This patch series implements basic Designated Vendor-Specific Extended
Capabilities (DVSEC) decode for Compute eXpress Link devices, a new CPU
interconnect building upon PCIe. As a basis for the CXL support it provides
PCI init handling for detection, decode, and caching of CXL device
capabilities. Moreover, it makes use of the DVSEC Vendor ID and DVSEC ID so
as to identify a CXL capable device. (PCIe r5.0, sec 7.9.6.2)
DocLink: https://www.computeexpresslink.org/
For your reference, a parallel series of patches have been submitted to enable
lspci decode of CXL DVSEC and may be tracked.
Link: https://lore.kernel.org/linux-pci/[email protected]/
This patch makes use of pending DVSEC related header additions and the
first patch of that series is included here. It can be sorted out when the
upstream merge is done.
Link: https://lore.kernel.org/linux-pci/[email protected]/
Sample dmesg output of a CXL Type 3 device (CXL.io, CXL.mem):
[ 2.997177] pci 0000:6b:00.0: CXL: Cache- IO+ Mem+ Viral- HDMCount 1
[ 2.997188] pci 0000:6b:00.0: CXL: cap ctrl status ctrl2 status2 lock
[ 2.997201] pci 0000:6b:00.0: CXL: 001e 0002 0000 0000 0000 0000
David E. Box (1):
pci: Add Designated Vendor Specific Capability
Sean V Kelley (1):
PCI: Add basic Compute eXpress Link DVSEC decode
drivers/pci/Kconfig | 9 ++++
drivers/pci/Makefile | 1 +
drivers/pci/cxl.c | 89 +++++++++++++++++++++++++++++++++++
drivers/pci/pci.h | 7 +++
drivers/pci/probe.c | 1 +
include/linux/pci.h | 1 +
include/uapi/linux/pci_regs.h | 5 ++
7 files changed, 113 insertions(+)
create mode 100644 drivers/pci/cxl.c
--
2.26.2
Compute eXpress Link is a new CPU interconnect created with
workload accelerators in mind. The interconnect relies on PCIe Electrical
and Physical interconnect for communication. CXL devices enumerate to the
OS as an ACPI-described PCIe Root Complex Integrated Endpoint.
This patch introduces the bare minimum support by simply looking for and
caching the DVSEC CXL Extended Capability. Currently, only CXL.io (which
is mandatory to be configured by BIOS) is enabled. In future, we will
also add support for CXL.cache and CXL.mem.
DocLink: https://www.computeexpresslink.org/
Originally-by: Felipe Balbi <[email protected]>
Signed-off-by: Sean V Kelley <[email protected]>
---
drivers/pci/Kconfig | 9 +++++
drivers/pci/Makefile | 1 +
drivers/pci/cxl.c | 89 ++++++++++++++++++++++++++++++++++++++++++++
drivers/pci/pci.h | 7 ++++
drivers/pci/probe.c | 1 +
include/linux/pci.h | 1 +
6 files changed, 108 insertions(+)
create mode 100644 drivers/pci/cxl.c
diff --git a/drivers/pci/Kconfig b/drivers/pci/Kconfig
index 4bef5c2bae9f..eafb200b320b 100644
--- a/drivers/pci/Kconfig
+++ b/drivers/pci/Kconfig
@@ -115,6 +115,15 @@ config XEN_PCIDEV_FRONTEND
The PCI device frontend driver allows the kernel to import arbitrary
PCI devices from a PCI backend to support PCI driver domains.
+config PCI_CXL
+ bool "Enable PCI Compute eXpress Link"
+ depends on PCI
+ help
+ Say Y here if you want the PCI core to detect CXL devices, decode, and
+ cache the DVSEC CXL Extended Capability as configured by BIOS.
+
+ When in doubt, say N.
+
config PCI_ATS
bool
diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile
index 522d2b974e91..465eee31e999 100644
--- a/drivers/pci/Makefile
+++ b/drivers/pci/Makefile
@@ -30,6 +30,7 @@ obj-$(CONFIG_PCI_PF_STUB) += pci-pf-stub.o
obj-$(CONFIG_PCI_ECAM) += ecam.o
obj-$(CONFIG_PCI_P2PDMA) += p2pdma.o
obj-$(CONFIG_XEN_PCIDEV_FRONTEND) += xen-pcifront.o
+obj-$(CONFIG_PCI_CXL) += cxl.o
# Endpoint library must be initialized before its users
obj-$(CONFIG_PCI_ENDPOINT) += endpoint/
diff --git a/drivers/pci/cxl.c b/drivers/pci/cxl.c
new file mode 100644
index 000000000000..f02bdd7b77f2
--- /dev/null
+++ b/drivers/pci/cxl.c
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Compute eXpress Link Support
+ */
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/pci.h>
+#include <linux/pci_regs.h>
+
+#define PCI_DVSEC_VENDOR_ID_CXL 0x1e98
+#define PCI_DVSEC_ID_CXL_DEV 0x0
+
+#define PCI_CXL_CAP 0x0a
+#define PCI_CXL_CTRL 0x0c
+#define PCI_CXL_STS 0x0e
+#define PCI_CXL_CTRL2 0x10
+#define PCI_CXL_STS2 0x12
+#define PCI_CXL_LOCK 0x14
+
+#define PCI_CXL_CACHE BIT(0)
+#define PCI_CXL_IO BIT(1)
+#define PCI_CXL_MEM BIT(2)
+#define PCI_CXL_HDM_COUNT(reg) (((reg) & (3 << 4)) >> 4)
+#define PCI_CXL_VIRAL BIT(14)
+
+/*
+ * pci_find_cxl_capability - Identify and return offset to Vendor-Specific
+ * capabilities.
+ *
+ * CXL makes use of Designated Vendor-Specific Extended Capability (DVSEC)
+ * to uniquely identify both DVSEC Vendor ID and DVSEC ID aligning with
+ * PCIe r5.0, sec 7.9.6.2
+ */
+static int pci_find_cxl_capability(struct pci_dev *dev)
+{
+ u16 vendor, id;
+ int pos = 0;
+
+ while ((pos = pci_find_next_ext_capability(dev, pos,
+ PCI_EXT_CAP_ID_DVSEC))) {
+ pci_read_config_word(dev, pos + PCI_DVSEC_HEADER1,
+ &vendor);
+ pci_read_config_word(dev, pos + PCI_DVSEC_HEADER2, &id);
+ if (vendor == PCI_DVSEC_VENDOR_ID_CXL &&
+ id == PCI_DVSEC_ID_CXL_DEV)
+ return pos;
+ }
+
+ return 0;
+}
+
+void pci_cxl_init(struct pci_dev *dev)
+{
+ u16 cap, ctrl, status, ctrl2, status2, lock;
+ int pos;
+
+ /* Only for PCIe */
+ if (!pci_is_pcie(dev))
+ return;
+
+ /* Only for Device 0 Function 0, Root Complex Integrated Endpoints */
+ if (dev->devfn != 0 || (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END))
+ return;
+
+ pos = pci_find_cxl_capability(dev);
+ if (!pos)
+ return;
+
+ dev->cxl_cap = pos;
+
+ pci_read_config_word(dev, pos + PCI_CXL_CAP, &cap);
+ pci_read_config_word(dev, pos + PCI_CXL_CTRL, &ctrl);
+ pci_read_config_word(dev, pos + PCI_CXL_STS, &status);
+ pci_read_config_word(dev, pos + PCI_CXL_CTRL2, &ctrl2);
+ pci_read_config_word(dev, pos + PCI_CXL_STS2, &status2);
+ pci_read_config_word(dev, pos + PCI_CXL_LOCK, &lock);
+
+ dev_info(&dev->dev, "CXL: Cache%c IO%c Mem%c Viral%c HDMCount %d\n",
+ (cap & PCI_CXL_CACHE) ? '+' : '-',
+ (cap & PCI_CXL_IO) ? '+' : '-',
+ (cap & PCI_CXL_MEM) ? '+' : '-',
+ (cap & PCI_CXL_VIRAL) ? '+' : '-',
+ PCI_CXL_HDM_COUNT(cap));
+
+ dev_info(&dev->dev, "CXL: cap ctrl status ctrl2 status2 lock\n");
+ dev_info(&dev->dev, "CXL: %04x %04x %04x %04x %04x %04x\n",
+ cap, ctrl, status, ctrl2, status2, lock);
+}
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index 6d3f75867106..d9905e2dee95 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -469,6 +469,13 @@ static inline void pci_ats_init(struct pci_dev *d) { }
static inline void pci_restore_ats_state(struct pci_dev *dev) { }
#endif /* CONFIG_PCI_ATS */
+#ifdef CONFIG_PCI_CXL
+/* Compute eXpress Link */
+void pci_cxl_init(struct pci_dev *dev);
+#else
+static inline void pci_cxl_init(struct pci_dev *dev) { }
+#endif
+
#ifdef CONFIG_PCI_PRI
void pci_pri_init(struct pci_dev *dev);
void pci_restore_pri_state(struct pci_dev *pdev);
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 77b8a145c39b..c55df0ae8f06 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -2371,6 +2371,7 @@ static void pci_init_capabilities(struct pci_dev *dev)
pci_ptm_init(dev); /* Precision Time Measurement */
pci_aer_init(dev); /* Advanced Error Reporting */
pci_dpc_init(dev); /* Downstream Port Containment */
+ pci_cxl_init(dev); /* Compute eXpress Link */
pcie_report_downtraining(dev);
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 83ce1cdf5676..0c9145f48456 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -315,6 +315,7 @@ struct pci_dev {
u16 aer_cap; /* AER capability offset */
struct aer_stats *aer_stats; /* AER stats for this device */
#endif
+ u16 cxl_cap; /* CXL capability offset */
u8 pcie_cap; /* PCIe capability offset */
u8 msi_cap; /* MSI capability offset */
u8 msix_cap; /* MSI-X capability offset */
--
2.26.2
From: "David E. Box" <[email protected]>
Add pcie dvsec extended capability id along with helper macros to
retrieve information from the headers.
https://members.pcisig.com/wg/PCI-SIG/document/12335
Signed-off-by: David E. Box <[email protected]>
---
include/uapi/linux/pci_regs.h | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
index f9701410d3b5..c96f08d1e711 100644
--- a/include/uapi/linux/pci_regs.h
+++ b/include/uapi/linux/pci_regs.h
@@ -720,6 +720,7 @@
#define PCI_EXT_CAP_ID_DPC 0x1D /* Downstream Port Containment */
#define PCI_EXT_CAP_ID_L1SS 0x1E /* L1 PM Substates */
#define PCI_EXT_CAP_ID_PTM 0x1F /* Precision Time Measurement */
+#define PCI_EXT_CAP_ID_DVSEC 0x23 /* Desinated Vendor-Specific */
#define PCI_EXT_CAP_ID_DLF 0x25 /* Data Link Feature */
#define PCI_EXT_CAP_ID_PL_16GT 0x26 /* Physical Layer 16.0 GT/s */
#define PCI_EXT_CAP_ID_MAX PCI_EXT_CAP_ID_PL_16GT
@@ -1062,6 +1063,10 @@
#define PCI_L1SS_CTL1_LTR_L12_TH_SCALE 0xe0000000 /* LTR_L1.2_THRESHOLD_Scale */
#define PCI_L1SS_CTL2 0x0c /* Control 2 Register */
+/* Designated Vendor-Specific (DVSEC, PCI_EXT_CAP_ID_DVSEC) */
+#define PCI_DVSEC_HEADER1 0x4 /* Vendor-Specific Header1 */
+#define PCI_DVSEC_HEADER2 0x8 /* Vendor-Specific Header2 */
+
/* Data Link Feature */
#define PCI_DLF_CAP 0x04 /* Capabilities Register */
#define PCI_DLF_EXCHANGE_ENABLE 0x80000000 /* Data Link Feature Exchange Enable */
--
2.26.2
On Fri, May 15, 2020 at 10:55:28AM -0700, Sean V Kelley wrote:
> Compute eXpress Link is a new CPU interconnect created with
> workload accelerators in mind. The interconnect relies on PCIe Electrical
> and Physical interconnect for communication. CXL devices enumerate to the
> OS as an ACPI-described PCIe Root Complex Integrated Endpoint.
>
> This patch introduces the bare minimum support by simply looking for and
> caching the DVSEC CXL Extended Capability. Currently, only CXL.io (which
> is mandatory to be configured by BIOS) is enabled. In future, we will
> also add support for CXL.cache and CXL.mem.
This looks fine, but AFAICT, it doesn't *do* anything yet (except
print a few things to dmesg). We don't normally merge code until it
adds some new functionality. So just FYI that I'll wait until that
new functionality comes along and then merge this as part of that
series. But let me know if I'm missing something.
> + dev_info(&dev->dev, "CXL: Cache%c IO%c Mem%c Viral%c HDMCount %d\n",
> + (cap & PCI_CXL_CACHE) ? '+' : '-',
> + (cap & PCI_CXL_IO) ? '+' : '-',
> + (cap & PCI_CXL_MEM) ? '+' : '-',
> + (cap & PCI_CXL_VIRAL) ? '+' : '-',
> + PCI_CXL_HDM_COUNT(cap));
These could use pci_info() and FLAG(), as in pcie_init().
> + dev_info(&dev->dev, "CXL: cap ctrl status ctrl2 status2 lock\n");
> + dev_info(&dev->dev, "CXL: %04x %04x %04x %04x %04x %04x\n",
> + cap, ctrl, status, ctrl2, status2, lock);
> +}
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -315,6 +315,7 @@ struct pci_dev {
> u16 aer_cap; /* AER capability offset */
> struct aer_stats *aer_stats; /* AER stats for this device */
> #endif
> + u16 cxl_cap; /* CXL capability offset */
Wrap in #ifdef PCI_CXL.
Hi,
On 15 May 2020, at 14:04, Bjorn Helgaas wrote:
> On Fri, May 15, 2020 at 10:55:28AM -0700, Sean V Kelley wrote:
>> Compute eXpress Link is a new CPU interconnect created with
>> workload accelerators in mind. The interconnect relies on PCIe
>> Electrical
>> and Physical interconnect for communication. CXL devices enumerate to
>> the
>> OS as an ACPI-described PCIe Root Complex Integrated Endpoint.
>>
>> This patch introduces the bare minimum support by simply looking for
>> and
>> caching the DVSEC CXL Extended Capability. Currently, only CXL.io
>> (which
>> is mandatory to be configured by BIOS) is enabled. In future, we will
>> also add support for CXL.cache and CXL.mem.
>
> This looks fine, but AFAICT, it doesn't *do* anything yet (except
> print a few things to dmesg). We don't normally merge code until it
> adds some new functionality. So just FYI that I'll wait until that
> new functionality comes along and then merge this as part of that
> series. But let me know if I'm missing something.
>
Correct. Understood. I’ve additional changes for CXL.mem/.cache.
I’ll queue those up.
>> + dev_info(&dev->dev, "CXL: Cache%c IO%c Mem%c Viral%c HDMCount
>> %d\n",
>> + (cap & PCI_CXL_CACHE) ? '+' : '-',
>> + (cap & PCI_CXL_IO) ? '+' : '-',
>> + (cap & PCI_CXL_MEM) ? '+' : '-',
>> + (cap & PCI_CXL_VIRAL) ? '+' : '-',
>> + PCI_CXL_HDM_COUNT(cap));
>
> These could use pci_info() and FLAG(), as in pcie_init().
Will do.
>
>> + dev_info(&dev->dev, "CXL: cap ctrl status ctrl2 status2 lock\n");
>> + dev_info(&dev->dev, "CXL: %04x %04x %04x %04x %04x %04x\n",
>> + cap, ctrl, status, ctrl2, status2, lock);
>> +}
>
>> --- a/include/linux/pci.h
>> +++ b/include/linux/pci.h
>> @@ -315,6 +315,7 @@ struct pci_dev {
>> u16 aer_cap; /* AER capability offset */
>> struct aer_stats *aer_stats; /* AER stats for this device */
>> #endif
>> + u16 cxl_cap; /* CXL capability offset */
>
> Wrap in #ifdef PCI_CXL.
Will do.
Thanks,
Sean