nb/intel/haswell;sb/intel/lynxpoint: Enable VT-d and X2APIC

We use the usual static addresses 0xfed90000/0xfed91000 for the GFX
IOMMU and the general IOMMU respectively. These addresses have to be
configured in MCHBAR registers and reserved from the OS.

GFXVTBAR/VTVC0BAR policy registers set to be consistent with
proprietary vendor firmwares on hardware of same platform 
(2 different vendor firmwares compared, found to be identical).

Change-Id: Ib8f2fed9ae08491779e76f7d1ddc1bd3eed45ac7
Signed-off-by: Matt DeVillier <matt.devillier@gmail.com>
Reviewed-on: https://review.coreboot.org/24983
Reviewed-by: Youness Alaoui <snifikino@gmail.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>

5 files changed, 81 insertions, 5 deletions

diff --git a/src/northbridge/intel/haswell/early_init.c b/src/northbridge/intel/haswell/early_init.c
index a2b0f7fb88..663812a5ce 100644
--- a/src/northbridge/intel/haswell/early_init.c
+++ b/src/northbridge/intel/haswell/early_init.c
@@ -94,11 +94,39 @@ static void haswell_setup_graphics(void)
 	MCHBAR32(0x5418) = reg32;
 }
 
+static void haswell_setup_iommu(void)
+{
+	const u32 capid0_a = pci_read_config32(PCI_DEV(0, 0, 0), CAPID0_A);
+
+	if (capid0_a & VTD_DISABLE)
+		return;
+
+	/* setup BARs: zeroize top 32 bits; set enable bit */
+	MCHBAR32(GFXVTBAR + 4) = GFXVT_BASE_ADDRESS >> 32;
+	MCHBAR32(GFXVTBAR) = GFXVT_BASE_ADDRESS | 1;
+	MCHBAR32(VTVC0BAR + 4) = VTVC0_BASE_ADDRESS >> 32;
+	MCHBAR32(VTVC0BAR) = VTVC0_BASE_ADDRESS | 1;
+
+	/* set L3HIT2PEND_DIS, lock GFXVTBAR policy cfg registers */
+	u32 reg32;
+	reg32 = read32((void *)(GFXVT_BASE_ADDRESS + ARCHDIS));
+	write32((void *)(GFXVT_BASE_ADDRESS + ARCHDIS),
+			reg32 | DMAR_LCKDN | L3HIT2PEND_DIS);
+	/* clear SPCAPCTRL */
+	reg32 = read32((void *)(VTVC0_BASE_ADDRESS + ARCHDIS)) & ~SPCAPCTRL;
+	/* set GLBIOTLBINV, GLBCTXTINV; lock VTVC0BAR policy cfg registers */
+	write32((void *)(VTVC0_BASE_ADDRESS + ARCHDIS),
+			reg32 | DMAR_LCKDN | GLBIOTLBINV | GLBCTXTINV);
+}
+
 void haswell_early_initialization(int chipset_type)
 {
 	/* Setup all BARs required for early PCIe and raminit */
 	haswell_setup_bars();
 
+	/* Setup IOMMU BARs */
+	haswell_setup_iommu();
+
 	/* Device Enable: IGD and Mini-HD Audio */
 	pci_write_config32(PCI_DEV(0, 0, 0), DEVEN,
 			   DEVEN_D0EN | DEVEN_D2EN | DEVEN_D3EN);
diff --git a/src/northbridge/intel/haswell/haswell.h b/src/northbridge/intel/haswell/haswell.h
index 25d88966de..5f1530aa05 100644
--- a/src/northbridge/intel/haswell/haswell.h
+++ b/src/northbridge/intel/haswell/haswell.h
@@ -35,6 +35,12 @@
 #endif
 #define DEFAULT_EPBAR		0xfed19000	/* 4 KB */
 
+#define GFXVT_BASE_ADDRESS	0xfed90000ULL
+#define GFXVT_BASE_SIZE		0x1000
+
+#define VTVC0_BASE_ADDRESS	0xfed91000ULL
+#define VTVC0_BASE_SIZE		0x1000
+
 #include <southbridge/intel/lynxpoint/pch.h>
 
 /* Everything below this line is ignored in the DSDT */
@@ -88,6 +94,16 @@
 
 #define SKPAD		0xdc	/* Scratchpad Data */
 
+#define CAPID0_A	0xe4
+#define  VTD_DISABLE	(1 << 23)
+#define ARCHDIS		0xff0	/* DMA Remap Engine Policy Control */
+#define  DMAR_LCKDN	(1 << 31)
+#define  SPCAPCTRL	(1 << 25)
+#define  L3HIT2PEND_DIS	(1 << 20)
+#define  PRSCAPDIS	(1 << 2)
+#define  GLBIOTLBINV	(1 << 1)
+#define  GLBCTXTINV	(1 << 0)
+
 /* Device 0:1.0 PCI configuration space (PCI Express) */
 
 #define BCTRL1		0x3e	/* 16bit */
@@ -107,6 +123,8 @@
 #define MCHBAR32_OR(x, or) MCHBAR32(x) = (MCHBAR32(x) | (or))
 
 #define BIOS_RESET_CPL	0x5da8	/* 8bit */
+#define GFXVTBAR	0x5400
+#define VTVC0BAR	0x5410
 
 /* Some power MSRs are also represented in MCHBAR */
 #define MCH_PKG_POWER_LIMIT_LO	0x59a0
diff --git a/src/northbridge/intel/haswell/northbridge.c b/src/northbridge/intel/haswell/northbridge.c
index f51eb35ee8..bbf1604b4c 100644
--- a/src/northbridge/intel/haswell/northbridge.c
+++ b/src/northbridge/intel/haswell/northbridge.c
@@ -290,7 +290,7 @@ static void mc_report_map_entries(device_t dev, uint64_t *values)
 	printk(BIOS_DEBUG, "MC MAP: GGC: 0x%x\n", pci_read_config16(dev, GGC));
 }
 
-static void mc_add_dram_resources(device_t dev)
+static void mc_add_dram_resources(device_t dev, int *resource_cnt)
 {
 	unsigned long base_k, size_k;
 	unsigned long touud_k;
@@ -332,7 +332,7 @@ static void mc_add_dram_resources(device_t dev)
 	 * The resource index starts low and should not meet or exceed
 	 * PCI_BASE_ADDRESS_0.
 	 */
-	index = 0;
+	index = *resource_cnt;
 
 	/* 0 - > 0xa0000 */
 	base_k = 0;
@@ -380,18 +380,31 @@ static void mc_add_dram_resources(device_t dev)
 			CONFIG_CHROMEOS_RAMOOPS_RAM_START >> 10,
 			CONFIG_CHROMEOS_RAMOOPS_RAM_SIZE >> 10);
 #endif
+	*resource_cnt = index;
 }
 
 static void mc_read_resources(device_t dev)
 {
+	int index = 0;
+	const bool vtd_capable =
+		!(pci_read_config32(dev, CAPID0_A) & VTD_DISABLE);
+
 	/* Read standard PCI resources. */
 	pci_dev_read_resources(dev);
 
 	/* Add all fixed MMIO resources. */
 	mc_add_fixed_mmio_resources(dev);
 
+	/* Add VT-d MMIO resources if capable */
+	if (vtd_capable) {
+		mmio_resource(dev, index++, GFXVT_BASE_ADDRESS / KiB,
+			GFXVT_BASE_SIZE / KiB);
+		mmio_resource(dev, index++, VTVC0_BASE_ADDRESS / KiB,
+			VTVC0_BASE_SIZE / KiB);
+	}
+
 	/* Calculate and add DRAM resources. */
-	mc_add_dram_resources(dev);
+	mc_add_dram_resources(dev, &index);
 }
 
 static void intel_set_subsystem(device_t dev, unsigned vendor, unsigned device)
diff --git a/src/southbridge/intel/lynxpoint/lpc.c b/src/southbridge/intel/lynxpoint/lpc.c
index a5c2351323..03a77d9b3e 100644
--- a/src/southbridge/intel/lynxpoint/lpc.c
+++ b/src/southbridge/intel/lynxpoint/lpc.c
@@ -51,6 +51,10 @@ static void pch_enable_ioapic(struct device *dev)
 {
 	u32 reg32;
 
+	/* Assign unique bus/dev/fn for I/O APIC */
+	pci_write_config16(dev, LPC_IBDF,
+		PCH_IOAPIC_PCI_BUS << 8 | PCH_IOAPIC_PCI_SLOT << 3);
+
 	/* Enable ACPI I/O range decode */
 	pci_write_config8(dev, ACPI_CNTL, ACPI_EN);
 
@@ -396,9 +400,15 @@ static void lpt_lp_pm_init(struct device *dev)
 	RCBA32_OR(0x33c8, (1 << 15));
 }
 
-static void enable_hpet(void)
+static void enable_hpet(struct device *const dev)
 {
 	u32 reg32;
+	size_t i;
+
+	/* Assign unique bus/dev/fn for each HPET */
+	for (i = 0; i < 8; ++i)
+		pci_write_config16(dev, LPC_HnBDF(i),
+			PCH_HPET_PCI_BUS << 8 | PCH_HPET_PCI_SLOT << 3 | i);
 
 	/* Move HPET to default address 0xfed00000 and enable it */
 	reg32 = RCBA32(HPTC);
@@ -570,7 +580,7 @@ static void lpc_init(struct device *dev)
 	isa_dma_init();
 
 	/* Initialize the High Precision Event Timers, if present. */
-	enable_hpet();
+	enable_hpet(dev);
 
 	setup_i8259();
 
diff --git a/src/southbridge/intel/lynxpoint/pch.h b/src/southbridge/intel/lynxpoint/pch.h
index af9e9542b5..f14a339d7b 100644
--- a/src/southbridge/intel/lynxpoint/pch.h
+++ b/src/southbridge/intel/lynxpoint/pch.h
@@ -282,6 +282,8 @@ void pch_enable_lpc(void);
 #define  LPT_LPC_EN		(1 << 2)  /* LPC_IO_DEC[9:8] */
 #define  COMB_LPC_EN		(1 << 1)  /* LPC_IO_DEC[6:4] */
 #define  COMA_LPC_EN		(1 << 0)  /* LPC_IO_DEC[2:0] */
+#define LPC_IBDF		0x6C /* I/O APIC bus/dev/fn */
+#define LPC_HnBDF(n)		(0x70 + n * 2) /* HPET n bus/dev/fn */
 #define LPC_GEN1_DEC		0x84 /* LPC IF Generic Decode Range 1 */
 #define LPC_GEN2_DEC		0x88 /* LPC IF Generic Decode Range 2 */
 #define LPC_GEN3_DEC		0x8c /* LPC IF Generic Decode Range 3 */
@@ -670,6 +672,11 @@ void pch_enable_lpc(void);
 #define PCH_DISABLE_MEI1	(1 << 1)
 #define PCH_ENABLE_DBDF		(1 << 0)
 
+#define PCH_IOAPIC_PCI_BUS	250
+#define PCH_IOAPIC_PCI_SLOT	31
+#define PCH_HPET_PCI_BUS	250
+#define PCH_HPET_PCI_SLOT	15
+
 /* ICH7 PMBASE */
 #define PM1_STS		0x00
 #define   WAK_STS	(1 << 15)