From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-4.2 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_PASS, UNWANTED_LANGUAGE_BODY autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 926F2C43381 for ; Mon, 25 Feb 2019 09:35:14 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 54FA92087C for ; Mon, 25 Feb 2019 09:35:14 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726590AbfBYJfN convert rfc822-to-8bit (ORCPT ); Mon, 25 Feb 2019 04:35:13 -0500 Received: from mga04.intel.com ([192.55.52.120]:49954 "EHLO mga04.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726498AbfBYJfM (ORCPT ); Mon, 25 Feb 2019 04:35:12 -0500 X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga104.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 25 Feb 2019 01:35:11 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.58,411,1544515200"; d="scan'208";a="146291219" Received: from lftan-mobl.gar.corp.intel.com (HELO ubuntu) ([10.226.248.73]) by fmsmga002.fm.intel.com with SMTP; 25 Feb 2019 01:35:10 -0800 Received: by ubuntu (sSMTP sendmail emulation); Mon, 25 Feb 2019 17:35:08 +0800 Message-ID: <1551087307.2534.16.camel@intel.com> Subject: Re: [PATCH v4 1/3] PCI: altera: Add Stratix 10 PCIe support From: Ley Foon Tan To: Lorenzo Pieralisi Cc: Bjorn Helgaas , linux-kernel@vger.kernel.org, linux-pci@vger.kernel.org, devicetree@vger.kernel.org, lftan.linux@gmail.com Date: Mon, 25 Feb 2019 17:35:07 +0800 In-Reply-To: <20190219162313.GB21833@e107981-ln.cambridge.arm.com> References: <1550157638-25112-1-git-send-email-ley.foon.tan@intel.com> <1550157638-25112-2-git-send-email-ley.foon.tan@intel.com> <20190219162313.GB21833@e107981-ln.cambridge.arm.com> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8BIT X-Mailer: Evolution 3.18.5.2-0ubuntu3.1 Mime-Version: 1.0 Sender: linux-pci-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-pci@vger.kernel.org On Tue, 2019-02-19 at 16:23 +0000, Lorenzo Pieralisi wrote: > On Thu, Feb 14, 2019 at 11:20:36PM +0800, Ley Foon Tan wrote: > > > > Add PCIe Root Port support for Stratix 10 device. > > > > Main differences: > Main differences with what ? We need to rewrite this commit log. Differences compare with Cyclone V and Arria 10 devices.  I will rewrite this. > > > > > - HIP interface to access Root Port configuration register. > > - TLP programming flow: > >   - One REG0 register > >   - Don't need to check alignment > > > > Signed-off-by: Ley Foon Tan > > --- > >  drivers/pci/controller/pcie-altera.c |  246 > > ++++++++++++++++++++++++++++++---- > >  1 files changed, 222 insertions(+), 24 deletions(-) > > > > diff --git a/drivers/pci/controller/pcie-altera.c > > b/drivers/pci/controller/pcie-altera.c > > index 7d05e51..76bb6a6 100644 > > --- a/drivers/pci/controller/pcie-altera.c > > +++ b/drivers/pci/controller/pcie-altera.c > > @@ -11,6 +11,7 @@ > >  #include > >  #include > >  #include > > +#include > >  #include > >  #include > >  #include > > @@ -37,7 +38,12 @@ > >  #define RP_LTSSM_MASK 0x1f > >  #define LTSSM_L0 0xf > >   > > -#define PCIE_CAP_OFFSET 0x80 > > +#define S10_RP_TX_CNTRL 0x2004 > > +#define S10_RP_RXCPL_REG 0x2008 > > +#define S10_RP_RXCPL_STATUS 0x200C > > +#define S10_RP_CFG_ADDR(pcie, reg) \ > > + (((pcie)->hip_base) + (reg) + (1 << 20)) > > + > >  /* TLP configuration type 0 and 1 */ > >  #define TLP_FMTTYPE_CFGRD0 0x04 /* > > Configuration Read Type 0 */ > >  #define TLP_FMTTYPE_CFGWR0 0x44 /* > > Configuration Write Type 0 */ > > @@ -49,18 +55,19 @@ > >  #define RP_DEVFN 0 > >  #define TLP_REQ_ID(bus, devfn) (((bus) << 8) | > > (devfn)) > >  #define TLP_CFGRD_DW0(pcie, bus) > > \ > > -    ((((bus == pcie->root_bus_nr) ? TLP_FMTTYPE_CFGRD0 > > \ > > -     : TLP_FMTTYPE_CFGRD1) << 24) | > > \ > > -     TLP_PAYLOAD_SIZE) > > + ((((bus == pcie->root_bus_nr) ? pcie->pcie_data->cfgrd0 > > \ > > + : pcie->pcie_data->cfgrd1) << 24) > > | \ > > + TLP_PAYLOAD_SIZE) > >  #define TLP_CFGWR_DW0(pcie, bus) > > \ > > -    ((((bus == pcie->root_bus_nr) ? TLP_FMTTYPE_CFGWR0 > > \ > > -     : TLP_FMTTYPE_CFGWR1) << 24) | > > \ > > -     TLP_PAYLOAD_SIZE) > > + ((((bus == pcie->root_bus_nr) ? pcie->pcie_data->cfgwr0 > > \ > > + : pcie->pcie_data->cfgwr1) << 24) > > | \ > > + TLP_PAYLOAD_SIZE) > >  #define TLP_CFG_DW1(pcie, tag, be) \ > > -    (((TLP_REQ_ID(pcie->root_bus_nr,  RP_DEVFN)) << 16) | (tag << > > 8) | (be)) > > + (((TLP_REQ_ID(pcie->root_bus_nr,  RP_DEVFN)) << 16) | (tag > > << 8) | (be)) > >  #define TLP_CFG_DW2(bus, devfn, offset) \ > >   (((bus) << 24) | ((devfn) << 16) | > > (offset)) > >  #define TLP_COMP_STATUS(s) (((s) >> 13) & 7) > > +#define TLP_BYTE_COUNT(s) (((s) >> 0) & 0xfff) > >  #define TLP_HDR_SIZE 3 > >  #define TLP_LOOP 500 > >   > > @@ -69,14 +76,43 @@ > >   > >  #define DWORD_MASK 3 > >   > > +#define S10_TLP_FMTTYPE_CFGRD0 0x05 > > +#define S10_TLP_FMTTYPE_CFGRD1 0x04 > > +#define S10_TLP_FMTTYPE_CFGWR0 0x45 > > +#define S10_TLP_FMTTYPE_CFGWR1 0x44 > > + > > +enum altera_pcie_version { > > + ALTERA_PCIE_V1 = 0, > > + ALTERA_PCIE_V2, > > +}; > > + > >  struct altera_pcie { > >   struct platform_device *pdev; > > - void __iomem *cra_base; /* DT Cra */ > > + void __iomem *cra_base; > > + void __iomem *hip_base; > >   int irq; > >   u8 root_bus_nr; > >   struct irq_domain *irq_domain; > >   struct resource bus_range; > >   struct list_head resources; > > + const struct altera_pcie_data *pcie_data; > > +}; > > + > > +struct altera_pcie_data { > > + int (*tlp_read_pkt)(struct altera_pcie *pcie, u32 *value); > > + void (*tlp_write_pkt)(struct altera_pcie *pcie, u32 > > *headers, > > +       u32 data, bool align); > > + bool (*get_link_status)(struct altera_pcie *pcie); > > + int (*rp_read_cfg)(struct altera_pcie *pcie, int where, > > +    int size, u32 *value); > > + int (*rp_write_cfg)(struct altera_pcie *pcie, u8 bus, int > > where, > > +     int size, u32 value); > > + enum altera_pcie_version version; > > + u32 cap_offset; /* PCIe capability > > structure register offset */ > This is a duplication of struct pci_dev.pcie_cap right ? This is register offset in PCIe Rootport IP to access PCIe Cap structure register of directly RP. This allow us to access PCIe cap struct register without pci_dev. > > > > > + u32 cfgrd0; > > + u32 cfgrd1; > > + u32 cfgwr0; > > + u32 cfgwr1; > >  }; > >   > >  struct tlp_rp_regpair_t { > > @@ -101,6 +137,15 @@ static bool altera_pcie_link_up(struct > > altera_pcie *pcie) > >   return !!((cra_readl(pcie, RP_LTSSM) & RP_LTSSM_MASK) == > > LTSSM_L0); > >  } > >   > > +static bool s10_altera_pcie_link_up(struct altera_pcie *pcie) > > +{ > > + void __iomem *addr = S10_RP_CFG_ADDR(pcie, > > +    pcie->pcie_data->cap_offset + > > +    PCI_EXP_LNKSTA); > > + > > + return !!(readw(addr) & PCI_EXP_LNKSTA_DLLLA); > > +} > > + > >  /* > >   * Altera PCIe port uses BAR0 of RC's configuration space as the > > translation > >   * from PCI bus to native BUS.  Entire DDR region is mapped into > > PCIe space > > @@ -128,12 +173,18 @@ static void tlp_write_tx(struct altera_pcie > > *pcie, > >   cra_writel(pcie, tlp_rp_regdata->ctrl, RP_TX_CNTRL); > >  } > >   > > +static void s10_tlp_write_tx(struct altera_pcie *pcie, u32 reg0, > > u32 ctrl) > > +{ > > + cra_writel(pcie, reg0, RP_TX_REG0); > > + cra_writel(pcie, ctrl, S10_RP_TX_CNTRL); > > +} > > + > >  static bool altera_pcie_valid_device(struct altera_pcie *pcie, > >        struct pci_bus *bus, int dev) > >  { > >   /* If there is no link, then there is no device */ > >   if (bus->number != pcie->root_bus_nr) { > > - if (!altera_pcie_link_up(pcie)) > > + if (!pcie->pcie_data->get_link_status(pcie)) > >   return false; > >   } > >   > > @@ -183,6 +234,46 @@ static int tlp_read_packet(struct altera_pcie > > *pcie, u32 *value) > >   return PCIBIOS_DEVICE_NOT_FOUND; > >  } > >   > > +static int s10_tlp_read_packet(struct altera_pcie *pcie, u32 > > *value) > > +{ > > + int i; > > + u32 ctrl; > > + u32 comp_status; > > + u32 dw[4]; > > + u32 count = 0; > > + > > + for (i = 0; i < TLP_LOOP; i++) { > > + ctrl = cra_readl(pcie, S10_RP_RXCPL_STATUS); > > + if (!(ctrl & RP_RXCPL_SOP)) > > + continue; > > + > > + /* Read first DW */ > > + dw[count++] = cra_readl(pcie, S10_RP_RXCPL_REG); > > + > > + /* Poll for EOP */ > > + for (i = 0; i < TLP_LOOP; i++) { > > + ctrl = cra_readl(pcie, > > S10_RP_RXCPL_STATUS); > > + dw[count++] = cra_readl(pcie, > > S10_RP_RXCPL_REG); > I think you'd better add a check on count lest it can overflow dw[]. Okay, will check that. > > > > > + if (ctrl & RP_RXCPL_EOP) { > > + comp_status = > > TLP_COMP_STATUS(dw[1]); > > + if (comp_status) > > + return > > PCIBIOS_DEVICE_NOT_FOUND; > > + > > + if (value && > > +     TLP_BYTE_COUNT(dw[1]) == > > sizeof(u32) && > > +     count >= 3) > > + *value = dw[3]; > > + > > + return PCIBIOS_SUCCESSFUL; > > + } > > + } > > + > > + udelay(5); > > + } > > + > > + return PCIBIOS_DEVICE_NOT_FOUND; > > +} > > + > >  static void tlp_write_packet(struct altera_pcie *pcie, u32 > > *headers, > >        u32 data, bool align) > >  { > > @@ -210,6 +301,15 @@ static void tlp_write_packet(struct > > altera_pcie *pcie, u32 *headers, > >   tlp_write_tx(pcie, &tlp_rp_regdata); > >  } > >   > > +static void s10_tlp_write_packet(struct altera_pcie *pcie, u32 > > *headers, > > +  u32 data, bool dummy) > > +{ > > + s10_tlp_write_tx(pcie, headers[0], RP_TX_SOP); > > + s10_tlp_write_tx(pcie, headers[1], 0); > > + s10_tlp_write_tx(pcie, headers[2], 0); > > + s10_tlp_write_tx(pcie, data, RP_TX_EOP); > > +} > > + > >  static int tlp_cfg_dword_read(struct altera_pcie *pcie, u8 bus, > > u32 devfn, > >         int where, u8 byte_en, u32 *value) > >  { > > @@ -219,9 +319,9 @@ static int tlp_cfg_dword_read(struct > > altera_pcie *pcie, u8 bus, u32 devfn, > >   headers[1] = TLP_CFG_DW1(pcie, TLP_READ_TAG, byte_en); > >   headers[2] = TLP_CFG_DW2(bus, devfn, where); > >   > > - tlp_write_packet(pcie, headers, 0, false); > > + pcie->pcie_data->tlp_write_pkt(pcie, headers, 0, false); > >   > > - return tlp_read_packet(pcie, value); > > + return pcie->pcie_data->tlp_read_pkt(pcie, value); > >  } > >   > >  static int tlp_cfg_dword_write(struct altera_pcie *pcie, u8 bus, > > u32 devfn, > > @@ -236,11 +336,11 @@ static int tlp_cfg_dword_write(struct > > altera_pcie *pcie, u8 bus, u32 devfn, > >   > >   /* check alignment to Qword */ > >   if ((where & 0x7) == 0) > > - tlp_write_packet(pcie, headers, value, true); > > + pcie->pcie_data->tlp_write_pkt(pcie, headers, > > value, true); > >   else > > - tlp_write_packet(pcie, headers, value, false); > > + pcie->pcie_data->tlp_write_pkt(pcie, headers, > > value, false); > >   > > - ret = tlp_read_packet(pcie, NULL); > > + ret = pcie->pcie_data->tlp_read_pkt(pcie, NULL); > >   if (ret != PCIBIOS_SUCCESSFUL) > >   return ret; > >   > > @@ -254,6 +354,53 @@ static int tlp_cfg_dword_write(struct > > altera_pcie *pcie, u8 bus, u32 devfn, > >   return PCIBIOS_SUCCESSFUL; > >  } > >   > > +static int s10_rp_read_cfg(struct altera_pcie *pcie, int where, > > +    int size, u32 *value) > > +{ > > + void *addr = S10_RP_CFG_ADDR(pcie, where); > > + > > + switch (size) { > > + case 1: > > + *value = readb(addr); > > + break; > > + case 2: > > + *value = readw(addr); > > + break; > > + default: > > + *value = readl(addr); > > + break; > > + } > This boilerplace could be avoided if you used generic config > accessors (ie pci_generic_config_read()/write()). Okay. I will change this. > > > > > + > > + return PCIBIOS_SUCCESSFUL; > > +} > > + > > +static int s10_rp_write_cfg(struct altera_pcie *pcie, u8 bus, int > > where, > > +     int size, u32 value) > > +{ > > + void *addr = S10_RP_CFG_ADDR(pcie, where); > > + > > + switch (size) { > > + case 1: > > + writeb(value, addr); > > + break; > > + case 2: > > + writew(value, addr); > > + break; > > + default: > > + writel(value, addr); > > + break; > > + } > Ditto. Noted. > > > > > + > > + /* > > +  * Monitor changes to PCI_PRIMARY_BUS register on root > > port > > +  * and update local copy of root bus number accordingly. > > +  */ > > + if (bus == pcie->root_bus_nr && where == PCI_PRIMARY_BUS) > > + pcie->root_bus_nr = value & 0xff; > > + > > + return PCIBIOS_SUCCESSFUL; > > +} > > + > >  static int _altera_pcie_cfg_read(struct altera_pcie *pcie, u8 > > busno, > >    unsigned int devfn, int where, > > int size, > >    u32 *value) > > @@ -262,6 +409,9 @@ static int _altera_pcie_cfg_read(struct > > altera_pcie *pcie, u8 busno, > >   u32 data; > >   u8 byte_en; > >   > > + if (busno == pcie->root_bus_nr && pcie->pcie_data- > > >rp_read_cfg) > > + return pcie->pcie_data->rp_read_cfg(pcie, where, > > size, value); > > + > >   switch (size) { > >   case 1: > >   byte_en = 1 << (where & 3); > > @@ -302,6 +452,10 @@ static int _altera_pcie_cfg_write(struct > > altera_pcie *pcie, u8 busno, > >   u32 shift = 8 * (where & 3); > >   u8 byte_en; > >   > > + if (busno == pcie->root_bus_nr && pcie->pcie_data- > > >rp_write_cfg) > > + return pcie->pcie_data->rp_write_cfg(pcie, busno, > > where, > > + size, > > value); > > + > >   switch (size) { > >   case 1: > >   data32 = (value & 0xff) << shift; > > @@ -365,7 +519,8 @@ static int altera_read_cap_word(struct > > altera_pcie *pcie, u8 busno, > >   int ret; > >   > >   ret = _altera_pcie_cfg_read(pcie, busno, devfn, > > -     PCIE_CAP_OFFSET + offset, > > sizeof(*value), > > +     pcie->pcie_data->cap_offset + > > offset, > > +     sizeof(*value), > >       &data); > >   *value = data; > >   return ret; > > @@ -375,7 +530,8 @@ static int altera_write_cap_word(struct > > altera_pcie *pcie, u8 busno, > >    unsigned int devfn, int offset, > > u16 value) > >  { > >   return _altera_pcie_cfg_write(pcie, busno, devfn, > > -       PCIE_CAP_OFFSET + offset, > > sizeof(value), > > +       pcie->pcie_data->cap_offset > > + offset, > > +       sizeof(value), > >         value); > >  } > >   > > @@ -403,7 +559,7 @@ static void altera_wait_link_retrain(struct > > altera_pcie *pcie) > >   /* Wait for link is up */ > >   start_jiffies = jiffies; > >   for (;;) { > > - if (altera_pcie_link_up(pcie)) > > + if (pcie->pcie_data->get_link_status(pcie)) > >   break; > >   > >   if (time_after(jiffies, start_jiffies + > > LINK_UP_TIMEOUT)) { > > @@ -418,7 +574,7 @@ static void altera_pcie_retrain(struct > > altera_pcie *pcie) > >  { > >   u16 linkcap, linkstat, linkctl; > >   > > - if (!altera_pcie_link_up(pcie)) > > + if (!pcie->pcie_data->get_link_status(pcie)) > >   return; > >   > >   /* > > @@ -540,12 +696,20 @@ static int altera_pcie_parse_dt(struct > > altera_pcie *pcie) > >   struct device *dev = &pcie->pdev->dev; > >   struct platform_device *pdev = pcie->pdev; > >   struct resource *cra; > > + struct resource *hip; > >   > >   cra = platform_get_resource_byname(pdev, IORESOURCE_MEM, > > "Cra"); > >   pcie->cra_base = devm_ioremap_resource(dev, cra); > >   if (IS_ERR(pcie->cra_base)) > >   return PTR_ERR(pcie->cra_base); > >   > > + if (pcie->pcie_data->version == ALTERA_PCIE_V2) { > > + hip = platform_get_resource_byname(pdev, > > IORESOURCE_MEM, "Hip"); > > + pcie->hip_base = devm_ioremap_resource(&pdev->dev, > > hip); > > + if (IS_ERR(pcie->hip_base)) > > + return PTR_ERR(pcie->hip_base); > > + } > > + > >   /* setup IRQ */ > >   pcie->irq = platform_get_irq(pdev, 0); > >   if (pcie->irq < 0) { > > @@ -562,6 +726,38 @@ static void altera_pcie_host_init(struct > > altera_pcie *pcie) > >   altera_pcie_retrain(pcie); > >  } > >   > > +static struct altera_pcie_data pci_1_0_data = { > > + .tlp_read_pkt = tlp_read_packet, > > + .tlp_write_pkt = tlp_write_packet, > > + .get_link_status = altera_pcie_link_up, > > + .cap_offset = 0x80, > > + .version = ALTERA_PCIE_V1, > > + .cfgrd0 = TLP_FMTTYPE_CFGRD0, > > + .cfgrd1 = TLP_FMTTYPE_CFGRD1, > > + .cfgwr0 = TLP_FMTTYPE_CFGWR0, > > + .cfgwr1 = TLP_FMTTYPE_CFGWR1, > > +}; > > + > > +static struct altera_pcie_data pci_2_0_data = { > > + .tlp_read_pkt = s10_tlp_read_packet, > > + .tlp_write_pkt = s10_tlp_write_packet, > > + .get_link_status = s10_altera_pcie_link_up, > > + .rp_read_cfg = s10_rp_read_cfg, > > + .rp_write_cfg = s10_rp_write_cfg, > > + .version = ALTERA_PCIE_V2, > > + .cap_offset = 0x70, > > + .cfgrd0 = S10_TLP_FMTTYPE_CFGRD0, > > + .cfgrd1 = S10_TLP_FMTTYPE_CFGRD1, > > + .cfgwr0 = S10_TLP_FMTTYPE_CFGWR0, > > + .cfgwr1 = S10_TLP_FMTTYPE_CFGWR1, > > +}; > I wonder if it is not better to have two different struct pci_ops > (and > have a pointer to them in struct altera_pcie_data) instead of all > these > parameters (most of which are there to differentiate struct pci_ops > methods). > > See also my comments above. > > In the end I guess it depends on how many host bridges will be based > on > this IP as design, what you did is OK but it is worth thinking about > it. > We have 2 versions of HW based on this IP as for now. May have new version in future. I can move all callback functions to struct pci_ops and have a pointer in struct altera_pcie_data. Thanks. Regards Ley Foon