From: Dong Wang <wangdong28@xxxxxxxxxx>
When enabling VMD function in UEFI setup, the physical slot of the M.2
NVMe device connected to the VMD device cannot be detected. Here is
the result from lspci ("Physical Slot" field is NOT shown):
10001:01:00.0 Non-Volatile memory controller: Intel Corporation NVMe
Datacenter SSD [3DNAND, Beta Rock Controller] (prog-if 02 [NVM Express])
Subsystem: Intel Corporation NVMe Datacenter SSD [3DNAND] SE M.2 (P4511)
Generally, the physical slot (/sys/bus/pci/slots) will be created via
either ACPI walking path during kernel init or hotplug path:
ACPI walking path:
pcibios_add_bus
acpi_pci_add_bus
acpi_pci_slot_enumerate
acpi_walk_namespace
register_slot
pci_create_slot
hotplug path:
__pci_hp_initialize
pci_create_slot
[M.2 NVMe Device]
A. VMD disabled
When VMD is disabled, NVMe will be discovered during bus scanning and
recognized as acpi device. In this case, the physical slot is created
via the ACPI walking path.
B. VMD enabled
vmd_enable_domain() invokes pcibios_add_bus(). This means that it goes
through the ACPI walking path. However, acpi_pci_add_bus() returns
directly becase the statment "!ACPI_HANDLE(bus->bridge)" is true.
See the following code snippet:
void acpi_pci_add_bus(struct pci_bus *bus)
{
...
if (acpi_pci_disabled || !bus->bridge || !ACPI_HANDLE(bus->bridge))
return;
...
}
Since VMD creates its own root bus and devices of VMD are attached to
the bus, those devices are non-ACPI devices. That's why
"!ACPI_HANDLE(bus->bridge)" returns true.
In addition, M.2 NVMe devices does not have the hotplug capability.
Here is the quote from PCI Express M.2 Specification (Revision 5.0,
Version 1.0):
CAUTION: M.2 Add-in Cards are not designed or intended to support
Hot-Swap or Hot-Plug connections. Performing Hot-Swap or Hot-Plug
may pose danger to the M.2 Add-in Card, to the system Platform,
and to the person performing this act.
M.2 NVMe devices (non-ACPI devices and no hotplug capability) connected
to the VMD device cannot meet the above-mentioned paths. The corresponding
slot info of the M.2 NVMe controller cannot be created in
/sys/bus/pci/slots.
Fix this issue by checking the available physical slot number in
slot capabilities register. If the physical slot number is available,
create the slot info accordingly. The following lspci output shows the
available slot info with applying this patch:
10001:01:00.0 Non-Volatile memory controller: Intel Corporation NVMe
Datacenter SSD [3DNAND, Beta Rock Controller] (prog-if 02 [NVM Express])
Subsystem: Intel Corporation NVMe Datacenter SSD [3DNAND] SE M.2 (P4511)
Physical Slot: 16
[U.2 NVMe device]
A. VMD disabled
Same as M.2 NVMe Device case "A".
B. VMD enabled
Same as M.2 NVMe Device case "B".
The hotplug of the U.2 device is optional (See "PCI Express SFF-8639 Module
Specification" for detail). The U.2 NVMe controller with hotplug capability
connected to the VMD device can meet the hotplug path, so the slot info can
be shown correctly via the lspci utility (without this patch):
10000:82:00.0 Non-Volatile memory controller: Intel Corporation NVMe
Datacenter SSD [3DNAND, Beta Rock Controller] (prog-if 02 [NVM Express])
Subsystem: Lenovo Thinksystem U.2 P4610 NVMe SSD
Physical Slot: 64
For U.2 NVMe controller without hotplug capability, this patch is needed
to fix the missing slot info.
Suggested-and-reviewed-by: Adrian Huang <ahuang12@xxxxxxxxxx>
Signed-off-by: Dong Wang <wangdong28@xxxxxxxxxx>
---
drivers/pci/pci-acpi.c | 30 +++++++++++++++++++++++++++++-
1 file changed, 29 insertions(+), 1 deletion(-)
diff --git a/drivers/pci/pci-acpi.c b/drivers/pci/pci-acpi.c
index 0045750..d7b3462 100644
--- a/drivers/pci/pci-acpi.c
+++ b/drivers/pci/pci-acpi.c
@@ -884,6 +884,27 @@ acpi_status pci_acpi_add_pm_notifier(struct acpi_device *dev,
return acpi_add_pm_notifier(dev, &pci_dev->dev, pci_acpi_wake_dev);
}
+#define SLOT_NAME_SIZE 5
+
+static void pci_check_extra_slot_register(struct pci_bus *bus)
+{
+ struct pci_dev *pdev = bus->self;
+ char slot_name[SLOT_NAME_SIZE];
+ struct pci_slot *pci_slot;
+ u32 slot_cap, slot_nr;
+
+ if (!is_vmd(bus) || !pdev || pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, &slot_cap))
+ return;
+
+ if (!(slot_cap & PCI_EXP_SLTCAP_HPC)) {
+ slot_nr = (slot_cap & PCI_EXP_SLTCAP_PSN) >> 19;
+ snprintf(slot_name, SLOT_NAME_SIZE, "%u", slot_nr);
+ pci_slot = pci_create_slot(bus, 0, slot_name, NULL);
+ if (IS_ERR(pci_slot))
+ pr_err("pci_create_slot returned %ld\n", PTR_ERR(pci_slot));
+ }
+}
+
/*
* _SxD returns the D-state with the highest power
* (lowest D-state number) supported in the S-state "x".
@@ -1202,9 +1223,16 @@ void acpi_pci_add_bus(struct pci_bus *bus)
union acpi_object *obj;
struct pci_host_bridge *bridge;
- if (acpi_pci_disabled || !bus->bridge || !ACPI_HANDLE(bus->bridge))
+ if (acpi_pci_disabled || !bus->bridge)
return;
+
+ if (!ACPI_HANDLE(bus->bridge)) {
+ pci_check_extra_slot_register(bus);
+ return;
+ }
+
+
acpi_pci_slot_enumerate(bus);
acpiphp_enumerate_slots(bus);
--
1.8.3.1
