From mboxrd@z Thu Jan 1 00:00:00 1970 From: tn@semihalf.com (Tomasz Nowicki) Date: Tue, 28 Mar 2017 21:28:51 +0200 Subject: [PATCH 2/2] PCI: Add legacy firmware support for Cavium ThunderX host controller In-Reply-To: <1490728896-30520-1-git-send-email-tn@semihalf.com> References: <1489564155-3881-3-git-send-email-tn@semihalf.com> <1490728896-30520-1-git-send-email-tn@semihalf.com> Message-ID: To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi Bjorn, Here is a complete patch to avoid manual patching. Please have a look and let me know if it is OK. Thanks, Tomasz On 28.03.2017 21:21, Tomasz Nowicki wrote: > During early days of PCI quirks support ThunderX firmware did not provide > PNP0c02 node with PCI configuration space and PEM-specific register ranges. > This means that for legacy FW we are not reserving these resources and > cannot gather PEM-specific resources for further PEM initialization. > > In order to support already deployed legacy FW calculate PEM-specific ranges > and provide resources reservation as fallback scenario into PEM driver when > we could not gather PEM reg base from ACPI tables. > > Signed-off-by: Tomasz Nowicki > Signed-off-by: Vadim Lomovtsev > --- > drivers/pci/host/pci-thunder-pem.c | 62 ++++++++++++++++++++++++++++++++++++-- > 1 file changed, 60 insertions(+), 2 deletions(-) > > diff --git a/drivers/pci/host/pci-thunder-pem.c b/drivers/pci/host/pci-thunder-pem.c > index e354010..cea5814 100644 > --- a/drivers/pci/host/pci-thunder-pem.c > +++ b/drivers/pci/host/pci-thunder-pem.c > @@ -14,6 +14,7 @@ > * Copyright (C) 2015 - 2016 Cavium, Inc. > */ > > +#include > #include > #include > #include > @@ -319,6 +320,49 @@ static int thunder_pem_init(struct device *dev, struct pci_config_window *cfg, > > #if defined(CONFIG_ACPI) && defined(CONFIG_PCI_QUIRKS) > > +#define PEM_RES_BASE 0x87e0c0000000UL > +#define PEM_NODE_MASK GENMASK(45, 44) > +#define PEM_INDX_MASK GENMASK(26, 24) > +#define PEM_MIN_DOM_IN_NODE 4 > +#define PEM_MAX_DOM_IN_NODE 10 > + > +static void thunder_pem_reserve_range(struct device *dev, int seg, > + struct resource *r) > +{ > + resource_size_t start = r->start, end = r->end; > + struct resource *res; > + const char *regionid; > + > + regionid = kasprintf(GFP_KERNEL, "PEM RC:%d", seg); > + if (!regionid) > + return; > + > + res = request_mem_region(start, end - start + 1, regionid); > + if (res) > + res->flags &= ~IORESOURCE_BUSY; > + else > + kfree(regionid); > + > + dev_info(dev, "%pR %s reserved\n", r, > + res ? "has been" : "could not be"); > +} > + > +static void thunder_pem_legacy_fw(struct acpi_pci_root *root, > + struct resource *res_pem) > +{ > + int node = acpi_get_node(root->device->handle); > + int index; > + > + if (node == NUMA_NO_NODE) > + node = 0; > + > + index = root->segment - PEM_MIN_DOM_IN_NODE; > + index -= node * PEM_MAX_DOM_IN_NODE; > + res_pem->start = PEM_RES_BASE | FIELD_PREP(PEM_NODE_MASK, node) | > + FIELD_PREP(PEM_INDX_MASK, index); > + res_pem->flags = IORESOURCE_MEM; > +} > + > static int thunder_pem_acpi_init(struct pci_config_window *cfg) > { > struct device *dev = cfg->parent; > @@ -332,9 +376,23 @@ static int thunder_pem_acpi_init(struct pci_config_window *cfg) > return -ENOMEM; > > ret = acpi_get_rc_resources(dev, "CAVA02B", root->segment, res_pem); > + > + /* > + * If we fail to gather resources it means that we run with old > + * FW where we need to calculate PEM-specific resources manually. > + */ > if (ret) { > - dev_err(dev, "can't get rc base address\n"); > - return ret; > + thunder_pem_legacy_fw(root, res_pem); > + /* > + * Reserve 64K size PEM specific resources. The full 16M range > + * size is required for thunder_pem_init() call. > + */ > + res_pem->end = res_pem->start + SZ_64K - 1; > + thunder_pem_reserve_range(dev, root->segment, res_pem); > + res_pem->end = res_pem->start + SZ_16M - 1; > + > + /* Reserve PCI configuration space as well. */ > + thunder_pem_reserve_range(dev, root->segment, &cfg->res); > } > > return thunder_pem_init(dev, cfg, res_pem); >