Re: [PATCH v2] cxl: add RAS status unmasking for CXL

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Tue, 13 Dec 2022 14:36:28 -0700
Dave Jiang <dave.jiang@intel.com> wrote:

> By default the CXL RAS mask registers bits are defaulted to 1's and
> suppress all error reporting. If the kernel has negotiated ownership
> of error handling for CXL then unmask the mask registers by writing 0s.
> 
> Signed-off-by: Dave Jiang <dave.jiang@intel.com>

A big oopsy from me.  As you've indirectly identified, the QEMU code completely
ignores the mask register.  I'll fix that up.

> 
> ---
> 
> Based on patch posted by Ira [1] to export CXL native error reporting control.
> 
> [1]: https://lore.kernel.org/linux-cxl/20221212070627.1372402-2-ira.weiny@intel.com/
> 
> v2:
> - Add definition of PCI_EXP_LNKSTA2_FLIT. (Dan)
> - Return error for cxl_pci_ras_unmask(). (Dan)
> - Add dev_dbg() for register bits to be cleared. (Dan)
> - Check Flex Port DVSEC status. (Dan)
> ---
>  drivers/cxl/cxl.h             |    1 +
>  drivers/cxl/cxlpci.h          |    4 +++
>  drivers/cxl/pci.c             |   61 +++++++++++++++++++++++++++++++++++++++++
>  include/uapi/linux/pci_regs.h |    1 +
>  4 files changed, 67 insertions(+)
> 
> diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h
> index 1b1cf459ac77..31e795c6d537 100644
> --- a/drivers/cxl/cxl.h
> +++ b/drivers/cxl/cxl.h
> @@ -130,6 +130,7 @@ static inline int ways_to_eiw(unsigned int ways, u8 *eiw)
>  #define   CXL_RAS_UNCORRECTABLE_STATUS_MASK (GENMASK(16, 14) | GENMASK(11, 0))
>  #define CXL_RAS_UNCORRECTABLE_MASK_OFFSET 0x4
>  #define   CXL_RAS_UNCORRECTABLE_MASK_MASK (GENMASK(16, 14) | GENMASK(11, 0))
> +#define   CXL_RAS_UNCORRECTABLE_MASK_F256B_MASK BIT(8)
>  #define CXL_RAS_UNCORRECTABLE_SEVERITY_OFFSET 0x8
>  #define   CXL_RAS_UNCORRECTABLE_SEVERITY_MASK (GENMASK(16, 14) | GENMASK(11, 0))
>  #define CXL_RAS_CORRECTABLE_STATUS_OFFSET 0xC
> diff --git a/drivers/cxl/cxlpci.h b/drivers/cxl/cxlpci.h
> index 920909791bb9..d138d9cd8e33 100644
> --- a/drivers/cxl/cxlpci.h
> +++ b/drivers/cxl/cxlpci.h
> @@ -45,6 +45,10 @@
>  
>  /* CXL 2.0 8.1.8: PCIe DVSEC for Flex Bus Port */
>  #define CXL_DVSEC_PCIE_FLEXBUS_PORT				7
> +#define    CXL_DVSEC_PORT_STATUS_OFFSET			0xE
> +#define      CXL_DVSEC_PORT_STATUS_CACHE_ENABLED	BIT(0)
> +#define      CXL_DVSEC_PORT_STATUS_IO_ENABLED		BIT(1)
> +#define      CXL_DVSEC_PORT_STATUS_MEM_ENABLED		BIT(2)
>  
>  /* CXL 2.0 8.1.9: Register Locator DVSEC */
>  #define CXL_DVSEC_REG_LOCATOR					8
> diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c
> index 33083a522fd1..03691570649d 100644
> --- a/drivers/cxl/pci.c
> +++ b/drivers/cxl/pci.c
> @@ -419,6 +419,66 @@ static void disable_aer(void *pdev)
>  	pci_disable_pcie_error_reporting(pdev);
>  }
>  
> +/*
> + * CXL v3.0 6.2.3 Table 6-4
> + * The table indicates that if PCIe Flit Mode is set, then CXL is in 256B flits
> + * mode, otherwise it's 68B flits mode.
> + */
> +static bool cxl_pci_flit_256(struct pci_dev *pdev)
> +{
> +	u32 lnksta2;
> +
> +	pcie_capability_read_dword(pdev, PCI_EXP_LNKSTA2, &lnksta2);
> +	return lnksta2 & PCI_EXP_LNKSTA2_FLIT;
> +}
> +
> +static int cxl_pci_ras_unmask(struct pci_dev *pdev)
> +{
> +	struct pci_host_bridge *host_bridge = pci_find_host_bridge(pdev->bus);
> +	struct cxl_dev_state *cxlds = pci_get_drvdata(pdev);
> +	void __iomem *addr;
> +	int dvsec, rc;
> +	u16 port_sta;
> +	u32 val;
> +
> +	if (!cxlds->regs.ras)
> +		return -ENODEV;
> +
> +	/* BIOS has CXL error control */
> +	if (!host_bridge->native_cxl_error)
> +		return -EOPNOTSUPP;
> +
> +	dvsec = pci_find_dvsec_capability(pdev, PCI_DVSEC_VENDOR_ID_CXL,
> +					  CXL_DVSEC_PCIE_FLEXBUS_PORT);
> +	if (!dvsec)
> +		return -ENODEV;
> +
> +	rc = pci_read_config_word(pdev, dvsec + CXL_DVSEC_PORT_STATUS_OFFSET,
> +				  &port_sta);
> +	if (rc)
> +		return rc;
> +
> +	if (!(port_sta & (CXL_DVSEC_PORT_STATUS_CACHE_ENABLED |
> +			  CXL_DVSEC_PORT_STATUS_IO_ENABLED |
> +			  CXL_DVSEC_PORT_STATUS_MEM_ENABLED)))
> +		return -EOPNOTSUPP;

confused. Why do we care about these?  If there is a spec reference
to say don't turn on RAS reporting unless these are true, fair enough,
but if we are in this state then far worse things are going on than
just not having RAS. 

> +
> +	addr = cxlds->regs.ras + CXL_RAS_UNCORRECTABLE_MASK_OFFSET;
> +	dev_dbg(&pdev->dev, "Unmasking Uncorrectable RAS errors: %#x\n",
> +		readl(addr));
> +	val = 0;

Awkward corner.  We kind of want to blanket unmask, but the other
bits are RsvdP so we should leave any we don't understand alone - in theory
a future spec version could use them for something other than error mask bits.

> +	if (!cxl_pci_flit_256(pdev))
> +		val |= CXL_RAS_UNCORRECTABLE_MASK_F256B_MASK;
> +	writel(val, addr);
> +
> +	addr = cxlds->regs.ras + CXL_RAS_CORRECTABLE_MASK_OFFSET;
> +	dev_dbg(&pdev->dev, "Unmasking Correctable RAS errors: %#x\n",
> +		readl(addr));
> +	val = 0;
Same here - don't touch the bits we don't undestand.

> +	writel(val, addr);
> +	return 0;
> +}
> +
>  static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>  {
>  	struct cxl_register_map map;
> @@ -498,6 +558,7 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
>  
>  	if (cxlds->regs.ras) {
>  		pci_enable_pcie_error_reporting(pdev);
> +		cxl_pci_ras_unmask(pdev);
>  		rc = devm_add_action_or_reset(&pdev->dev, disable_aer, pdev);
>  		if (rc)
>  			return rc;
> diff --git a/include/uapi/linux/pci_regs.h b/include/uapi/linux/pci_regs.h
> index 82a03ea954af..576ee2ec973f 100644
> --- a/include/uapi/linux/pci_regs.h
> +++ b/include/uapi/linux/pci_regs.h
> @@ -693,6 +693,7 @@
>  #define  PCI_EXP_LNKCTL2_TX_MARGIN	0x0380 /* Transmit Margin */
>  #define  PCI_EXP_LNKCTL2_HASD		0x0020 /* HW Autonomous Speed Disable */
>  #define PCI_EXP_LNKSTA2		0x32	/* Link Status 2 */
> +#define  PCI_EXP_LNKSTA2_FLIT		BIT(10) /* Flit Mode Status */
>  #define PCI_CAP_EXP_ENDPOINT_SIZEOF_V2	0x32	/* end of v2 EPs w/ link */
>  #define PCI_EXP_SLTCAP2		0x34	/* Slot Capabilities 2 */
>  #define  PCI_EXP_SLTCAP2_IBPD	0x00000001 /* In-band PD Disable Supported */
> 
>