_pm_not_required(dev);
> @@ -240,27 +242,52 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
>  	return cxl_nvd;
>  }
>  
> -static void cxl_nvd_unregister(void *dev)
> +static void cxl_nvd_unregister(void *_cxl_nvd)
>  {
> -	device_unregister(dev);
> +	struct cxl_nvdimm *cxl_nvd = _cxl_nvd;
> +	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
> +
> +	device_lock_assert(&cxlmd->cxl_nvb->dev);

Locally it's not immediately obvious if that is always the same
as 
	device_lock_assert(&cxl_nvb->dev);
If not, a comment, if it is maybe just change to that.

> +	cxl_nvd->cxlmd = NULL;
> +	cxlmd->cxl_nvd = NULL;
> +	device_unregister(&cxl_nvd->dev);
> +}
> +
> +static void cxlmd_release_nvdimm(void *_cxlmd)
> +{
> +	struct cxl_memdev *cxlmd = _cxlmd;
> +	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
> +
> +	device_lock(&cxl_nvb->dev);
> +	if (cxlmd->cxl_nvd)
> +		devm_release_action(&cxl_nvb->dev, cxl_nvd_unregister,
> +				    cxlmd->cxl_nvd);
> +	device_unlock(&cxl_nvb->dev);
> +	put_device(&cxl_nvb->dev);
>  }
>  
>  /**
>   * devm_cxl_add_nvdimm() - add a bridge between a cxl_memdev and an nvdimm
> - * @host: same host as @cxlmd
>   * @cxlmd: cxl_memdev instance that will perform LIBNVDIMM operations
>   *
>   * Return: 0 on success negative error code on failure.
>   */
> -int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
> +int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd)
>  {
> +	struct cxl_nvdimm_bridge *cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);

Another cosmetic change, but I'd prefer the actual
	cxl_nvb = cxl_find_nvdimm_bridge();

to be just above the error check rather than up here.

>  	struct cxl_nvdimm *cxl_nvd;
>  	struct device *dev;
>  	int rc;
>  
> -	cxl_nvd = cxl_nvdimm_alloc(cxlmd);
> -	if (IS_ERR(cxl_nvd))
> -		return PTR_ERR(cxl_nvd);
> +	if (!cxl_nvb)
> +		return -ENODEV;
> +
> +	cxl_nvd = cxl_nvdimm_alloc(cxl_nvb, cxlmd);
> +	if (IS_ERR(cxl_nvd)) {
> +		rc = PTR_ERR(cxl_nvd);
> +		goto err_alloc;
> +	}
> +	cxlmd->cxl_nvb = cxl_nvb;
>  
>  	dev = &cxl_nvd->dev;
>  	rc = dev_set_name(dev, "pmem%d", cxlmd->id);
> @@ -271,13 +298,34 @@ int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
>  	if (rc)
>  		goto err;
>  
> -	dev_dbg(host, "%s: register %s\n", dev_name(dev->parent),
> -		dev_name(dev));
> +	dev_dbg(&cxlmd->dev, "register %s\n", dev_name(dev));
>  
> -	return devm_add_action_or_reset(host, cxl_nvd_unregister, dev);
> +	/*
> +	 * Remove this nvdimm connection if either the top-level PMEM
> +	 * bridge goes down, or the endpoint device goes through
> +	 * ->remove().
> +	 */

Perhaps move this comment down to inside the if (cxl_nvb->dev.driver)
block as it only refers (I think) to the devm_add_action_or_reset(),
not the surrounding driver binding checks.

> +	device_lock(&cxl_nvb->dev);
> +	if (cxl_nvb->dev.driver)
> +		rc = devm_add_action_or_reset(&cxl_nvb->dev, cxl_nvd_unregister,
> +					      cxl_nvd);
> +	else
> +		rc = -ENXIO;
> +	device_unlock(&cxl_nvb->dev);
> +
> +	if (rc)
> +		goto err_alloc;
> +
> +	/* @cxlmd carries a reference on @cxl_nvb until cxlmd_release_nvdimm */
> +	return devm_add_action_or_reset(&cxlmd->dev, cxlmd_release_nvdimm, cxlmd);
>  
>  err:
>  	put_device(dev);
> +err_alloc:
> +	put_device(&cxl_nvb->dev);

Is this ordering necessary? It's not reverse of the setup above, so if we can reordering
to be so, that is probably a good thing. (move these NULL setting above the put_device(&cxl_nvb->dev)).

> +	cxlmd->cxl_nvb = NULL;
> +	cxlmd->cxl_nvd = NULL;
> +
>  	return rc;
>  }
>  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_nvdimm, CXL);
> diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> index f9ae5ad284ff..e73bec828032 100644
> --- a/drivers/cxl/core/region.c
> +++ b/drivers/cxl/core/region.c
> @@ -1812,6 +1812,7 @@ static struct lock_class_key cxl_pmem_region_key;
>  static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
>  {
>  	struct cxl_region_params *p = &cxlr->params;
> +	struct cxl_nvdimm_bridge *cxl_nvb;
>  	struct cxl_pmem_region *cxlr_pmem;
>  	struct device *dev;
>  	int i;
> @@ -1839,6 +1840,14 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
>  		struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
>  		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
>  
> +		if (i == 0) {

Whilst kind of obvious, maybe a comment in here that for end points in the region the
cxl_nvb will be the same hence we just look it up for the first one?

> +			cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
> +			if (!cxl_nvb) {
> +				cxlr_pmem = ERR_PTR(-ENODEV);
> +				goto out;
> +			}
> +			cxlr->cxl_nvb = cxl_nvb;
> +		}
>  		m->cxlmd = cxlmd;
>  		get_device(&cxlmd->dev);
>  		m->start = cxled->dpa_res->start;
> @@ -1848,6 +1857,7 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
>  
>  	dev = &cxlr_pmem->dev;
>  	cxlr_pmem->cxlr = cxlr;
> +	cxlr->cxlr_pmem = cxlr_pmem;
>  	device_initialize(dev);
>  	lockdep_set_class(&dev->mutex, &cxl_pmem_region_key);
>  	device_set_pm_not_required(dev);
> @@ -1860,9 +1870,30 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
>  	return cxlr_pmem;
>  }
>  
> -static void cxlr_pmem_unregister(void *dev)
> +static void cxlr_pmem_unregister(void *_cxlr_pmem)
> +{
> +	struct cxl_pmem_region *cxlr_pmem = _cxlr_pmem;
> +	struct cxl_region *cxlr = cxlr_pmem->cxlr;
> +	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
> +
> +	device_lock_assert(&cxl_nvb->dev);

This scheme is obvious in this patch, but probably less so when just
looking at the resulting code. Perhaps worth a comment
here on why we care about that particular lock?

> +	cxlr->cxlr_pmem = NULL;
> +	cxlr_pmem->cxlr = NULL;
> +	device_unregister(&cxlr_pmem->dev);
> +}
> +
> +static void cxlr_release_nvdimm(void *_cxlr)
>  {
> -	device_unregister(dev);
> +	struct cxl_region *cxlr = _cxlr;
> +	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
> +
> +	device_lock(&cxl_nvb->dev);
> +	if (cxlr->cxlr_pmem)
> +		devm_release_action(&cxl_nvb->dev, cxlr_pmem_unregister,
> +				    cxlr->cxlr_pmem);
> +	device_unlock(&cxl_nvb->dev);
> +	cxlr->cxl_nvb = NULL;
> +	put_device(&cxl_nvb->dev);
>  }
>  
>  /**
> @@ -1874,12 +1905,14 @@ static void cxlr_pmem_unregister(void *dev)
>  static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
>  {
>  	struct cxl_pmem_region *cxlr_pmem;
> +	struct cxl_nvdimm_bridge *cxl_nvb;
>  	struct device *dev;
>  	int rc;
>  
>  	cxlr_pmem = cxl_pmem_region_alloc(cxlr);
>  	if (IS_ERR(cxlr_pmem))
>  		return PTR_ERR(cxlr_pmem);
> +	cxl_nvb = cxlr->cxl_nvb;
>  
>  	dev = &cxlr_pmem->dev;
>  	rc = dev_set_name(dev, "pmem_region%d", cxlr->id);
> @@ -1893,10 +1926,25 @@ static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
>  	dev_dbg(&cxlr->dev, "%s: register %s\n", dev_name(dev->parent),
>  		dev_name(dev));
>  
> -	return devm_add_action_or_reset(&cxlr->dev, cxlr_pmem_unregister, dev);
> +	device_lock(&cxl_nvb->dev);
> +	if (cxl_nvb->dev.driver)
> +		rc = devm_add_action_or_reset(&cxl_nvb->dev,
> +					      cxlr_pmem_unregister, cxlr_pmem);
> +	else
> +		rc = -ENXIO;
> +	device_unlock(&cxl_nvb->dev);
> +
> +	if (rc)
> +		goto err_bridge;
> +
> +	/* @cxlr carries a reference on @cxl_nvb until cxlr_release_nvdimm */
> +	return devm_add_action_or_reset(&cxlr->dev, cxlr_release_nvdimm, cxlr);
>  
>  err:
>  	put_device(dev);
> +err_bridge:
> +	put_device(&cxl_nvb->dev);
> +	cxlr->cxl_nvb = NULL;
>  	return rc;
>  }
>  
> diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> index 4ac7938eaf6c..9b5ba9626636 100644
> --- a/drivers/cxl/cxl.h
> +++ b/drivers/cxl/cxl.h
> @@ -386,6 +386,8 @@ struct cxl_region_params {
>   * @id: This region's id. Id is globally unique across all regions
>   * @mode: Endpoint decoder allocation / access mode
>   * @type: Endpoint decoder target type
> + * @cxl_nvb: nvdimm bridge for coordinating @cxlr_pmem shutdown

I was going to suggest just carrying the struct device around, but this
comment isn't quite true.  I think cxl_region->cxl_nvb is also used in
cxl_pmem_region_probe() to get the nvdimm_buss for nvdimm_pmem_region_create()


> + * @cxlr_pmem: (for pmem regions) cached copy of the nvdimm bridge
>   * @params: active + config params for the region
>   */
>  struct cxl_region {
> @@ -393,6 +395,8 @@ struct cxl_region {
>  	int id;
>  	enum cxl_decoder_mode mode;
>  	enum cxl_decoder_type type;
> +	struct cxl_nvdimm_bridge *cxl_nvb;
> +	struct cxl_pmem_region *cxlr_pmem;
>  	struct cxl_region_params params;
>  };
>  
> @@ -438,7 +442,6 @@ struct cxl_pmem_region {
>  	struct device dev;
>  	struct cxl_region *cxlr;
>  	struct nd_region *nd_region;
> -	struct cxl_nvdimm_bridge *bridge;
>  	struct range hpa_range;
>  	int nr_mappings;
>  	struct cxl_pmem_region_mapping mapping[];
> @@ -637,7 +640,7 @@ struct cxl_nvdimm_bridge *devm_cxl_add_nvdimm_bridge(struct device *host,
>  struct cxl_nvdimm *to_cxl_nvdimm(struct device *dev);
>  bool is_cxl_nvdimm(struct device *dev);
>  bool is_cxl_nvdimm_bridge(struct device *dev);
> -int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd);
> +int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd);
>  struct cxl_nvdimm_bridge *cxl_find_nvdimm_bridge(struct device *dev);
>  
>  #ifdef CONFIG_CXL_REGION
> diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> index 88e3a8e54b6a..c1c9960ab05f 100644
> --- a/drivers/cxl/cxlmem.h
> +++ b/drivers/cxl/cxlmem.h
> @@ -35,6 +35,8 @@
>   * @cdev: char dev core object for ioctl operations
>   * @cxlds: The device state backing this device
>   * @detach_work: active memdev lost a port in its ancestry
> + * @cxl_nvb: coordinate removal of @cxl_nvd if present
> + * @cxl_nvd: optional bridge to an nvdimm if the device supports pmem
>   * @id: id number of this memdev instance.
>   */
>  struct cxl_memdev {
> @@ -42,6 +44,8 @@ struct cxl_memdev {
>  	struct cdev cdev;
>  	struct cxl_dev_state *cxlds;
>  	struct work_struct detach_work;
> +	struct cxl_nvdimm_bridge *cxl_nvb;
> +	struct cxl_nvdimm *cxl_nvd;
>  	int id;
>  };
>  

> diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
> index 652f00fc68ca..73357d0c3f25 100644
> --- a/drivers/cxl/pmem.c
> +++ b/drivers/cxl/pmem.c


>  static struct cxl_driver cxl_nvdimm_driver = {
> @@ -200,6 +182,16 @@ static int cxl_pmem_ctl(struct nvdimm_bus_descriptor *nd_desc,
>  	return cxl_pmem_nvdimm_ctl(nvdimm, cmd, buf, buf_len);
>  }
>  
> +static void unregister_nvdimm_bus(void *_cxl_nvb)
> +{
> +	struct cxl_nvdimm_bridge *cxl_nvb = _cxl_nvb;
> +	struct nvdimm_bus *nvdimm_bus = cxl_nvb->nvdimm_bus;
> +
> +	cxl_nvb->nvdimm_bus = NULL;
> +	nvdimm_bus_unregister(nvdimm_bus);
> +}
> +

Single blank line.

> +
>  static bool online_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb)
>  {
>  	if (cxl_nvb->nvdimm_bus)
> @@ -303,23 +295,21 @@ static int cxl_nvdimm_bridge_probe(struct device *dev)
>  {
>  	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
>  
> -	if (cxl_nvb->state == CXL_NVB_DEAD)
> -		return -ENXIO;
> +	cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor){
) {
matches existing style in this file.

> +		.provider_name = "CXL",
> +		.module = THIS_MODULE,
> +		.ndctl = cxl_pmem_ctl,
> +	};
>  
> -	if (cxl_nvb->state == CXL_NVB_NEW) {
> -		cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor) {
> -			.provider_name = "CXL",
> -			.module = THIS_MODULE,
> -			.ndctl = cxl_pmem_ctl,
> -		};
> +	cxl_nvb->nvdimm_bus =
> +		nvdimm_bus_register(&cxl_nvb->dev, &cxl_nvb->nd_desc);
>  
> -		INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
> -	}
> +	if (!cxl_nvb->nvdimm_bus)
> +		return -ENOMEM;
>  
> -	cxl_nvb->state = CXL_NVB_ONLINE;
> -	cxl_nvdimm_bridge_state_work(cxl_nvb);
> +	INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
>  
> -	return 0;
> +	return devm_add_action_or_reset(dev, unregister_nvdimm_bus, cxl_nvb);

Re: [PATCH v4 03/12] cxl/pmem: Refactor nvdimm device registration, delete the workqueue

[Dan Williams]

Jonathan Cameron wrote:
> On Thu, 24 Nov 2022 10:34:52 -0800
> Dan Williams <dan.j.williams@intel.com> wrote:
> 
> > The three objects 'struct cxl_nvdimm_bridge', 'struct cxl_nvdimm', and
> > 'struct cxl_pmem_region' manage CXL persistent memory resources. The
> > bridge represents base platform resources, the nvdimm represents one or
> > more endpoints, and the region is a collection of nvdimms that
> > contribute to an assembled address range.
> > 
> > Their relationship is such that a region is torn down if any component
> > endpoints are removed. All regions and endpoints are torn down if the
> > foundational bridge device goes down.
> > 
> > A workqueue was deployed to manage these interdependencies, but it is
> > difficult to reason about, and fragile. A recent attempt to take the CXL
> > root device lock in the cxl_mem driver was reported by lockdep as
> > colliding with the flush_work() in the cxl_pmem flows.
> > 
> > Instead of the workqueue, arrange for all pmem/nvdimm devices to be torn
> > down immediately and hierarchically. A similar change is made to both
> > the 'cxl_nvdimm' and 'cxl_pmem_region' objects. For bisect-ability both
> > changes are made in the same patch which unfortunately makes the patch
> > bigger than desired.
> > 
> > Arrange for cxl_memdev and cxl_region to register a cxl_nvdimm and
> > cxl_pmem_region as a devres release action of the bridge device.
> > Additionally, include a devres release action of the cxl_memdev or
> > cxl_region device that triggers the bridge's release action if an endpoint
> > exits before the bridge. I.e. this allows either unplugging the bridge,
> > or unplugging and endpoint to result in the same cleanup actions.
> > 
> > To keep the patch smaller the cleanup of the now defunct workqueue
> > infrastructure is saved for a follow-on patch.
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> 
> Hi Dan,
> 
> This is fiddly to follow, but then so was the original.

Fiddly is a charitable comment for the original. This patch is larger
than I would have liked, but I did not see a way to unwind the
complexity without causing a bisect break.

> A few minor comments inline.

Thanks for taking a look.

> 
> Jonathan
> 
> > ---
> >  drivers/cxl/core/pmem.c      |   70 ++++++++++++++++++++----
> >  drivers/cxl/core/region.c    |   54 ++++++++++++++++++-
> >  drivers/cxl/cxl.h            |    7 ++
> >  drivers/cxl/cxlmem.h         |    4 +
> >  drivers/cxl/mem.c            |    9 +++
> >  drivers/cxl/pci.c            |    3 -
> >  drivers/cxl/pmem.c           |  122 ++++++++++++------------------------------
> >  tools/testing/cxl/test/mem.c |    3 -
> >  8 files changed, 164 insertions(+), 108 deletions(-)
> > 
> > diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
> > index 1d12a8206444..647b3a30638e 100644
> > --- a/drivers/cxl/core/pmem.c
> > +++ b/drivers/cxl/core/pmem.c
> > @@ -219,7 +219,8 @@ EXPORT_SYMBOL_NS_GPL(to_cxl_nvdimm, CXL);
> >  
> >  static struct lock_class_key cxl_nvdimm_key;
> >  
> > -static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
> > +static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_nvdimm_bridge *cxl_nvb,
> > +					   struct cxl_memdev *cxlmd)
> >  {
> >  	struct cxl_nvdimm *cxl_nvd;
> >  	struct device *dev;
> > @@ -230,6 +231,7 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
> >  
> >  	dev = &cxl_nvd->dev;
> >  	cxl_nvd->cxlmd = cxlmd;
> > +	cxlmd->cxl_nvd = cxl_nvd;
> >  	device_initialize(dev);
> >  	lockdep_set_class(&dev->mutex, &cxl_nvdimm_key);
> >  	device_set_pm_not_required(dev);
> > @@ -240,27 +242,52 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
> >  	return cxl_nvd;
> >  }
> >  
> > -static void cxl_nvd_unregister(void *dev)
> > +static void cxl_nvd_unregister(void *_cxl_nvd)
> >  {
> > -	device_unregister(dev);
> > +	struct cxl_nvdimm *cxl_nvd = _cxl_nvd;
> > +	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
> > +
> > +	device_lock_assert(&cxlmd->cxl_nvb->dev);
> 
> Locally it's not immediately obvious if that is always the same
> as 
> 	device_lock_assert(&cxl_nvb->dev);
> If not, a comment, if it is maybe just change to that.

Sure, yes, it is always the latter.

> 
> > +	cxl_nvd->cxlmd = NULL;
> > +	cxlmd->cxl_nvd = NULL;
> > +	device_unregister(&cxl_nvd->dev);
> > +}
> > +
> > +static void cxlmd_release_nvdimm(void *_cxlmd)
> > +{
> > +	struct cxl_memdev *cxlmd = _cxlmd;
> > +	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
> > +
> > +	device_lock(&cxl_nvb->dev);
> > +	if (cxlmd->cxl_nvd)
> > +		devm_release_action(&cxl_nvb->dev, cxl_nvd_unregister,
> > +				    cxlmd->cxl_nvd);
> > +	device_unlock(&cxl_nvb->dev);
> > +	put_device(&cxl_nvb->dev);
> >  }
> >  
> >  /**
> >   * devm_cxl_add_nvdimm() - add a bridge between a cxl_memdev and an nvdimm
> > - * @host: same host as @cxlmd
> >   * @cxlmd: cxl_memdev instance that will perform LIBNVDIMM operations
> >   *
> >   * Return: 0 on success negative error code on failure.
> >   */
> > -int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
> > +int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd)
> >  {
> > +	struct cxl_nvdimm_bridge *cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
> 
> Another cosmetic change, but I'd prefer the actual
> 	cxl_nvb = cxl_find_nvdimm_bridge();
> 
> to be just above the error check rather than up here.

ok.

> 
> >  	struct cxl_nvdimm *cxl_nvd;
> >  	struct device *dev;
> >  	int rc;
> >  
> > -	cxl_nvd = cxl_nvdimm_alloc(cxlmd);
> > -	if (IS_ERR(cxl_nvd))
> > -		return PTR_ERR(cxl_nvd);
> > +	if (!cxl_nvb)
> > +		return -ENODEV;
> > +
> > +	cxl_nvd = cxl_nvdimm_alloc(cxl_nvb, cxlmd);
> > +	if (IS_ERR(cxl_nvd)) {
> > +		rc = PTR_ERR(cxl_nvd);
> > +		goto err_alloc;
> > +	}
> > +	cxlmd->cxl_nvb = cxl_nvb;
> >  
> >  	dev = &cxl_nvd->dev;
> >  	rc = dev_set_name(dev, "pmem%d", cxlmd->id);
> > @@ -271,13 +298,34 @@ int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
> >  	if (rc)
> >  		goto err;
> >  
> > -	dev_dbg(host, "%s: register %s\n", dev_name(dev->parent),
> > -		dev_name(dev));
> > +	dev_dbg(&cxlmd->dev, "register %s\n", dev_name(dev));
> >  
> > -	return devm_add_action_or_reset(host, cxl_nvd_unregister, dev);
> > +	/*
> > +	 * Remove this nvdimm connection if either the top-level PMEM
> > +	 * bridge goes down, or the endpoint device goes through
> > +	 * ->remove().
> > +	 */
> 
> Perhaps move this comment down to inside the if (cxl_nvb->dev.driver)
> block as it only refers (I think) to the devm_add_action_or_reset(),
> not the surrounding driver binding checks.

It refers to the whole rest of this function because this action:

    devm_add_action_or_reset(&cxl_nvb->dev, cxl_nvd_unregister, cxl_nvd)

...deletes the cxl_nvdimm when the bridge is removed, and this action:

    devm_add_action_or_reset(&cxlmd->dev, cxlmd_release_nvdimm, cxlmd)

...removes the nvdimm, and the above release action, if the endpoint is
removed *before* the bridge.

At a minimum the comment needs strengthening if that detail was not
conveyed.

> 
> > +	device_lock(&cxl_nvb->dev);
> > +	if (cxl_nvb->dev.driver)
> > +		rc = devm_add_action_or_reset(&cxl_nvb->dev, cxl_nvd_unregister,
> > +					      cxl_nvd);
> > +	else
> > +		rc = -ENXIO;
> > +	device_unlock(&cxl_nvb->dev);
> > +
> > +	if (rc)
> > +		goto err_alloc;
> > +
> > +	/* @cxlmd carries a reference on @cxl_nvb until cxlmd_release_nvdimm */
> > +	return devm_add_action_or_reset(&cxlmd->dev, cxlmd_release_nvdimm, cxlmd);
> >  
> >  err:
> >  	put_device(dev);
> > +err_alloc:
> > +	put_device(&cxl_nvb->dev);
> 
> Is this ordering necessary? It's not reverse of the setup above, so if we can reordering
> to be so, that is probably a good thing. (move these NULL setting above the put_device(&cxl_nvb->dev)).

Sure, they can move above.

> 
> > +	cxlmd->cxl_nvb = NULL;
> > +	cxlmd->cxl_nvd = NULL;
> > +
> >  	return rc;
> >  }
> >  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_nvdimm, CXL);
> > diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
> > index f9ae5ad284ff..e73bec828032 100644
> > --- a/drivers/cxl/core/region.c
> > +++ b/drivers/cxl/core/region.c
> > @@ -1812,6 +1812,7 @@ static struct lock_class_key cxl_pmem_region_key;
> >  static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
> >  {
> >  	struct cxl_region_params *p = &cxlr->params;
> > +	struct cxl_nvdimm_bridge *cxl_nvb;
> >  	struct cxl_pmem_region *cxlr_pmem;
> >  	struct device *dev;
> >  	int i;
> > @@ -1839,6 +1840,14 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
> >  		struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
> >  		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
> >  
> > +		if (i == 0) {
> 
> Whilst kind of obvious, maybe a comment in here that for end points in the region the
> cxl_nvb will be the same hence we just look it up for the first one?

ok.

> 
> > +			cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
> > +			if (!cxl_nvb) {
> > +				cxlr_pmem = ERR_PTR(-ENODEV);
> > +				goto out;
> > +			}
> > +			cxlr->cxl_nvb = cxl_nvb;
> > +		}
> >  		m->cxlmd = cxlmd;
> >  		get_device(&cxlmd->dev);
> >  		m->start = cxled->dpa_res->start;
> > @@ -1848,6 +1857,7 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
> >  
> >  	dev = &cxlr_pmem->dev;
> >  	cxlr_pmem->cxlr = cxlr;
> > +	cxlr->cxlr_pmem = cxlr_pmem;
> >  	device_initialize(dev);
> >  	lockdep_set_class(&dev->mutex, &cxl_pmem_region_key);
> >  	device_set_pm_not_required(dev);
> > @@ -1860,9 +1870,30 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
> >  	return cxlr_pmem;
> >  }
> >  
> > -static void cxlr_pmem_unregister(void *dev)
> > +static void cxlr_pmem_unregister(void *_cxlr_pmem)
> > +{
> > +	struct cxl_pmem_region *cxlr_pmem = _cxlr_pmem;
> > +	struct cxl_region *cxlr = cxlr_pmem->cxlr;
> > +	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
> > +
> > +	device_lock_assert(&cxl_nvb->dev);
> 
> This scheme is obvious in this patch, but probably less so when just
> looking at the resulting code. Perhaps worth a comment
> here on why we care about that particular lock?

ok, and I'll make a similar comment in cxl_nvd_unregister.

> 
> > +	cxlr->cxlr_pmem = NULL;
> > +	cxlr_pmem->cxlr = NULL;
> > +	device_unregister(&cxlr_pmem->dev);
> > +}
> > +
> > +static void cxlr_release_nvdimm(void *_cxlr)
> >  {
> > -	device_unregister(dev);
> > +	struct cxl_region *cxlr = _cxlr;
> > +	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
> > +
> > +	device_lock(&cxl_nvb->dev);
> > +	if (cxlr->cxlr_pmem)
> > +		devm_release_action(&cxl_nvb->dev, cxlr_pmem_unregister,
> > +				    cxlr->cxlr_pmem);
> > +	device_unlock(&cxl_nvb->dev);
> > +	cxlr->cxl_nvb = NULL;
> > +	put_device(&cxl_nvb->dev);
> >  }
> >  
> >  /**
> > @@ -1874,12 +1905,14 @@ static void cxlr_pmem_unregister(void *dev)
> >  static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
> >  {
> >  	struct cxl_pmem_region *cxlr_pmem;
> > +	struct cxl_nvdimm_bridge *cxl_nvb;
> >  	struct device *dev;
> >  	int rc;
> >  
> >  	cxlr_pmem = cxl_pmem_region_alloc(cxlr);
> >  	if (IS_ERR(cxlr_pmem))
> >  		return PTR_ERR(cxlr_pmem);
> > +	cxl_nvb = cxlr->cxl_nvb;
> >  
> >  	dev = &cxlr_pmem->dev;
> >  	rc = dev_set_name(dev, "pmem_region%d", cxlr->id);
> > @@ -1893,10 +1926,25 @@ static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
> >  	dev_dbg(&cxlr->dev, "%s: register %s\n", dev_name(dev->parent),
> >  		dev_name(dev));
> >  
> > -	return devm_add_action_or_reset(&cxlr->dev, cxlr_pmem_unregister, dev);
> > +	device_lock(&cxl_nvb->dev);
> > +	if (cxl_nvb->dev.driver)
> > +		rc = devm_add_action_or_reset(&cxl_nvb->dev,
> > +					      cxlr_pmem_unregister, cxlr_pmem);
> > +	else
> > +		rc = -ENXIO;
> > +	device_unlock(&cxl_nvb->dev);
> > +
> > +	if (rc)
> > +		goto err_bridge;
> > +
> > +	/* @cxlr carries a reference on @cxl_nvb until cxlr_release_nvdimm */
> > +	return devm_add_action_or_reset(&cxlr->dev, cxlr_release_nvdimm, cxlr);
> >  
> >  err:
> >  	put_device(dev);
> > +err_bridge:
> > +	put_device(&cxl_nvb->dev);
> > +	cxlr->cxl_nvb = NULL;
> >  	return rc;
> >  }
> >  
> > diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> > index 4ac7938eaf6c..9b5ba9626636 100644
> > --- a/drivers/cxl/cxl.h
> > +++ b/drivers/cxl/cxl.h
> > @@ -386,6 +386,8 @@ struct cxl_region_params {
> >   * @id: This region's id. Id is globally unique across all regions
> >   * @mode: Endpoint decoder allocation / access mode
> >   * @type: Endpoint decoder target type
> > + * @cxl_nvb: nvdimm bridge for coordinating @cxlr_pmem shutdown
> 
> I was going to suggest just carrying the struct device around, but this
> comment isn't quite true.  I think cxl_region->cxl_nvb is also used in
> cxl_pmem_region_probe() to get the nvdimm_buss for nvdimm_pmem_region_create()

Perhaps just make the comment less specific and say "nvdimm bridge for
coordinating @cxlr_pmem setup and shutdown".

> 
> 
> > + * @cxlr_pmem: (for pmem regions) cached copy of the nvdimm bridge
> >   * @params: active + config params for the region
> >   */
> >  struct cxl_region {
> > @@ -393,6 +395,8 @@ struct cxl_region {
> >  	int id;
> >  	enum cxl_decoder_mode mode;
> >  	enum cxl_decoder_type type;
> > +	struct cxl_nvdimm_bridge *cxl_nvb;
> > +	struct cxl_pmem_region *cxlr_pmem;
> >  	struct cxl_region_params params;
> >  };
> >  
> > @@ -438,7 +442,6 @@ struct cxl_pmem_region {
> >  	struct device dev;
> >  	struct cxl_region *cxlr;
> >  	struct nd_region *nd_region;
> > -	struct cxl_nvdimm_bridge *bridge;
> >  	struct range hpa_range;
> >  	int nr_mappings;
> >  	struct cxl_pmem_region_mapping mapping[];
> > @@ -637,7 +640,7 @@ struct cxl_nvdimm_bridge *devm_cxl_add_nvdimm_bridge(struct device *host,
> >  struct cxl_nvdimm *to_cxl_nvdimm(struct device *dev);
> >  bool is_cxl_nvdimm(struct device *dev);
> >  bool is_cxl_nvdimm_bridge(struct device *dev);
> > -int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd);
> > +int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd);
> >  struct cxl_nvdimm_bridge *cxl_find_nvdimm_bridge(struct device *dev);
> >  
> >  #ifdef CONFIG_CXL_REGION
> > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > index 88e3a8e54b6a..c1c9960ab05f 100644
> > --- a/drivers/cxl/cxlmem.h
> > +++ b/drivers/cxl/cxlmem.h
> > @@ -35,6 +35,8 @@
> >   * @cdev: char dev core object for ioctl operations
> >   * @cxlds: The device state backing this device
> >   * @detach_work: active memdev lost a port in its ancestry
> > + * @cxl_nvb: coordinate removal of @cxl_nvd if present
> > + * @cxl_nvd: optional bridge to an nvdimm if the device supports pmem
> >   * @id: id number of this memdev instance.
> >   */
> >  struct cxl_memdev {
> > @@ -42,6 +44,8 @@ struct cxl_memdev {
> >  	struct cdev cdev;
> >  	struct cxl_dev_state *cxlds;
> >  	struct work_struct detach_work;
> > +	struct cxl_nvdimm_bridge *cxl_nvb;
> > +	struct cxl_nvdimm *cxl_nvd;
> >  	int id;
> >  };
> >  
> 
> > diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
> > index 652f00fc68ca..73357d0c3f25 100644
> > --- a/drivers/cxl/pmem.c
> > +++ b/drivers/cxl/pmem.c
> 
> 
> >  static struct cxl_driver cxl_nvdimm_driver = {
> > @@ -200,6 +182,16 @@ static int cxl_pmem_ctl(struct nvdimm_bus_descriptor *nd_desc,
> >  	return cxl_pmem_nvdimm_ctl(nvdimm, cmd, buf, buf_len);
> >  }
> >  
> > +static void unregister_nvdimm_bus(void *_cxl_nvb)
> > +{
> > +	struct cxl_nvdimm_bridge *cxl_nvb = _cxl_nvb;
> > +	struct nvdimm_bus *nvdimm_bus = cxl_nvb->nvdimm_bus;
> > +
> > +	cxl_nvb->nvdimm_bus = NULL;
> > +	nvdimm_bus_unregister(nvdimm_bus);
> > +}
> > +
> 
> Single blank line.
> 
> > +
> >  static bool online_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb)
> >  {
> >  	if (cxl_nvb->nvdimm_bus)
> > @@ -303,23 +295,21 @@ static int cxl_nvdimm_bridge_probe(struct device *dev)
> >  {
> >  	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
> >  
> > -	if (cxl_nvb->state == CXL_NVB_DEAD)
> > -		return -ENXIO;
> > +	cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor){
> ) {
> matches existing style in this file.

clang-format removes that space, will add it back.

> 
> > +		.provider_name = "CXL",
> > +		.module = THIS_MODULE,
> > +		.ndctl = cxl_pmem_ctl,
> > +	};
> >  
> > -	if (cxl_nvb->state == CXL_NVB_NEW) {
> > -		cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor) {
> > -			.provider_name = "CXL",
> > -			.module = THIS_MODULE,
> > -			.ndctl = cxl_pmem_ctl,
> > -		};
> > +	cxl_nvb->nvdimm_bus =
> > +		nvdimm_bus_register(&cxl_nvb->dev, &cxl_nvb->nd_desc);
> >  
> > -		INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
> > -	}
> > +	if (!cxl_nvb->nvdimm_bus)
> > +		return -ENOMEM;
> >  
> > -	cxl_nvb->state = CXL_NVB_ONLINE;
> > -	cxl_nvdimm_bridge_state_work(cxl_nvb);
> > +	INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
> >  
> > -	return 0;
> > +	return devm_add_action_or_reset(dev, unregister_nvdimm_bus, cxl_nvb);

[PATCH v5 03/12] cxl/pmem: Refactor nvdimm device registration, delete the workqueue

[Dan Williams]

The three objects 'struct cxl_nvdimm_bridge', 'struct cxl_nvdimm', and
'struct cxl_pmem_region' manage CXL persistent memory resources. The
bridge represents base platform resources, the nvdimm represents one or
more endpoints, and the region is a collection of nvdimms that
contribute to an assembled address range.

Their relationship is such that a region is torn down if any component
endpoints are removed. All regions and endpoints are torn down if the
foundational bridge device goes down.

A workqueue was deployed to manage these interdependencies, but it is
difficult to reason about, and fragile. A recent attempt to take the CXL
root device lock in the cxl_mem driver was reported by lockdep as
colliding with the flush_work() in the cxl_pmem flows.

Instead of the workqueue, arrange for all pmem/nvdimm devices to be torn
down immediately and hierarchically. A similar change is made to both
the 'cxl_nvdimm' and 'cxl_pmem_region' objects. For bisect-ability both
changes are made in the same patch which unfortunately makes the patch
bigger than desired.

Arrange for cxl_memdev and cxl_region to register a cxl_nvdimm and
cxl_pmem_region as a devres release action of the bridge device.
Additionally, include a devres release action of the cxl_memdev or
cxl_region device that triggers the bridge's release action if an endpoint
exits before the bridge. I.e. this allows either unplugging the bridge,
or unplugging and endpoint to result in the same cleanup actions.

To keep the patch smaller the cleanup of the now defunct workqueue
infrastructure is saved for a follow-on patch.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
---
Changes since v4 from Jonathan's feedback:
- Clarify cxl_nvd_unregister()
- Cosmetic and error exit ordering fixups in devm_cxl_add_nvdimm()
  (Jonathan)
- Clarify devm release action comment in devm_cxl_add_nvdimm()
- Clarify that each region member shares the same nvdimm-bridge
- Add similar clarifications to cxlr_pmem_unregister() as
  cxl_nvd_unregister()
- Clarify what the @cxl_nvb attribute of a cxl_region is used for
- Whitespace fixups

 drivers/cxl/core/pmem.c      |   78 +++++++++++++++++++++++----
 drivers/cxl/core/region.c    |   64 +++++++++++++++++++++-
 drivers/cxl/cxl.h            |    7 ++
 drivers/cxl/cxlmem.h         |    4 +
 drivers/cxl/mem.c            |    9 +++
 drivers/cxl/pci.c            |    3 -
 drivers/cxl/pmem.c           |  121 ++++++++++++------------------------------
 tools/testing/cxl/test/mem.c |    3 -
 8 files changed, 181 insertions(+), 108 deletions(-)

diff --git a/drivers/cxl/core/pmem.c b/drivers/cxl/core/pmem.c
index 1d12a8206444..4d36805079ad 100644
--- a/drivers/cxl/core/pmem.c
+++ b/drivers/cxl/core/pmem.c
@@ -219,7 +219,8 @@ EXPORT_SYMBOL_NS_GPL(to_cxl_nvdimm, CXL);
 
 static struct lock_class_key cxl_nvdimm_key;
 
-static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
+static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_nvdimm_bridge *cxl_nvb,
+					   struct cxl_memdev *cxlmd)
 {
 	struct cxl_nvdimm *cxl_nvd;
 	struct device *dev;
@@ -230,6 +231,7 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
 
 	dev = &cxl_nvd->dev;
 	cxl_nvd->cxlmd = cxlmd;
+	cxlmd->cxl_nvd = cxl_nvd;
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_nvdimm_key);
 	device_set_pm_not_required(dev);
@@ -240,27 +242,60 @@ static struct cxl_nvdimm *cxl_nvdimm_alloc(struct cxl_memdev *cxlmd)
 	return cxl_nvd;
 }
 
-static void cxl_nvd_unregister(void *dev)
+static void cxl_nvd_unregister(void *_cxl_nvd)
 {
-	device_unregister(dev);
+	struct cxl_nvdimm *cxl_nvd = _cxl_nvd;
+	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
+
+	/*
+	 * Either the bridge is in ->remove() context under the device_lock(),
+	 * or cxlmd_release_nvdimm() is cancelling the bridge's release action
+	 * for @cxl_nvd and doing it itself (while manually holding the bridge
+	 * lock).
+	 */
+	device_lock_assert(&cxl_nvb->dev);
+	cxl_nvd->cxlmd = NULL;
+	cxlmd->cxl_nvd = NULL;
+	device_unregister(&cxl_nvd->dev);
+}
+
+static void cxlmd_release_nvdimm(void *_cxlmd)
+{
+	struct cxl_memdev *cxlmd = _cxlmd;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
+
+	device_lock(&cxl_nvb->dev);
+	if (cxlmd->cxl_nvd)
+		devm_release_action(&cxl_nvb->dev, cxl_nvd_unregister,
+				    cxlmd->cxl_nvd);
+	device_unlock(&cxl_nvb->dev);
+	put_device(&cxl_nvb->dev);
 }
 
 /**
  * devm_cxl_add_nvdimm() - add a bridge between a cxl_memdev and an nvdimm
- * @host: same host as @cxlmd
  * @cxlmd: cxl_memdev instance that will perform LIBNVDIMM operations
  *
  * Return: 0 on success negative error code on failure.
  */
-int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
+int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd)
 {
+	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct cxl_nvdimm *cxl_nvd;
 	struct device *dev;
 	int rc;
 
-	cxl_nvd = cxl_nvdimm_alloc(cxlmd);
-	if (IS_ERR(cxl_nvd))
-		return PTR_ERR(cxl_nvd);
+	cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
+	if (!cxl_nvb)
+		return -ENODEV;
+
+	cxl_nvd = cxl_nvdimm_alloc(cxl_nvb, cxlmd);
+	if (IS_ERR(cxl_nvd)) {
+		rc = PTR_ERR(cxl_nvd);
+		goto err_alloc;
+	}
+	cxlmd->cxl_nvb = cxl_nvb;
 
 	dev = &cxl_nvd->dev;
 	rc = dev_set_name(dev, "pmem%d", cxlmd->id);
@@ -271,13 +306,34 @@ int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd)
 	if (rc)
 		goto err;
 
-	dev_dbg(host, "%s: register %s\n", dev_name(dev->parent),
-		dev_name(dev));
+	dev_dbg(&cxlmd->dev, "register %s\n", dev_name(dev));
+
+	/*
+	 * The two actions below arrange for @cxl_nvd to be deleted when either
+	 * the top-level PMEM bridge goes down, or the endpoint device goes
+	 * through ->remove().
+	 */
+	device_lock(&cxl_nvb->dev);
+	if (cxl_nvb->dev.driver)
+		rc = devm_add_action_or_reset(&cxl_nvb->dev, cxl_nvd_unregister,
+					      cxl_nvd);
+	else
+		rc = -ENXIO;
+	device_unlock(&cxl_nvb->dev);
+
+	if (rc)
+		goto err_alloc;
 
-	return devm_add_action_or_reset(host, cxl_nvd_unregister, dev);
+	/* @cxlmd carries a reference on @cxl_nvb until cxlmd_release_nvdimm */
+	return devm_add_action_or_reset(&cxlmd->dev, cxlmd_release_nvdimm, cxlmd);
 
 err:
 	put_device(dev);
+err_alloc:
+	cxlmd->cxl_nvb = NULL;
+	cxlmd->cxl_nvd = NULL;
+	put_device(&cxl_nvb->dev);
+
 	return rc;
 }
 EXPORT_SYMBOL_NS_GPL(devm_cxl_add_nvdimm, CXL);
diff --git a/drivers/cxl/core/region.c b/drivers/cxl/core/region.c
index f9ae5ad284ff..1e61d1bafc0c 100644
--- a/drivers/cxl/core/region.c
+++ b/drivers/cxl/core/region.c
@@ -1812,6 +1812,7 @@ static struct lock_class_key cxl_pmem_region_key;
 static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 {
 	struct cxl_region_params *p = &cxlr->params;
+	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct cxl_pmem_region *cxlr_pmem;
 	struct device *dev;
 	int i;
@@ -1839,6 +1840,18 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 		struct cxl_memdev *cxlmd = cxled_to_memdev(cxled);
 		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
 
+		/*
+		 * Regions never span CXL root devices, so by definition the
+		 * bridge for one device is the same for all.
+		 */
+		if (i == 0) {
+			cxl_nvb = cxl_find_nvdimm_bridge(&cxlmd->dev);
+			if (!cxl_nvb) {
+				cxlr_pmem = ERR_PTR(-ENODEV);
+				goto out;
+			}
+			cxlr->cxl_nvb = cxl_nvb;
+		}
 		m->cxlmd = cxlmd;
 		get_device(&cxlmd->dev);
 		m->start = cxled->dpa_res->start;
@@ -1848,6 +1861,7 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 
 	dev = &cxlr_pmem->dev;
 	cxlr_pmem->cxlr = cxlr;
+	cxlr->cxlr_pmem = cxlr_pmem;
 	device_initialize(dev);
 	lockdep_set_class(&dev->mutex, &cxl_pmem_region_key);
 	device_set_pm_not_required(dev);
@@ -1860,9 +1874,36 @@ static struct cxl_pmem_region *cxl_pmem_region_alloc(struct cxl_region *cxlr)
 	return cxlr_pmem;
 }
 
-static void cxlr_pmem_unregister(void *dev)
+static void cxlr_pmem_unregister(void *_cxlr_pmem)
 {
-	device_unregister(dev);
+	struct cxl_pmem_region *cxlr_pmem = _cxlr_pmem;
+	struct cxl_region *cxlr = cxlr_pmem->cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
+
+	/*
+	 * Either the bridge is in ->remove() context under the device_lock(),
+	 * or cxlr_release_nvdimm() is cancelling the bridge's release action
+	 * for @cxlr_pmem and doing it itself (while manually holding the bridge
+	 * lock).
+	 */
+	device_lock_assert(&cxl_nvb->dev);
+	cxlr->cxlr_pmem = NULL;
+	cxlr_pmem->cxlr = NULL;
+	device_unregister(&cxlr_pmem->dev);
+}
+
+static void cxlr_release_nvdimm(void *_cxlr)
+{
+	struct cxl_region *cxlr = _cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
+
+	device_lock(&cxl_nvb->dev);
+	if (cxlr->cxlr_pmem)
+		devm_release_action(&cxl_nvb->dev, cxlr_pmem_unregister,
+				    cxlr->cxlr_pmem);
+	device_unlock(&cxl_nvb->dev);
+	cxlr->cxl_nvb = NULL;
+	put_device(&cxl_nvb->dev);
 }
 
 /**
@@ -1874,12 +1915,14 @@ static void cxlr_pmem_unregister(void *dev)
 static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
 {
 	struct cxl_pmem_region *cxlr_pmem;
+	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct device *dev;
 	int rc;
 
 	cxlr_pmem = cxl_pmem_region_alloc(cxlr);
 	if (IS_ERR(cxlr_pmem))
 		return PTR_ERR(cxlr_pmem);
+	cxl_nvb = cxlr->cxl_nvb;
 
 	dev = &cxlr_pmem->dev;
 	rc = dev_set_name(dev, "pmem_region%d", cxlr->id);
@@ -1893,10 +1936,25 @@ static int devm_cxl_add_pmem_region(struct cxl_region *cxlr)
 	dev_dbg(&cxlr->dev, "%s: register %s\n", dev_name(dev->parent),
 		dev_name(dev));
 
-	return devm_add_action_or_reset(&cxlr->dev, cxlr_pmem_unregister, dev);
+	device_lock(&cxl_nvb->dev);
+	if (cxl_nvb->dev.driver)
+		rc = devm_add_action_or_reset(&cxl_nvb->dev,
+					      cxlr_pmem_unregister, cxlr_pmem);
+	else
+		rc = -ENXIO;
+	device_unlock(&cxl_nvb->dev);
+
+	if (rc)
+		goto err_bridge;
+
+	/* @cxlr carries a reference on @cxl_nvb until cxlr_release_nvdimm */
+	return devm_add_action_or_reset(&cxlr->dev, cxlr_release_nvdimm, cxlr);
 
 err:
 	put_device(dev);
+err_bridge:
+	put_device(&cxl_nvb->dev);
+	cxlr->cxl_nvb = NULL;
 	return rc;
 }
 
diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
index 4ac7938eaf6c..fc6083b0e467 100644
--- a/drivers/cxl/cxl.h
+++ b/drivers/cxl/cxl.h
@@ -386,6 +386,8 @@ struct cxl_region_params {
  * @id: This region's id. Id is globally unique across all regions
  * @mode: Endpoint decoder allocation / access mode
  * @type: Endpoint decoder target type
+ * @cxl_nvb: nvdimm bridge for coordinating @cxlr_pmem setup / shutdown
+ * @cxlr_pmem: (for pmem regions) cached copy of the nvdimm bridge
  * @params: active + config params for the region
  */
 struct cxl_region {
@@ -393,6 +395,8 @@ struct cxl_region {
 	int id;
 	enum cxl_decoder_mode mode;
 	enum cxl_decoder_type type;
+	struct cxl_nvdimm_bridge *cxl_nvb;
+	struct cxl_pmem_region *cxlr_pmem;
 	struct cxl_region_params params;
 };
 
@@ -438,7 +442,6 @@ struct cxl_pmem_region {
 	struct device dev;
 	struct cxl_region *cxlr;
 	struct nd_region *nd_region;
-	struct cxl_nvdimm_bridge *bridge;
 	struct range hpa_range;
 	int nr_mappings;
 	struct cxl_pmem_region_mapping mapping[];
@@ -637,7 +640,7 @@ struct cxl_nvdimm_bridge *devm_cxl_add_nvdimm_bridge(struct device *host,
 struct cxl_nvdimm *to_cxl_nvdimm(struct device *dev);
 bool is_cxl_nvdimm(struct device *dev);
 bool is_cxl_nvdimm_bridge(struct device *dev);
-int devm_cxl_add_nvdimm(struct device *host, struct cxl_memdev *cxlmd);
+int devm_cxl_add_nvdimm(struct cxl_memdev *cxlmd);
 struct cxl_nvdimm_bridge *cxl_find_nvdimm_bridge(struct device *dev);
 
 #ifdef CONFIG_CXL_REGION
diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
index 88e3a8e54b6a..c1c9960ab05f 100644
--- a/drivers/cxl/cxlmem.h
+++ b/drivers/cxl/cxlmem.h
@@ -35,6 +35,8 @@
  * @cdev: char dev core object for ioctl operations
  * @cxlds: The device state backing this device
  * @detach_work: active memdev lost a port in its ancestry
+ * @cxl_nvb: coordinate removal of @cxl_nvd if present
+ * @cxl_nvd: optional bridge to an nvdimm if the device supports pmem
  * @id: id number of this memdev instance.
  */
 struct cxl_memdev {
@@ -42,6 +44,8 @@ struct cxl_memdev {
 	struct cdev cdev;
 	struct cxl_dev_state *cxlds;
 	struct work_struct detach_work;
+	struct cxl_nvdimm_bridge *cxl_nvb;
+	struct cxl_nvdimm *cxl_nvd;
 	int id;
 };
 
diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
index 64ccf053d32c..549b6b499bae 100644
--- a/drivers/cxl/mem.c
+++ b/drivers/cxl/mem.c
@@ -48,6 +48,7 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
 static int cxl_mem_probe(struct device *dev)
 {
 	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
+	struct cxl_dev_state *cxlds = cxlmd->cxlds;
 	struct cxl_port *parent_port;
 	struct cxl_dport *dport;
 	struct dentry *dentry;
@@ -95,6 +96,14 @@ static int cxl_mem_probe(struct device *dev)
 	if (rc)
 		return rc;
 
+	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM)) {
+		rc = devm_cxl_add_nvdimm(cxlmd);
+		if (rc == -ENODEV)
+			dev_info(dev, "PMEM disabled by platform\n");
+		else
+			return rc;
+	}
+
 	/*
 	 * The kernel may be operating out of CXL memory on this device,
 	 * there is no spec defined way to determine whether this device
diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
index 621a0522b554..e15da405b948 100644
--- a/drivers/cxl/pci.c
+++ b/drivers/cxl/pci.c
@@ -503,9 +503,6 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (IS_ERR(cxlmd))
 		return PTR_ERR(cxlmd);
 
-	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM))
-		rc = devm_cxl_add_nvdimm(&pdev->dev, cxlmd);
-
 	return rc;
 }
 
diff --git a/drivers/cxl/pmem.c b/drivers/cxl/pmem.c
index 652f00fc68ca..76cf54eeb310 100644
--- a/drivers/cxl/pmem.c
+++ b/drivers/cxl/pmem.c
@@ -34,26 +34,16 @@ static int cxl_nvdimm_probe(struct device *dev)
 {
 	struct cxl_nvdimm *cxl_nvd = to_cxl_nvdimm(dev);
 	struct cxl_memdev *cxlmd = cxl_nvd->cxlmd;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlmd->cxl_nvb;
 	unsigned long flags = 0, cmd_mask = 0;
 	struct cxl_dev_state *cxlds = cxlmd->cxlds;
-	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct nvdimm *nvdimm;
 	int rc;
 
-	cxl_nvb = cxl_find_nvdimm_bridge(dev);
-	if (!cxl_nvb)
-		return -ENXIO;
-
-	device_lock(&cxl_nvb->dev);
-	if (!cxl_nvb->nvdimm_bus) {
-		rc = -ENXIO;
-		goto out;
-	}
-
 	set_exclusive_cxl_commands(cxlds, exclusive_cmds);
 	rc = devm_add_action_or_reset(dev, clear_exclusive, cxlds);
 	if (rc)
-		goto out;
+		return rc;
 
 	set_bit(NDD_LABELING, &flags);
 	set_bit(ND_CMD_GET_CONFIG_SIZE, &cmd_mask);
@@ -61,19 +51,11 @@ static int cxl_nvdimm_probe(struct device *dev)
 	set_bit(ND_CMD_SET_CONFIG_DATA, &cmd_mask);
 	nvdimm = nvdimm_create(cxl_nvb->nvdimm_bus, cxl_nvd, NULL, flags,
 			       cmd_mask, 0, NULL);
-	if (!nvdimm) {
-		rc = -ENOMEM;
-		goto out;
-	}
+	if (!nvdimm)
+		return -ENOMEM;
 
 	dev_set_drvdata(dev, nvdimm);
-	cxl_nvd->bridge = cxl_nvb;
-	rc = devm_add_action_or_reset(dev, unregister_nvdimm, nvdimm);
-out:
-	device_unlock(&cxl_nvb->dev);
-	put_device(&cxl_nvb->dev);
-
-	return rc;
+	return devm_add_action_or_reset(dev, unregister_nvdimm, nvdimm);
 }
 
 static struct cxl_driver cxl_nvdimm_driver = {
@@ -200,6 +182,15 @@ static int cxl_pmem_ctl(struct nvdimm_bus_descriptor *nd_desc,
 	return cxl_pmem_nvdimm_ctl(nvdimm, cmd, buf, buf_len);
 }
 
+static void unregister_nvdimm_bus(void *_cxl_nvb)
+{
+	struct cxl_nvdimm_bridge *cxl_nvb = _cxl_nvb;
+	struct nvdimm_bus *nvdimm_bus = cxl_nvb->nvdimm_bus;
+
+	cxl_nvb->nvdimm_bus = NULL;
+	nvdimm_bus_unregister(nvdimm_bus);
+}
+
 static bool online_nvdimm_bus(struct cxl_nvdimm_bridge *cxl_nvb)
 {
 	if (cxl_nvb->nvdimm_bus)
@@ -303,23 +294,21 @@ static int cxl_nvdimm_bridge_probe(struct device *dev)
 {
 	struct cxl_nvdimm_bridge *cxl_nvb = to_cxl_nvdimm_bridge(dev);
 
-	if (cxl_nvb->state == CXL_NVB_DEAD)
-		return -ENXIO;
+	cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor) {
+		.provider_name = "CXL",
+		.module = THIS_MODULE,
+		.ndctl = cxl_pmem_ctl,
+	};
 
-	if (cxl_nvb->state == CXL_NVB_NEW) {
-		cxl_nvb->nd_desc = (struct nvdimm_bus_descriptor) {
-			.provider_name = "CXL",
-			.module = THIS_MODULE,
-			.ndctl = cxl_pmem_ctl,
-		};
+	cxl_nvb->nvdimm_bus =
+		nvdimm_bus_register(&cxl_nvb->dev, &cxl_nvb->nd_desc);
 
-		INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
-	}
+	if (!cxl_nvb->nvdimm_bus)
+		return -ENOMEM;
 
-	cxl_nvb->state = CXL_NVB_ONLINE;
-	cxl_nvdimm_bridge_state_work(cxl_nvb);
+	INIT_WORK(&cxl_nvb->state_work, cxl_nvb_update_state);
 
-	return 0;
+	return devm_add_action_or_reset(dev, unregister_nvdimm_bus, cxl_nvb);
 }
 
 static struct cxl_driver cxl_nvdimm_bridge_driver = {
@@ -332,11 +321,6 @@ static struct cxl_driver cxl_nvdimm_bridge_driver = {
 	},
 };
 
-static int match_cxl_nvdimm(struct device *dev, void *data)
-{
-	return is_cxl_nvdimm(dev);
-}
-
 static void unregister_nvdimm_region(void *nd_region)
 {
 	nvdimm_region_delete(nd_region);
@@ -357,8 +341,8 @@ static int cxl_pmem_region_probe(struct device *dev)
 	struct nd_mapping_desc mappings[CXL_DECODER_MAX_INTERLEAVE];
 	struct cxl_pmem_region *cxlr_pmem = to_cxl_pmem_region(dev);
 	struct cxl_region *cxlr = cxlr_pmem->cxlr;
+	struct cxl_nvdimm_bridge *cxl_nvb = cxlr->cxl_nvb;
 	struct cxl_pmem_region_info *info = NULL;
-	struct cxl_nvdimm_bridge *cxl_nvb;
 	struct nd_interleave_set *nd_set;
 	struct nd_region_desc ndr_desc;
 	struct cxl_nvdimm *cxl_nvd;
@@ -366,28 +350,12 @@ static int cxl_pmem_region_probe(struct device *dev)
 	struct resource *res;
 	int rc, i = 0;
 
-	cxl_nvb = cxl_find_nvdimm_bridge(&cxlr_pmem->mapping[0].cxlmd->dev);
-	if (!cxl_nvb) {
-		dev_dbg(dev, "bridge not found\n");
-		return -ENXIO;
-	}
-	cxlr_pmem->bridge = cxl_nvb;
-
-	device_lock(&cxl_nvb->dev);
-	if (!cxl_nvb->nvdimm_bus) {
-		dev_dbg(dev, "nvdimm bus not found\n");
-		rc = -ENXIO;
-		goto out_nvb;
-	}
-
 	memset(&mappings, 0, sizeof(mappings));
 	memset(&ndr_desc, 0, sizeof(ndr_desc));
 
 	res = devm_kzalloc(dev, sizeof(*res), GFP_KERNEL);
-	if (!res) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!res)
+		return -ENOMEM;
 
 	res->name = "Persistent Memory";
 	res->start = cxlr_pmem->hpa_range.start;
@@ -397,11 +365,11 @@ static int cxl_pmem_region_probe(struct device *dev)
 
 	rc = insert_resource(&iomem_resource, res);
 	if (rc)
-		goto out_nvb;
+		return rc;
 
 	rc = devm_add_action_or_reset(dev, cxlr_pmem_remove_resource, res);
 	if (rc)
-		goto out_nvb;
+		return rc;
 
 	ndr_desc.res = res;
 	ndr_desc.provider_data = cxlr_pmem;
@@ -415,39 +383,23 @@ static int cxl_pmem_region_probe(struct device *dev)
 	}
 
 	nd_set = devm_kzalloc(dev, sizeof(*nd_set), GFP_KERNEL);
-	if (!nd_set) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!nd_set)
+		return -ENOMEM;
 
 	ndr_desc.memregion = cxlr->id;
 	set_bit(ND_REGION_CXL, &ndr_desc.flags);
 	set_bit(ND_REGION_PERSIST_MEMCTRL, &ndr_desc.flags);
 
 	info = kmalloc_array(cxlr_pmem->nr_mappings, sizeof(*info), GFP_KERNEL);
-	if (!info) {
-		rc = -ENOMEM;
-		goto out_nvb;
-	}
+	if (!info)
+		return -ENOMEM;
 
 	for (i = 0; i < cxlr_pmem->nr_mappings; i++) {
 		struct cxl_pmem_region_mapping *m = &cxlr_pmem->mapping[i];
 		struct cxl_memdev *cxlmd = m->cxlmd;
 		struct cxl_dev_state *cxlds = cxlmd->cxlds;
-		struct device *d;
 
-		d = device_find_child(&cxlmd->dev, NULL, match_cxl_nvdimm);
-		if (!d) {
-			dev_dbg(dev, "[%d]: %s: no cxl_nvdimm found\n", i,
-				dev_name(&cxlmd->dev));
-			rc = -ENODEV;
-			goto out_nvd;
-		}
-
-		/* safe to drop ref now with bridge lock held */
-		put_device(d);
-
-		cxl_nvd = to_cxl_nvdimm(d);
+		cxl_nvd = cxlmd->cxl_nvd;
 		nvdimm = dev_get_drvdata(&cxl_nvd->dev);
 		if (!nvdimm) {
 			dev_dbg(dev, "[%d]: %s: no nvdimm found\n", i,
@@ -488,9 +440,6 @@ static int cxl_pmem_region_probe(struct device *dev)
 				      cxlr_pmem->nd_region);
 out_nvd:
 	kfree(info);
-out_nvb:
-	device_unlock(&cxl_nvb->dev);
-	put_device(&cxl_nvb->dev);
 
 	return rc;
 }
diff --git a/tools/testing/cxl/test/mem.c b/tools/testing/cxl/test/mem.c
index aa2df3a15051..a4ee8e61dd60 100644
--- a/tools/testing/cxl/test/mem.c
+++ b/tools/testing/cxl/test/mem.c
@@ -285,9 +285,6 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
 	if (IS_ERR(cxlmd))
 		return PTR_ERR(cxlmd);
 
-	if (resource_size(&cxlds->pmem_res) && IS_ENABLED(CONFIG_CXL_PMEM))
-		rc = devm_cxl_add_nvdimm(dev, cxlmd);
-
 	return 0;
 }
 

Re: [PATCH v4 06/12] tools/testing/cxl: Make mock CEDT parsing more robust

[Robert Richter]

On 24.11.22 10:35:10, Dan Williams wrote:
> Accept any cxl_test topology device as the first argument in
> cxl_chbs_context. This is in preparation for reworking the detection of
> the component registers across VH and RCH topologies. Move
> mock_acpi_table_parse_cedt() beneath the definition of is_mock_port()
> and use is_mock_port() instead of the explicit mock cxl_acpi device
> check.
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Reviewed-by: Robert Richter <rrichter@amd.com>

> ---
>  tools/testing/cxl/test/cxl.c |   80 +++++++++++++++++++++---------------------
>  1 file changed, 40 insertions(+), 40 deletions(-)

Re: [PATCH v4 07/12] cxl/ACPI: Register CXL host ports by bridge device

[Robert Richter]

On 24.11.22 10:35:16, Dan Williams wrote:
> From: Robert Richter <rrichter@amd.com>
> 
> A port of a CXL host bridge links to the bridge's ACPI device
> (&adev->dev) with its corresponding uport/dport device (uport_dev and
> dport_dev respectively). The device is not a direct parent device in
> the PCI topology as pdev->dev.parent points to a PCI bridge's (struct
> pci_host_bridge) device. The following CXL memory device hierarchy
> would be valid for an endpoint once an RCD EP would be enabled (note
> this will be done in a later patch):
> 
> VH mode:
> 
>  cxlmd->dev.parent->parent
>         ^^^\^^^^^^\ ^^^^^^\
>             \      \       pci_dev (Type 1, Downstream Port)
>              \      pci_dev (Type 0, PCI Express Endpoint)
>               cxl mem device
> 
> RCD mode:
> 
>  cxlmd->dev.parent->parent
>         ^^^\^^^^^^\ ^^^^^^\
>             \      \       pci_host_bridge
>              \      pci_dev (Type 0, RCiEP)
>               cxl mem device
> 
> In VH mode a downstream port is created by port enumeration and thus
> always exists.
> 
> Now, in RCD mode the host bridge also already exists but it references
> to an ACPI device. A port lookup by the PCI device's parent device
> will fail as a direct link to the registered port is missing. The ACPI
> device of the bridge must be determined first.
> 
> To prevent this, change port registration of a CXL host to use the
> bridge device instead. Do this also for the VH case as port topology
> will better reflect the PCI topology then.
> 
> Signed-off-by: Robert Richter <rrichter@amd.com>
> [djbw: rebase on brige mocking]
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Patch looks unchanged compared to v3.

Reviewed-by: Robert Richter <rrichter@amd.com>

> ---
>  drivers/cxl/acpi.c |   35 +++++++++++++++++++----------------
>  1 file changed, 19 insertions(+), 16 deletions(-)

Re: [PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Robert Richter]

On 24.11.22 10:35:21, Dan Williams wrote:
> From: Robert Richter <rrichter@amd.com>
> 
> A downstream port must be connected to a component register block.
> For restricted hosts the base address is determined from the RCRB. The
> RCRB is provided by the host's CEDT CHBS entry. Rework CEDT parser to
> get the RCRB and add code to extract the component register block from
> it.
> 
> RCRB's BAR[0..1] point to the component block containing CXL subsystem
> component registers. MEMBAR extraction follows the PCI base spec here,
> esp. 64 bit extraction and memory range alignment (6.0, 7.5.1.2.1). The
> RCRB base address is cached in the cxl_dport per-host bridge so that the
> upstream port component registers can be retrieved later by an RCD
> (RCIEP) associated with the host bridge.
> 
> Note: Right now the component register block is used for HDM decoder
> capability only which is optional for RCDs. If unsupported by the RCD,
> the HDM init will fail. It is future work to bypass it in this case.
> 
> Co-developed-by: Terry Bowman <terry.bowman@amd.com>
> Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> Signed-off-by: Robert Richter <rrichter@amd.com>
> [djbw: introduce devm_cxl_add_rch_dport()]
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  drivers/cxl/acpi.c            |   54 ++++++++++++++++++++++++++++++++--------
>  drivers/cxl/core/port.c       |   42 +++++++++++++++++++++++++++----
>  drivers/cxl/core/regs.c       |   56 +++++++++++++++++++++++++++++++++++++++++
>  drivers/cxl/cxl.h             |   16 ++++++++++++
>  tools/testing/cxl/Kbuild      |    1 +
>  tools/testing/cxl/test/cxl.c  |   10 +++++++
>  tools/testing/cxl/test/mock.c |   19 ++++++++++++++
>  tools/testing/cxl/test/mock.h |    3 ++
>  8 files changed, 186 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> index 50d82376097c..1224add13529 100644
> --- a/drivers/cxl/acpi.c
> +++ b/drivers/cxl/acpi.c
> @@ -9,6 +9,8 @@
>  #include "cxlpci.h"
>  #include "cxl.h"
>  
> +#define CXL_RCRB_SIZE	SZ_8K
> +
>  static unsigned long cfmws_to_decoder_flags(int restrictions)
>  {
>  	unsigned long flags = CXL_DECODER_F_ENABLE;
> @@ -215,6 +217,11 @@ static int add_host_bridge_uport(struct device *match, void *arg)
>  	if (rc)
>  		return rc;
>  
> +	if (dport->rch) {
> +		dev_info(bridge, "host supports CXL (restricted)\n");
> +		return 0;
> +	}

This change comes after devm_cxl_register_pci_bus() to serve the
cxl_port_to_pci_bus() in devm_cxl_port_enumerate_dports() in
cxl_port_probe(). A root port is not probed and
devm_cxl_port_enumerate_dports() will be never called, so we could
jump out before devm_cxl_register_pci_bus().

On the other side we might want to be ready to use
cxl_port_to_pci_bus() elsewhere in later changes. RCHs would not work
then.

> +
>  	port = devm_cxl_add_port(host, bridge, dport->component_reg_phys,
>  				 dport);
>  	if (IS_ERR(port))
> @@ -228,27 +235,46 @@ static int add_host_bridge_uport(struct device *match, void *arg)
>  struct cxl_chbs_context {
>  	struct device *dev;
>  	unsigned long long uid;
> -	resource_size_t chbcr;
> +	struct acpi_cedt_chbs chbs;
>  };
>  
> -static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
> -			 const unsigned long end)
> +static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
> +			const unsigned long end)
>  {
>  	struct cxl_chbs_context *ctx = arg;
>  	struct acpi_cedt_chbs *chbs;
>  
> -	if (ctx->chbcr)
> +	if (ctx->chbs.base)
>  		return 0;
>  
>  	chbs = (struct acpi_cedt_chbs *) header;
>  
>  	if (ctx->uid != chbs->uid)
>  		return 0;
> -	ctx->chbcr = chbs->base;
> +	ctx->chbs = *chbs;
>  
>  	return 0;
>  }
>  
> +static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
> +{
> +	struct acpi_cedt_chbs *chbs = &ctx->chbs;
> +
> +	if (!chbs->base)
> +		return CXL_RESOURCE_NONE;
> +
> +	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
> +		return chbs->base;
> +
> +	if (chbs->length != CXL_RCRB_SIZE)
> +		return CXL_RESOURCE_NONE;
> +
> +	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
> +		&chbs->base);
> +
> +	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
> +}
> +

I have an improved version of this code which squashes cxl_get_chbcr()
into cxl_get_chbs() (basically extends the original cxl_get_chbcr()
function).

>  static int add_host_bridge_dport(struct device *match, void *arg)
>  {
>  	acpi_status status;
> @@ -258,6 +284,7 @@ static int add_host_bridge_dport(struct device *match, void *arg)
>  	struct cxl_chbs_context ctx;
>  	struct acpi_pci_root *pci_root;
>  	struct cxl_port *root_port = arg;
> +	resource_size_t component_reg_phys;
>  	struct device *host = root_port->dev.parent;
>  	struct acpi_device *hb = to_cxl_host_bridge(host, match);
>  
> @@ -274,21 +301,28 @@ static int add_host_bridge_dport(struct device *match, void *arg)
>  	dev_dbg(match, "UID found: %lld\n", uid);
>  
>  	ctx = (struct cxl_chbs_context) {
> -		.dev = host,
> +		.dev = match,
>  		.uid = uid,
>  	};
> -	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbcr, &ctx);
> +	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbs, &ctx);
>  
> -	if (ctx.chbcr == 0) {
> +	component_reg_phys = cxl_get_chbcr(&ctx);
> +	if (component_reg_phys == CXL_RESOURCE_NONE) {
>  		dev_warn(match, "No CHBS found for Host Bridge (UID %lld)\n", uid);
>  		return 0;
>  	}
>  
> -	dev_dbg(match, "CHBCR found: 0x%08llx\n", (u64)ctx.chbcr);
> +	dev_dbg(match, "CHBCR found: %pa\n", &component_reg_phys);
>  
>  	pci_root = acpi_pci_find_root(hb->handle);
>  	bridge = pci_root->bus->bridge;
> -	dport = devm_cxl_add_dport(root_port, bridge, uid, ctx.chbcr);
> +	if (ctx.chbs.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
> +		dport = devm_cxl_add_rch_dport(root_port, bridge, uid,
> +					       component_reg_phys,
> +					       ctx.chbs.base);

Yes, this new function makes the rcrb handling much more simpler.

> +	else
> +		dport = devm_cxl_add_dport(root_port, bridge, uid,
> +					   component_reg_phys);
>  	if (IS_ERR(dport))
>  		return PTR_ERR(dport);
>  
> diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> index d225267c69bb..d9fe06e1462f 100644
> --- a/drivers/cxl/core/port.c
> +++ b/drivers/cxl/core/port.c
> @@ -628,6 +628,8 @@ static struct cxl_port *cxl_port_alloc(struct device *uport,
>  			iter = to_cxl_port(iter->dev.parent);
>  		if (iter->host_bridge)
>  			port->host_bridge = iter->host_bridge;
> +		else if (parent_dport->rch)
> +			port->host_bridge = parent_dport->dport;

Yes, looks good. This makes the endpoint a child of a CXL root port,
not the ACPI0017 the root device.

>  		else
>  			port->host_bridge = iter->uport;
>  		dev_dbg(uport, "host-bridge: %s\n", dev_name(port->host_bridge));
> @@ -899,10 +901,15 @@ static void cxl_dport_unlink(void *data)
>  	sysfs_remove_link(&port->dev.kobj, link_name);
>  }
>  
> -static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
> -					      struct device *dport_dev,
> -					      int port_id,
> -					      resource_size_t component_reg_phys)
> +enum cxl_dport_mode {
> +	CXL_DPORT_VH,
> +	CXL_DPORT_RCH,
> +};
> +
> +static struct cxl_dport *
> +__devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
> +		     int port_id, resource_size_t component_reg_phys,
> +		     enum cxl_dport_mode mode, resource_size_t rcrb)
>  {
>  	char link_name[CXL_TARGET_STRLEN];
>  	struct cxl_dport *dport;
> @@ -932,6 +939,9 @@ static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
>  	dport->port_id = port_id;
>  	dport->component_reg_phys = component_reg_phys;
>  	dport->port = port;
> +	if (mode == CXL_DPORT_RCH)
> +		dport->rch = true;

Alternatively an inline function could be added which checks
dport->rcrb for a valid address.

> +	dport->rcrb = rcrb;
>  
>  	cond_cxl_root_lock(port);
>  	rc = add_dport(port, dport);
> @@ -973,7 +983,8 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
>  	struct cxl_dport *dport;
>  
>  	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> -				     component_reg_phys);
> +				     component_reg_phys, CXL_DPORT_VH,
> +				     CXL_RESOURCE_NONE);
>  	if (IS_ERR(dport)) {
>  		dev_dbg(dport_dev, "failed to add dport to %s: %ld\n",
>  			dev_name(&port->dev), PTR_ERR(dport));
> @@ -986,6 +997,27 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
>  }
>  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
>  
> +struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
> +					 struct device *dport_dev, int port_id,
> +					 resource_size_t component_reg_phys,
> +					 resource_size_t rcrb)

The documentation header is missing for that.

> +{
> +	struct cxl_dport *dport;
> +
> +	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> +				     component_reg_phys, CXL_DPORT_RCH, rcrb);
> +	if (IS_ERR(dport)) {
> +		dev_dbg(dport_dev, "failed to add RCH dport to %s: %ld\n",
> +			dev_name(&port->dev), PTR_ERR(dport));
> +	} else {
> +		dev_dbg(dport_dev, "RCH dport added to %s\n",
> +			dev_name(&port->dev));
> +	}
> +
> +	return dport;
> +}
> +EXPORT_SYMBOL_NS_GPL(devm_cxl_add_rch_dport, CXL);
> +
>  static int add_ep(struct cxl_ep *new)
>  {
>  	struct cxl_port *port = new->dport->port;
> diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c
> index ec178e69b18f..7c2a85dc4125 100644
> --- a/drivers/cxl/core/regs.c
> +++ b/drivers/cxl/core/regs.c
> @@ -307,3 +307,59 @@ int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
>  	return -ENODEV;
>  }
>  EXPORT_SYMBOL_NS_GPL(cxl_find_regblock, CXL);
> +
> +resource_size_t cxl_rcrb_to_component(struct device *dev,
> +				      resource_size_t rcrb,
> +				      enum cxl_rcrb which)
> +{
> +	resource_size_t component_reg_phys;
> +	u32 bar0, bar1;
> +	void *addr;
> +	u16 cmd;
> +
> +	if (which == CXL_RCRB_UPSTREAM)
> +		rcrb += SZ_4K;
> +
> +	/*
> +	 * RCRB's BAR[0..1] point to component block containing CXL
> +	 * subsystem component registers. MEMBAR extraction follows
> +	 * the PCI Base spec here, esp. 64 bit extraction and memory
> +	 * ranges alignment (6.0, 7.5.1.2.1).
> +	 */
> +	if (!request_mem_region(rcrb, SZ_4K, "CXL RCRB"))
> +		return CXL_RESOURCE_NONE;
> +	addr = ioremap(rcrb, SZ_4K);
> +	if (!addr) {
> +		dev_err(dev, "Failed to map region %pr\n", addr);
> +		release_mem_region(rcrb, SZ_4K);
> +		return CXL_RESOURCE_NONE;
> +	}
> +
> +	cmd = readw(addr + PCI_COMMAND);
> +	bar0 = readl(addr + PCI_BASE_ADDRESS_0);
> +	bar1 = readl(addr + PCI_BASE_ADDRESS_1);
> +	iounmap(addr);
> +	release_mem_region(rcrb, SZ_4K);
> +
> +	/* sanity check */
> +	if (cmd == 0xffff)
> +		return CXL_RESOURCE_NONE;

The spec says offset 0 should be checked (32 bit) which is always
non-FF if implemented. This requires another read.

cmd is most of the cases also non-zero, so probably checking cmd
instead will have the same effect. Still worth changing that.

If the downstream port's rcrb is all FFs, it is a FW bug. Could be
worth a message.

> +	if ((cmd & PCI_COMMAND_MEMORY) == 0)

I like the following more, but that's a flavor:

	if (!(cmd & PCI_COMMAND_MEMORY))

> +		return CXL_RESOURCE_NONE;
> +	if (bar0 & (PCI_BASE_ADDRESS_MEM_TYPE_1M | PCI_BASE_ADDRESS_SPACE_IO))
> +		return CXL_RESOURCE_NONE;
> +
> +	component_reg_phys = bar0 & PCI_BASE_ADDRESS_MEM_MASK;
> +	if (bar0 & PCI_BASE_ADDRESS_MEM_TYPE_64)
> +		component_reg_phys |= ((u64)bar1) << 32;
> +
> +	if (!component_reg_phys)
> +		return CXL_RESOURCE_NONE;
> +
> +	/* MEMBAR is block size (64k) aligned. */
> +	if (!IS_ALIGNED(component_reg_phys, CXL_COMPONENT_REG_BLOCK_SIZE))
> +		return CXL_RESOURCE_NONE;
> +
> +	return component_reg_phys;
> +}
> +EXPORT_SYMBOL_NS_GPL(cxl_rcrb_to_component, CXL);
> diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> index 9b33ae4b2aec..43c43d1ec069 100644
> --- a/drivers/cxl/cxl.h
> +++ b/drivers/cxl/cxl.h
> @@ -223,6 +223,14 @@ enum cxl_regloc_type;
>  int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
>  		      struct cxl_register_map *map);
>  
> +enum cxl_rcrb {
> +	CXL_RCRB_DOWNSTREAM,
> +	CXL_RCRB_UPSTREAM,
> +};
> +resource_size_t cxl_rcrb_to_component(struct device *dev,
> +				      resource_size_t rcrb,
> +				      enum cxl_rcrb which);
> +
>  #define CXL_RESOURCE_NONE ((resource_size_t) -1)
>  #define CXL_TARGET_STRLEN 20
>  
> @@ -486,12 +494,16 @@ cxl_find_dport_by_dev(struct cxl_port *port, const struct device *dport_dev)
>   * @dport: PCI bridge or firmware device representing the downstream link
>   * @port_id: unique hardware identifier for dport in decoder target list
>   * @component_reg_phys: downstream port component registers
> + * @rcrb: base address for the Root Complex Register Block
> + * @rch: Indicate whether this dport was enumerated in RCH or VH mode
>   * @port: reference to cxl_port that contains this downstream port
>   */
>  struct cxl_dport {
>  	struct device *dport;
>  	int port_id;
>  	resource_size_t component_reg_phys;
> +	resource_size_t rcrb;
> +	bool rch;
>  	struct cxl_port *port;
>  };
>  
> @@ -561,6 +573,10 @@ bool schedule_cxl_memdev_detach(struct cxl_memdev *cxlmd);
>  struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
>  				     struct device *dport, int port_id,
>  				     resource_size_t component_reg_phys);
> +struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
> +					 struct device *dport_dev, int port_id,
> +					 resource_size_t component_reg_phys,
> +					 resource_size_t rcrb);
>  
>  struct cxl_decoder *to_cxl_decoder(struct device *dev);
>  struct cxl_root_decoder *to_cxl_root_decoder(struct device *dev);
> diff --git a/tools/testing/cxl/Kbuild b/tools/testing/cxl/Kbuild
> index 500be85729cc..9e4d94e81723 100644
> --- a/tools/testing/cxl/Kbuild
> +++ b/tools/testing/cxl/Kbuild
> @@ -10,6 +10,7 @@ ldflags-y += --wrap=devm_cxl_add_passthrough_decoder
>  ldflags-y += --wrap=devm_cxl_enumerate_decoders
>  ldflags-y += --wrap=cxl_await_media_ready
>  ldflags-y += --wrap=cxl_hdm_decode_init
> +ldflags-y += --wrap=cxl_rcrb_to_component
>  
>  DRIVERS := ../../../drivers
>  CXL_SRC := $(DRIVERS)/cxl
> diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> index 42a34650dd2f..1823c61d7ba3 100644
> --- a/tools/testing/cxl/test/cxl.c
> +++ b/tools/testing/cxl/test/cxl.c
> @@ -694,6 +694,15 @@ static int mock_cxl_port_enumerate_dports(struct cxl_port *port)
>  	return 0;
>  }
>  
> +resource_size_t mock_cxl_rcrb_to_component(struct device *dev,
> +					   resource_size_t rcrb,
> +					   enum cxl_rcrb which)
> +{
> +	dev_dbg(dev, "rcrb: %pa which: %d\n", &rcrb, which);
> +
> +	return 0;
> +}
> +
>  static struct cxl_mock_ops cxl_mock_ops = {
>  	.is_mock_adev = is_mock_adev,
>  	.is_mock_bridge = is_mock_bridge,
> @@ -702,6 +711,7 @@ static struct cxl_mock_ops cxl_mock_ops = {
>  	.is_mock_dev = is_mock_dev,
>  	.acpi_table_parse_cedt = mock_acpi_table_parse_cedt,
>  	.acpi_evaluate_integer = mock_acpi_evaluate_integer,
> +	.cxl_rcrb_to_component = mock_cxl_rcrb_to_component,
>  	.acpi_pci_find_root = mock_acpi_pci_find_root,
>  	.devm_cxl_port_enumerate_dports = mock_cxl_port_enumerate_dports,
>  	.devm_cxl_setup_hdm = mock_cxl_setup_hdm,
> diff --git a/tools/testing/cxl/test/mock.c b/tools/testing/cxl/test/mock.c
> index bce6a21df0d5..5dface08e0de 100644
> --- a/tools/testing/cxl/test/mock.c
> +++ b/tools/testing/cxl/test/mock.c
> @@ -224,6 +224,25 @@ int __wrap_cxl_hdm_decode_init(struct cxl_dev_state *cxlds,
>  }
>  EXPORT_SYMBOL_NS_GPL(__wrap_cxl_hdm_decode_init, CXL);
>  
> +resource_size_t __wrap_cxl_rcrb_to_component(struct device *dev,
> +					     resource_size_t rcrb,
> +					     enum cxl_rcrb which)
> +{
> +	int index;
> +	resource_size_t component_reg_phys;
> +	struct cxl_mock_ops *ops = get_cxl_mock_ops(&index);
> +
> +	if (ops && ops->is_mock_port(dev))
> +		component_reg_phys =
> +			ops->cxl_rcrb_to_component(dev, rcrb, which);
> +	else
> +		component_reg_phys = cxl_rcrb_to_component(dev, rcrb, which);
> +	put_cxl_mock_ops(index);
> +
> +	return component_reg_phys;
> +}
> +EXPORT_SYMBOL_NS_GPL(__wrap_cxl_rcrb_to_component, CXL);
> +
>  MODULE_LICENSE("GPL v2");
>  MODULE_IMPORT_NS(ACPI);
>  MODULE_IMPORT_NS(CXL);
> diff --git a/tools/testing/cxl/test/mock.h b/tools/testing/cxl/test/mock.h
> index 738f24e3988a..ef33f159375e 100644
> --- a/tools/testing/cxl/test/mock.h
> +++ b/tools/testing/cxl/test/mock.h
> @@ -15,6 +15,9 @@ struct cxl_mock_ops {
>  					     acpi_string pathname,
>  					     struct acpi_object_list *arguments,
>  					     unsigned long long *data);
> +	resource_size_t (*cxl_rcrb_to_component)(struct device *dev,
> +						 resource_size_t rcrb,
> +						 enum cxl_rcrb which);
>  	struct acpi_pci_root *(*acpi_pci_find_root)(acpi_handle handle);
>  	bool (*is_mock_bus)(struct pci_bus *bus);
>  	bool (*is_mock_port)(struct device *dev);
> 

The mock part looks good to me.

-Robert

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dave Jiang]

On 11/24/2022 11:35 AM, Dan Williams wrote:
> Unlike a CXL memory expander in a VH topology that has at least one
> intervening 'struct cxl_port' instance between itself and the CXL root
> device, an RCD attaches one-level higher. For example:
> 
>                 VH
>            ┌──────────┐
>            │ ACPI0017 │
>            │  root0   │
>            └─────┬────┘
>                  │
>            ┌─────┴────┐
>            │  dport0  │
>      ┌─────┤ ACPI0016 ├─────┐
>      │     │  port1   │     │
>      │     └────┬─────┘     │
>      │          │           │
>   ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
>   │dport0│   │dport1│   │dport2│
>   │ RP0  │   │ RP1  │   │ RP2  │
>   └──────┘   └──┬───┘   └──────┘
>                 │
>             ┌───┴─────┐
>             │endpoint0│
>             │  port2  │
>             └─────────┘
> 
> ...vs:
> 
>                RCH
>            ┌──────────┐
>            │ ACPI0017 │
>            │  root0   │
>            └────┬─────┘
>                 │
>             ┌───┴────┐
>             │ dport0 │
>             │ACPI0016│
>             └───┬────┘
>                 │
>            ┌────┴─────┐
>            │endpoint0 │
>            │  port1   │
>            └──────────┘
> 
> So arrange for endpoint port in the RCH/RCD case to appear directly
> connected to the host-bridge in its singular role as a dport. Compare
> that to the VH case where the host-bridge serves a dual role as a
> 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> instances in the CXL topology.
> 
> Another deviation from the VH case is that RCDs may need to look up
> their component registers from the Root Complex Register Block (RCRB).
> That platform firmware specified RCRB area is cached by the cxl_acpi
> driver and conveyed via the host-bridge dport to the cxl_mem driver to
> perform the cxl_rcrb_to_component() lookup for the endpoint port
> (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> upstream port component registers).
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>   drivers/cxl/core/port.c |   11 +++++++++--
>   drivers/cxl/cxlmem.h    |    2 ++
>   drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
>   drivers/cxl/pci.c       |   10 ++++++++++
>   4 files changed, 45 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> index c7f58282b2c1..2385ee00eb9a 100644
> --- a/drivers/cxl/core/port.c
> +++ b/drivers/cxl/core/port.c
> @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
>   {
>   	struct device *dev = &cxlmd->dev;
>   	struct device *iter;
> +	struct cxl_dport *dport;
> +	struct cxl_port *port;
>   	int rc;
>   
> +	/*
> +	 * Skip intermediate port enumeration in the RCH case, there
> +	 * are no ports in between a host bridge and an endpoint.
> +	 */
> +	if (cxlmd->cxlds->rcd)
> +		return 0;
> +
>   	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
>   	if (rc)
>   		return rc;
> @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
>   	for (iter = dev; iter; iter = grandparent(iter)) {
>   		struct device *dport_dev = grandparent(iter);
>   		struct device *uport_dev;
> -		struct cxl_dport *dport;
> -		struct cxl_port *port;
>   
>   		if (!dport_dev)
>   			return 0;
> diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> index e082991bc58c..35d485d041f0 100644
> --- a/drivers/cxl/cxlmem.h
> +++ b/drivers/cxl/cxlmem.h
> @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
>    * @dev: The device associated with this CXL state
>    * @regs: Parsed register blocks
>    * @cxl_dvsec: Offset to the PCIe device DVSEC
> + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
>    * @payload_size: Size of space for payload
>    *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
>    * @lsa_size: Size of Label Storage Area
> @@ -235,6 +236,7 @@ struct cxl_dev_state {
>   	struct cxl_regs regs;
>   	int cxl_dvsec;
>   
> +	bool rcd;
>   	size_t payload_size;
>   	size_t lsa_size;
>   	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> index aa63ce8c7ca6..9a655b4b5e52 100644
> --- a/drivers/cxl/mem.c
> +++ b/drivers/cxl/mem.c
> @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
>   	return 0;
>   }
>   
> -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
>   				 struct cxl_dport *parent_dport)
>   {
>   	struct cxl_port *parent_port = parent_dport->port;
>   	struct cxl_dev_state *cxlds = cxlmd->cxlds;
>   	struct cxl_port *endpoint, *iter, *down;
> +	resource_size_t component_reg_phys;
>   	int rc;
>   
>   	/*
> @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
>   		ep->next = down;
>   	}
>   
> -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> -				     cxlds->component_reg_phys, parent_dport);
> +	/*
> +	 * The component registers for an RCD might come from the
> +	 * host-bridge RCRB if they are not already mapped via the
> +	 * typical register locator mechanism.
> +	 */
> +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> +		component_reg_phys = cxl_rcrb_to_component(
> +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);

Should this be CXL_RCRB_UPSTREAM?

> +	else
> +		component_reg_phys = cxlds->component_reg_phys;
> +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> +				     parent_dport);
>   	if (IS_ERR(endpoint))
>   		return PTR_ERR(endpoint);
>   
> @@ -87,6 +98,7 @@ static int cxl_mem_probe(struct device *dev)
>   {
>   	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
>   	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> +	struct device *endpoint_parent;
>   	struct cxl_port *parent_port;
>   	struct cxl_dport *dport;
>   	struct dentry *dentry;
> @@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
>   		return -ENXIO;
>   	}
>   
> -	device_lock(&parent_port->dev);
> -	if (!parent_port->dev.driver) {
> +	if (dport->rch)
> +		endpoint_parent = parent_port->uport;
> +	else
> +		endpoint_parent = &parent_port->dev;
> +
> +	device_lock(endpoint_parent);
> +	if (!endpoint_parent->driver) {
>   		dev_err(dev, "CXL port topology %s not enabled\n",
>   			dev_name(&parent_port->dev));
>   		rc = -ENXIO;
>   		goto unlock;
>   	}
>   
> -	rc = devm_cxl_add_endpoint(cxlmd, dport);
> +	rc = devm_cxl_add_endpoint(endpoint_parent, cxlmd, dport);
>   unlock:
> -	device_unlock(&parent_port->dev);
> +	device_unlock(endpoint_parent);
>   	put_device(&parent_port->dev);
>   	if (rc)
>   		return rc;
> diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
> index e15da405b948..73ff6c33a0c0 100644
> --- a/drivers/cxl/pci.c
> +++ b/drivers/cxl/pci.c
> @@ -433,6 +433,15 @@ static void devm_cxl_pci_create_doe(struct cxl_dev_state *cxlds)
>   	}
>   }
>   
> +/*
> + * Assume that any RCIEP that emits the CXL memory expander class code
> + * is an RCD
> + */
> +static bool is_cxl_restricted(struct pci_dev *pdev)
> +{
> +	return pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END;
> +}
> +
>   static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>   {
>   	struct cxl_register_map map;
> @@ -455,6 +464,7 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>   	if (IS_ERR(cxlds))
>   		return PTR_ERR(cxlds);
>   
> +	cxlds->rcd = is_cxl_restricted(pdev);
>   	cxlds->serial = pci_get_dsn(pdev);
>   	cxlds->cxl_dvsec = pci_find_dvsec_capability(
>   		pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_PCIE_DEVICE);
> 

Re: [PATCH v4 06/12] tools/testing/cxl: Make mock CEDT parsing more robust

[Alison Schofield]

On Thu, Nov 24, 2022 at 10:35:10AM -0800, Dan Williams wrote:
> Accept any cxl_test topology device as the first argument in
> cxl_chbs_context. This is in preparation for reworking the detection of
> the component registers across VH and RCH topologies. Move
> mock_acpi_table_parse_cedt() beneath the definition of is_mock_port()
> and use is_mock_port() instead of the explicit mock cxl_acpi device
> check.

I'll ACK this change, alhtough I don't appreciate the code move and modify
in the same patch. It hides the diff.

The commit msg seems needlessly vague. Why not state what was done?
Something like: 'Accept any topology device in cxl_chbs_context'

> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  tools/testing/cxl/test/cxl.c |   80 +++++++++++++++++++++---------------------
>  1 file changed, 40 insertions(+), 40 deletions(-)
> 
> diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> index facfcd11cb67..42a34650dd2f 100644
> --- a/tools/testing/cxl/test/cxl.c
> +++ b/tools/testing/cxl/test/cxl.c
> @@ -320,46 +320,6 @@ static int populate_cedt(void)
>  	return 0;
>  }
>  
> -/*
> - * WARNING, this hack assumes the format of 'struct
> - * cxl_cfmws_context' and 'struct cxl_chbs_context' share the property that
> - * the first struct member is the device being probed by the cxl_acpi
> - * driver.
> - */
> -struct cxl_cedt_context {
> -	struct device *dev;
> -};
> -
> -static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id,
> -				      acpi_tbl_entry_handler_arg handler_arg,
> -				      void *arg)
> -{
> -	struct cxl_cedt_context *ctx = arg;
> -	struct device *dev = ctx->dev;
> -	union acpi_subtable_headers *h;
> -	unsigned long end;
> -	int i;
> -
> -	if (dev != &cxl_acpi->dev)
> -		return acpi_table_parse_cedt(id, handler_arg, arg);
> -
> -	if (id == ACPI_CEDT_TYPE_CHBS)
> -		for (i = 0; i < ARRAY_SIZE(mock_cedt.chbs); i++) {
> -			h = (union acpi_subtable_headers *)&mock_cedt.chbs[i];
> -			end = (unsigned long)&mock_cedt.chbs[i + 1];
> -			handler_arg(h, arg, end);
> -		}
> -
> -	if (id == ACPI_CEDT_TYPE_CFMWS)
> -		for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
> -			h = (union acpi_subtable_headers *) mock_cfmws[i];
> -			end = (unsigned long) h + mock_cfmws[i]->header.length;
> -			handler_arg(h, arg, end);
> -		}
> -
> -	return 0;
> -}
> -
>  static bool is_mock_bridge(struct device *dev)
>  {
>  	int i;
> @@ -410,6 +370,46 @@ static bool is_mock_port(struct device *dev)
>  	return false;
>  }
>  
> +/*
> + * WARNING, this hack assumes the format of 'struct cxl_cfmws_context'
> + * and 'struct cxl_chbs_context' share the property that the first
> + * struct member is cxl_test device being probed by the cxl_acpi
> + * driver.
> + */
> +struct cxl_cedt_context {
> +	struct device *dev;
> +};
> +
> +static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id,
> +				      acpi_tbl_entry_handler_arg handler_arg,
> +				      void *arg)
> +{
> +	struct cxl_cedt_context *ctx = arg;
> +	struct device *dev = ctx->dev;
> +	union acpi_subtable_headers *h;
> +	unsigned long end;
> +	int i;
> +
> +	if (!is_mock_port(dev) && !is_mock_dev(dev))
> +		return acpi_table_parse_cedt(id, handler_arg, arg);
> +
> +	if (id == ACPI_CEDT_TYPE_CHBS)
> +		for (i = 0; i < ARRAY_SIZE(mock_cedt.chbs); i++) {
> +			h = (union acpi_subtable_headers *)&mock_cedt.chbs[i];
> +			end = (unsigned long)&mock_cedt.chbs[i + 1];
> +			handler_arg(h, arg, end);
> +		}
> +
> +	if (id == ACPI_CEDT_TYPE_CFMWS)
> +		for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
> +			h = (union acpi_subtable_headers *) mock_cfmws[i];
> +			end = (unsigned long) h + mock_cfmws[i]->header.length;
> +			handler_arg(h, arg, end);
> +		}
> +
> +	return 0;
> +}
> +
>  static int host_bridge_index(struct acpi_device *adev)
>  {
>  	return adev - host_bridge;
> 

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dan Williams]

Dave Jiang wrote:
> 
> 
> On 11/24/2022 11:35 AM, Dan Williams wrote:
> > Unlike a CXL memory expander in a VH topology that has at least one
> > intervening 'struct cxl_port' instance between itself and the CXL root
> > device, an RCD attaches one-level higher. For example:
> > 
> >                 VH
> >            ┌──────────┐
> >            │ ACPI0017 │
> >            │  root0   │
> >            └─────┬────┘
> >                  │
> >            ┌─────┴────┐
> >            │  dport0  │
> >      ┌─────┤ ACPI0016 ├─────┐
> >      │     │  port1   │     │
> >      │     └────┬─────┘     │
> >      │          │           │
> >   ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> >   │dport0│   │dport1│   │dport2│
> >   │ RP0  │   │ RP1  │   │ RP2  │
> >   └──────┘   └──┬───┘   └──────┘
> >                 │
> >             ┌───┴─────┐
> >             │endpoint0│
> >             │  port2  │
> >             └─────────┘
> > 
> > ...vs:
> > 
> >                RCH
> >            ┌──────────┐
> >            │ ACPI0017 │
> >            │  root0   │
> >            └────┬─────┘
> >                 │
> >             ┌───┴────┐
> >             │ dport0 │
> >             │ACPI0016│
> >             └───┬────┘
> >                 │
> >            ┌────┴─────┐
> >            │endpoint0 │
> >            │  port1   │
> >            └──────────┘
> > 
> > So arrange for endpoint port in the RCH/RCD case to appear directly
> > connected to the host-bridge in its singular role as a dport. Compare
> > that to the VH case where the host-bridge serves a dual role as a
> > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > instances in the CXL topology.
> > 
> > Another deviation from the VH case is that RCDs may need to look up
> > their component registers from the Root Complex Register Block (RCRB).
> > That platform firmware specified RCRB area is cached by the cxl_acpi
> > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > upstream port component registers).
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >   drivers/cxl/core/port.c |   11 +++++++++--
> >   drivers/cxl/cxlmem.h    |    2 ++
> >   drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> >   drivers/cxl/pci.c       |   10 ++++++++++
> >   4 files changed, 45 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > index c7f58282b2c1..2385ee00eb9a 100644
> > --- a/drivers/cxl/core/port.c
> > +++ b/drivers/cxl/core/port.c
> > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> >   {
> >   	struct device *dev = &cxlmd->dev;
> >   	struct device *iter;
> > +	struct cxl_dport *dport;
> > +	struct cxl_port *port;
> >   	int rc;
> >   
> > +	/*
> > +	 * Skip intermediate port enumeration in the RCH case, there
> > +	 * are no ports in between a host bridge and an endpoint.
> > +	 */
> > +	if (cxlmd->cxlds->rcd)
> > +		return 0;
> > +
> >   	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> >   	if (rc)
> >   		return rc;
> > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> >   	for (iter = dev; iter; iter = grandparent(iter)) {
> >   		struct device *dport_dev = grandparent(iter);
> >   		struct device *uport_dev;
> > -		struct cxl_dport *dport;
> > -		struct cxl_port *port;
> >   
> >   		if (!dport_dev)
> >   			return 0;
> > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > index e082991bc58c..35d485d041f0 100644
> > --- a/drivers/cxl/cxlmem.h
> > +++ b/drivers/cxl/cxlmem.h
> > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> >    * @dev: The device associated with this CXL state
> >    * @regs: Parsed register blocks
> >    * @cxl_dvsec: Offset to the PCIe device DVSEC
> > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> >    * @payload_size: Size of space for payload
> >    *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> >    * @lsa_size: Size of Label Storage Area
> > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> >   	struct cxl_regs regs;
> >   	int cxl_dvsec;
> >   
> > +	bool rcd;
> >   	size_t payload_size;
> >   	size_t lsa_size;
> >   	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > index aa63ce8c7ca6..9a655b4b5e52 100644
> > --- a/drivers/cxl/mem.c
> > +++ b/drivers/cxl/mem.c
> > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> >   	return 0;
> >   }
> >   
> > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> >   				 struct cxl_dport *parent_dport)
> >   {
> >   	struct cxl_port *parent_port = parent_dport->port;
> >   	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> >   	struct cxl_port *endpoint, *iter, *down;
> > +	resource_size_t component_reg_phys;
> >   	int rc;
> >   
> >   	/*
> > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> >   		ep->next = down;
> >   	}
> >   
> > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > -				     cxlds->component_reg_phys, parent_dport);
> > +	/*
> > +	 * The component registers for an RCD might come from the
> > +	 * host-bridge RCRB if they are not already mapped via the
> > +	 * typical register locator mechanism.
> > +	 */
> > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > +		component_reg_phys = cxl_rcrb_to_component(
> > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> 
> Should this be CXL_RCRB_UPSTREAM?

Indeed it should, good catch.

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Alison Schofield]

On Thu, Nov 24, 2022 at 10:35:38AM -0800, Dan Williams wrote:
> In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> the represents the memory expander. Unlike a VH topology there is no
> CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> host-bridge as a dport, per usual, but then that dport directly hosts
> the endpoint port.
> 
> Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> device instance as its immediate child.
> 

Reviewed-by: Alison Schofield <alison.schofield@intel.com>

How can this host bridge and device be used? Should I be looking in
the spec to see the limitations on a 1.1 usage? Expect it to behave
like VH topo? Expect graceful failure where it doesn't?  I'm just
after a starting point for understanding how the 1.1 world fits in.


> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  tools/testing/cxl/test/cxl.c |  151 +++++++++++++++++++++++++++++++++++++++---
>  tools/testing/cxl/test/mem.c |   37 ++++++++++
>  2 files changed, 176 insertions(+), 12 deletions(-)
> 
> diff --git a/tools/testing/cxl/test/cxl.c b/tools/testing/cxl/test/cxl.c
> index 1823c61d7ba3..b1c362090b92 100644
> --- a/tools/testing/cxl/test/cxl.c
> +++ b/tools/testing/cxl/test/cxl.c
> @@ -13,9 +13,11 @@
>  
>  #define NR_CXL_HOST_BRIDGES 2
>  #define NR_CXL_SINGLE_HOST 1
> +#define NR_CXL_RCH 1
>  #define NR_CXL_ROOT_PORTS 2
>  #define NR_CXL_SWITCH_PORTS 2
>  #define NR_CXL_PORT_DECODERS 8
> +#define NR_BRIDGES (NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + NR_CXL_RCH)
>  
>  static struct platform_device *cxl_acpi;
>  static struct platform_device *cxl_host_bridge[NR_CXL_HOST_BRIDGES];
> @@ -35,6 +37,8 @@ static struct platform_device *cxl_swd_single[NR_MEM_SINGLE];
>  struct platform_device *cxl_mem[NR_MEM_MULTI];
>  struct platform_device *cxl_mem_single[NR_MEM_SINGLE];
>  
> +static struct platform_device *cxl_rch[NR_CXL_RCH];
> +static struct platform_device *cxl_rcd[NR_CXL_RCH];
>  
>  static inline bool is_multi_bridge(struct device *dev)
>  {
> @@ -57,7 +61,7 @@ static inline bool is_single_bridge(struct device *dev)
>  }
>  
>  static struct acpi_device acpi0017_mock;
> -static struct acpi_device host_bridge[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST] = {
> +static struct acpi_device host_bridge[NR_BRIDGES] = {
>  	[0] = {
>  		.handle = &host_bridge[0],
>  	},
> @@ -67,7 +71,9 @@ static struct acpi_device host_bridge[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST]
>  	[2] = {
>  		.handle = &host_bridge[2],
>  	},
> -
> +	[3] = {
> +		.handle = &host_bridge[3],
> +	},
>  };
>  
>  static bool is_mock_dev(struct device *dev)
> @@ -80,6 +86,9 @@ static bool is_mock_dev(struct device *dev)
>  	for (i = 0; i < ARRAY_SIZE(cxl_mem_single); i++)
>  		if (dev == &cxl_mem_single[i]->dev)
>  			return true;
> +	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++)
> +		if (dev == &cxl_rcd[i]->dev)
> +			return true;
>  	if (dev == &cxl_acpi->dev)
>  		return true;
>  	return false;
> @@ -101,7 +110,7 @@ static bool is_mock_adev(struct acpi_device *adev)
>  
>  static struct {
>  	struct acpi_table_cedt cedt;
> -	struct acpi_cedt_chbs chbs[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST];
> +	struct acpi_cedt_chbs chbs[NR_BRIDGES];
>  	struct {
>  		struct acpi_cedt_cfmws cfmws;
>  		u32 target[1];
> @@ -122,6 +131,10 @@ static struct {
>  		struct acpi_cedt_cfmws cfmws;
>  		u32 target[1];
>  	} cfmws4;
> +	struct {
> +		struct acpi_cedt_cfmws cfmws;
> +		u32 target[1];
> +	} cfmws5;
>  } __packed mock_cedt = {
>  	.cedt = {
>  		.header = {
> @@ -154,6 +167,14 @@ static struct {
>  		.uid = 2,
>  		.cxl_version = ACPI_CEDT_CHBS_VERSION_CXL20,
>  	},
> +	.chbs[3] = {
> +		.header = {
> +			.type = ACPI_CEDT_TYPE_CHBS,
> +			.length = sizeof(mock_cedt.chbs[0]),
> +		},
> +		.uid = 3,
> +		.cxl_version = ACPI_CEDT_CHBS_VERSION_CXL11,
> +	},
>  	.cfmws0 = {
>  		.cfmws = {
>  			.header = {
> @@ -229,6 +250,21 @@ static struct {
>  		},
>  		.target = { 2 },
>  	},
> +	.cfmws5 = {
> +		.cfmws = {
> +			.header = {
> +				.type = ACPI_CEDT_TYPE_CFMWS,
> +				.length = sizeof(mock_cedt.cfmws5),
> +			},
> +			.interleave_ways = 0,
> +			.granularity = 4,
> +			.restrictions = ACPI_CEDT_CFMWS_RESTRICT_TYPE3 |
> +					ACPI_CEDT_CFMWS_RESTRICT_VOLATILE,
> +			.qtg_id = 5,
> +			.window_size = SZ_256M,
> +		},
> +		.target = { 3 },
> +	},
>  };
>  
>  struct acpi_cedt_cfmws *mock_cfmws[] = {
> @@ -237,6 +273,7 @@ struct acpi_cedt_cfmws *mock_cfmws[] = {
>  	[2] = &mock_cedt.cfmws2.cfmws,
>  	[3] = &mock_cedt.cfmws3.cfmws,
>  	[4] = &mock_cedt.cfmws4.cfmws,
> +	[5] = &mock_cedt.cfmws5.cfmws,
>  };
>  
>  struct cxl_mock_res {
> @@ -262,11 +299,11 @@ static void depopulate_all_mock_resources(void)
>  	mutex_unlock(&mock_res_lock);
>  }
>  
> -static struct cxl_mock_res *alloc_mock_res(resource_size_t size)
> +static struct cxl_mock_res *alloc_mock_res(resource_size_t size, int align)
>  {
>  	struct cxl_mock_res *res = kzalloc(sizeof(*res), GFP_KERNEL);
>  	struct genpool_data_align data = {
> -		.align = SZ_256M,
> +		.align = align,
>  	};
>  	unsigned long phys;
>  
> @@ -301,7 +338,7 @@ static int populate_cedt(void)
>  		else
>  			size = ACPI_CEDT_CHBS_LENGTH_CXL11;
>  
> -		res = alloc_mock_res(size);
> +		res = alloc_mock_res(size, size);
>  		if (!res)
>  			return -ENOMEM;
>  		chbs->base = res->range.start;
> @@ -311,7 +348,7 @@ static int populate_cedt(void)
>  	for (i = 0; i < ARRAY_SIZE(mock_cfmws); i++) {
>  		struct acpi_cedt_cfmws *window = mock_cfmws[i];
>  
> -		res = alloc_mock_res(window->window_size);
> +		res = alloc_mock_res(window->window_size, SZ_256M);
>  		if (!res)
>  			return -ENOMEM;
>  		window->base_hpa = res->range.start;
> @@ -330,6 +367,10 @@ static bool is_mock_bridge(struct device *dev)
>  	for (i = 0; i < ARRAY_SIZE(cxl_hb_single); i++)
>  		if (dev == &cxl_hb_single[i]->dev)
>  			return true;
> +	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++)
> +		if (dev == &cxl_rch[i]->dev)
> +			return true;
> +
>  	return false;
>  }
>  
> @@ -439,7 +480,7 @@ mock_acpi_evaluate_integer(acpi_handle handle, acpi_string pathname,
>  	return AE_OK;
>  }
>  
> -static struct pci_bus mock_pci_bus[NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST];
> +static struct pci_bus mock_pci_bus[NR_BRIDGES];
>  static struct acpi_pci_root mock_pci_root[ARRAY_SIZE(mock_pci_bus)] = {
>  	[0] = {
>  		.bus = &mock_pci_bus[0],
> @@ -450,7 +491,9 @@ static struct acpi_pci_root mock_pci_root[ARRAY_SIZE(mock_pci_bus)] = {
>  	[2] = {
>  		.bus = &mock_pci_bus[2],
>  	},
> -
> +	[3] = {
> +		.bus = &mock_pci_bus[3],
> +	},
>  };
>  
>  static bool is_mock_bus(struct pci_bus *bus)
> @@ -736,6 +779,87 @@ static void mock_companion(struct acpi_device *adev, struct device *dev)
>  #define SZ_512G (SZ_64G * 8)
>  #endif
>  
> +static __init int cxl_rch_init(void)
> +{
> +	int rc, i;
> +
> +	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++) {
> +		int idx = NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + i;
> +		struct acpi_device *adev = &host_bridge[idx];
> +		struct platform_device *pdev;
> +
> +		pdev = platform_device_alloc("cxl_host_bridge", idx);
> +		if (!pdev)
> +			goto err_bridge;
> +
> +		mock_companion(adev, &pdev->dev);
> +		rc = platform_device_add(pdev);
> +		if (rc) {
> +			platform_device_put(pdev);
> +			goto err_bridge;
> +		}
> +
> +		cxl_rch[i] = pdev;
> +		mock_pci_bus[idx].bridge = &pdev->dev;
> +		rc = sysfs_create_link(&pdev->dev.kobj, &pdev->dev.kobj,
> +				       "firmware_node");
> +		if (rc)
> +			goto err_bridge;
> +	}
> +
> +	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++) {
> +		int idx = NR_MEM_MULTI + NR_MEM_SINGLE + i;
> +		struct platform_device *rch = cxl_rch[i];
> +		struct platform_device *pdev;
> +
> +		pdev = platform_device_alloc("cxl_rcd", idx);
> +		if (!pdev)
> +			goto err_mem;
> +		pdev->dev.parent = &rch->dev;
> +		set_dev_node(&pdev->dev, i % 2);
> +
> +		rc = platform_device_add(pdev);
> +		if (rc) {
> +			platform_device_put(pdev);
> +			goto err_mem;
> +		}
> +		cxl_rcd[i] = pdev;
> +	}
> +
> +	return 0;
> +
> +err_mem:
> +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> +		platform_device_unregister(cxl_rcd[i]);
> +err_bridge:
> +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> +		struct platform_device *pdev = cxl_rch[i];
> +
> +		if (!pdev)
> +			continue;
> +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> +		platform_device_unregister(cxl_rch[i]);
> +	}
> +
> +	return rc;
> +}
> +
> +static void cxl_rch_exit(void)
> +{
> +	int i;
> +
> +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> +		platform_device_unregister(cxl_rcd[i]);
> +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> +		struct platform_device *pdev = cxl_rch[i];
> +
> +		if (!pdev)
> +			continue;
> +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> +		platform_device_unregister(cxl_rch[i]);
> +	}
> +}
> +
>  static __init int cxl_single_init(void)
>  {
>  	int i, rc;
> @@ -1008,9 +1132,13 @@ static __init int cxl_test_init(void)
>  	if (rc)
>  		goto err_mem;
>  
> +	rc = cxl_rch_init();
> +	if (rc)
> +		goto err_single;
> +
>  	cxl_acpi = platform_device_alloc("cxl_acpi", 0);
>  	if (!cxl_acpi)
> -		goto err_single;
> +		goto err_rch;
>  
>  	mock_companion(&acpi0017_mock, &cxl_acpi->dev);
>  	acpi0017_mock.dev.bus = &platform_bus_type;
> @@ -1023,6 +1151,8 @@ static __init int cxl_test_init(void)
>  
>  err_add:
>  	platform_device_put(cxl_acpi);
> +err_rch:
> +	cxl_rch_exit();
>  err_single:
>  	cxl_single_exit();
>  err_mem:
> @@ -1060,6 +1190,7 @@ static __exit void cxl_test_exit(void)
>  	int i;
>  
>  	platform_device_unregister(cxl_acpi);
> +	cxl_rch_exit();
>  	cxl_single_exit();
>  	for (i = ARRAY_SIZE(cxl_mem) - 1; i >= 0; i--)
>  		platform_device_unregister(cxl_mem[i]);
> diff --git a/tools/testing/cxl/test/mem.c b/tools/testing/cxl/test/mem.c
> index a4ee8e61dd60..b59c5976b2d9 100644
> --- a/tools/testing/cxl/test/mem.c
> +++ b/tools/testing/cxl/test/mem.c
> @@ -100,6 +100,24 @@ static int mock_get_log(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
>  	return 0;
>  }
>  
> +static int mock_rcd_id(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
> +{
> +	struct cxl_mbox_identify id = {
> +		.fw_revision = { "mock fw v1 " },
> +		.total_capacity =
> +			cpu_to_le64(DEV_SIZE / CXL_CAPACITY_MULTIPLIER),
> +		.volatile_capacity =
> +			cpu_to_le64(DEV_SIZE / CXL_CAPACITY_MULTIPLIER),
> +	};
> +
> +	if (cmd->size_out < sizeof(id))
> +		return -EINVAL;
> +
> +	memcpy(cmd->payload_out, &id, sizeof(id));
> +
> +	return 0;
> +}
> +
>  static int mock_id(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *cmd)
>  {
>  	struct cxl_mbox_identify id = {
> @@ -216,7 +234,10 @@ static int cxl_mock_mbox_send(struct cxl_dev_state *cxlds, struct cxl_mbox_cmd *
>  		rc = mock_get_log(cxlds, cmd);
>  		break;
>  	case CXL_MBOX_OP_IDENTIFY:
> -		rc = mock_id(cxlds, cmd);
> +		if (cxlds->rcd)
> +			rc = mock_rcd_id(cxlds, cmd);
> +		else
> +			rc = mock_id(cxlds, cmd);
>  		break;
>  	case CXL_MBOX_OP_GET_LSA:
>  		rc = mock_get_lsa(cxlds, cmd);
> @@ -245,6 +266,13 @@ static void label_area_release(void *lsa)
>  	vfree(lsa);
>  }
>  
> +static bool is_rcd(struct platform_device *pdev)
> +{
> +	const struct platform_device_id *id = platform_get_device_id(pdev);
> +
> +	return !!id->driver_data;
> +}
> +
>  static int cxl_mock_mem_probe(struct platform_device *pdev)
>  {
>  	struct device *dev = &pdev->dev;
> @@ -268,6 +296,10 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
>  	cxlds->serial = pdev->id;
>  	cxlds->mbox_send = cxl_mock_mbox_send;
>  	cxlds->payload_size = SZ_4K;
> +	if (is_rcd(pdev)) {
> +		cxlds->rcd = true;
> +		cxlds->component_reg_phys = CXL_RESOURCE_NONE;
> +	}
>  
>  	rc = cxl_enumerate_cmds(cxlds);
>  	if (rc)
> @@ -289,7 +321,8 @@ static int cxl_mock_mem_probe(struct platform_device *pdev)
>  }
>  
>  static const struct platform_device_id cxl_mock_mem_ids[] = {
> -	{ .name = "cxl_mem", },
> +	{ .name = "cxl_mem", 0 },
> +	{ .name = "cxl_rcd", 1 },
>  	{ },
>  };
>  MODULE_DEVICE_TABLE(platform, cxl_mock_mem_ids);
> 

Re: [PATCH v4 09/12] cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem

[Robert Richter]

On 24.11.22 10:35:27, Dan Williams wrote:
> In preparation for devm_cxl_add_endpoint() to call out to
> cxl_rcrb_to_component() to map the upstream port component registers,
> move devm_cxl_add_endpoint() from the cxl_core to the cxl_mem driver.
> This is due to the organization of cxl_test that mandates that the
> cxl_core not call out to any mocked symbols. It also cleans up the
> export of devm_cxl_add_endpoint() which is just a wrapper around
> devm_cxl_add_port().

The last sentense is unclear to me? I see you just make it static as
there is only a single user, do you mean that here? The change itself
looks good.

> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

Reviewed-by: Robert Richter <rrichter@amd.com>

> ---
>  drivers/cxl/core/core.h |    8 --------
>  drivers/cxl/core/port.c |   39 ---------------------------------------
>  drivers/cxl/cxl.h       |    2 --
>  drivers/cxl/cxlmem.h    |    9 +++++++++
>  drivers/cxl/mem.c       |   38 ++++++++++++++++++++++++++++++++++++++
>  5 files changed, 47 insertions(+), 49 deletions(-)

Re: [PATCH v4 09/12] cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem

[Robert Richter]

On 28.11.22 20:50:16, Robert Richter wrote:
> On 24.11.22 10:35:27, Dan Williams wrote:
> > In preparation for devm_cxl_add_endpoint() to call out to
> > cxl_rcrb_to_component() to map the upstream port component registers,
> > move devm_cxl_add_endpoint() from the cxl_core to the cxl_mem driver.
> > This is due to the organization of cxl_test that mandates that the
> > cxl_core not call out to any mocked symbols. It also cleans up the
> > export of devm_cxl_add_endpoint() which is just a wrapper around
> > devm_cxl_add_port().
> 
> The last sentense is unclear to me? I see you just make it static as
> there is only a single user, do you mean that here? The change itself
> looks good.

Looking at the next patch I think you mean the change of the function
i/f of devm_cxl_add_endpoint() here. Please update description.

> 
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> 
> Reviewed-by: Robert Richter <rrichter@amd.com>
> 
> > ---
> >  drivers/cxl/core/core.h |    8 --------
> >  drivers/cxl/core/port.c |   39 ---------------------------------------
> >  drivers/cxl/cxl.h       |    2 --
> >  drivers/cxl/cxlmem.h    |    9 +++++++++
> >  drivers/cxl/mem.c       |   38 ++++++++++++++++++++++++++++++++++++++
> >  5 files changed, 47 insertions(+), 49 deletions(-)

Re: [PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:21, Dan Williams wrote:
> > From: Robert Richter <rrichter@amd.com>
> > 
> > A downstream port must be connected to a component register block.
> > For restricted hosts the base address is determined from the RCRB. The
> > RCRB is provided by the host's CEDT CHBS entry. Rework CEDT parser to
> > get the RCRB and add code to extract the component register block from
> > it.
> > 
> > RCRB's BAR[0..1] point to the component block containing CXL subsystem
> > component registers. MEMBAR extraction follows the PCI base spec here,
> > esp. 64 bit extraction and memory range alignment (6.0, 7.5.1.2.1). The
> > RCRB base address is cached in the cxl_dport per-host bridge so that the
> > upstream port component registers can be retrieved later by an RCD
> > (RCIEP) associated with the host bridge.
> > 
> > Note: Right now the component register block is used for HDM decoder
> > capability only which is optional for RCDs. If unsupported by the RCD,
> > the HDM init will fail. It is future work to bypass it in this case.
> > 
> > Co-developed-by: Terry Bowman <terry.bowman@amd.com>
> > Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> > Signed-off-by: Robert Richter <rrichter@amd.com>
> > [djbw: introduce devm_cxl_add_rch_dport()]
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >  drivers/cxl/acpi.c            |   54 ++++++++++++++++++++++++++++++++--------
> >  drivers/cxl/core/port.c       |   42 +++++++++++++++++++++++++++----
> >  drivers/cxl/core/regs.c       |   56 +++++++++++++++++++++++++++++++++++++++++
> >  drivers/cxl/cxl.h             |   16 ++++++++++++
> >  tools/testing/cxl/Kbuild      |    1 +
> >  tools/testing/cxl/test/cxl.c  |   10 +++++++
> >  tools/testing/cxl/test/mock.c |   19 ++++++++++++++
> >  tools/testing/cxl/test/mock.h |    3 ++
> >  8 files changed, 186 insertions(+), 15 deletions(-)
> > 
> > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > index 50d82376097c..1224add13529 100644
> > --- a/drivers/cxl/acpi.c
> > +++ b/drivers/cxl/acpi.c
> > @@ -9,6 +9,8 @@
> >  #include "cxlpci.h"
> >  #include "cxl.h"
> >  
> > +#define CXL_RCRB_SIZE	SZ_8K
> > +
> >  static unsigned long cfmws_to_decoder_flags(int restrictions)
> >  {
> >  	unsigned long flags = CXL_DECODER_F_ENABLE;
> > @@ -215,6 +217,11 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> >  	if (rc)
> >  		return rc;
> >  
> > +	if (dport->rch) {
> > +		dev_info(bridge, "host supports CXL (restricted)\n");
> > +		return 0;
> > +	}
> 
> This change comes after devm_cxl_register_pci_bus() to serve the
> cxl_port_to_pci_bus() in devm_cxl_port_enumerate_dports() in
> cxl_port_probe(). A root port is not probed and
> devm_cxl_port_enumerate_dports() will be never called, so we could
> jump out before devm_cxl_register_pci_bus().

Good point.

> On the other side we might want to be ready to use
> cxl_port_to_pci_bus() elsewhere in later changes. RCHs would not work
> then.

At least cxl_port_to_pci_bus() as is will work fine for the RCD
endpoint-cxl-port, so I think it is ok to leave this alone for now.

> 
> > +
> >  	port = devm_cxl_add_port(host, bridge, dport->component_reg_phys,
> >  				 dport);
> >  	if (IS_ERR(port))
> > @@ -228,27 +235,46 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> >  struct cxl_chbs_context {
> >  	struct device *dev;
> >  	unsigned long long uid;
> > -	resource_size_t chbcr;
> > +	struct acpi_cedt_chbs chbs;
> >  };
> >  
> > -static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
> > -			 const unsigned long end)
> > +static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
> > +			const unsigned long end)
> >  {
> >  	struct cxl_chbs_context *ctx = arg;
> >  	struct acpi_cedt_chbs *chbs;
> >  
> > -	if (ctx->chbcr)
> > +	if (ctx->chbs.base)
> >  		return 0;
> >  
> >  	chbs = (struct acpi_cedt_chbs *) header;
> >  
> >  	if (ctx->uid != chbs->uid)
> >  		return 0;
> > -	ctx->chbcr = chbs->base;
> > +	ctx->chbs = *chbs;
> >  
> >  	return 0;
> >  }
> >  
> > +static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
> > +{
> > +	struct acpi_cedt_chbs *chbs = &ctx->chbs;
> > +
> > +	if (!chbs->base)
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
> > +		return chbs->base;
> > +
> > +	if (chbs->length != CXL_RCRB_SIZE)
> > +		return CXL_RESOURCE_NONE;
> > +
> > +	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
> > +		&chbs->base);
> > +
> > +	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
> > +}
> > +
> 
> I have an improved version of this code which squashes cxl_get_chbcr()
> into cxl_get_chbs() (basically extends the original cxl_get_chbcr()
> function).

Care to send it? If I see it before the next posting I can fold it in,
otherwise it can be a follow-on cleanup.

> 
> >  static int add_host_bridge_dport(struct device *match, void *arg)
> >  {
> >  	acpi_status status;
> > @@ -258,6 +284,7 @@ static int add_host_bridge_dport(struct device *match, void *arg)
> >  	struct cxl_chbs_context ctx;
> >  	struct acpi_pci_root *pci_root;
> >  	struct cxl_port *root_port = arg;
> > +	resource_size_t component_reg_phys;
> >  	struct device *host = root_port->dev.parent;
> >  	struct acpi_device *hb = to_cxl_host_bridge(host, match);
> >  
> > @@ -274,21 +301,28 @@ static int add_host_bridge_dport(struct device *match, void *arg)
> >  	dev_dbg(match, "UID found: %lld\n", uid);
> >  
> >  	ctx = (struct cxl_chbs_context) {
> > -		.dev = host,
> > +		.dev = match,
> >  		.uid = uid,
> >  	};
> > -	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbcr, &ctx);
> > +	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbs, &ctx);
> >  
> > -	if (ctx.chbcr == 0) {
> > +	component_reg_phys = cxl_get_chbcr(&ctx);
> > +	if (component_reg_phys == CXL_RESOURCE_NONE) {
> >  		dev_warn(match, "No CHBS found for Host Bridge (UID %lld)\n", uid);
> >  		return 0;
> >  	}
> >  
> > -	dev_dbg(match, "CHBCR found: 0x%08llx\n", (u64)ctx.chbcr);
> > +	dev_dbg(match, "CHBCR found: %pa\n", &component_reg_phys);
> >  
> >  	pci_root = acpi_pci_find_root(hb->handle);
> >  	bridge = pci_root->bus->bridge;
> > -	dport = devm_cxl_add_dport(root_port, bridge, uid, ctx.chbcr);
> > +	if (ctx.chbs.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
> > +		dport = devm_cxl_add_rch_dport(root_port, bridge, uid,
> > +					       component_reg_phys,
> > +					       ctx.chbs.base);
> 
> Yes, this new function makes the rcrb handling much more simpler.
> 
> > +	else
> > +		dport = devm_cxl_add_dport(root_port, bridge, uid,
> > +					   component_reg_phys);
> >  	if (IS_ERR(dport))
> >  		return PTR_ERR(dport);
> >  
> > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > index d225267c69bb..d9fe06e1462f 100644
> > --- a/drivers/cxl/core/port.c
> > +++ b/drivers/cxl/core/port.c
> > @@ -628,6 +628,8 @@ static struct cxl_port *cxl_port_alloc(struct device *uport,
> >  			iter = to_cxl_port(iter->dev.parent);
> >  		if (iter->host_bridge)
> >  			port->host_bridge = iter->host_bridge;
> > +		else if (parent_dport->rch)
> > +			port->host_bridge = parent_dport->dport;
> 
> Yes, looks good. This makes the endpoint a child of a CXL root port,
> not the ACPI0017 the root device.
> 
> >  		else
> >  			port->host_bridge = iter->uport;
> >  		dev_dbg(uport, "host-bridge: %s\n", dev_name(port->host_bridge));
> > @@ -899,10 +901,15 @@ static void cxl_dport_unlink(void *data)
> >  	sysfs_remove_link(&port->dev.kobj, link_name);
> >  }
> >  
> > -static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
> > -					      struct device *dport_dev,
> > -					      int port_id,
> > -					      resource_size_t component_reg_phys)
> > +enum cxl_dport_mode {
> > +	CXL_DPORT_VH,
> > +	CXL_DPORT_RCH,
> > +};
> > +
> > +static struct cxl_dport *
> > +__devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
> > +		     int port_id, resource_size_t component_reg_phys,
> > +		     enum cxl_dport_mode mode, resource_size_t rcrb)
> >  {
> >  	char link_name[CXL_TARGET_STRLEN];
> >  	struct cxl_dport *dport;
> > @@ -932,6 +939,9 @@ static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
> >  	dport->port_id = port_id;
> >  	dport->component_reg_phys = component_reg_phys;
> >  	dport->port = port;
> > +	if (mode == CXL_DPORT_RCH)
> > +		dport->rch = true;
> 
> Alternatively an inline function could be added which checks
> dport->rcrb for a valid address.

I like it. Especially because a valid RCRB is needed to register the
dport in the first instance. I think it looks ok without an inline
function:

@@ -901,15 +901,10 @@ static void cxl_dport_unlink(void *data)
        sysfs_remove_link(&port->dev.kobj, link_name);
 }
 
-enum cxl_dport_mode {
-       CXL_DPORT_VH,
-       CXL_DPORT_RCH,
-};
-
 static struct cxl_dport *
 __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
                     int port_id, resource_size_t component_reg_phys,
-                    enum cxl_dport_mode mode, resource_size_t rcrb)
+                    resource_size_t rcrb)
 {
        char link_name[CXL_TARGET_STRLEN];
        struct cxl_dport *dport;
@@ -939,7 +934,7 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
        dport->port_id = port_id;
        dport->component_reg_phys = component_reg_phys;
        dport->port = port;
-       if (mode == CXL_DPORT_RCH)
+       if (rcrb != CXL_RESOURCE_NONE)
                dport->rch = true;
        dport->rcrb = rcrb;
 

> 
> > +	dport->rcrb = rcrb;
> >  
> >  	cond_cxl_root_lock(port);
> >  	rc = add_dport(port, dport);
> > @@ -973,7 +983,8 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> >  	struct cxl_dport *dport;
> >  
> >  	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> > -				     component_reg_phys);
> > +				     component_reg_phys, CXL_DPORT_VH,
> > +				     CXL_RESOURCE_NONE);
> >  	if (IS_ERR(dport)) {
> >  		dev_dbg(dport_dev, "failed to add dport to %s: %ld\n",
> >  			dev_name(&port->dev), PTR_ERR(dport));
> > @@ -986,6 +997,27 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> >  }
> >  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
> >  
> > +struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
> > +					 struct device *dport_dev, int port_id,
> > +					 resource_size_t component_reg_phys,
> > +					 resource_size_t rcrb)
> 
> The documentation header is missing for that.

Added the following and clarified that @rcrb is mandatory:

@@ -966,7 +961,7 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
 }
 
 /**
- * devm_cxl_add_dport - append downstream port data to a cxl_port
+ * devm_cxl_add_dport - append VH downstream port data to a cxl_port
  * @port: the cxl_port that references this dport
  * @dport_dev: firmware or PCI device representing the dport
  * @port_id: identifier for this dport in a decoder's target list
@@ -997,6 +991,16 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
 }
 EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
 
+/**
+ * devm_cxl_add_rch_dport - append RCH downstream port data to a cxl_port
+ * @port: the cxl_port that references this dport
+ * @dport_dev: firmware or PCI device representing the dport
+ * @port_id: identifier for this dport in a decoder's target list
+ * @component_reg_phys: optional location of CXL component registers
+ * @rcrb: mandatory location of a Root Complex Register Block
+ *
+ * See CXL 3.0 9.11.8 CXL Devices Attached to an RCH
+ */
 struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
                                         struct device *dport_dev, int port_id,
                                         resource_size_t component_reg_phys,

> 
> > +{
> > +	struct cxl_dport *dport;
> > +
> > +	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> > +				     component_reg_phys, CXL_DPORT_RCH, rcrb);
> > +	if (IS_ERR(dport)) {
> > +		dev_dbg(dport_dev, "failed to add RCH dport to %s: %ld\n",
> > +			dev_name(&port->dev), PTR_ERR(dport));
> > +	} else {
> > +		dev_dbg(dport_dev, "RCH dport added to %s\n",
> > +			dev_name(&port->dev));
> > +	}
> > +
> > +	return dport;
> > +}
> > +EXPORT_SYMBOL_NS_GPL(devm_cxl_add_rch_dport, CXL);
> > +
> >  static int add_ep(struct cxl_ep *new)
> >  {
> >  	struct cxl_port *port = new->dport->port;
> > diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c
> > index ec178e69b18f..7c2a85dc4125 100644
> > --- a/drivers/cxl/core/regs.c
> > +++ b/drivers/cxl/core/regs.c
> > @@ -307,3 +307,59 @@ int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
> >  	return -ENODEV;
> >  }
> >  EXPORT_SYMBOL_NS_GPL(cxl_find_regblock, CXL);
> > +
> > +resource_size_t cxl_rcrb_to_component(struct device *dev,
> > +				      resource_size_t rcrb,
> > +				      enum cxl_rcrb which)
> > +{
> > +	resource_size_t component_reg_phys;
> > +	u32 bar0, bar1;
> > +	void *addr;
> > +	u16 cmd;
> > +
> > +	if (which == CXL_RCRB_UPSTREAM)
> > +		rcrb += SZ_4K;
> > +
> > +	/*
> > +	 * RCRB's BAR[0..1] point to component block containing CXL
> > +	 * subsystem component registers. MEMBAR extraction follows
> > +	 * the PCI Base spec here, esp. 64 bit extraction and memory
> > +	 * ranges alignment (6.0, 7.5.1.2.1).
> > +	 */
> > +	if (!request_mem_region(rcrb, SZ_4K, "CXL RCRB"))
> > +		return CXL_RESOURCE_NONE;
> > +	addr = ioremap(rcrb, SZ_4K);
> > +	if (!addr) {
> > +		dev_err(dev, "Failed to map region %pr\n", addr);
> > +		release_mem_region(rcrb, SZ_4K);
> > +		return CXL_RESOURCE_NONE;
> > +	}
> > +
> > +	cmd = readw(addr + PCI_COMMAND);
> > +	bar0 = readl(addr + PCI_BASE_ADDRESS_0);
> > +	bar1 = readl(addr + PCI_BASE_ADDRESS_1);
> > +	iounmap(addr);
> > +	release_mem_region(rcrb, SZ_4K);
> > +
> > +	/* sanity check */
> > +	if (cmd == 0xffff)
> > +		return CXL_RESOURCE_NONE;
> 
> The spec says offset 0 should be checked (32 bit) which is always
> non-FF if implemented. This requires another read.
> 
> cmd is most of the cases also non-zero, so probably checking cmd
> instead will have the same effect. Still worth changing that.
> 
> If the downstream port's rcrb is all FFs, it is a FW bug. Could be
> worth a message.

Ok, makes sense, added:

@@ -335,15 +336,22 @@ resource_size_t cxl_rcrb_to_component(struct device *dev,
                return CXL_RESOURCE_NONE;
        }
 
+       id = readl(addr + PCI_VENDOR_ID);
        cmd = readw(addr + PCI_COMMAND);
        bar0 = readl(addr + PCI_BASE_ADDRESS_0);
        bar1 = readl(addr + PCI_BASE_ADDRESS_1);
        iounmap(addr);
        release_mem_region(rcrb, SZ_4K);
 
-       /* sanity check */
-       if (cmd == 0xffff)
+       /*
+        * Sanity check, see CXL 3.0 Figure 9-8 CXL Device that Does Not
+        * Remap Upstream Port and Component Registers
+        */
+       if (id == (u32) -1) {
+               if (which == CXL_RCRB_DOWNSTREAM)
+                       dev_err(dev, "Failed to access Downstream Port RCRB\n");
                return CXL_RESOURCE_NONE;
+       }
        if ((cmd & PCI_COMMAND_MEMORY) == 0)
                return CXL_RESOURCE_NONE;
        if (bar0 & (PCI_BASE_ADDRESS_MEM_TYPE_1M | PCI_BASE_ADDRESS_SPACE_IO))

Re: [PATCH v4 06/12] tools/testing/cxl: Make mock CEDT parsing more robust

[Dan Williams]

Alison Schofield wrote:
> On Thu, Nov 24, 2022 at 10:35:10AM -0800, Dan Williams wrote:
> > Accept any cxl_test topology device as the first argument in
> > cxl_chbs_context. This is in preparation for reworking the detection of
> > the component registers across VH and RCH topologies. Move
> > mock_acpi_table_parse_cedt() beneath the definition of is_mock_port()
> > and use is_mock_port() instead of the explicit mock cxl_acpi device
> > check.
> 
> I'll ACK this change, alhtough I don't appreciate the code move and modify
> in the same patch. It hides the diff.

That's fair. I'll go ahead and just forward declare is_mock_port() which
makes the diff easier to read:

@@ -320,10 +320,12 @@ static int populate_cedt(void)
        return 0;
 }
 
+static bool is_mock_port(struct device *dev);
+
 /*
- * WARNING, this hack assumes the format of 'struct
- * cxl_cfmws_context' and 'struct cxl_chbs_context' share the property that
- * the first struct member is the device being probed by the cxl_acpi
+ * WARNING, this hack assumes the format of 'struct cxl_cfmws_context'
+ * and 'struct cxl_chbs_context' share the property that the first
+ * struct member is cxl_test device being probed by the cxl_acpi
  * driver.
  */
 struct cxl_cedt_context {
@@ -340,7 +342,7 @@ static int mock_acpi_table_parse_cedt(enum acpi_cedt_type id,
        unsigned long end;
        int i;
 
-       if (dev != &cxl_acpi->dev)
+       if (!is_mock_port(dev) && !is_mock_dev(dev))
                return acpi_table_parse_cedt(id, handler_arg, arg);
 
        if (id == ACPI_CEDT_TYPE_CHBS)


> The commit msg seems needlessly vague. Why not state what was done?
> Something like: 'Accept any topology device in cxl_chbs_context'

I do start off with : "Accept any cxl_test topology device as the first
argument in cxl_chbs_context", but I can move the rest to its own
paragraph.

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Dan Williams]

Alison Schofield wrote:
> On Thu, Nov 24, 2022 at 10:35:38AM -0800, Dan Williams wrote:
> > In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> > the represents the memory expander. Unlike a VH topology there is no
> > CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> > as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> > host-bridge as a dport, per usual, but then that dport directly hosts
> > the endpoint port.
> > 
> > Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> > device instance as its immediate child.
> > 
> 
> Reviewed-by: Alison Schofield <alison.schofield@intel.com>
> 
> How can this host bridge and device be used? Should I be looking in
> the spec to see the limitations on a 1.1 usage?

Yes, CXL 3.0 9.11.8 "CXL Devices Attached to an RCH" is the best
starting point.

> Expect it to behave like VH topo? Expect graceful failure where it
> doesn't?  I'm just after a starting point for understanding how the
> 1.1 world fits in.

The expectation is that an RCH topology is a strict subset of a VH
topology. I think of it as VH minus device-interleaving and hotplug.
Once you have found the CXL component register block the enabling can be
shared between 1.1 and 2.0. So, for example see all the different
component registers listed in CXL 3.0 Table 8-22. Where that block is
available to "D1" and "D2" the driver enabling can be shared. The only
difference is how to find those component registers.

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Dan Williams]

Alison Schofield wrote:
> On Thu, Nov 24, 2022 at 10:35:38AM -0800, Dan Williams wrote:
> > In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> > the represents the memory expander. Unlike a VH topology there is no
> > CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> > as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> > host-bridge as a dport, per usual, but then that dport directly hosts
> > the endpoint port.
> > 
> > Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> > device instance as its immediate child.
> > 
> 
> Reviewed-by: Alison Schofield <alison.schofield@intel.com>
> 
> How can this host bridge and device be used?

Answering the direct question... it's not good for much more than
testing enumeration. The expectation is that RCH hosts will be
configured by BIOS and most likely Linux driver only ever needs to read
the configuration, not change it.  So most of the excitement from a
cxl_test perspective is in the enumeration. The rest of the RCH enabling
will be for error handling for errors that impact regions set up by
BIOS. That testing will need hardware, or QEMU, but I do not expect RCH
topologies to show up in QEMU any time soon, if ever.

Re: [PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Robert Richter]

On 28.11.22 13:58:55, Dan Williams wrote:
> Robert Richter wrote:
> > On 24.11.22 10:35:21, Dan Williams wrote:
> > > From: Robert Richter <rrichter@amd.com>
> > > 
> > > A downstream port must be connected to a component register block.
> > > For restricted hosts the base address is determined from the RCRB. The
> > > RCRB is provided by the host's CEDT CHBS entry. Rework CEDT parser to
> > > get the RCRB and add code to extract the component register block from
> > > it.
> > > 
> > > RCRB's BAR[0..1] point to the component block containing CXL subsystem
> > > component registers. MEMBAR extraction follows the PCI base spec here,
> > > esp. 64 bit extraction and memory range alignment (6.0, 7.5.1.2.1). The
> > > RCRB base address is cached in the cxl_dport per-host bridge so that the
> > > upstream port component registers can be retrieved later by an RCD
> > > (RCIEP) associated with the host bridge.
> > > 
> > > Note: Right now the component register block is used for HDM decoder
> > > capability only which is optional for RCDs. If unsupported by the RCD,
> > > the HDM init will fail. It is future work to bypass it in this case.
> > > 
> > > Co-developed-by: Terry Bowman <terry.bowman@amd.com>
> > > Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> > > Signed-off-by: Robert Richter <rrichter@amd.com>
> > > [djbw: introduce devm_cxl_add_rch_dport()]
> > > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > > ---
> > >  drivers/cxl/acpi.c            |   54 ++++++++++++++++++++++++++++++++--------
> > >  drivers/cxl/core/port.c       |   42 +++++++++++++++++++++++++++----
> > >  drivers/cxl/core/regs.c       |   56 +++++++++++++++++++++++++++++++++++++++++
> > >  drivers/cxl/cxl.h             |   16 ++++++++++++
> > >  tools/testing/cxl/Kbuild      |    1 +
> > >  tools/testing/cxl/test/cxl.c  |   10 +++++++
> > >  tools/testing/cxl/test/mock.c |   19 ++++++++++++++
> > >  tools/testing/cxl/test/mock.h |    3 ++
> > >  8 files changed, 186 insertions(+), 15 deletions(-)
> > > 
> > > diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
> > > index 50d82376097c..1224add13529 100644
> > > --- a/drivers/cxl/acpi.c
> > > +++ b/drivers/cxl/acpi.c
> > > @@ -9,6 +9,8 @@
> > >  #include "cxlpci.h"
> > >  #include "cxl.h"
> > >  
> > > +#define CXL_RCRB_SIZE	SZ_8K
> > > +
> > >  static unsigned long cfmws_to_decoder_flags(int restrictions)
> > >  {
> > >  	unsigned long flags = CXL_DECODER_F_ENABLE;
> > > @@ -215,6 +217,11 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> > >  	if (rc)
> > >  		return rc;
> > >  
> > > +	if (dport->rch) {
> > > +		dev_info(bridge, "host supports CXL (restricted)\n");
> > > +		return 0;
> > > +	}
> > 
> > This change comes after devm_cxl_register_pci_bus() to serve the
> > cxl_port_to_pci_bus() in devm_cxl_port_enumerate_dports() in
> > cxl_port_probe(). A root port is not probed and
> > devm_cxl_port_enumerate_dports() will be never called, so we could
> > jump out before devm_cxl_register_pci_bus().
> 
> Good point.
> 
> > On the other side we might want to be ready to use
> > cxl_port_to_pci_bus() elsewhere in later changes. RCHs would not work
> > then.
> 
> At least cxl_port_to_pci_bus() as is will work fine for the RCD
> endpoint-cxl-port, so I think it is ok to leave this alone for now.

Yes, should work as is.

> 
> > 
> > > +
> > >  	port = devm_cxl_add_port(host, bridge, dport->component_reg_phys,
> > >  				 dport);
> > >  	if (IS_ERR(port))
> > > @@ -228,27 +235,46 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> > >  struct cxl_chbs_context {
> > >  	struct device *dev;
> > >  	unsigned long long uid;
> > > -	resource_size_t chbcr;
> > > +	struct acpi_cedt_chbs chbs;
> > >  };
> > >  
> > > -static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
> > > -			 const unsigned long end)
> > > +static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
> > > +			const unsigned long end)
> > >  {
> > >  	struct cxl_chbs_context *ctx = arg;
> > >  	struct acpi_cedt_chbs *chbs;
> > >  
> > > -	if (ctx->chbcr)
> > > +	if (ctx->chbs.base)
> > >  		return 0;
> > >  
> > >  	chbs = (struct acpi_cedt_chbs *) header;
> > >  
> > >  	if (ctx->uid != chbs->uid)
> > >  		return 0;
> > > -	ctx->chbcr = chbs->base;
> > > +	ctx->chbs = *chbs;
> > >  
> > >  	return 0;
> > >  }
> > >  
> > > +static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
> > > +{
> > > +	struct acpi_cedt_chbs *chbs = &ctx->chbs;
> > > +
> > > +	if (!chbs->base)
> > > +		return CXL_RESOURCE_NONE;
> > > +
> > > +	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
> > > +		return chbs->base;
> > > +
> > > +	if (chbs->length != CXL_RCRB_SIZE)
> > > +		return CXL_RESOURCE_NONE;
> > > +
> > > +	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
> > > +		&chbs->base);
> > > +
> > > +	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
> > > +}
> > > +
> > 
> > I have an improved version of this code which squashes cxl_get_chbcr()
> > into cxl_get_chbs() (basically extends the original cxl_get_chbcr()
> > function).
> 
> Care to send it? If I see it before the next posting I can fold it in,
> otherwise it can be a follow-on cleanup.

Yes, will try to send tomorrow. Need to rework, still doing some
testing.

> 
> > 
> > >  static int add_host_bridge_dport(struct device *match, void *arg)
> > >  {
> > >  	acpi_status status;
> > > @@ -258,6 +284,7 @@ static int add_host_bridge_dport(struct device *match, void *arg)
> > >  	struct cxl_chbs_context ctx;
> > >  	struct acpi_pci_root *pci_root;
> > >  	struct cxl_port *root_port = arg;
> > > +	resource_size_t component_reg_phys;
> > >  	struct device *host = root_port->dev.parent;
> > >  	struct acpi_device *hb = to_cxl_host_bridge(host, match);
> > >  
> > > @@ -274,21 +301,28 @@ static int add_host_bridge_dport(struct device *match, void *arg)
> > >  	dev_dbg(match, "UID found: %lld\n", uid);
> > >  
> > >  	ctx = (struct cxl_chbs_context) {
> > > -		.dev = host,
> > > +		.dev = match,
> > >  		.uid = uid,
> > >  	};
> > > -	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbcr, &ctx);
> > > +	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbs, &ctx);
> > >  
> > > -	if (ctx.chbcr == 0) {
> > > +	component_reg_phys = cxl_get_chbcr(&ctx);
> > > +	if (component_reg_phys == CXL_RESOURCE_NONE) {
> > >  		dev_warn(match, "No CHBS found for Host Bridge (UID %lld)\n", uid);
> > >  		return 0;
> > >  	}
> > >  
> > > -	dev_dbg(match, "CHBCR found: 0x%08llx\n", (u64)ctx.chbcr);
> > > +	dev_dbg(match, "CHBCR found: %pa\n", &component_reg_phys);
> > >  
> > >  	pci_root = acpi_pci_find_root(hb->handle);
> > >  	bridge = pci_root->bus->bridge;
> > > -	dport = devm_cxl_add_dport(root_port, bridge, uid, ctx.chbcr);
> > > +	if (ctx.chbs.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
> > > +		dport = devm_cxl_add_rch_dport(root_port, bridge, uid,
> > > +					       component_reg_phys,
> > > +					       ctx.chbs.base);
> > 
> > Yes, this new function makes the rcrb handling much more simpler.
> > 
> > > +	else
> > > +		dport = devm_cxl_add_dport(root_port, bridge, uid,
> > > +					   component_reg_phys);
> > >  	if (IS_ERR(dport))
> > >  		return PTR_ERR(dport);
> > >  
> > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > index d225267c69bb..d9fe06e1462f 100644
> > > --- a/drivers/cxl/core/port.c
> > > +++ b/drivers/cxl/core/port.c
> > > @@ -628,6 +628,8 @@ static struct cxl_port *cxl_port_alloc(struct device *uport,
> > >  			iter = to_cxl_port(iter->dev.parent);
> > >  		if (iter->host_bridge)
> > >  			port->host_bridge = iter->host_bridge;
> > > +		else if (parent_dport->rch)
> > > +			port->host_bridge = parent_dport->dport;
> > 
> > Yes, looks good. This makes the endpoint a child of a CXL root port,
> > not the ACPI0017 the root device.
> > 
> > >  		else
> > >  			port->host_bridge = iter->uport;
> > >  		dev_dbg(uport, "host-bridge: %s\n", dev_name(port->host_bridge));
> > > @@ -899,10 +901,15 @@ static void cxl_dport_unlink(void *data)
> > >  	sysfs_remove_link(&port->dev.kobj, link_name);
> > >  }
> > >  
> > > -static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
> > > -					      struct device *dport_dev,
> > > -					      int port_id,
> > > -					      resource_size_t component_reg_phys)
> > > +enum cxl_dport_mode {
> > > +	CXL_DPORT_VH,
> > > +	CXL_DPORT_RCH,
> > > +};
> > > +
> > > +static struct cxl_dport *
> > > +__devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
> > > +		     int port_id, resource_size_t component_reg_phys,
> > > +		     enum cxl_dport_mode mode, resource_size_t rcrb)
> > >  {
> > >  	char link_name[CXL_TARGET_STRLEN];
> > >  	struct cxl_dport *dport;
> > > @@ -932,6 +939,9 @@ static struct cxl_dport *__devm_cxl_add_dport(struct cxl_port *port,
> > >  	dport->port_id = port_id;
> > >  	dport->component_reg_phys = component_reg_phys;
> > >  	dport->port = port;
> > > +	if (mode == CXL_DPORT_RCH)
> > > +		dport->rch = true;
> > 
> > Alternatively an inline function could be added which checks
> > dport->rcrb for a valid address.
> 
> I like it. Especially because a valid RCRB is needed to register the
> dport in the first instance. I think it looks ok without an inline
> function:
> 
> @@ -901,15 +901,10 @@ static void cxl_dport_unlink(void *data)
>         sysfs_remove_link(&port->dev.kobj, link_name);
>  }
>  
> -enum cxl_dport_mode {
> -       CXL_DPORT_VH,
> -       CXL_DPORT_RCH,
> -};
> -
>  static struct cxl_dport *
>  __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
>                      int port_id, resource_size_t component_reg_phys,
> -                    enum cxl_dport_mode mode, resource_size_t rcrb)
> +                    resource_size_t rcrb)
>  {
>         char link_name[CXL_TARGET_STRLEN];
>         struct cxl_dport *dport;
> @@ -939,7 +934,7 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
>         dport->port_id = port_id;
>         dport->component_reg_phys = component_reg_phys;
>         dport->port = port;
> -       if (mode == CXL_DPORT_RCH)
> +       if (rcrb != CXL_RESOURCE_NONE)
>                 dport->rch = true;
>         dport->rcrb = rcrb;

Change looks good.

I was even thinking to even use ->rcrb to check for rch:

static inline bool is_rch(struct cxl_dport *dport)
{
	return (dport->rcrb != CXL_RESOURCE_NONE);
}

But "if (dport->rch) ..." looks nice too, esp. with its counterpart
cxlds->rcd.

>  
> 
> > 
> > > +	dport->rcrb = rcrb;
> > >  
> > >  	cond_cxl_root_lock(port);
> > >  	rc = add_dport(port, dport);
> > > @@ -973,7 +983,8 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> > >  	struct cxl_dport *dport;
> > >  
> > >  	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> > > -				     component_reg_phys);
> > > +				     component_reg_phys, CXL_DPORT_VH,
> > > +				     CXL_RESOURCE_NONE);
> > >  	if (IS_ERR(dport)) {
> > >  		dev_dbg(dport_dev, "failed to add dport to %s: %ld\n",
> > >  			dev_name(&port->dev), PTR_ERR(dport));
> > > @@ -986,6 +997,27 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
> > >  }
> > >  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
> > >  
> > > +struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
> > > +					 struct device *dport_dev, int port_id,
> > > +					 resource_size_t component_reg_phys,
> > > +					 resource_size_t rcrb)
> > 
> > The documentation header is missing for that.
> 
> Added the following and clarified that @rcrb is mandatory:
> 
> @@ -966,7 +961,7 @@ __devm_cxl_add_dport(struct cxl_port *port, struct device *dport_dev,
>  }
>  
>  /**
> - * devm_cxl_add_dport - append downstream port data to a cxl_port
> + * devm_cxl_add_dport - append VH downstream port data to a cxl_port
>   * @port: the cxl_port that references this dport
>   * @dport_dev: firmware or PCI device representing the dport
>   * @port_id: identifier for this dport in a decoder's target list
> @@ -997,6 +991,16 @@ struct cxl_dport *devm_cxl_add_dport(struct cxl_port *port,
>  }
>  EXPORT_SYMBOL_NS_GPL(devm_cxl_add_dport, CXL);
>  
> +/**
> + * devm_cxl_add_rch_dport - append RCH downstream port data to a cxl_port
> + * @port: the cxl_port that references this dport
> + * @dport_dev: firmware or PCI device representing the dport
> + * @port_id: identifier for this dport in a decoder's target list
> + * @component_reg_phys: optional location of CXL component registers
> + * @rcrb: mandatory location of a Root Complex Register Block
> + *
> + * See CXL 3.0 9.11.8 CXL Devices Attached to an RCH
> + */

Looks good. The component_reg_phys arg could be used for Register
Locator DVSEC in case of a VH device layout (9.11.8).

>  struct cxl_dport *devm_cxl_add_rch_dport(struct cxl_port *port,
>                                          struct device *dport_dev, int port_id,
>                                          resource_size_t component_reg_phys,
> 
> > 
> > > +{
> > > +	struct cxl_dport *dport;
> > > +
> > > +	dport = __devm_cxl_add_dport(port, dport_dev, port_id,
> > > +				     component_reg_phys, CXL_DPORT_RCH, rcrb);
> > > +	if (IS_ERR(dport)) {
> > > +		dev_dbg(dport_dev, "failed to add RCH dport to %s: %ld\n",
> > > +			dev_name(&port->dev), PTR_ERR(dport));
> > > +	} else {
> > > +		dev_dbg(dport_dev, "RCH dport added to %s\n",
> > > +			dev_name(&port->dev));
> > > +	}
> > > +
> > > +	return dport;
> > > +}
> > > +EXPORT_SYMBOL_NS_GPL(devm_cxl_add_rch_dport, CXL);
> > > +
> > >  static int add_ep(struct cxl_ep *new)
> > >  {
> > >  	struct cxl_port *port = new->dport->port;
> > > diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c
> > > index ec178e69b18f..7c2a85dc4125 100644
> > > --- a/drivers/cxl/core/regs.c
> > > +++ b/drivers/cxl/core/regs.c
> > > @@ -307,3 +307,59 @@ int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type,
> > >  	return -ENODEV;
> > >  }
> > >  EXPORT_SYMBOL_NS_GPL(cxl_find_regblock, CXL);
> > > +
> > > +resource_size_t cxl_rcrb_to_component(struct device *dev,
> > > +				      resource_size_t rcrb,
> > > +				      enum cxl_rcrb which)
> > > +{
> > > +	resource_size_t component_reg_phys;
> > > +	u32 bar0, bar1;
> > > +	void *addr;
> > > +	u16 cmd;
> > > +
> > > +	if (which == CXL_RCRB_UPSTREAM)
> > > +		rcrb += SZ_4K;
> > > +
> > > +	/*
> > > +	 * RCRB's BAR[0..1] point to component block containing CXL
> > > +	 * subsystem component registers. MEMBAR extraction follows
> > > +	 * the PCI Base spec here, esp. 64 bit extraction and memory
> > > +	 * ranges alignment (6.0, 7.5.1.2.1).
> > > +	 */
> > > +	if (!request_mem_region(rcrb, SZ_4K, "CXL RCRB"))
> > > +		return CXL_RESOURCE_NONE;
> > > +	addr = ioremap(rcrb, SZ_4K);
> > > +	if (!addr) {
> > > +		dev_err(dev, "Failed to map region %pr\n", addr);
> > > +		release_mem_region(rcrb, SZ_4K);
> > > +		return CXL_RESOURCE_NONE;
> > > +	}
> > > +
> > > +	cmd = readw(addr + PCI_COMMAND);
> > > +	bar0 = readl(addr + PCI_BASE_ADDRESS_0);
> > > +	bar1 = readl(addr + PCI_BASE_ADDRESS_1);
> > > +	iounmap(addr);
> > > +	release_mem_region(rcrb, SZ_4K);
> > > +
> > > +	/* sanity check */
> > > +	if (cmd == 0xffff)
> > > +		return CXL_RESOURCE_NONE;
> > 
> > The spec says offset 0 should be checked (32 bit) which is always
> > non-FF if implemented. This requires another read.
> > 
> > cmd is most of the cases also non-zero, so probably checking cmd
> > instead will have the same effect. Still worth changing that.
> > 
> > If the downstream port's rcrb is all FFs, it is a FW bug. Could be
> > worth a message.
> 
> Ok, makes sense, added:
> 
> @@ -335,15 +336,22 @@ resource_size_t cxl_rcrb_to_component(struct device *dev,
>                 return CXL_RESOURCE_NONE;
>         }
>  
> +       id = readl(addr + PCI_VENDOR_ID);
>         cmd = readw(addr + PCI_COMMAND);
>         bar0 = readl(addr + PCI_BASE_ADDRESS_0);
>         bar1 = readl(addr + PCI_BASE_ADDRESS_1);
>         iounmap(addr);
>         release_mem_region(rcrb, SZ_4K);
>  
> -       /* sanity check */
> -       if (cmd == 0xffff)
> +       /*
> +        * Sanity check, see CXL 3.0 Figure 9-8 CXL Device that Does Not
> +        * Remap Upstream Port and Component Registers
> +        */
> +       if (id == (u32) -1) {

U32_MAX? Or, cheating there: ((u32)~0U).

> +               if (which == CXL_RCRB_DOWNSTREAM)
> +                       dev_err(dev, "Failed to access Downstream Port RCRB\n");
>                 return CXL_RESOURCE_NONE;
> +       }
>         if ((cmd & PCI_COMMAND_MEMORY) == 0)
>                 return CXL_RESOURCE_NONE;
>         if (bar0 & (PCI_BASE_ADDRESS_MEM_TYPE_1M | PCI_BASE_ADDRESS_SPACE_IO))
> 

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Robert Richter]

On 24.11.22 10:35:32, Dan Williams wrote:
> Unlike a CXL memory expander in a VH topology that has at least one
> intervening 'struct cxl_port' instance between itself and the CXL root
> device, an RCD attaches one-level higher. For example:
> 
>                VH
>           ┌──────────┐
>           │ ACPI0017 │
>           │  root0   │
>           └─────┬────┘
>                 │
>           ┌─────┴────┐
>           │  dport0  │
>     ┌─────┤ ACPI0016 ├─────┐
>     │     │  port1   │     │
>     │     └────┬─────┘     │
>     │          │           │
>  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
>  │dport0│   │dport1│   │dport2│
>  │ RP0  │   │ RP1  │   │ RP2  │
>  └──────┘   └──┬───┘   └──────┘
>                │
>            ┌───┴─────┐
>            │endpoint0│
>            │  port2  │
>            └─────────┘
> 
> ...vs:
> 
>               RCH
>           ┌──────────┐
>           │ ACPI0017 │
>           │  root0   │
>           └────┬─────┘
>                │
>            ┌───┴────┐
>            │ dport0 │
>            │ACPI0016│
>            └───┬────┘
>                │
>           ┌────┴─────┐
>           │endpoint0 │
>           │  port1   │
>           └──────────┘
> 
> So arrange for endpoint port in the RCH/RCD case to appear directly
> connected to the host-bridge in its singular role as a dport. Compare
> that to the VH case where the host-bridge serves a dual role as a
> 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> instances in the CXL topology.
> 
> Another deviation from the VH case is that RCDs may need to look up
> their component registers from the Root Complex Register Block (RCRB).
> That platform firmware specified RCRB area is cached by the cxl_acpi
> driver and conveyed via the host-bridge dport to the cxl_mem driver to
> perform the cxl_rcrb_to_component() lookup for the endpoint port
> (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> upstream port component registers).
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  drivers/cxl/core/port.c |   11 +++++++++--
>  drivers/cxl/cxlmem.h    |    2 ++
>  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
>  drivers/cxl/pci.c       |   10 ++++++++++
>  4 files changed, 45 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> index c7f58282b2c1..2385ee00eb9a 100644
> --- a/drivers/cxl/core/port.c
> +++ b/drivers/cxl/core/port.c
> @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
>  {
>  	struct device *dev = &cxlmd->dev;
>  	struct device *iter;
> +	struct cxl_dport *dport;
> +	struct cxl_port *port;

There is no direct need to move that code here.

If you want to clean that up in this patch too, then leave a comment
in the change log?

>  	int rc;
>  
> +	/*
> +	 * Skip intermediate port enumeration in the RCH case, there
> +	 * are no ports in between a host bridge and an endpoint.
> +	 */
> +	if (cxlmd->cxlds->rcd)
> +		return 0;
> +
>  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
>  	if (rc)
>  		return rc;
> @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
>  	for (iter = dev; iter; iter = grandparent(iter)) {
>  		struct device *dport_dev = grandparent(iter);
>  		struct device *uport_dev;
> -		struct cxl_dport *dport;
> -		struct cxl_port *port;
>  
>  		if (!dport_dev)
>  			return 0;
> diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> index e082991bc58c..35d485d041f0 100644
> --- a/drivers/cxl/cxlmem.h
> +++ b/drivers/cxl/cxlmem.h
> @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
>   * @dev: The device associated with this CXL state
>   * @regs: Parsed register blocks
>   * @cxl_dvsec: Offset to the PCIe device DVSEC
> + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
>   * @payload_size: Size of space for payload
>   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
>   * @lsa_size: Size of Label Storage Area
> @@ -235,6 +236,7 @@ struct cxl_dev_state {
>  	struct cxl_regs regs;
>  	int cxl_dvsec;
>  
> +	bool rcd;
>  	size_t payload_size;
>  	size_t lsa_size;
>  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> index aa63ce8c7ca6..9a655b4b5e52 100644
> --- a/drivers/cxl/mem.c
> +++ b/drivers/cxl/mem.c
> @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
>  	return 0;
>  }
>  
> -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
>  				 struct cxl_dport *parent_dport)
>  {
>  	struct cxl_port *parent_port = parent_dport->port;
>  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
>  	struct cxl_port *endpoint, *iter, *down;
> +	resource_size_t component_reg_phys;
>  	int rc;
>  
>  	/*
> @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
>  		ep->next = down;
>  	}
>  
> -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> -				     cxlds->component_reg_phys, parent_dport);
> +	/*
> +	 * The component registers for an RCD might come from the
> +	 * host-bridge RCRB if they are not already mapped via the
> +	 * typical register locator mechanism.
> +	 */
> +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> +		component_reg_phys = cxl_rcrb_to_component(
> +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);

As already commented: this must be the upstream RCRB here.

> +	else
> +		component_reg_phys = cxlds->component_reg_phys;
> +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> +				     parent_dport);

Looking at CXL 3.0 spec, table 8-22, there are the various sources of
component registers listed. For RCD we need: D1, DP1, UP1 (optional
R).

	D1:	endpoint->component_reg_phys;
	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
	DP1:	parent_dport->component_reg_phys;

I don't see how all of them could be stored in this data layout as the
cxl host port is missing.

>  	if (IS_ERR(endpoint))
>  		return PTR_ERR(endpoint);
>  
> @@ -87,6 +98,7 @@ static int cxl_mem_probe(struct device *dev)
>  {
>  	struct cxl_memdev *cxlmd = to_cxl_memdev(dev);
>  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> +	struct device *endpoint_parent;
>  	struct cxl_port *parent_port;
>  	struct cxl_dport *dport;
>  	struct dentry *dentry;
> @@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
>  		return -ENXIO;
>  	}
>  
> -	device_lock(&parent_port->dev);
> -	if (!parent_port->dev.driver) {
> +	if (dport->rch)
> +		endpoint_parent = parent_port->uport;
> +	else
> +		endpoint_parent = &parent_port->dev;
> +
> +	device_lock(endpoint_parent);
> +	if (!endpoint_parent->driver) {
>  		dev_err(dev, "CXL port topology %s not enabled\n",
>  			dev_name(&parent_port->dev));
>  		rc = -ENXIO;
>  		goto unlock;
>  	}
>  
> -	rc = devm_cxl_add_endpoint(cxlmd, dport);
> +	rc = devm_cxl_add_endpoint(endpoint_parent, cxlmd, dport);
>  unlock:
> -	device_unlock(&parent_port->dev);
> +	device_unlock(endpoint_parent);
>  	put_device(&parent_port->dev);
>  	if (rc)
>  		return rc;
> diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
> index e15da405b948..73ff6c33a0c0 100644
> --- a/drivers/cxl/pci.c
> +++ b/drivers/cxl/pci.c
> @@ -433,6 +433,15 @@ static void devm_cxl_pci_create_doe(struct cxl_dev_state *cxlds)
>  	}
>  }
>  
> +/*
> + * Assume that any RCIEP that emits the CXL memory expander class code
> + * is an RCD
> + */
> +static bool is_cxl_restricted(struct pci_dev *pdev)
> +{
> +	return pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END;
> +}
> +
>  static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>  {
>  	struct cxl_register_map map;
> @@ -455,6 +464,7 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>  	if (IS_ERR(cxlds))
>  		return PTR_ERR(cxlds);
>  
> +	cxlds->rcd = is_cxl_restricted(pdev);
>  	cxlds->serial = pci_get_dsn(pdev);
>  	cxlds->cxl_dvsec = pci_find_dvsec_capability(
>  		pdev, PCI_DVSEC_VENDOR_ID_CXL, CXL_DVSEC_PCIE_DEVICE);
> 

Re: [PATCH v4 09/12] cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:27, Dan Williams wrote:
> > In preparation for devm_cxl_add_endpoint() to call out to
> > cxl_rcrb_to_component() to map the upstream port component registers,
> > move devm_cxl_add_endpoint() from the cxl_core to the cxl_mem driver.
> > This is due to the organization of cxl_test that mandates that the
> > cxl_core not call out to any mocked symbols. It also cleans up the
> > export of devm_cxl_add_endpoint() which is just a wrapper around
> > devm_cxl_add_port().
> 
> The last sentense is unclear to me? I see you just make it static as
> there is only a single user, do you mean that here? The change itself
> looks good.

Yeah, it does read strangely. I changed it to:

    cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem
    
    tl;dr: Clean up an unnecessary export and enable cxl_test.
    
    An RCD (Restricted CXL Device), in contrast to a typical CXL device in
    a VH topology, obtains its component registers from the bottom half of
    the associated CXL host bridge RCRB (Root Complex Register Block). In
    turn this means that cxl_rcrb_to_component() needs to be called from
    devm_cxl_add_endpoint().
    
    Presently devm_cxl_add_endpoint() is part of the CXL core, but the only
    user is the CXL mem module. Move it from cxl_core to cxl_mem to not only
    get rid of an unnecessary export, but to also enable its call out to
    cxl_rcrb_to_component(), in a subsequent patch, to be mocked by
    cxl_test. Recall that cxl_test can only mock exported symbols, and since
    cxl_rcrb_to_component() is itself inside the core, all callers must be
    outside of cxl_core to allow cxl_test to mock it.

> Reviewed-by: Robert Richter <rrichter@amd.com>

Re: [PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Dan Williams]

Robert Richter wrote:
> On 28.11.22 13:58:55, Dan Williams wrote:
> > @@ -335,15 +336,22 @@ resource_size_t cxl_rcrb_to_component(struct device *dev,
> >                 return CXL_RESOURCE_NONE;
> >         }
> >  
> > +       id = readl(addr + PCI_VENDOR_ID);
> >         cmd = readw(addr + PCI_COMMAND);
> >         bar0 = readl(addr + PCI_BASE_ADDRESS_0);
> >         bar1 = readl(addr + PCI_BASE_ADDRESS_1);
> >         iounmap(addr);
> >         release_mem_region(rcrb, SZ_4K);
> >  
> > -       /* sanity check */
> > -       if (cmd == 0xffff)
> > +       /*
> > +        * Sanity check, see CXL 3.0 Figure 9-8 CXL Device that Does Not
> > +        * Remap Upstream Port and Component Registers
> > +        */
> > +       if (id == (u32) -1) {
> 
> U32_MAX? Or, cheating there: ((u32)~0U).
> 

U32_MAX looks good to me.

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:32, Dan Williams wrote:
> > Unlike a CXL memory expander in a VH topology that has at least one
> > intervening 'struct cxl_port' instance between itself and the CXL root
> > device, an RCD attaches one-level higher. For example:
> > 
> >                VH
> >           ┌──────────┐
> >           │ ACPI0017 │
> >           │  root0   │
> >           └─────┬────┘
> >                 │
> >           ┌─────┴────┐
> >           │  dport0  │
> >     ┌─────┤ ACPI0016 ├─────┐
> >     │     │  port1   │     │
> >     │     └────┬─────┘     │
> >     │          │           │
> >  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> >  │dport0│   │dport1│   │dport2│
> >  │ RP0  │   │ RP1  │   │ RP2  │
> >  └──────┘   └──┬───┘   └──────┘
> >                │
> >            ┌───┴─────┐
> >            │endpoint0│
> >            │  port2  │
> >            └─────────┘
> > 
> > ...vs:
> > 
> >               RCH
> >           ┌──────────┐
> >           │ ACPI0017 │
> >           │  root0   │
> >           └────┬─────┘
> >                │
> >            ┌───┴────┐
> >            │ dport0 │
> >            │ACPI0016│
> >            └───┬────┘
> >                │
> >           ┌────┴─────┐
> >           │endpoint0 │
> >           │  port1   │
> >           └──────────┘
> > 
> > So arrange for endpoint port in the RCH/RCD case to appear directly
> > connected to the host-bridge in its singular role as a dport. Compare
> > that to the VH case where the host-bridge serves a dual role as a
> > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > instances in the CXL topology.
> > 
> > Another deviation from the VH case is that RCDs may need to look up
> > their component registers from the Root Complex Register Block (RCRB).
> > That platform firmware specified RCRB area is cached by the cxl_acpi
> > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > upstream port component registers).
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >  drivers/cxl/core/port.c |   11 +++++++++--
> >  drivers/cxl/cxlmem.h    |    2 ++
> >  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> >  drivers/cxl/pci.c       |   10 ++++++++++
> >  4 files changed, 45 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > index c7f58282b2c1..2385ee00eb9a 100644
> > --- a/drivers/cxl/core/port.c
> > +++ b/drivers/cxl/core/port.c
> > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> >  {
> >  	struct device *dev = &cxlmd->dev;
> >  	struct device *iter;
> > +	struct cxl_dport *dport;
> > +	struct cxl_port *port;
> 
> There is no direct need to move that code here.
> 
> If you want to clean that up in this patch too, then leave a comment
> in the change log?

Oh, good point, must have been left over from an earlier revision of the
patch, dropped it.

> 
> >  	int rc;
> >  
> > +	/*
> > +	 * Skip intermediate port enumeration in the RCH case, there
> > +	 * are no ports in between a host bridge and an endpoint.
> > +	 */
> > +	if (cxlmd->cxlds->rcd)
> > +		return 0;
> > +
> >  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> >  	if (rc)
> >  		return rc;
> > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> >  	for (iter = dev; iter; iter = grandparent(iter)) {
> >  		struct device *dport_dev = grandparent(iter);
> >  		struct device *uport_dev;
> > -		struct cxl_dport *dport;
> > -		struct cxl_port *port;
> >  
> >  		if (!dport_dev)
> >  			return 0;
> > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > index e082991bc58c..35d485d041f0 100644
> > --- a/drivers/cxl/cxlmem.h
> > +++ b/drivers/cxl/cxlmem.h
> > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> >   * @dev: The device associated with this CXL state
> >   * @regs: Parsed register blocks
> >   * @cxl_dvsec: Offset to the PCIe device DVSEC
> > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> >   * @payload_size: Size of space for payload
> >   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> >   * @lsa_size: Size of Label Storage Area
> > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> >  	struct cxl_regs regs;
> >  	int cxl_dvsec;
> >  
> > +	bool rcd;
> >  	size_t payload_size;
> >  	size_t lsa_size;
> >  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > index aa63ce8c7ca6..9a655b4b5e52 100644
> > --- a/drivers/cxl/mem.c
> > +++ b/drivers/cxl/mem.c
> > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> >  	return 0;
> >  }
> >  
> > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> >  				 struct cxl_dport *parent_dport)
> >  {
> >  	struct cxl_port *parent_port = parent_dport->port;
> >  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> >  	struct cxl_port *endpoint, *iter, *down;
> > +	resource_size_t component_reg_phys;
> >  	int rc;
> >  
> >  	/*
> > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> >  		ep->next = down;
> >  	}
> >  
> > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > -				     cxlds->component_reg_phys, parent_dport);
> > +	/*
> > +	 * The component registers for an RCD might come from the
> > +	 * host-bridge RCRB if they are not already mapped via the
> > +	 * typical register locator mechanism.
> > +	 */
> > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > +		component_reg_phys = cxl_rcrb_to_component(
> > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> 
> As already commented: this must be the upstream RCRB here.
> 
> > +	else
> > +		component_reg_phys = cxlds->component_reg_phys;
> > +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> > +				     parent_dport);
> 
> Looking at CXL 3.0 spec, table 8-22, there are the various sources of
> component registers listed. For RCD we need: D1, DP1, UP1 (optional
> R).
> 
> 	D1:	endpoint->component_reg_phys;
> 	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
> 	DP1:	parent_dport->component_reg_phys;
> 
> I don't see how all of them could be stored in this data layout as the
> cxl host port is missing.

If I am understanding your concern correctly, that's handled here:

    if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)

In the D1 case cxlds->component_reg_phys will be valid since the
component registers were visible via the register locator DVSEC and
retrieved by cxl_pci. In the UP1 case cxlds->component_reg_phys is
invalid and the driver falls back to the RCRB. DP1 is handled in
cxl_acpi. I.e. the D1 and UP1 cases do not co-exist.

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Jonathan Cameron]

On Mon, 28 Nov 2022 14:41:49 -0800
Dan Williams <dan.j.williams@intel.com> wrote:

> Alison Schofield wrote:
> > On Thu, Nov 24, 2022 at 10:35:38AM -0800, Dan Williams wrote:  
> > > In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> > > the represents the memory expander. Unlike a VH topology there is no
> > > CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> > > as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> > > host-bridge as a dport, per usual, but then that dport directly hosts
> > > the endpoint port.
> > > 
> > > Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> > > device instance as its immediate child.
> > >   
> > 
> > Reviewed-by: Alison Schofield <alison.schofield@intel.com>
> > 
> > How can this host bridge and device be used?  
> 
> Answering the direct question... it's not good for much more than
> testing enumeration. The expectation is that RCH hosts will be
> configured by BIOS and most likely Linux driver only ever needs to read
> the configuration, not change it.  So most of the excitement from a
> cxl_test perspective is in the enumeration. The rest of the RCH enabling
> will be for error handling for errors that impact regions set up by
> BIOS. That testing will need hardware, or QEMU, but I do not expect RCH
> topologies to show up in QEMU any time soon, if ever.

I wasn't planning on QEMU support, but if someone wants to do it I'll be
happy to review!  Patches welcome etc.  As Dan says, it'll be a job for BIOS
configuration and so far we don't have a suitable BIOS to run against
the QEMU emulation - there is code for ARM's own models (FVP) as talked about
at Plumbers CXL UConf, but I don't know of anyone looking at getting that
up and running on top of QEMU yet.  That would be best done first on CXL 2.0+
but once it is place, a full setup for RCH would be possible.

Given there is hardware and new features aren't going to be built on RCH
topologies, I'd guess no one will care enough to write QEMU emulation
any time soon.

Jonathan

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Robert Richter]

On 24.11.22 10:35:38, Dan Williams wrote:
> In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> the represents the memory expander. Unlike a VH topology there is no
> CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> host-bridge as a dport, per usual, but then that dport directly hosts
> the endpoint port.
> 
> Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> device instance as its immediate child.
> 
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> ---
>  tools/testing/cxl/test/cxl.c |  151 +++++++++++++++++++++++++++++++++++++++---
>  tools/testing/cxl/test/mem.c |   37 ++++++++++
>  2 files changed, 176 insertions(+), 12 deletions(-)

One comment below.

> @@ -736,6 +779,87 @@ static void mock_companion(struct acpi_device *adev, struct device *dev)
>  #define SZ_512G (SZ_64G * 8)
>  #endif
>  
> +static __init int cxl_rch_init(void)
> +{
> +	int rc, i;
> +
> +	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++) {
> +		int idx = NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + i;
> +		struct acpi_device *adev = &host_bridge[idx];
> +		struct platform_device *pdev;
> +
> +		pdev = platform_device_alloc("cxl_host_bridge", idx);
> +		if (!pdev)
> +			goto err_bridge;
> +
> +		mock_companion(adev, &pdev->dev);
> +		rc = platform_device_add(pdev);
> +		if (rc) {
> +			platform_device_put(pdev);
> +			goto err_bridge;
> +		}
> +
> +		cxl_rch[i] = pdev;
> +		mock_pci_bus[idx].bridge = &pdev->dev;
> +		rc = sysfs_create_link(&pdev->dev.kobj, &pdev->dev.kobj,
> +				       "firmware_node");
> +		if (rc)
> +			goto err_bridge;
> +	}
> +
> +	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++) {
> +		int idx = NR_MEM_MULTI + NR_MEM_SINGLE + i;
> +		struct platform_device *rch = cxl_rch[i];
> +		struct platform_device *pdev;
> +
> +		pdev = platform_device_alloc("cxl_rcd", idx);
> +		if (!pdev)
> +			goto err_mem;
> +		pdev->dev.parent = &rch->dev;
> +		set_dev_node(&pdev->dev, i % 2);
> +
> +		rc = platform_device_add(pdev);
> +		if (rc) {
> +			platform_device_put(pdev);
> +			goto err_mem;
> +		}
> +		cxl_rcd[i] = pdev;
> +	}
> +
> +	return 0;
> +
> +err_mem:
> +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> +		platform_device_unregister(cxl_rcd[i]);
> +err_bridge:

platform_device_unregister() is safe to be used with NULL, so we can
have a single entry of this unregister code ...

> +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> +		struct platform_device *pdev = cxl_rch[i];
> +
> +		if (!pdev)
> +			continue;
> +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> +		platform_device_unregister(cxl_rch[i]);
> +	}
> +
> +	return rc;
> +}
> +
> +static void cxl_rch_exit(void)
> +{
> +	int i;
> +
> +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> +		platform_device_unregister(cxl_rcd[i]);
> +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> +		struct platform_device *pdev = cxl_rch[i];
> +
> +		if (!pdev)
> +			continue;
> +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> +		platform_device_unregister(cxl_rch[i]);
> +	}
> +}

... and have a single function for both. This reduces code
duplication here.

-Robert

Re: [PATCH v4 09/12] cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem

[Robert Richter]

On 28.11.22 15:42:38, Dan Williams wrote:
> Robert Richter wrote:
> > On 24.11.22 10:35:27, Dan Williams wrote:
> > > In preparation for devm_cxl_add_endpoint() to call out to
> > > cxl_rcrb_to_component() to map the upstream port component registers,
> > > move devm_cxl_add_endpoint() from the cxl_core to the cxl_mem driver.
> > > This is due to the organization of cxl_test that mandates that the
> > > cxl_core not call out to any mocked symbols. It also cleans up the
> > > export of devm_cxl_add_endpoint() which is just a wrapper around
> > > devm_cxl_add_port().
> > 
> > The last sentense is unclear to me? I see you just make it static as
> > there is only a single user, do you mean that here? The change itself
> > looks good.
> 
> Yeah, it does read strangely. I changed it to:
> 
>     cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem
>     
>     tl;dr: Clean up an unnecessary export and enable cxl_test.
>     
>     An RCD (Restricted CXL Device), in contrast to a typical CXL device in
>     a VH topology, obtains its component registers from the bottom half of
>     the associated CXL host bridge RCRB (Root Complex Register Block). In
>     turn this means that cxl_rcrb_to_component() needs to be called from
>     devm_cxl_add_endpoint().
>     
>     Presently devm_cxl_add_endpoint() is part of the CXL core, but the only
>     user is the CXL mem module. Move it from cxl_core to cxl_mem to not only
>     get rid of an unnecessary export, but to also enable its call out to
>     cxl_rcrb_to_component(), in a subsequent patch, to be mocked by
>     cxl_test. Recall that cxl_test can only mock exported symbols, and since
>     cxl_rcrb_to_component() is itself inside the core, all callers must be
>     outside of cxl_core to allow cxl_test to mock it.
> 
> > Reviewed-by: Robert Richter <rrichter@amd.com>

Yep, thanks for changing.

-Robert

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Robert Richter]

On 28.11.22 16:25:18, Dan Williams wrote:
> Robert Richter wrote:
> > On 24.11.22 10:35:32, Dan Williams wrote:
> > > Unlike a CXL memory expander in a VH topology that has at least one
> > > intervening 'struct cxl_port' instance between itself and the CXL root
> > > device, an RCD attaches one-level higher. For example:
> > > 
> > >                VH
> > >           ┌──────────┐
> > >           │ ACPI0017 │
> > >           │  root0   │
> > >           └─────┬────┘
> > >                 │
> > >           ┌─────┴────┐
> > >           │  dport0  │
> > >     ┌─────┤ ACPI0016 ├─────┐
> > >     │     │  port1   │     │
> > >     │     └────┬─────┘     │
> > >     │          │           │
> > >  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> > >  │dport0│   │dport1│   │dport2│
> > >  │ RP0  │   │ RP1  │   │ RP2  │
> > >  └──────┘   └──┬───┘   └──────┘
> > >                │
> > >            ┌───┴─────┐
> > >            │endpoint0│
> > >            │  port2  │
> > >            └─────────┘
> > > 
> > > ...vs:
> > > 
> > >               RCH
> > >           ┌──────────┐
> > >           │ ACPI0017 │
> > >           │  root0   │
> > >           └────┬─────┘
> > >                │
> > >            ┌───┴────┐
> > >            │ dport0 │
> > >            │ACPI0016│
> > >            └───┬────┘
> > >                │
> > >           ┌────┴─────┐
> > >           │endpoint0 │
> > >           │  port1   │
> > >           └──────────┘
> > > 
> > > So arrange for endpoint port in the RCH/RCD case to appear directly
> > > connected to the host-bridge in its singular role as a dport. Compare
> > > that to the VH case where the host-bridge serves a dual role as a
> > > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > > instances in the CXL topology.
> > > 
> > > Another deviation from the VH case is that RCDs may need to look up
> > > their component registers from the Root Complex Register Block (RCRB).
> > > That platform firmware specified RCRB area is cached by the cxl_acpi
> > > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > > upstream port component registers).
> > > 
> > > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > > ---
> > >  drivers/cxl/core/port.c |   11 +++++++++--
> > >  drivers/cxl/cxlmem.h    |    2 ++
> > >  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> > >  drivers/cxl/pci.c       |   10 ++++++++++
> > >  4 files changed, 45 insertions(+), 9 deletions(-)
> > > 
> > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > index c7f58282b2c1..2385ee00eb9a 100644
> > > --- a/drivers/cxl/core/port.c
> > > +++ b/drivers/cxl/core/port.c
> > > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > >  {
> > >  	struct device *dev = &cxlmd->dev;
> > >  	struct device *iter;
> > > +	struct cxl_dport *dport;
> > > +	struct cxl_port *port;
> > 
> > There is no direct need to move that code here.
> > 
> > If you want to clean that up in this patch too, then leave a comment
> > in the change log?
> 
> Oh, good point, must have been left over from an earlier revision of the
> patch, dropped it.
> 
> > 
> > >  	int rc;
> > >  
> > > +	/*
> > > +	 * Skip intermediate port enumeration in the RCH case, there
> > > +	 * are no ports in between a host bridge and an endpoint.
> > > +	 */
> > > +	if (cxlmd->cxlds->rcd)
> > > +		return 0;
> > > +
> > >  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> > >  	if (rc)
> > >  		return rc;
> > > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > >  	for (iter = dev; iter; iter = grandparent(iter)) {
> > >  		struct device *dport_dev = grandparent(iter);
> > >  		struct device *uport_dev;
> > > -		struct cxl_dport *dport;
> > > -		struct cxl_port *port;
> > >  
> > >  		if (!dport_dev)
> > >  			return 0;
> > > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > > index e082991bc58c..35d485d041f0 100644
> > > --- a/drivers/cxl/cxlmem.h
> > > +++ b/drivers/cxl/cxlmem.h
> > > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> > >   * @dev: The device associated with this CXL state
> > >   * @regs: Parsed register blocks
> > >   * @cxl_dvsec: Offset to the PCIe device DVSEC
> > > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> > >   * @payload_size: Size of space for payload
> > >   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> > >   * @lsa_size: Size of Label Storage Area
> > > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> > >  	struct cxl_regs regs;
> > >  	int cxl_dvsec;
> > >  
> > > +	bool rcd;
> > >  	size_t payload_size;
> > >  	size_t lsa_size;
> > >  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > > index aa63ce8c7ca6..9a655b4b5e52 100644
> > > --- a/drivers/cxl/mem.c
> > > +++ b/drivers/cxl/mem.c
> > > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> > >  	return 0;
> > >  }
> > >  
> > > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> > >  				 struct cxl_dport *parent_dport)
> > >  {
> > >  	struct cxl_port *parent_port = parent_dport->port;
> > >  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > >  	struct cxl_port *endpoint, *iter, *down;
> > > +	resource_size_t component_reg_phys;
> > >  	int rc;
> > >  
> > >  	/*
> > > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > >  		ep->next = down;
> > >  	}
> > >  
> > > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > > -				     cxlds->component_reg_phys, parent_dport);
> > > +	/*
> > > +	 * The component registers for an RCD might come from the
> > > +	 * host-bridge RCRB if they are not already mapped via the
> > > +	 * typical register locator mechanism.
> > > +	 */
> > > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > +		component_reg_phys = cxl_rcrb_to_component(
> > > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> > 
> > As already commented: this must be the upstream RCRB here.
> > 
> > > +	else
> > > +		component_reg_phys = cxlds->component_reg_phys;
> > > +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> > > +				     parent_dport);
> > 
> > Looking at CXL 3.0 spec, table 8-22, there are the various sources of
> > component registers listed. For RCD we need: D1, DP1, UP1 (optional
> > R).
> > 
> > 	D1:	endpoint->component_reg_phys;
> > 	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
> > 	DP1:	parent_dport->component_reg_phys;
> > 
> > I don't see how all of them could be stored in this data layout as the
> > cxl host port is missing.
> 
> If I am understanding your concern correctly, that's handled here:
> 
>     if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> 
> In the D1 case cxlds->component_reg_phys will be valid since the
> component registers were visible via the register locator DVSEC and
> retrieved by cxl_pci. In the UP1 case cxlds->component_reg_phys is
> invalid and the driver falls back to the RCRB. DP1 is handled in
> cxl_acpi. I.e. the D1 and UP1 cases do not co-exist.

What I mean is we must store all 3 component reg base addresses for
later access. E.g., if there is an AER error of a pci dev or the host,
we must (depending on the error details) access CXL RAS status of
either D1, UP1 or DP1. So for all 3 of them there must be a way to
determine this walking through the port hierarchy. In the above list
of locations I don't where UP1's component reg base address is stored.

-Robert

Re: [PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[Robert Richter]

Dan,

On 24.11.22 10:34:35, Dan Williams wrote:
> Changes since v3 [1]:
> - Rework / simplify CXL to LIBNVDIMM coordination to remove a
>   flush_work() locking dependency from underneath the root device lock.
> - Move the root device rescan to a workqueue
> - Connect RCDs directly as endpoints reachable through a CXL host bridge
>   as a dport, i.e. drop the extra dport indirection from v3
> - Add unit test infrastructure for an RCD configuration

thank you for this posting.

Patches #1-#6 are not really prerequisites (except for a trivial
conflict), right? I only reviewed them starting with #6.

Thanks,

-Robert

Re: [PATCH v4 12/12] cxl/acpi: Set ACPI's CXL _OSC to indicate CXL1.1 support

[Robert Richter]

On 24.11.22 10:35:43, Dan Williams wrote:
> From: Terry Bowman <terry.bowman@amd.com>
> 
> ACPI includes a CXL _OSC for the OS to communicate what it knows of CXL
> device topologies. To date Linux has added support for CXL 2.0 (VH) port
> topologies, hotplug, and error handling. Now that the driver also know
> how to enumerate CXL 1.1 (RCH) port topologies, indicate that capability
> via CXL _OSC. See CXL3.0 Table 9-26 'Interpretation of CXL _OSC Support
> Field'
> 
> Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> Signed-off-by: Robert Richter <rrichter@amd.com>
> Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> [djbw: wordsmith changelog]
> Signed-off-by: Dan Williams <dan.j.williams@intel.com>

I have had a reworded version of this in the pipe too, esp. version
strings were dropped in favor of VH and RCD mode. You might want to
consider that one:

-- >8 --

From 260e04d5d34c6d37a1866b73a5e229d1ceddf272 Mon Sep 17 00:00:00 2001
From: Terry Bowman <terry.bowman@amd.com>
Date: Mon, 14 Nov 2022 10:03:30 -0600
Subject: [PATCH v5] cxl/acpi: Set ACPI's CXL _OSC to indicate RCD mode support

ACPI uses the CXL _OSC support method to communicate the available CXL
functionality to FW. The CXL _OSC support method includes a field to
indicate the OS is capable of RCD mode. FW can potentially change it's
operation depending on the _OSC support method reported by the OS.

The ACPI driver currently only sets the ACPI _OSC support method to
indicate CXL VH mode. Change the capability reported to also include
CXL RCD mode.

[1] CXL3.0 Table 9-26 'Interpretation of CXL _OSC Support Field'

Signed-off-by: Terry Bowman <terry.bowman@amd.com>
[rrichter@amd.com: Reworded patch description.]
Signed-off-by: Robert Richter <rrichter@amd.com>
---
 drivers/acpi/pci_root.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/acpi/pci_root.c b/drivers/acpi/pci_root.c
index 4e3db20e9cbb..b3c202d2a433 100644
--- a/drivers/acpi/pci_root.c
+++ b/drivers/acpi/pci_root.c
@@ -493,6 +493,7 @@ static u32 calculate_cxl_support(void)
 	u32 support;
 
 	support = OSC_CXL_2_0_PORT_DEV_REG_ACCESS_SUPPORT;
+	support |= OSC_CXL_1_1_PORT_REG_ACCESS_SUPPORT;
 	if (pci_aer_available())
 		support |= OSC_CXL_PROTOCOL_ERR_REPORTING_SUPPORT;
 	if (IS_ENABLED(CONFIG_HOTPLUG_PCI_PCIE))
-- 
2.30.2

Re: [PATCH v4 08/12] cxl/acpi: Extract component registers of restricted hosts from RCRB

[Robert Richter]

On 28.11.22 13:58:55, Dan Williams wrote:
> Robert Richter wrote:
> > On 24.11.22 10:35:21, Dan Williams wrote:
> > > From: Robert Richter <rrichter@amd.com>

> > > @@ -228,27 +235,46 @@ static int add_host_bridge_uport(struct device *match, void *arg)
> > >  struct cxl_chbs_context {
> > >  	struct device *dev;
> > >  	unsigned long long uid;
> > > -	resource_size_t chbcr;
> > > +	struct acpi_cedt_chbs chbs;
> > >  };
> > >  
> > > -static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
> > > -			 const unsigned long end)
> > > +static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
> > > +			const unsigned long end)
> > >  {
> > >  	struct cxl_chbs_context *ctx = arg;
> > >  	struct acpi_cedt_chbs *chbs;
> > >  
> > > -	if (ctx->chbcr)
> > > +	if (ctx->chbs.base)
> > >  		return 0;
> > >  
> > >  	chbs = (struct acpi_cedt_chbs *) header;
> > >  
> > >  	if (ctx->uid != chbs->uid)
> > >  		return 0;
> > > -	ctx->chbcr = chbs->base;
> > > +	ctx->chbs = *chbs;
> > >  
> > >  	return 0;
> > >  }
> > >  
> > > +static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
> > > +{
> > > +	struct acpi_cedt_chbs *chbs = &ctx->chbs;
> > > +
> > > +	if (!chbs->base)
> > > +		return CXL_RESOURCE_NONE;
> > > +
> > > +	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
> > > +		return chbs->base;
> > > +
> > > +	if (chbs->length != CXL_RCRB_SIZE)
> > > +		return CXL_RESOURCE_NONE;
> > > +
> > > +	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
> > > +		&chbs->base);
> > > +
> > > +	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
> > > +}
> > > +
> > 
> > I have an improved version of this code which squashes cxl_get_chbcr()
> > into cxl_get_chbs() (basically extends the original cxl_get_chbcr()
> > function).
> 
> Care to send it? If I see it before the next posting I can fold it in,
> otherwise it can be a follow-on cleanup.

Here comes the delta patch of that change. Since you probably reworked
this patch already I hope the changes will apply cleanly or with small
conflicts only.

-- >8 --

From 65a6e03e53f2298c46d3fe0c16150aa7d539cfca Mon Sep 17 00:00:00 2001
From: Robert Richter <rrichter@amd.com>
Date: Wed, 30 Nov 2022 14:39:08 +0100
Subject: [PATCH v5] delta

Signed-off-by: Robert Richter <rrichter@amd.com>
---
 drivers/cxl/acpi.c | 59 +++++++++++++++++++++++-----------------------
 1 file changed, 30 insertions(+), 29 deletions(-)

diff --git a/drivers/cxl/acpi.c b/drivers/cxl/acpi.c
index f13b702e9fd5..1eb564e697fb 100644
--- a/drivers/cxl/acpi.c
+++ b/drivers/cxl/acpi.c
@@ -233,46 +233,47 @@ static int add_host_bridge_uport(struct device *match, void *arg)
 }
 
 struct cxl_chbs_context {
-	struct device *dev;
-	unsigned long long uid;
-	struct acpi_cedt_chbs chbs;
+	struct device		*dev;
+	unsigned long long	uid;
+	resource_size_t		rcrb;
+	resource_size_t		chbcr;
+	u32			cxl_version;
 };
 
-static int cxl_get_chbs(union acpi_subtable_headers *header, void *arg,
-			const unsigned long end)
+static int cxl_get_chbcr(union acpi_subtable_headers *header, void *arg,
+			 const unsigned long end)
 {
 	struct cxl_chbs_context *ctx = arg;
 	struct acpi_cedt_chbs *chbs;
 
-	if (ctx->chbs.base)
+	if (ctx->chbcr)
 		return 0;
 
 	chbs = (struct acpi_cedt_chbs *) header;
 
 	if (ctx->uid != chbs->uid)
 		return 0;
-	ctx->chbs = *chbs;
-
-	return 0;
-}
 
-static resource_size_t cxl_get_chbcr(struct cxl_chbs_context *ctx)
-{
-	struct acpi_cedt_chbs *chbs = &ctx->chbs;
+	ctx->cxl_version = chbs->cxl_version;
+	ctx->rcrb = CXL_RESOURCE_NONE;
+	ctx->chbcr = CXL_RESOURCE_NONE;
 
 	if (!chbs->base)
-		return CXL_RESOURCE_NONE;
+		return 0;
 
-	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11)
-		return chbs->base;
+	if (chbs->cxl_version != ACPI_CEDT_CHBS_VERSION_CXL11) {
+		ctx->chbcr = chbs->base;
+		return 0;
+	}
 
 	if (chbs->length != CXL_RCRB_SIZE)
-		return CXL_RESOURCE_NONE;
+		return 0;
 
-	dev_dbg(ctx->dev, "RCRB found for UID %lld: %pa\n", ctx->uid,
-		&chbs->base);
+	ctx->rcrb = chbs->base;
+	ctx->chbcr = cxl_rcrb_to_component(ctx->dev, chbs->base,
+					   CXL_RCRB_DOWNSTREAM);
 
-	return cxl_rcrb_to_component(ctx->dev, chbs->base, CXL_RCRB_DOWNSTREAM);
+	return 0;
 }
 
 static int add_host_bridge_dport(struct device *match, void *arg)
@@ -284,7 +285,6 @@ static int add_host_bridge_dport(struct device *match, void *arg)
 	struct cxl_chbs_context ctx;
 	struct acpi_pci_root *pci_root;
 	struct cxl_port *root_port = arg;
-	resource_size_t component_reg_phys;
 	struct device *host = root_port->dev.parent;
 	struct acpi_device *hb = to_cxl_host_bridge(host, match);
 
@@ -304,25 +304,26 @@ static int add_host_bridge_dport(struct device *match, void *arg)
 		.dev = match,
 		.uid = uid,
 	};
-	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbs, &ctx);
+	acpi_table_parse_cedt(ACPI_CEDT_TYPE_CHBS, cxl_get_chbcr, &ctx);
+
+	if (ctx.rcrb != CXL_RESOURCE_NONE)
+		dev_dbg(match, "RCRB found for UID %lld: %pa\n", uid, &ctx.rcrb);
 
-	component_reg_phys = cxl_get_chbcr(&ctx);
-	if (component_reg_phys == CXL_RESOURCE_NONE) {
+	if (ctx.chbcr == CXL_RESOURCE_NONE) {
 		dev_warn(match, "No CHBS found for Host Bridge (UID %lld)\n", uid);
 		return 0;
 	}
 
-	dev_dbg(match, "CHBCR found: %pa\n", &component_reg_phys);
+	dev_dbg(match, "CHBCR found: %pa\n", &ctx.chbcr);
 
 	pci_root = acpi_pci_find_root(hb->handle);
 	bridge = pci_root->bus->bridge;
-	if (ctx.chbs.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
+	if (ctx.cxl_version == ACPI_CEDT_CHBS_VERSION_CXL11)
 		dport = devm_cxl_add_rch_dport(root_port, bridge, uid,
-					       component_reg_phys,
-					       ctx.chbs.base);
+					       ctx.chbcr, ctx.rcrb);
 	else
 		dport = devm_cxl_add_dport(root_port, bridge, uid,
-					   component_reg_phys);
+					   ctx.chbcr);
 	if (IS_ERR(dport))
 		return PTR_ERR(dport);
 
-- 
2.30.2

Re: [PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[Robert Richter]

On 24.11.22 10:34:35, Dan Williams wrote:
> Changes since v3 [1]:
> - Rework / simplify CXL to LIBNVDIMM coordination to remove a
>   flush_work() locking dependency from underneath the root device lock.
> - Move the root device rescan to a workqueue
> - Connect RCDs directly as endpoints reachable through a CXL host bridge
>   as a dport, i.e. drop the extra dport indirection from v3
> - Add unit test infrastructure for an RCD configuration
> 
> [1]: http://lore.kernel.org/r/20221109104059.766720-1-rrichter@amd.com/
> 
> ---
> 
> >From [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

> ---
> 
> Dan Williams (9):
>       cxl/acpi: Simplify cxl_nvdimm_bridge probing
>       cxl/region: Drop redundant pmem region release handling
>       cxl/pmem: Refactor nvdimm device registration, delete the workqueue
>       cxl/pmem: Remove the cxl_pmem_wq and related infrastructure
>       cxl/acpi: Move rescan to the workqueue
>       tools/testing/cxl: Make mock CEDT parsing more robust
>       cxl/mem: Move devm_cxl_add_endpoint() from cxl_core to cxl_mem
>       cxl/port: Add RCD endpoint port enumeration
>       tools/testing/cxl: Add an RCH topology
> 
> Robert Richter (2):
>       cxl/ACPI: Register CXL host ports by bridge device
>       cxl/acpi: Extract component registers of restricted hosts from RCRB
> 
> Terry Bowman (1):
>       cxl/acpi: Set ACPI's CXL _OSC to indicate CXL1.1 support
> 
> 
>  drivers/acpi/pci_root.c       |    1 
>  drivers/cxl/acpi.c            |  105 +++++++++---
>  drivers/cxl/core/core.h       |    8 -
>  drivers/cxl/core/pmem.c       |   94 +++++++----
>  drivers/cxl/core/port.c       |  111 +++++++------
>  drivers/cxl/core/region.c     |   54 ++++++
>  drivers/cxl/core/regs.c       |   56 +++++++
>  drivers/cxl/cxl.h             |   46 +++--
>  drivers/cxl/cxlmem.h          |   15 ++
>  drivers/cxl/mem.c             |   72 ++++++++
>  drivers/cxl/pci.c             |   13 +-
>  drivers/cxl/pmem.c            |  351 +++++------------------------------------
>  tools/testing/cxl/Kbuild      |    1 
>  tools/testing/cxl/test/cxl.c  |  241 ++++++++++++++++++++++------
>  tools/testing/cxl/test/mem.c  |   40 ++++-
>  tools/testing/cxl/test/mock.c |   19 ++
>  tools/testing/cxl/test/mock.h |    3 
>  17 files changed, 712 insertions(+), 518 deletions(-)

I have tested this series and the enumeration is as expected (see
sysfs dump below). I see an HDM failure which I am investigating, but
that seems unrelated to this series.

You can add to this series my:

Tested-by: Robert Richter <rrichter@amd.com>

The drawback of this topology decision is that the endpoint's port is
a child of root0, which is the ACPI0017's device (not the CXL host
bridge):

 /sys/bus/cxl/devices/root0/endpoint1

I understand this is the CXL's RCD view on the pcie hierarchy here.
Logically there would be a cxl host port in between, like this:

 /sys/bus/cxl/devices/root0/port1/endpoint2

Esp. if there are multiple hosts in the system the port relations will
be lost and can only be determined using the pci dev's parent bridge.

port1 could then also hold the uport's component regs with the
upstream port capabilities to access e.g. status regs for RAS.

Anyway, let's see how this approach flies.

Thanks for looing into this.

-Robert

---

# find /sys/bus/cxl/devices/ -ls
   265293      0 drwxr-xr-x   2 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/
   265493      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/root0 -> ../../../devices/platform/ACPI0017:00/root0
   265664      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/endpoint1 -> ../../../devices/platform/ACPI0017:00/root0/endpoint1
   265584      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/mem0 -> ../../../devices/pci0000:7f/0000:7f:00.0/mem0
# find /sys/bus/cxl/devices/*/ -ls 
   265660      0 drwxr-xr-x   2 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/endpoint1/
   265661      0 -rw-r--r--   1 root     root         4096 Nov 30 14:18 /sys/bus/cxl/devices/endpoint1/uevent
   265667      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/endpoint1/driver -> ../../../../../bus/cxl/drivers/cxl_port
   265669      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/endpoint1/uport -> ../../../../pci0000:7f/0000:7f:00.0/mem0
   265668      0 -r--------   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/endpoint1/CDAT
   265665      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/endpoint1/subsystem -> ../../../../../bus/cxl
   265662      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/endpoint1/devtype
   265663      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/endpoint1/modalias
   265573      0 drwxr-xr-x   4 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/mem0/
   265574      0 -rw-r--r--   1 root     root         4096 Nov 30 14:18 /sys/bus/cxl/devices/mem0/uevent
   265578      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/label_storage_size
   265582      0 drwxr-xr-x   2 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/mem0/pmem
   265583      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/pmem/size
   265576      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/firmware_version
   265579      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/numa_node
   265586      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/dev
   265659      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/mem0/driver -> ../../../../bus/cxl/drivers/cxl_mem
   265580      0 drwxr-xr-x   2 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/mem0/ram
   265581      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/ram/size
   265585      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/mem0/subsystem -> ../../../../bus/cxl
   265575      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/serial
   265577      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/mem0/payload_max
   265489      0 drwxr-xr-x   3 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/root0/
   265490      0 -rw-r--r--   1 root     root         4096 Nov 30 14:18 /sys/bus/cxl/devices/root0/uevent
   265660      0 drwxr-xr-x   2 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/root0/endpoint1
   265661      0 -rw-r--r--   1 root     root         4096 Nov 30 14:18 /sys/bus/cxl/devices/root0/endpoint1/uevent
   265667      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/root0/endpoint1/driver -> ../../../../../bus/cxl/drivers/cxl_port
   265669      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/root0/endpoint1/uport -> ../../../../pci0000:7f/0000:7f:00.0/mem0
   265668      0 -r--------   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/root0/endpoint1/CDAT
   265665      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/root0/endpoint1/subsystem -> ../../../../../bus/cxl
   265662      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/root0/endpoint1/devtype
   265663      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/root0/endpoint1/modalias
   265495      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/root0/uport -> ../../ACPI0017:00
   265496      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:21 /sys/bus/cxl/devices/root0/dport4 -> ../../../pci0000:7f
   265494      0 lrwxrwxrwx   1 root     root            0 Nov 30 14:18 /sys/bus/cxl/devices/root0/subsystem -> ../../../../bus/cxl
   265491      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/root0/devtype
   265492      0 -r--r--r--   1 root     root         4096 Nov 30 14:21 /sys/bus/cxl/devices/root0/modalias

Re: [PATCH v4 11/12] tools/testing/cxl: Add an RCH topology

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:38, Dan Williams wrote:
> > In an RCH topology a CXL host-bridge as Root Complex Integrated Endpoint
> > the represents the memory expander. Unlike a VH topology there is no
> > CXL/PCIE Root Port that host the endpoint. The CXL subsystem maps this
> > as the CXL root object (ACPI0017 on ACPI based systems) targeting the
> > host-bridge as a dport, per usual, but then that dport directly hosts
> > the endpoint port.
> > 
> > Mock up that configuration with a 4th host-bridge that has a 'cxl_rcd'
> > device instance as its immediate child.
> > 
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > ---
> >  tools/testing/cxl/test/cxl.c |  151 +++++++++++++++++++++++++++++++++++++++---
> >  tools/testing/cxl/test/mem.c |   37 ++++++++++
> >  2 files changed, 176 insertions(+), 12 deletions(-)
> 
> One comment below.
> 
> > @@ -736,6 +779,87 @@ static void mock_companion(struct acpi_device *adev, struct device *dev)
> >  #define SZ_512G (SZ_64G * 8)
> >  #endif
> >  
> > +static __init int cxl_rch_init(void)
> > +{
> > +	int rc, i;
> > +
> > +	for (i = 0; i < ARRAY_SIZE(cxl_rch); i++) {
> > +		int idx = NR_CXL_HOST_BRIDGES + NR_CXL_SINGLE_HOST + i;
> > +		struct acpi_device *adev = &host_bridge[idx];
> > +		struct platform_device *pdev;
> > +
> > +		pdev = platform_device_alloc("cxl_host_bridge", idx);
> > +		if (!pdev)
> > +			goto err_bridge;
> > +
> > +		mock_companion(adev, &pdev->dev);
> > +		rc = platform_device_add(pdev);
> > +		if (rc) {
> > +			platform_device_put(pdev);
> > +			goto err_bridge;
> > +		}
> > +
> > +		cxl_rch[i] = pdev;
> > +		mock_pci_bus[idx].bridge = &pdev->dev;
> > +		rc = sysfs_create_link(&pdev->dev.kobj, &pdev->dev.kobj,
> > +				       "firmware_node");
> > +		if (rc)
> > +			goto err_bridge;
> > +	}
> > +
> > +	for (i = 0; i < ARRAY_SIZE(cxl_rcd); i++) {
> > +		int idx = NR_MEM_MULTI + NR_MEM_SINGLE + i;
> > +		struct platform_device *rch = cxl_rch[i];
> > +		struct platform_device *pdev;
> > +
> > +		pdev = platform_device_alloc("cxl_rcd", idx);
> > +		if (!pdev)
> > +			goto err_mem;
> > +		pdev->dev.parent = &rch->dev;
> > +		set_dev_node(&pdev->dev, i % 2);
> > +
> > +		rc = platform_device_add(pdev);
> > +		if (rc) {
> > +			platform_device_put(pdev);
> > +			goto err_mem;
> > +		}
> > +		cxl_rcd[i] = pdev;
> > +	}
> > +
> > +	return 0;
> > +
> > +err_mem:
> > +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> > +		platform_device_unregister(cxl_rcd[i]);
> > +err_bridge:
> 
> platform_device_unregister() is safe to be used with NULL, so we can
> have a single entry of this unregister code ...
> 
> > +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> > +		struct platform_device *pdev = cxl_rch[i];
> > +
> > +		if (!pdev)
> > +			continue;
> > +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> > +		platform_device_unregister(cxl_rch[i]);
> > +	}
> > +
> > +	return rc;
> > +}
> > +
> > +static void cxl_rch_exit(void)
> > +{
> > +	int i;
> > +
> > +	for (i = ARRAY_SIZE(cxl_rcd) - 1; i >= 0; i--)
> > +		platform_device_unregister(cxl_rcd[i]);
> > +	for (i = ARRAY_SIZE(cxl_rch) - 1; i >= 0; i--) {
> > +		struct platform_device *pdev = cxl_rch[i];
> > +
> > +		if (!pdev)
> > +			continue;
> > +		sysfs_remove_link(&pdev->dev.kobj, "firmware_node");
> > +		platform_device_unregister(cxl_rch[i]);
> > +	}
> > +}
> 
> ... and have a single function for both. This reduces code
> duplication here.
> 

That's true. This was cargo culted from the other parts of cxl_test, but
those can be cleaned up too. I will do this with a follow-on patch and
clean up both.

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dan Williams]

Robert Richter wrote:
> On 28.11.22 16:25:18, Dan Williams wrote:
> > Robert Richter wrote:
> > > On 24.11.22 10:35:32, Dan Williams wrote:
> > > > Unlike a CXL memory expander in a VH topology that has at least one
> > > > intervening 'struct cxl_port' instance between itself and the CXL root
> > > > device, an RCD attaches one-level higher. For example:
> > > > 
> > > >                VH
> > > >           ┌──────────┐
> > > >           │ ACPI0017 │
> > > >           │  root0   │
> > > >           └─────┬────┘
> > > >                 │
> > > >           ┌─────┴────┐
> > > >           │  dport0  │
> > > >     ┌─────┤ ACPI0016 ├─────┐
> > > >     │     │  port1   │     │
> > > >     │     └────┬─────┘     │
> > > >     │          │           │
> > > >  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> > > >  │dport0│   │dport1│   │dport2│
> > > >  │ RP0  │   │ RP1  │   │ RP2  │
> > > >  └──────┘   └──┬───┘   └──────┘
> > > >                │
> > > >            ┌───┴─────┐
> > > >            │endpoint0│
> > > >            │  port2  │
> > > >            └─────────┘
> > > > 
> > > > ...vs:
> > > > 
> > > >               RCH
> > > >           ┌──────────┐
> > > >           │ ACPI0017 │
> > > >           │  root0   │
> > > >           └────┬─────┘
> > > >                │
> > > >            ┌───┴────┐
> > > >            │ dport0 │
> > > >            │ACPI0016│
> > > >            └───┬────┘
> > > >                │
> > > >           ┌────┴─────┐
> > > >           │endpoint0 │
> > > >           │  port1   │
> > > >           └──────────┘
> > > > 
> > > > So arrange for endpoint port in the RCH/RCD case to appear directly
> > > > connected to the host-bridge in its singular role as a dport. Compare
> > > > that to the VH case where the host-bridge serves a dual role as a
> > > > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > > > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > > > instances in the CXL topology.
> > > > 
> > > > Another deviation from the VH case is that RCDs may need to look up
> > > > their component registers from the Root Complex Register Block (RCRB).
> > > > That platform firmware specified RCRB area is cached by the cxl_acpi
> > > > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > > > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > > > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > > > upstream port component registers).
> > > > 
> > > > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > > > ---
> > > >  drivers/cxl/core/port.c |   11 +++++++++--
> > > >  drivers/cxl/cxlmem.h    |    2 ++
> > > >  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> > > >  drivers/cxl/pci.c       |   10 ++++++++++
> > > >  4 files changed, 45 insertions(+), 9 deletions(-)
> > > > 
> > > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > > index c7f58282b2c1..2385ee00eb9a 100644
> > > > --- a/drivers/cxl/core/port.c
> > > > +++ b/drivers/cxl/core/port.c
> > > > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > >  {
> > > >  	struct device *dev = &cxlmd->dev;
> > > >  	struct device *iter;
> > > > +	struct cxl_dport *dport;
> > > > +	struct cxl_port *port;
> > > 
> > > There is no direct need to move that code here.
> > > 
> > > If you want to clean that up in this patch too, then leave a comment
> > > in the change log?
> > 
> > Oh, good point, must have been left over from an earlier revision of the
> > patch, dropped it.
> > 
> > > 
> > > >  	int rc;
> > > >  
> > > > +	/*
> > > > +	 * Skip intermediate port enumeration in the RCH case, there
> > > > +	 * are no ports in between a host bridge and an endpoint.
> > > > +	 */
> > > > +	if (cxlmd->cxlds->rcd)
> > > > +		return 0;
> > > > +
> > > >  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> > > >  	if (rc)
> > > >  		return rc;
> > > > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > >  	for (iter = dev; iter; iter = grandparent(iter)) {
> > > >  		struct device *dport_dev = grandparent(iter);
> > > >  		struct device *uport_dev;
> > > > -		struct cxl_dport *dport;
> > > > -		struct cxl_port *port;
> > > >  
> > > >  		if (!dport_dev)
> > > >  			return 0;
> > > > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > > > index e082991bc58c..35d485d041f0 100644
> > > > --- a/drivers/cxl/cxlmem.h
> > > > +++ b/drivers/cxl/cxlmem.h
> > > > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> > > >   * @dev: The device associated with this CXL state
> > > >   * @regs: Parsed register blocks
> > > >   * @cxl_dvsec: Offset to the PCIe device DVSEC
> > > > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> > > >   * @payload_size: Size of space for payload
> > > >   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> > > >   * @lsa_size: Size of Label Storage Area
> > > > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> > > >  	struct cxl_regs regs;
> > > >  	int cxl_dvsec;
> > > >  
> > > > +	bool rcd;
> > > >  	size_t payload_size;
> > > >  	size_t lsa_size;
> > > >  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > > > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > > > index aa63ce8c7ca6..9a655b4b5e52 100644
> > > > --- a/drivers/cxl/mem.c
> > > > +++ b/drivers/cxl/mem.c
> > > > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> > > >  	return 0;
> > > >  }
> > > >  
> > > > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> > > >  				 struct cxl_dport *parent_dport)
> > > >  {
> > > >  	struct cxl_port *parent_port = parent_dport->port;
> > > >  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > > >  	struct cxl_port *endpoint, *iter, *down;
> > > > +	resource_size_t component_reg_phys;
> > > >  	int rc;
> > > >  
> > > >  	/*
> > > > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > >  		ep->next = down;
> > > >  	}
> > > >  
> > > > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > > > -				     cxlds->component_reg_phys, parent_dport);
> > > > +	/*
> > > > +	 * The component registers for an RCD might come from the
> > > > +	 * host-bridge RCRB if they are not already mapped via the
> > > > +	 * typical register locator mechanism.
> > > > +	 */
> > > > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > > +		component_reg_phys = cxl_rcrb_to_component(
> > > > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> > > 
> > > As already commented: this must be the upstream RCRB here.
> > > 
> > > > +	else
> > > > +		component_reg_phys = cxlds->component_reg_phys;
> > > > +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> > > > +				     parent_dport);
> > > 
> > > Looking at CXL 3.0 spec, table 8-22, there are the various sources of
> > > component registers listed. For RCD we need: D1, DP1, UP1 (optional
> > > R).
> > > 
> > > 	D1:	endpoint->component_reg_phys;
> > > 	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
> > > 	DP1:	parent_dport->component_reg_phys;
> > > 
> > > I don't see how all of them could be stored in this data layout as the
> > > cxl host port is missing.
> > 
> > If I am understanding your concern correctly, that's handled here:
> > 
> >     if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > 
> > In the D1 case cxlds->component_reg_phys will be valid since the
> > component registers were visible via the register locator DVSEC and
> > retrieved by cxl_pci. In the UP1 case cxlds->component_reg_phys is
> > invalid and the driver falls back to the RCRB. DP1 is handled in
> > cxl_acpi. I.e. the D1 and UP1 cases do not co-exist.
> 
> What I mean is we must store all 3 component reg base addresses for
> later access. E.g., if there is an AER error of a pci dev or the host,
> we must (depending on the error details) access CXL RAS status of
> either D1, UP1 or DP1. So for all 3 of them there must be a way to
> determine this walking through the port hierarchy. In the above list
> of locations I don't where UP1's component reg base address is stored.

So I think we are reading the specification differently. I am comparing
Figure 9-7 "CXL Device Remaps Upstream Port and Component Registers" and
Figure Figure 9-8 "CXL Device that Does Not Remap Upstream Port and
Component Registers" and noting that there is never a case where three
sets of component registers are visible. It is either DP1 connected to
UP1 (Figure 9-7) or DP1 connected to D1 (Figure 9-8). There is never a
case where the code needs to consider UP1 and D1 component registers at
the same time because those things are identical just enumerated
differently depending on how the endpoint is implemented.

Re: [PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[Dan Williams]

Robert Richter wrote:
> Dan,
> 
> On 24.11.22 10:34:35, Dan Williams wrote:
> > Changes since v3 [1]:
> > - Rework / simplify CXL to LIBNVDIMM coordination to remove a
> >   flush_work() locking dependency from underneath the root device lock.
> > - Move the root device rescan to a workqueue
> > - Connect RCDs directly as endpoints reachable through a CXL host bridge
> >   as a dport, i.e. drop the extra dport indirection from v3
> > - Add unit test infrastructure for an RCD configuration
> 
> thank you for this posting.
> 
> Patches #1-#6 are not really prerequisites (except for a trivial
> conflict), right? I only reviewed them starting with #6.

In fact they are pre-requisites because of this hunk in:

[PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

http://lore.kernel.org/r/166931493266.2104015.8062923429837042172.stgit@dwillia2-xfh.jf.intel.com/

@@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
                return -ENXIO;
        }
 
-       device_lock(&parent_port->dev);
-       if (!parent_port->dev.driver) {
+       if (dport->rch)
+               endpoint_parent = parent_port->uport;
+       else
+               endpoint_parent = &parent_port->dev;
+
+       device_lock(endpoint_parent);
+       if (!endpoint_parent->driver) {
                dev_err(dev, "CXL port topology %s not enabled\n",
                        dev_name(&parent_port->dev));
                rc = -ENXIO;
                goto unlock;
        }
 
-       rc = devm_cxl_add_endpoint(cxlmd, dport);
+       rc = devm_cxl_add_endpoint(endpoint_parent, cxlmd, dport);
 unlock:
-       device_unlock(&parent_port->dev);
+       device_unlock(endpoint_parent);
        put_device(&parent_port->dev);
        if (rc)
                return rc;

That device_lock(endpoint_parent) in the RCH case locks the ACPI0017
device so that a devm action can be serialized against ACPI0017 being
unbound from its driver. Before RCH support this path only locked
cxl_port objects to syncrhonize with the cxl_port driver. In the RCH
case the port that needs to be locked while the endpoint is attached is
the one owned by the cxl_acpi driver.

This all happens while holding device_lock(&cxlmd->dev). It means
lockdep complains about cxl_bus_rescan() being done under
device_lock(&ACPI0017->dev), since rescan may take
device_lock(&cxlmd->dev), and it complains about the flush_work() in
unregister_nvb() for a similar entanglement. So the initial patches in
this series are all about making it possible setup an unregistration
chain when the root CXL device goes through ->remove().

Re: [PATCH v4 12/12] cxl/acpi: Set ACPI's CXL _OSC to indicate CXL1.1 support

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:43, Dan Williams wrote:
> > From: Terry Bowman <terry.bowman@amd.com>
> > 
> > ACPI includes a CXL _OSC for the OS to communicate what it knows of CXL
> > device topologies. To date Linux has added support for CXL 2.0 (VH) port
> > topologies, hotplug, and error handling. Now that the driver also know
> > how to enumerate CXL 1.1 (RCH) port topologies, indicate that capability
> > via CXL _OSC. See CXL3.0 Table 9-26 'Interpretation of CXL _OSC Support
> > Field'
> > 
> > Signed-off-by: Terry Bowman <terry.bowman@amd.com>
> > Signed-off-by: Robert Richter <rrichter@amd.com>
> > Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> > [djbw: wordsmith changelog]
> > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> 
> I have had a reworded version of this in the pipe too, esp. version
> strings were dropped in favor of VH and RCD mode. You might want to
> consider that one:

Sure, looks good to me.

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Robert Richter]

On 30.11.22 12:39:00, Dan Williams wrote:
> Robert Richter wrote:
> > On 28.11.22 16:25:18, Dan Williams wrote:
> > > Robert Richter wrote:
> > > > On 24.11.22 10:35:32, Dan Williams wrote:
> > > > > Unlike a CXL memory expander in a VH topology that has at least one
> > > > > intervening 'struct cxl_port' instance between itself and the CXL root
> > > > > device, an RCD attaches one-level higher. For example:
> > > > > 
> > > > >                VH
> > > > >           ┌──────────┐
> > > > >           │ ACPI0017 │
> > > > >           │  root0   │
> > > > >           └─────┬────┘
> > > > >                 │
> > > > >           ┌─────┴────┐
> > > > >           │  dport0  │
> > > > >     ┌─────┤ ACPI0016 ├─────┐
> > > > >     │     │  port1   │     │
> > > > >     │     └────┬─────┘     │
> > > > >     │          │           │
> > > > >  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> > > > >  │dport0│   │dport1│   │dport2│
> > > > >  │ RP0  │   │ RP1  │   │ RP2  │
> > > > >  └──────┘   └──┬───┘   └──────┘
> > > > >                │
> > > > >            ┌───┴─────┐
> > > > >            │endpoint0│
> > > > >            │  port2  │
> > > > >            └─────────┘
> > > > > 
> > > > > ...vs:
> > > > > 
> > > > >               RCH
> > > > >           ┌──────────┐
> > > > >           │ ACPI0017 │
> > > > >           │  root0   │
> > > > >           └────┬─────┘
> > > > >                │
> > > > >            ┌───┴────┐
> > > > >            │ dport0 │
> > > > >            │ACPI0016│
> > > > >            └───┬────┘
> > > > >                │
> > > > >           ┌────┴─────┐
> > > > >           │endpoint0 │
> > > > >           │  port1   │
> > > > >           └──────────┘
> > > > > 
> > > > > So arrange for endpoint port in the RCH/RCD case to appear directly
> > > > > connected to the host-bridge in its singular role as a dport. Compare
> > > > > that to the VH case where the host-bridge serves a dual role as a
> > > > > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > > > > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > > > > instances in the CXL topology.
> > > > > 
> > > > > Another deviation from the VH case is that RCDs may need to look up
> > > > > their component registers from the Root Complex Register Block (RCRB).
> > > > > That platform firmware specified RCRB area is cached by the cxl_acpi
> > > > > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > > > > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > > > > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > > > > upstream port component registers).
> > > > > 
> > > > > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > > > > ---
> > > > >  drivers/cxl/core/port.c |   11 +++++++++--
> > > > >  drivers/cxl/cxlmem.h    |    2 ++
> > > > >  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> > > > >  drivers/cxl/pci.c       |   10 ++++++++++
> > > > >  4 files changed, 45 insertions(+), 9 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > > > index c7f58282b2c1..2385ee00eb9a 100644
> > > > > --- a/drivers/cxl/core/port.c
> > > > > +++ b/drivers/cxl/core/port.c
> > > > > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > > >  {
> > > > >  	struct device *dev = &cxlmd->dev;
> > > > >  	struct device *iter;
> > > > > +	struct cxl_dport *dport;
> > > > > +	struct cxl_port *port;
> > > > 
> > > > There is no direct need to move that code here.
> > > > 
> > > > If you want to clean that up in this patch too, then leave a comment
> > > > in the change log?
> > > 
> > > Oh, good point, must have been left over from an earlier revision of the
> > > patch, dropped it.
> > > 
> > > > 
> > > > >  	int rc;
> > > > >  
> > > > > +	/*
> > > > > +	 * Skip intermediate port enumeration in the RCH case, there
> > > > > +	 * are no ports in between a host bridge and an endpoint.
> > > > > +	 */
> > > > > +	if (cxlmd->cxlds->rcd)
> > > > > +		return 0;
> > > > > +
> > > > >  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> > > > >  	if (rc)
> > > > >  		return rc;
> > > > > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > > >  	for (iter = dev; iter; iter = grandparent(iter)) {
> > > > >  		struct device *dport_dev = grandparent(iter);
> > > > >  		struct device *uport_dev;
> > > > > -		struct cxl_dport *dport;
> > > > > -		struct cxl_port *port;
> > > > >  
> > > > >  		if (!dport_dev)
> > > > >  			return 0;
> > > > > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > > > > index e082991bc58c..35d485d041f0 100644
> > > > > --- a/drivers/cxl/cxlmem.h
> > > > > +++ b/drivers/cxl/cxlmem.h
> > > > > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> > > > >   * @dev: The device associated with this CXL state
> > > > >   * @regs: Parsed register blocks
> > > > >   * @cxl_dvsec: Offset to the PCIe device DVSEC
> > > > > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> > > > >   * @payload_size: Size of space for payload
> > > > >   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> > > > >   * @lsa_size: Size of Label Storage Area
> > > > > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> > > > >  	struct cxl_regs regs;
> > > > >  	int cxl_dvsec;
> > > > >  
> > > > > +	bool rcd;
> > > > >  	size_t payload_size;
> > > > >  	size_t lsa_size;
> > > > >  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > > > > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > > > > index aa63ce8c7ca6..9a655b4b5e52 100644
> > > > > --- a/drivers/cxl/mem.c
> > > > > +++ b/drivers/cxl/mem.c
> > > > > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> > > > >  	return 0;
> > > > >  }
> > > > >  
> > > > > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > > > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> > > > >  				 struct cxl_dport *parent_dport)
> > > > >  {
> > > > >  	struct cxl_port *parent_port = parent_dport->port;
> > > > >  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > > > >  	struct cxl_port *endpoint, *iter, *down;
> > > > > +	resource_size_t component_reg_phys;
> > > > >  	int rc;
> > > > >  
> > > > >  	/*
> > > > > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > > >  		ep->next = down;
> > > > >  	}
> > > > >  
> > > > > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > > > > -				     cxlds->component_reg_phys, parent_dport);
> > > > > +	/*
> > > > > +	 * The component registers for an RCD might come from the
> > > > > +	 * host-bridge RCRB if they are not already mapped via the
> > > > > +	 * typical register locator mechanism.
> > > > > +	 */
> > > > > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > > > +		component_reg_phys = cxl_rcrb_to_component(
> > > > > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> > > > 
> > > > As already commented: this must be the upstream RCRB here.
> > > > 
> > > > > +	else
> > > > > +		component_reg_phys = cxlds->component_reg_phys;
> > > > > +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> > > > > +				     parent_dport);
> > > > 
> > > > Looking at CXL 3.0 spec, table 8-22, there are the various sources of
> > > > component registers listed. For RCD we need: D1, DP1, UP1 (optional
> > > > R).
> > > > 
> > > > 	D1:	endpoint->component_reg_phys;
> > > > 	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
> > > > 	DP1:	parent_dport->component_reg_phys;
> > > > 
> > > > I don't see how all of them could be stored in this data layout as the
> > > > cxl host port is missing.
> > > 
> > > If I am understanding your concern correctly, that's handled here:
> > > 
> > >     if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > 
> > > In the D1 case cxlds->component_reg_phys will be valid since the
> > > component registers were visible via the register locator DVSEC and
> > > retrieved by cxl_pci. In the UP1 case cxlds->component_reg_phys is
> > > invalid and the driver falls back to the RCRB. DP1 is handled in
> > > cxl_acpi. I.e. the D1 and UP1 cases do not co-exist.
> > 
> > What I mean is we must store all 3 component reg base addresses for
> > later access. E.g., if there is an AER error of a pci dev or the host,
> > we must (depending on the error details) access CXL RAS status of
> > either D1, UP1 or DP1. So for all 3 of them there must be a way to
> > determine this walking through the port hierarchy. In the above list
> > of locations I don't where UP1's component reg base address is stored.
> 
> So I think we are reading the specification differently. I am comparing
> Figure 9-7 "CXL Device Remaps Upstream Port and Component Registers" and
> Figure Figure 9-8 "CXL Device that Does Not Remap Upstream Port and
> Component Registers" and noting that there is never a case where three
> sets of component registers are visible. It is either DP1 connected to
> UP1 (Figure 9-7) or DP1 connected to D1 (Figure 9-8). There is never a
> case where the code needs to consider UP1 and D1 component registers at
> the same time because those things are identical just enumerated
> differently depending on how the endpoint is implemented.

Yes, the spec is ambiguous here. Looking at CXL 3.0, Figure 12-3 I
also tend to agree the RCiEP's CXL RAS cap is used for that ("UP Z
sends an error message to all CXL.io Functions that are affected by
this error."). Unfortunately 12.2.1.2 does not state where the CXL RAS
Capability that logs the error resides.

-Robert

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Robert Richter]

On 01.12.22 16:23:46, Robert Richter wrote:
> On 30.11.22 12:39:00, Dan Williams wrote:
> > Robert Richter wrote:
> > > On 28.11.22 16:25:18, Dan Williams wrote:
> > > > Robert Richter wrote:
> > > > > On 24.11.22 10:35:32, Dan Williams wrote:
> > > > > > Unlike a CXL memory expander in a VH topology that has at least one
> > > > > > intervening 'struct cxl_port' instance between itself and the CXL root
> > > > > > device, an RCD attaches one-level higher. For example:
> > > > > > 
> > > > > >                VH
> > > > > >           ┌──────────┐
> > > > > >           │ ACPI0017 │
> > > > > >           │  root0   │
> > > > > >           └─────┬────┘
> > > > > >                 │
> > > > > >           ┌─────┴────┐
> > > > > >           │  dport0  │
> > > > > >     ┌─────┤ ACPI0016 ├─────┐
> > > > > >     │     │  port1   │     │
> > > > > >     │     └────┬─────┘     │
> > > > > >     │          │           │
> > > > > >  ┌──┴───┐   ┌──┴───┐   ┌───┴──┐
> > > > > >  │dport0│   │dport1│   │dport2│
> > > > > >  │ RP0  │   │ RP1  │   │ RP2  │
> > > > > >  └──────┘   └──┬───┘   └──────┘
> > > > > >                │
> > > > > >            ┌───┴─────┐
> > > > > >            │endpoint0│
> > > > > >            │  port2  │
> > > > > >            └─────────┘
> > > > > > 
> > > > > > ...vs:
> > > > > > 
> > > > > >               RCH
> > > > > >           ┌──────────┐
> > > > > >           │ ACPI0017 │
> > > > > >           │  root0   │
> > > > > >           └────┬─────┘
> > > > > >                │
> > > > > >            ┌───┴────┐
> > > > > >            │ dport0 │
> > > > > >            │ACPI0016│
> > > > > >            └───┬────┘
> > > > > >                │
> > > > > >           ┌────┴─────┐
> > > > > >           │endpoint0 │
> > > > > >           │  port1   │
> > > > > >           └──────────┘
> > > > > > 
> > > > > > So arrange for endpoint port in the RCH/RCD case to appear directly
> > > > > > connected to the host-bridge in its singular role as a dport. Compare
> > > > > > that to the VH case where the host-bridge serves a dual role as a
> > > > > > 'cxl_dport' for the CXL root device *and* a 'cxl_port' upstream port for
> > > > > > the Root Ports in the Root Complex that are modeled as 'cxl_dport'
> > > > > > instances in the CXL topology.
> > > > > > 
> > > > > > Another deviation from the VH case is that RCDs may need to look up
> > > > > > their component registers from the Root Complex Register Block (RCRB).
> > > > > > That platform firmware specified RCRB area is cached by the cxl_acpi
> > > > > > driver and conveyed via the host-bridge dport to the cxl_mem driver to
> > > > > > perform the cxl_rcrb_to_component() lookup for the endpoint port
> > > > > > (See 9.11.8 CXL Devices Attached to an RCH for the lookup of the
> > > > > > upstream port component registers).
> > > > > > 
> > > > > > Signed-off-by: Dan Williams <dan.j.williams@intel.com>
> > > > > > ---
> > > > > >  drivers/cxl/core/port.c |   11 +++++++++--
> > > > > >  drivers/cxl/cxlmem.h    |    2 ++
> > > > > >  drivers/cxl/mem.c       |   31 ++++++++++++++++++++++++-------
> > > > > >  drivers/cxl/pci.c       |   10 ++++++++++
> > > > > >  4 files changed, 45 insertions(+), 9 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c
> > > > > > index c7f58282b2c1..2385ee00eb9a 100644
> > > > > > --- a/drivers/cxl/core/port.c
> > > > > > +++ b/drivers/cxl/core/port.c
> > > > > > @@ -1358,8 +1358,17 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > > > >  {
> > > > > >  	struct device *dev = &cxlmd->dev;
> > > > > >  	struct device *iter;
> > > > > > +	struct cxl_dport *dport;
> > > > > > +	struct cxl_port *port;
> > > > > 
> > > > > There is no direct need to move that code here.
> > > > > 
> > > > > If you want to clean that up in this patch too, then leave a comment
> > > > > in the change log?
> > > > 
> > > > Oh, good point, must have been left over from an earlier revision of the
> > > > patch, dropped it.
> > > > 
> > > > > 
> > > > > >  	int rc;
> > > > > >  
> > > > > > +	/*
> > > > > > +	 * Skip intermediate port enumeration in the RCH case, there
> > > > > > +	 * are no ports in between a host bridge and an endpoint.
> > > > > > +	 */
> > > > > > +	if (cxlmd->cxlds->rcd)
> > > > > > +		return 0;
> > > > > > +
> > > > > >  	rc = devm_add_action_or_reset(&cxlmd->dev, cxl_detach_ep, cxlmd);
> > > > > >  	if (rc)
> > > > > >  		return rc;
> > > > > > @@ -1373,8 +1382,6 @@ int devm_cxl_enumerate_ports(struct cxl_memdev *cxlmd)
> > > > > >  	for (iter = dev; iter; iter = grandparent(iter)) {
> > > > > >  		struct device *dport_dev = grandparent(iter);
> > > > > >  		struct device *uport_dev;
> > > > > > -		struct cxl_dport *dport;
> > > > > > -		struct cxl_port *port;
> > > > > >  
> > > > > >  		if (!dport_dev)
> > > > > >  			return 0;
> > > > > > diff --git a/drivers/cxl/cxlmem.h b/drivers/cxl/cxlmem.h
> > > > > > index e082991bc58c..35d485d041f0 100644
> > > > > > --- a/drivers/cxl/cxlmem.h
> > > > > > +++ b/drivers/cxl/cxlmem.h
> > > > > > @@ -201,6 +201,7 @@ struct cxl_endpoint_dvsec_info {
> > > > > >   * @dev: The device associated with this CXL state
> > > > > >   * @regs: Parsed register blocks
> > > > > >   * @cxl_dvsec: Offset to the PCIe device DVSEC
> > > > > > + * @rcd: operating in RCD mode (CXL 3.0 9.11.8 CXL Devices Attached to an RCH)
> > > > > >   * @payload_size: Size of space for payload
> > > > > >   *                (CXL 2.0 8.2.8.4.3 Mailbox Capabilities Register)
> > > > > >   * @lsa_size: Size of Label Storage Area
> > > > > > @@ -235,6 +236,7 @@ struct cxl_dev_state {
> > > > > >  	struct cxl_regs regs;
> > > > > >  	int cxl_dvsec;
> > > > > >  
> > > > > > +	bool rcd;
> > > > > >  	size_t payload_size;
> > > > > >  	size_t lsa_size;
> > > > > >  	struct mutex mbox_mutex; /* Protects device mailbox and firmware */
> > > > > > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > > > > > index aa63ce8c7ca6..9a655b4b5e52 100644
> > > > > > --- a/drivers/cxl/mem.c
> > > > > > +++ b/drivers/cxl/mem.c
> > > > > > @@ -45,12 +45,13 @@ static int cxl_mem_dpa_show(struct seq_file *file, void *data)
> > > > > >  	return 0;
> > > > > >  }
> > > > > >  
> > > > > > -static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > > > > +static int devm_cxl_add_endpoint(struct device *host, struct cxl_memdev *cxlmd,
> > > > > >  				 struct cxl_dport *parent_dport)
> > > > > >  {
> > > > > >  	struct cxl_port *parent_port = parent_dport->port;
> > > > > >  	struct cxl_dev_state *cxlds = cxlmd->cxlds;
> > > > > >  	struct cxl_port *endpoint, *iter, *down;
> > > > > > +	resource_size_t component_reg_phys;
> > > > > >  	int rc;
> > > > > >  
> > > > > >  	/*
> > > > > > @@ -65,8 +66,18 @@ static int devm_cxl_add_endpoint(struct cxl_memdev *cxlmd,
> > > > > >  		ep->next = down;
> > > > > >  	}
> > > > > >  
> > > > > > -	endpoint = devm_cxl_add_port(&parent_port->dev, &cxlmd->dev,
> > > > > > -				     cxlds->component_reg_phys, parent_dport);
> > > > > > +	/*
> > > > > > +	 * The component registers for an RCD might come from the
> > > > > > +	 * host-bridge RCRB if they are not already mapped via the
> > > > > > +	 * typical register locator mechanism.
> > > > > > +	 */
> > > > > > +	if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > > > > +		component_reg_phys = cxl_rcrb_to_component(
> > > > > > +			&cxlmd->dev, parent_dport->rcrb, CXL_RCRB_DOWNSTREAM);
> > > > > 
> > > > > As already commented: this must be the upstream RCRB here.
> > > > > 
> > > > > > +	else
> > > > > > +		component_reg_phys = cxlds->component_reg_phys;
> > > > > > +	endpoint = devm_cxl_add_port(host, &cxlmd->dev, component_reg_phys,
> > > > > > +				     parent_dport);
> > > > > 
> > > > > Looking at CXL 3.0 spec, table 8-22, there are the various sources of
> > > > > component registers listed. For RCD we need: D1, DP1, UP1 (optional
> > > > > R).
> > > > > 
> > > > > 	D1:	endpoint->component_reg_phys;
> > > > > 	UP1:	parent_port-component_reg_phys; (missing in RCH topology)
> > > > > 	DP1:	parent_dport->component_reg_phys;
> > > > > 
> > > > > I don't see how all of them could be stored in this data layout as the
> > > > > cxl host port is missing.
> > > > 
> > > > If I am understanding your concern correctly, that's handled here:
> > > > 
> > > >     if (parent_dport->rch && cxlds->component_reg_phys == CXL_RESOURCE_NONE)
> > > > 
> > > > In the D1 case cxlds->component_reg_phys will be valid since the
> > > > component registers were visible via the register locator DVSEC and
> > > > retrieved by cxl_pci. In the UP1 case cxlds->component_reg_phys is
> > > > invalid and the driver falls back to the RCRB. DP1 is handled in
> > > > cxl_acpi. I.e. the D1 and UP1 cases do not co-exist.
> > > 
> > > What I mean is we must store all 3 component reg base addresses for
> > > later access. E.g., if there is an AER error of a pci dev or the host,
> > > we must (depending on the error details) access CXL RAS status of
> > > either D1, UP1 or DP1. So for all 3 of them there must be a way to
> > > determine this walking through the port hierarchy. In the above list
> > > of locations I don't where UP1's component reg base address is stored.
> > 
> > So I think we are reading the specification differently. I am comparing
> > Figure 9-7 "CXL Device Remaps Upstream Port and Component Registers" and
> > Figure Figure 9-8 "CXL Device that Does Not Remap Upstream Port and
> > Component Registers" and noting that there is never a case where three
> > sets of component registers are visible. It is either DP1 connected to
> > UP1 (Figure 9-7) or DP1 connected to D1 (Figure 9-8). There is never a
> > case where the code needs to consider UP1 and D1 component registers at
> > the same time because those things are identical just enumerated
> > differently depending on how the endpoint is implemented.
> 
> Yes, the spec is ambiguous here. Looking at CXL 3.0, Figure 12-3 I
> also tend to agree the RCiEP's CXL RAS cap is used for that ("UP Z
> sends an error message to all CXL.io Functions that are affected by
> this error."). Unfortunately 12.2.1.2 does not state where the CXL RAS
> Capability that logs the error resides.

I found it now in 8.2.2 of the CXL 3.0 spec: For an RCD upstream port
the RCRB should be used to locate the component registers. If the UP's
RCRB is not implemented, then the Register Locator DVSEC (of the
RCiEP) should be used. So right, UP1 and D1 component registers are
never used at the same time.

Thanks for clarification,

-Robert

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Robert Richter]

On 24.11.22 10:35:32, Dan Williams wrote:

> diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c

> @@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
>  		return -ENXIO;
>  	}
>  
> -	device_lock(&parent_port->dev);
> -	if (!parent_port->dev.driver) {
> +	if (dport->rch)
> +		endpoint_parent = parent_port->uport;
> +	else
> +		endpoint_parent = &parent_port->dev;
> +
> +	device_lock(endpoint_parent);
> +	if (!endpoint_parent->driver) {
>  		dev_err(dev, "CXL port topology %s not enabled\n",
>  			dev_name(&parent_port->dev));

This must be dev_name(endpoint_parent) here.

>  		rc = -ENXIO;
>  		goto unlock;
>  	}
>  
> -	rc = devm_cxl_add_endpoint(cxlmd, dport);
> +	rc = devm_cxl_add_endpoint(endpoint_parent, cxlmd, dport);
>  unlock:
> -	device_unlock(&parent_port->dev);
> +	device_unlock(endpoint_parent);
>  	put_device(&parent_port->dev);

This is correct and the counterpart to cxl_mem_find_port().

>  	if (rc)
>  		return rc;

-Robert

Re: [PATCH v4 00/12] cxl: Add support for Restricted CXL hosts (RCD mode)

[Robert Richter]

On 30.11.22 12:59:28, Dan Williams wrote:
> Robert Richter wrote:
> > Dan,
> > 
> > On 24.11.22 10:34:35, Dan Williams wrote:
> > > Changes since v3 [1]:
> > > - Rework / simplify CXL to LIBNVDIMM coordination to remove a
> > >   flush_work() locking dependency from underneath the root device lock.
> > > - Move the root device rescan to a workqueue
> > > - Connect RCDs directly as endpoints reachable through a CXL host bridge
> > >   as a dport, i.e. drop the extra dport indirection from v3
> > > - Add unit test infrastructure for an RCD configuration
> > 
> > thank you for this posting.
> > 
> > Patches #1-#6 are not really prerequisites (except for a trivial
> > conflict), right? I only reviewed them starting with #6.
> 
> In fact they are pre-requisites because of this hunk in:

Thanks for the explanation of using device_lock() here.

-Robert

Re: [PATCH v4 10/12] cxl/port: Add RCD endpoint port enumeration

[Dan Williams]

Robert Richter wrote:
> On 24.11.22 10:35:32, Dan Williams wrote:
> 
> > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> 
> > @@ -119,17 +131,22 @@ static int cxl_mem_probe(struct device *dev)
> >  		return -ENXIO;
> >  	}
> >  
> > -	device_lock(&parent_port->dev);
> > -	if (!parent_port->dev.driver) {
> > +	if (dport->rch)
> > +		endpoint_parent = parent_port->uport;
> > +	else
> > +		endpoint_parent = &parent_port->dev;
> > +
> > +	device_lock(endpoint_parent);
> > +	if (!endpoint_parent->driver) {
> >  		dev_err(dev, "CXL port topology %s not enabled\n",
> >  			dev_name(&parent_port->dev));
> 
> This must be dev_name(endpoint_parent) here.

Indeed, good catch.