Re: [PATCH v8 2/3] PCI: iproc: retry request when CRS returned from EP

From: Bjorn Helgaas
Date: Mon Aug 28 2017 - 17:48:01 EST


On Thu, Aug 24, 2017 at 10:34:25AM +0530, Oza Pawandeep wrote:
> PCIe spec r3.1, sec 2.3.2
> If CRS software visibility is not enabled, the RC must reissue the
> config request as a new request.
>
> - If CRS software visibility is enabled,
> - for a config read of Vendor ID, the RC must return 0x0001 data
> - for all other config reads/writes, the RC must reissue the
> request
>
> iproc PCIe Controller spec:
> 4.7.3.3. Retry Status On Configuration Cycle
> Endpoints are allowed to generate retry status on configuration
> cycles. In this case, the RC needs to re-issue the request. The IP
> does not handle this because the number of configuration cycles needed
> will probably be less than the total number of non-posted operations
> needed.
>
> When a retry status is received on the User RX interface for a
> configuration request that was sent on the User TX interface,
> it will be indicated with a completion with the CMPL_STATUS field set
> to 2=CRS, and the user will have to find the address and data values
> and send a new transaction on the User TX interface.
> When the internal configuration space returns a retry status during a
> configuration cycle (user_cscfg = 1) on the Command/Status interface,
> the pcie_cscrs will assert with the pcie_csack signal to indicate the
> CRS status.
> When the CRS Software Visibility Enable register in the Root Control
> register is enabled, the IP will return the data value to 0x0001 for
> the Vendor ID value and 0xffff (all 1âs) for the rest of the data in
> the request for reads of offset 0 that return with CRS status. This
> is true for both the User RX Interface and for the Command/Status
> interface. When CRS Software Visibility is enabled, the CMPL_STATUS
> field of the completion on the User RX Interface will not be 2=CRS and
> the pcie_cscrs signal will not assert on the Command/Status interface.
>
> Per PCIe r3.1, sec 2.3.2, config requests that receive completions
> with Configuration Request Retry Status (CRS) should be reissued by
> the hardware except reads of the Vendor ID when CRS Software
> Visibility is enabled.
>
> This hardware never reissues configuration requests when it receives
> CRS completions.
> Note that, neither PCIe host bridge nor PCIe core re-issues the
> request for any configuration offset.
>
> For config reads, this hardware returns CFG_RETRY_STATUS data when
> it receives a CRS completion for a config read, regardless of the
> address of the read or the CRS Software Visibility Enable bit.
>
> This patch implements iproc_pcie_config_read which gets called for
> Stingray, if it receives a CRS completion, it retries reading it again.
> In case of timeout, it returns 0xffffffff.
> For other iproc based SOC, it falls back to PCI generic APIs.
>
> Signed-off-by: Oza Pawandeep <oza.oza@xxxxxxxxxxxx>
>
> diff --git a/drivers/pci/host/pcie-iproc.c b/drivers/pci/host/pcie-iproc.c
> index 61d9be6..37f4adf 100644
> --- a/drivers/pci/host/pcie-iproc.c
> +++ b/drivers/pci/host/pcie-iproc.c
> @@ -68,6 +68,9 @@
> #define APB_ERR_EN_SHIFT 0
> #define APB_ERR_EN BIT(APB_ERR_EN_SHIFT)
>
> +#define CFG_RETRY_STATUS 0xffff0001
> +#define CFG_RETRY_STATUS_TIMEOUT_US 500000 /* 500 milli-seconds. */
> +
> /* derive the enum index of the outbound/inbound mapping registers */
> #define MAP_REG(base_reg, index) ((base_reg) + (index) * 2)
>
> @@ -473,6 +476,64 @@ static void __iomem *iproc_pcie_map_ep_cfg_reg(struct iproc_pcie *pcie,
> return (pcie->base + offset);
> }
>
> +static unsigned int iproc_pcie_cfg_retry(void __iomem *cfg_data_p)
> +{
> + int timeout = CFG_RETRY_STATUS_TIMEOUT_US;
> + unsigned int data;
> +
> + /*
> + * As per PCIe spec r3.1, sec 2.3.2, CRS Software
> + * Visibility only affects config read of the Vendor ID.
> + * For config write or any other config read the Root must
> + * automatically re-issue configuration request again as a
> + * new request.
> + *
> + * For config reads, this hardware returns CFG_RETRY_STATUS data when
> + * it receives a CRS completion for a config read, regardless of the
> + * address of the read or the CRS Software Visibility Enable bit. As a
> + * partial workaround for this, we retry in software any read that
> + * returns CFG_RETRY_STATUS.
> + */
> + data = readl(cfg_data_p);
> + while (data == CFG_RETRY_STATUS && timeout--) {
> + udelay(1);
> + data = readl(cfg_data_p);
> + }
> +
> + if (data == CFG_RETRY_STATUS)
> + data = 0xffffffff;
> +
> + return data;
> +}
> +
> +static int iproc_pcie_config_read(struct pci_bus *bus, unsigned int devfn,
> + int where, int size, u32 *val)
> +{
> + struct iproc_pcie *pcie = iproc_data(bus);
> + unsigned int slot = PCI_SLOT(devfn);
> + unsigned int fn = PCI_FUNC(devfn);
> + unsigned int busno = bus->number;
> + void __iomem *cfg_data_p;
> + unsigned int data;
> +
> + /* root complex access. */
> + if (busno == 0)
> + return pci_generic_config_read32(bus, devfn, where, size, val);

It sounds like Stingray advertises CRS SV support in its Root Capabilities
register. I think we should mask out PCI_EXP_RTCAP_CRSVIS so we don't
advertise it. That will keep Linux from trying to enable it. I know the
hardware doesn't look at PCI_EXP_RTCTL_CRSSVE, but there's no point in
confusing users reading the lspci output.

We did something similar with f09f8735fb9c ("PCI: xgene: Disable
Configuration Request Retry Status for v1 silicon").

I tried to do this in the patch I pushed to pci/host-iproc.

> +
> + cfg_data_p = iproc_pcie_map_ep_cfg_reg(pcie, busno, slot, fn, where);
> +
> + if (!cfg_data_p)
> + return PCIBIOS_DEVICE_NOT_FOUND;
> +
> + data = iproc_pcie_cfg_retry(cfg_data_p);
> +
> + *val = data;
> + if (size <= 2)
> + *val = (data >> (8 * (where & 3))) & ((1 << (size * 8)) - 1);
> +
> + return PCIBIOS_SUCCESSFUL;
> +}
> +
> /**
> * Note access to the configuration registers are protected at the higher layer
> * by 'pci_lock' in drivers/pci/access.c
> @@ -567,8 +628,13 @@ static int iproc_pcie_config_read32(struct pci_bus *bus, unsigned int devfn,
> int where, int size, u32 *val)
> {
> int ret;
> + struct iproc_pcie *pcie = iproc_data(bus);
>
> iproc_pcie_apb_err_disable(bus, true);
> + if (pcie->type == IPROC_PCIE_PAXB_V2)
> + ret = iproc_pcie_config_read(bus, devfn, where, size, val);
> + else
> + ret = pci_generic_config_read32(bus, devfn, where, size, val);
> ret = pci_generic_config_read32(bus, devfn, where, size, val);

This last pci_generic_config_read32() call looks like a duplicate.

> iproc_pcie_apb_err_disable(bus, false);
>
> --
> 1.9.1
>