[PATCH v2 0/9] Apple PCIe/XHCI support
This series adds support for the PCIe controller found on Apple M1 and M2 machines and enables support for PCIe XHCI controllers. This makes the type-A USB ports on the M1 Mac mini work. Since the use of Apples DART IOMMU is mandatory (these PCIe DARTs don't support bypass mode), this adds DMA mapping operations to the IOMMU uclass and implements them for the Apple DART. It modifies the XHCI driver code to go map DMA buffers through the IOMMU if there is one. Since the M1 Mac mini now has two types of XHCI controllers with different number of ports (2 for the DWC3 controllers, 8 for the Fresco Logic PCIe controller) this uncovered an issue in with the way the hub descriptor is implemented in the XHCI driver.
ChangeLog:
v2: - Add test for new operations and PCI IOMMU mapping
Mark Kettenis (9): iommu: Add DMA mapping operations test: Add test for IOMMU uclass map/unmap ops iommu: apple: Implement DMA mapping operations for Apple DART usb: xhci: Implement DMA mapping iommu: Implement mapping IOMMUs for PCI devices test: Add test for mapping IOMMUs for PCI devices pci: Add Apple PCIe controller driver arm: apple: Enable PCIe USB controller usb: xhci: Fix root hub descriptor
MAINTAINERS | 1 + arch/arm/Kconfig | 2 + arch/sandbox/dts/test.dts | 2 + configs/apple_m1_defconfig | 1 + drivers/iommu/apple_dart.c | 311 +++++++++++++++++++++++++---- drivers/iommu/iommu-uclass.c | 93 +++++++++ drivers/iommu/sandbox_iommu.c | 58 ++++++ drivers/pci/Kconfig | 9 + drivers/pci/Makefile | 1 + drivers/pci/pcie_apple.c | 354 ++++++++++++++++++++++++++++++++++ drivers/usb/host/xhci-mem.c | 84 +++++--- drivers/usb/host/xhci-ring.c | 76 +++++--- drivers/usb/host/xhci.c | 25 +-- include/dm/device.h | 3 + include/iommu.h | 24 +++ include/usb/xhci.h | 28 ++- test/dm/iommu.c | 74 ++++++- 17 files changed, 1037 insertions(+), 109 deletions(-) create mode 100644 drivers/pci/pcie_apple.c
In order to support IOMMUs in non-bypass mode we need device ops to map and unmap DMA memory. The map operation enters a mapping for a region specified by CPU address and size into the translation table of the IOMMU and returns a DMA address suitable for programming the device to do DMA. The unmap operation removes this mapping from the translation table of the IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- drivers/iommu/iommu-uclass.c | 28 ++++++++++++++++++++++++++++ include/dm/device.h | 3 +++ include/iommu.h | 24 ++++++++++++++++++++++++ 3 files changed, 55 insertions(+)
diff --git a/drivers/iommu/iommu-uclass.c b/drivers/iommu/iommu-uclass.c index ed917b3c3e..f6b1457736 100644 --- a/drivers/iommu/iommu-uclass.c +++ b/drivers/iommu/iommu-uclass.c @@ -7,6 +7,9 @@
#include <common.h> #include <dm.h> +#include <iommu.h> +#include <phys2bus.h> +#include <asm/io.h>
#if (CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA)) int dev_iommu_enable(struct udevice *dev) @@ -33,12 +36,37 @@ int dev_iommu_enable(struct udevice *dev) __func__, ret); return ret; } + dev->iommu = dev_iommu; }
return 0; } #endif
+dma_addr_t dev_iommu_dma_map(struct udevice *dev, void *addr, size_t size) +{ + const struct iommu_ops *ops; + + if (dev->iommu) { + ops = device_get_ops(dev->iommu); + if (ops && ops->map) + return ops->map(dev->iommu, addr, size); + } + + return dev_phys_to_bus(dev, virt_to_phys(addr)); +} + +void dev_iommu_dma_unmap(struct udevice *dev, dma_addr_t addr, size_t size) +{ + const struct iommu_ops *ops; + + if (dev->iommu) { + ops = device_get_ops(dev->iommu); + if (ops && ops->unmap) + ops->unmap(dev->iommu, addr, size); + } +} + UCLASS_DRIVER(iommu) = { .id = UCLASS_IOMMU, .name = "iommu", diff --git a/include/dm/device.h b/include/dm/device.h index f3f953c9af..abe1927ecd 100644 --- a/include/dm/device.h +++ b/include/dm/device.h @@ -194,6 +194,9 @@ struct udevice { #if CONFIG_IS_ENABLED(DM_DMA) ulong dma_offset; #endif +#if CONFIG_IS_ENABLED(IOMMU) + struct udevice *iommu; +#endif };
static inline int dm_udevice_size(void) diff --git a/include/iommu.h b/include/iommu.h index 6c46adf449..cf9719c5e9 100644 --- a/include/iommu.h +++ b/include/iommu.h @@ -3,6 +3,27 @@
struct udevice;
+struct iommu_ops { + /** + * map() - map DMA memory + * + * @dev: device for which to map DMA memory + * @addr: CPU address of the memory + * @size: size of the memory + * @return DMA address for the device + */ + dma_addr_t (*map)(struct udevice *dev, void *addr, size_t size); + + /** + * unmap() - unmap DMA memory + * + * @dev: device for which to unmap DMA memory + * @addr: DMA address of the memory + * @size: size of the memory + */ + void (*unmap)(struct udevice *dev, dma_addr_t addr, size_t size); +}; + #if (CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA)) && \ CONFIG_IS_ENABLED(IOMMU) int dev_iommu_enable(struct udevice *dev); @@ -13,4 +34,7 @@ static inline int dev_iommu_enable(struct udevice *dev) } #endif
+dma_addr_t dev_iommu_dma_map(struct udevice *dev, void *addr, size_t size); +void dev_iommu_dma_unmap(struct udevice *dev, dma_addr_t addr, size_t size); + #endif
On Sat, Jan 21, 2023 at 08:27:52PM +0100, Mark Kettenis wrote:
In order to support IOMMUs in non-bypass mode we need device ops to map and unmap DMA memory. The map operation enters a mapping for a region specified by CPU address and size into the translation table of the IOMMU and returns a DMA address suitable for programming the device to do DMA. The unmap operation removes this mapping from the translation table of the IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org
For the series, applied to u-boot/master, thanks!
Note that in this specific patch I also added the following comment to include/dm/device.h: * @iommu: IOMMU device associated with this device as the doc targets were noting that this was not described. If my description here is wrong / insufficient, please follow-up with a correction.
Date: Fri, 27 Jan 2023 17:48:32 -0500 From: Tom Rini trini@konsulko.com
On Sat, Jan 21, 2023 at 08:27:52PM +0100, Mark Kettenis wrote:
In order to support IOMMUs in non-bypass mode we need device ops to map and unmap DMA memory. The map operation enters a mapping for a region specified by CPU address and size into the translation table of the IOMMU and returns a DMA address suitable for programming the device to do DMA. The unmap operation removes this mapping from the translation table of the IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org
For the series, applied to u-boot/master, thanks!
Note that in this specific patch I also added the following comment to include/dm/device.h:
- @iommu: IOMMU device associated with this device
as the doc targets were noting that this was not described. If my description here is wrong / insufficient, please follow-up with a correction.
That reads absolutely fine.
Thanks!
Mark
Test that the map and unmap operations work for devices that have DMA translated by an IOMMU and devices that don't have DMA translated by an IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- drivers/iommu/sandbox_iommu.c | 58 +++++++++++++++++++++++++++++++++++ test/dm/iommu.c | 44 +++++++++++++++++++++++++- 2 files changed, 101 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/sandbox_iommu.c b/drivers/iommu/sandbox_iommu.c index c8161a40ae..6ceb7fd5ec 100644 --- a/drivers/iommu/sandbox_iommu.c +++ b/drivers/iommu/sandbox_iommu.c @@ -5,6 +5,61 @@
#include <common.h> #include <dm.h> +#include <iommu.h> +#include <lmb.h> +#include <asm/io.h> +#include <linux/sizes.h> + +#define IOMMU_PAGE_SIZE SZ_4K + +struct sandbox_iommu_priv { + struct lmb lmb; +}; + +static dma_addr_t sandbox_iommu_map(struct udevice *dev, void *addr, + size_t size) +{ + struct sandbox_iommu_priv *priv = dev_get_priv(dev); + phys_addr_t paddr, dva; + phys_size_t psize, off; + + paddr = ALIGN_DOWN(virt_to_phys(addr), IOMMU_PAGE_SIZE); + off = virt_to_phys(addr) - paddr; + psize = ALIGN(size + off, IOMMU_PAGE_SIZE); + + dva = lmb_alloc(&priv->lmb, psize, IOMMU_PAGE_SIZE); + + return dva + off; +} + +static void sandbox_iommu_unmap(struct udevice *dev, dma_addr_t addr, + size_t size) +{ + struct sandbox_iommu_priv *priv = dev_get_priv(dev); + phys_addr_t dva; + phys_size_t psize; + + dva = ALIGN_DOWN(addr, IOMMU_PAGE_SIZE); + psize = size + (addr - dva); + psize = ALIGN(psize, IOMMU_PAGE_SIZE); + + lmb_free(&priv->lmb, dva, psize); +} + +static struct iommu_ops sandbox_iommu_ops = { + .map = sandbox_iommu_map, + .unmap = sandbox_iommu_unmap, +}; + +static int sandbox_iommu_probe(struct udevice *dev) +{ + struct sandbox_iommu_priv *priv = dev_get_priv(dev); + + lmb_init(&priv->lmb); + lmb_add(&priv->lmb, 0x89abc000, SZ_16K); + + return 0; +}
static const struct udevice_id sandbox_iommu_ids[] = { { .compatible = "sandbox,iommu" }, @@ -15,4 +70,7 @@ U_BOOT_DRIVER(sandbox_iommu) = { .name = "sandbox_iommu", .id = UCLASS_IOMMU, .of_match = sandbox_iommu_ids, + .priv_auto = sizeof(struct sandbox_iommu_priv), + .ops = &sandbox_iommu_ops, + .probe = sandbox_iommu_probe, }; diff --git a/test/dm/iommu.c b/test/dm/iommu.c index 94174a7482..6f932ef233 100644 --- a/test/dm/iommu.c +++ b/test/dm/iommu.c @@ -8,12 +8,16 @@ #include <dm/test.h> #include <dm/uclass-internal.h> #include <iommu.h> +#include <malloc.h> #include <test/test.h> #include <test/ut.h> +#include <asm/io.h>
static int dm_test_iommu(struct unit_test_state *uts) { struct udevice *dev; + dma_addr_t dva; + void *addr;
ut_assertok(uclass_find_device(UCLASS_IOMMU, 0, &dev)); ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); @@ -22,7 +26,45 @@ static int dm_test_iommu(struct unit_test_state *uts) ut_assertok(uclass_probe_all(UCLASS_USB)); ut_assert(dev_get_flags(dev) & DM_FLAG_ACTIVATED);
+ addr = malloc(256); + ut_assert(addr); + + ut_assertok(uclass_find_device(UCLASS_USB, 0, &dev)); + dva = dev_iommu_dma_map(dev, addr, 256); + ut_assert(dva >= 0x89abc000 && dva < 0x89ac00000); + + dev_iommu_dma_unmap(dev, dva, 256); + + free(addr); + return 0; } - DM_TEST(dm_test_iommu, UT_TESTF_SCAN_FDT); + +static int dm_test_iommu_noiommu(struct unit_test_state *uts) +{ + struct udevice *dev; + dma_addr_t dva; + void *addr; + + ut_assertok(uclass_find_device(UCLASS_IOMMU, 0, &dev)); + ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); + + /* Probing ethernet should not probe the IOMMU */ + ut_assertok(uclass_probe_all(UCLASS_ETH)); + ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); + + addr = malloc(256); + ut_assert(addr); + + ut_assertok(uclass_find_device(UCLASS_ETH, 0, &dev)); + dva = dev_iommu_dma_map(dev, addr, 256); + ut_assert(dva == virt_to_phys(addr)); + + dev_iommu_dma_unmap(dev, dva, 256); + + free(addr); + + return 0; +} +DM_TEST(dm_test_iommu_noiommu, UT_TESTF_SCAN_FDT);
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
Test that the map and unmap operations work for devices that have DMA translated by an IOMMU and devices that don't have DMA translated by an IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org
drivers/iommu/sandbox_iommu.c | 58 +++++++++++++++++++++++++++++++++++ test/dm/iommu.c | 44 +++++++++++++++++++++++++- 2 files changed, 101 insertions(+), 1 deletion(-)
Reviewed-by: Simon Glass sjg@chromium.org
BTW I noticed something like CONFIG_IS_ENABLED(ARCH_APPLE) but it should be IS_ENABLED() since in SPL we would still use the same architecture. Could you please send a clean-up patch?
Regards, Simon
From: Simon Glass sjg@chromium.org Date: Mon, 23 Jan 2023 11:42:25 -0700
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
Test that the map and unmap operations work for devices that have DMA translated by an IOMMU and devices that don't have DMA translated by an IOMMU.
Signed-off-by: Mark Kettenis kettenis@openbsd.org
drivers/iommu/sandbox_iommu.c | 58 +++++++++++++++++++++++++++++++++++ test/dm/iommu.c | 44 +++++++++++++++++++++++++- 2 files changed, 101 insertions(+), 1 deletion(-)
Reviewed-by: Simon Glass sjg@chromium.org
BTW I noticed something like CONFIG_IS_ENABLED(ARCH_APPLE) but it should be IS_ENABLED() since in SPL we would still use the same architecture. Could you please send a clean-up patch?
Hi Simon,
Just sent a separate patch for that.
Cheers,
Mark
Implement translation table support for all the variations of Apple's DART IOMMU that can be found on Apple's M1 and M2 SoCs.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- drivers/iommu/apple_dart.c | 311 +++++++++++++++++++++++++++++++++---- 1 file changed, 277 insertions(+), 34 deletions(-)
diff --git a/drivers/iommu/apple_dart.c b/drivers/iommu/apple_dart.c index 2faacb8f3b..6ecd84303b 100644 --- a/drivers/iommu/apple_dart.c +++ b/drivers/iommu/apple_dart.c @@ -6,57 +6,296 @@ #include <common.h> #include <cpu_func.h> #include <dm.h> +#include <iommu.h> +#include <lmb.h> +#include <memalign.h> #include <asm/io.h>
#define DART_PARAMS2 0x0004 #define DART_PARAMS2_BYPASS_SUPPORT BIT(0) -#define DART_TLB_OP 0x0020 -#define DART_TLB_OP_OPMASK (0xfff << 20) -#define DART_TLB_OP_FLUSH (0x001 << 20) -#define DART_TLB_OP_BUSY BIT(2) -#define DART_TLB_OP_SIDMASK 0x0034 -#define DART_ERROR_STATUS 0x0040 -#define DART_TCR(sid) (0x0100 + 4 * (sid)) -#define DART_TCR_TRANSLATE_ENABLE BIT(7) -#define DART_TCR_BYPASS_DART BIT(8) -#define DART_TCR_BYPASS_DAPF BIT(12) -#define DART_TTBR(sid, idx) (0x0200 + 16 * (sid) + 4 * (idx)) -#define DART_TTBR_VALID BIT(31) -#define DART_TTBR_SHIFT 12
-#define DART_T8110_TCR(sid) (0x1000 + 4 * (sid)) +#define DART_T8020_TLB_CMD 0x0020 +#define DART_T8020_TLB_CMD_FLUSH BIT(20) +#define DART_T8020_TLB_CMD_BUSY BIT(2) +#define DART_T8020_TLB_SIDMASK 0x0034 +#define DART_T8020_ERROR 0x0040 +#define DART_T8020_ERROR_ADDR_LO 0x0050 +#define DART_T8020_ERROR_ADDR_HI 0x0054 +#define DART_T8020_CONFIG 0x0060 +#define DART_T8020_CONFIG_LOCK BIT(15) +#define DART_T8020_SID_ENABLE 0x00fc +#define DART_T8020_TCR_BASE 0x0100 +#define DART_T8020_TCR_TRANSLATE_ENABLE BIT(7) +#define DART_T8020_TCR_BYPASS_DART BIT(8) +#define DART_T8020_TCR_BYPASS_DAPF BIT(12) +#define DART_T8020_TTBR_BASE 0x0200 +#define DART_T8020_TTBR_VALID BIT(31) + +#define DART_T8110_PARAMS4 0x000c +#define DART_T8110_PARAMS4_NSID_MASK (0x1ff << 0) +#define DART_T8110_TLB_CMD 0x0080 +#define DART_T8110_TLB_CMD_BUSY BIT(31) +#define DART_T8110_TLB_CMD_FLUSH_ALL BIT(8) +#define DART_T8110_ERROR 0x0100 +#define DART_T8110_ERROR_MASK 0x0104 +#define DART_T8110_ERROR_ADDR_LO 0x0170 +#define DART_T8110_ERROR_ADDR_HI 0x0174 +#define DART_T8110_PROTECT 0x0200 +#define DART_T8110_PROTECT_TTBR_TCR BIT(0) +#define DART_T8110_SID_ENABLE_BASE 0x0c00 +#define DART_T8110_TCR_BASE 0x1000 #define DART_T8110_TCR_BYPASS_DAPF BIT(2) #define DART_T8110_TCR_BYPASS_DART BIT(1) #define DART_T8110_TCR_TRANSLATE_ENABLE BIT(0) -#define DART_T8110_TTBR(sid) (0x1400 + 4 * (sid)) +#define DART_T8110_TTBR_BASE 0x1400 +#define DART_T8110_TTBR_VALID BIT(0) + +#define DART_SID_ENABLE(priv, idx) \ + ((priv)->sid_enable_base + 4 * (idx)) +#define DART_TCR(priv, sid) ((priv)->tcr_base + 4 * (sid)) +#define DART_TTBR(priv, sid, idx) \ + ((priv)->ttbr_base + 4 * (priv)->nttbr * (sid) + 4 * (idx)) +#define DART_TTBR_SHIFT 12 + +#define DART_ALL_STREAMS(priv) ((1U << (priv)->nsid) - 1) + +#define DART_PAGE_SIZE SZ_16K +#define DART_PAGE_MASK (DART_PAGE_SIZE - 1) + +#define DART_L1_TABLE 0x3 +#define DART_L2_INVAL 0 +#define DART_L2_VALID BIT(0) +#define DART_L2_FULL_PAGE BIT(1) +#define DART_L2_START(addr) ((((addr) & DART_PAGE_MASK) >> 2) << 52) +#define DART_L2_END(addr) ((((addr) & DART_PAGE_MASK) >> 2) << 40) + +struct apple_dart_priv { + void *base; + struct lmb lmb; + u64 *l1, *l2; + int bypass, shift; + + dma_addr_t dvabase; + dma_addr_t dvaend; + + int nsid; + int nttbr; + int sid_enable_base; + int tcr_base; + u32 tcr_translate_enable; + u32 tcr_bypass; + int ttbr_base; + u32 ttbr_valid; + void (*flush_tlb)(struct apple_dart_priv *priv); +}; + +static void apple_dart_t8020_flush_tlb(struct apple_dart_priv *priv) +{ + dsb(); + + writel(DART_ALL_STREAMS(priv), priv->base + DART_T8020_TLB_SIDMASK); + writel(DART_T8020_TLB_CMD_FLUSH, priv->base + DART_T8020_TLB_CMD); + while (readl(priv->base + DART_T8020_TLB_CMD) & + DART_T8020_TLB_CMD_BUSY) + continue; +} + +static void apple_dart_t8110_flush_tlb(struct apple_dart_priv *priv) +{ + dsb(); + + writel(DART_T8110_TLB_CMD_FLUSH_ALL, + priv->base + DART_T8110_TLB_CMD_FLUSH_ALL); + while (readl(priv->base + DART_T8110_TLB_CMD) & + DART_T8110_TLB_CMD_BUSY) + continue; +} + +static dma_addr_t apple_dart_map(struct udevice *dev, void *addr, size_t size) +{ + struct apple_dart_priv *priv = dev_get_priv(dev); + phys_addr_t paddr, dva; + phys_size_t psize, off; + int i, idx; + + if (priv->bypass) + return (phys_addr_t)addr; + + paddr = ALIGN_DOWN((phys_addr_t)addr, DART_PAGE_SIZE); + off = (phys_addr_t)addr - paddr; + psize = ALIGN(size + off, DART_PAGE_SIZE); + + dva = lmb_alloc(&priv->lmb, psize, DART_PAGE_SIZE); + + idx = dva / DART_PAGE_SIZE; + for (i = 0; i < psize / DART_PAGE_SIZE; i++) { + priv->l2[idx + i] = (paddr >> priv->shift) | DART_L2_VALID | + DART_L2_START(0LL) | DART_L2_END(~0LL); + paddr += DART_PAGE_SIZE; + } + flush_dcache_range((unsigned long)&priv->l2[idx], + (unsigned long)&priv->l2[idx + i]); + priv->flush_tlb(priv); + + return dva + off; +} + +static void apple_dart_unmap(struct udevice *dev, dma_addr_t addr, size_t size) +{ + struct apple_dart_priv *priv = dev_get_priv(dev); + phys_addr_t dva; + phys_size_t psize; + int i, idx; + + if (priv->bypass) + return; + + dva = ALIGN_DOWN(addr, DART_PAGE_SIZE); + psize = size + (addr - dva); + psize = ALIGN(psize, DART_PAGE_SIZE); + + idx = dva / DART_PAGE_SIZE; + for (i = 0; i < psize / DART_PAGE_SIZE; i++) + priv->l2[idx + i] = DART_L2_INVAL; + flush_dcache_range((unsigned long)&priv->l2[idx], + (unsigned long)&priv->l2[idx + i]); + priv->flush_tlb(priv); + + lmb_free(&priv->lmb, dva, psize); +} + +static struct iommu_ops apple_dart_ops = { + .map = apple_dart_map, + .unmap = apple_dart_unmap, +};
static int apple_dart_probe(struct udevice *dev) { - void *base; + struct apple_dart_priv *priv = dev_get_priv(dev); + dma_addr_t addr; + phys_addr_t l2; + int ntte, nl1, nl2; int sid, i; + u32 params2, params4;
- base = dev_read_addr_ptr(dev); - if (!base) + priv->base = dev_read_addr_ptr(dev); + if (!priv->base) return -EINVAL;
- u32 params2 = readl(base + DART_PARAMS2); - if (!(params2 & DART_PARAMS2_BYPASS_SUPPORT)) - return 0; + if (device_is_compatible(dev, "apple,t8110-dart")) { + params4 = readl(priv->base + DART_T8110_PARAMS4); + priv->nsid = params4 & DART_T8110_PARAMS4_NSID_MASK; + priv->nttbr = 1; + priv->sid_enable_base = DART_T8110_SID_ENABLE_BASE; + priv->tcr_base = DART_T8110_TCR_BASE; + priv->tcr_translate_enable = DART_T8110_TCR_TRANSLATE_ENABLE; + priv->tcr_bypass = + DART_T8110_TCR_BYPASS_DAPF | DART_T8110_TCR_BYPASS_DART; + priv->ttbr_base = DART_T8110_TTBR_BASE; + priv->ttbr_valid = DART_T8110_TTBR_VALID; + priv->flush_tlb = apple_dart_t8110_flush_tlb; + } else { + priv->nsid = 16; + priv->nttbr = 4; + priv->sid_enable_base = DART_T8020_SID_ENABLE; + priv->tcr_base = DART_T8020_TCR_BASE; + priv->tcr_translate_enable = DART_T8020_TCR_TRANSLATE_ENABLE; + priv->tcr_bypass = + DART_T8020_TCR_BYPASS_DAPF | DART_T8020_TCR_BYPASS_DART; + priv->ttbr_base = DART_T8020_TTBR_BASE; + priv->ttbr_valid = DART_T8020_TTBR_VALID; + priv->flush_tlb = apple_dart_t8020_flush_tlb; + } + + if (device_is_compatible(dev, "apple,t6000-dart") || + device_is_compatible(dev, "apple,t8110-dart")) + priv->shift = 4;
- if (device_is_compatible(dev, "apple,t8112-dart")) { - for (sid = 0; sid < 256; sid++) { - writel(DART_T8110_TCR_BYPASS_DART | DART_T8110_TCR_BYPASS_DAPF, - base + DART_T8110_TCR(sid)); - writel(0, base + DART_T8110_TTBR(sid)); + priv->dvabase = DART_PAGE_SIZE; + priv->dvaend = SZ_4G - DART_PAGE_SIZE; + + lmb_init(&priv->lmb); + lmb_add(&priv->lmb, priv->dvabase, priv->dvaend - priv->dvabase); + + /* Disable translations. */ + for (sid = 0; sid < priv->nsid; sid++) + writel(0, priv->base + DART_TCR(priv, sid)); + + /* Remove page tables. */ + for (sid = 0; sid < priv->nsid; sid++) { + for (i = 0; i < priv->nttbr; i++) + writel(0, priv->base + DART_TTBR(priv, sid, i)); + } + priv->flush_tlb(priv); + + params2 = readl(priv->base + DART_PARAMS2); + if (params2 & DART_PARAMS2_BYPASS_SUPPORT) { + for (sid = 0; sid < priv->nsid; sid++) { + writel(priv->tcr_bypass, + priv->base + DART_TCR(priv, sid)); } - } else { - for (sid = 0; sid < 16; sid++) { - writel(DART_TCR_BYPASS_DART | DART_TCR_BYPASS_DAPF, - base + DART_TCR(sid)); - for (i = 0; i < 4; i++) - writel(0, base + DART_TTBR(sid, i)); + priv->bypass = 1; + return 0; + } + + ntte = DIV_ROUND_UP(priv->dvaend, DART_PAGE_SIZE); + nl2 = DIV_ROUND_UP(ntte, DART_PAGE_SIZE / sizeof(u64)); + nl1 = DIV_ROUND_UP(nl2, DART_PAGE_SIZE / sizeof(u64)); + + priv->l2 = memalign(DART_PAGE_SIZE, nl2 * DART_PAGE_SIZE); + memset(priv->l2, 0, nl2 * DART_PAGE_SIZE); + flush_dcache_range((unsigned long)priv->l2, + (unsigned long)priv->l2 + nl2 * DART_PAGE_SIZE); + + priv->l1 = memalign(DART_PAGE_SIZE, nl1 * DART_PAGE_SIZE); + memset(priv->l1, 0, nl1 * DART_PAGE_SIZE); + l2 = (phys_addr_t)priv->l2; + for (i = 0; i < nl2; i++) { + priv->l1[i] = (l2 >> priv->shift) | DART_L1_TABLE; + l2 += DART_PAGE_SIZE; + } + flush_dcache_range((unsigned long)priv->l1, + (unsigned long)priv->l1 + nl1 * DART_PAGE_SIZE); + + /* Install page tables. */ + for (sid = 0; sid < priv->nsid; sid++) { + addr = (phys_addr_t)priv->l1; + for (i = 0; i < nl1; i++) { + writel(addr >> DART_TTBR_SHIFT | priv->ttbr_valid, + priv->base + DART_TTBR(priv, sid, i)); + addr += DART_PAGE_SIZE; } } + priv->flush_tlb(priv); + + /* Enable all streams. */ + for (i = 0; i < priv->nsid / 32; i++) + writel(~0, priv->base + DART_SID_ENABLE(priv, i)); + + /* Enable translations. */ + for (sid = 0; sid < priv->nsid; sid++) { + writel(priv->tcr_translate_enable, + priv->base + DART_TCR(priv, sid)); + } + + return 0; +} + +static int apple_dart_remove(struct udevice *dev) +{ + struct apple_dart_priv *priv = dev_get_priv(dev); + int sid, i; + + /* Disable translations. */ + for (sid = 0; sid < priv->nsid; sid++) + writel(0, priv->base + DART_TCR(priv, sid)); + + /* Remove page tables. */ + for (sid = 0; sid < priv->nsid; sid++) { + for (i = 0; i < priv->nttbr; i++) + writel(0, priv->base + DART_TTBR(priv, sid, i)); + } + priv->flush_tlb(priv);
return 0; } @@ -64,7 +303,7 @@ static int apple_dart_probe(struct udevice *dev) static const struct udevice_id apple_dart_ids[] = { { .compatible = "apple,t8103-dart" }, { .compatible = "apple,t6000-dart" }, - { .compatible = "apple,t8112-dart" }, + { .compatible = "apple,t8110-dart" }, { /* sentinel */ } };
@@ -72,5 +311,9 @@ U_BOOT_DRIVER(apple_dart) = { .name = "apple_dart", .id = UCLASS_IOMMU, .of_match = apple_dart_ids, - .probe = apple_dart_probe + .priv_auto = sizeof(struct apple_dart_priv), + .ops = &apple_dart_ops, + .probe = apple_dart_probe, + .remove = apple_dart_remove, + .flags = DM_FLAG_OS_PREPARE };
An XHCI controller that sits behind an IOMMU needs to map and unmap its memory buffers to do DMA. Implement this by inroducing new xhci_dma_map() and xhci_dma_unmap() helper functions. The xhci_dma_map() function replaces the existing xhci_virt_to_bus() function in the sense that it returns the bus address in the case of simple address translation in the absence of an IOMMU. The xhci_bus_to_virt() function is eliminated by storing the CPU address of the allocated scratchpad memory in struct xhci_ctrl.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de --- drivers/usb/host/xhci-mem.c | 84 +++++++++++++++++++++++------------- drivers/usb/host/xhci-ring.c | 76 ++++++++++++++++++++------------ drivers/usb/host/xhci.c | 10 +++-- include/usb/xhci.h | 27 +++++++++--- 4 files changed, 130 insertions(+), 67 deletions(-)
diff --git a/drivers/usb/host/xhci-mem.c b/drivers/usb/host/xhci-mem.c index 21cd03b662..72b7530626 100644 --- a/drivers/usb/host/xhci-mem.c +++ b/drivers/usb/host/xhci-mem.c @@ -64,8 +64,9 @@ void xhci_inval_cache(uintptr_t addr, u32 len) * @param ptr pointer to "segement" to be freed * Return: none */ -static void xhci_segment_free(struct xhci_segment *seg) +static void xhci_segment_free(struct xhci_ctrl *ctrl, struct xhci_segment *seg) { + xhci_dma_unmap(ctrl, seg->dma, SEGMENT_SIZE); free(seg->trbs); seg->trbs = NULL;
@@ -78,7 +79,7 @@ static void xhci_segment_free(struct xhci_segment *seg) * @param ptr pointer to "ring" to be freed * Return: none */ -static void xhci_ring_free(struct xhci_ring *ring) +static void xhci_ring_free(struct xhci_ctrl *ctrl, struct xhci_ring *ring) { struct xhci_segment *seg; struct xhci_segment *first_seg; @@ -89,10 +90,10 @@ static void xhci_ring_free(struct xhci_ring *ring) seg = first_seg->next; while (seg != first_seg) { struct xhci_segment *next = seg->next; - xhci_segment_free(seg); + xhci_segment_free(ctrl, seg); seg = next; } - xhci_segment_free(first_seg); + xhci_segment_free(ctrl, first_seg);
free(ring); } @@ -105,12 +106,20 @@ static void xhci_ring_free(struct xhci_ring *ring) */ static void xhci_scratchpad_free(struct xhci_ctrl *ctrl) { + struct xhci_hccr *hccr = ctrl->hccr; + int num_sp; + if (!ctrl->scratchpad) return;
+ num_sp = HCS_MAX_SCRATCHPAD(xhci_readl(&hccr->cr_hcsparams2)); + xhci_dma_unmap(ctrl, ctrl->scratchpad->sp_array[0], + num_sp * ctrl->page_size); + xhci_dma_unmap(ctrl, ctrl->dcbaa->dev_context_ptrs[0], + num_sp * sizeof(u64)); ctrl->dcbaa->dev_context_ptrs[0] = 0;
- free(xhci_bus_to_virt(ctrl, le64_to_cpu(ctrl->scratchpad->sp_array[0]))); + free(ctrl->scratchpad->scratchpad); free(ctrl->scratchpad->sp_array); free(ctrl->scratchpad); ctrl->scratchpad = NULL; @@ -122,8 +131,10 @@ static void xhci_scratchpad_free(struct xhci_ctrl *ctrl) * @param ptr pointer to "xhci_container_ctx" to be freed * Return: none */ -static void xhci_free_container_ctx(struct xhci_container_ctx *ctx) +static void xhci_free_container_ctx(struct xhci_ctrl *ctrl, + struct xhci_container_ctx *ctx) { + xhci_dma_unmap(ctrl, ctx->dma, ctx->size); free(ctx->bytes); free(ctx); } @@ -153,12 +164,12 @@ static void xhci_free_virt_devices(struct xhci_ctrl *ctrl)
for (i = 0; i < 31; ++i) if (virt_dev->eps[i].ring) - xhci_ring_free(virt_dev->eps[i].ring); + xhci_ring_free(ctrl, virt_dev->eps[i].ring);
if (virt_dev->in_ctx) - xhci_free_container_ctx(virt_dev->in_ctx); + xhci_free_container_ctx(ctrl, virt_dev->in_ctx); if (virt_dev->out_ctx) - xhci_free_container_ctx(virt_dev->out_ctx); + xhci_free_container_ctx(ctrl, virt_dev->out_ctx);
free(virt_dev); /* make sure we are pointing to NULL */ @@ -174,11 +185,15 @@ static void xhci_free_virt_devices(struct xhci_ctrl *ctrl) */ void xhci_cleanup(struct xhci_ctrl *ctrl) { - xhci_ring_free(ctrl->event_ring); - xhci_ring_free(ctrl->cmd_ring); + xhci_ring_free(ctrl, ctrl->event_ring); + xhci_ring_free(ctrl, ctrl->cmd_ring); xhci_scratchpad_free(ctrl); xhci_free_virt_devices(ctrl); + xhci_dma_unmap(ctrl, ctrl->erst.erst_dma_addr, + sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS); free(ctrl->erst.entries); + xhci_dma_unmap(ctrl, ctrl->dcbaa->dma, + sizeof(struct xhci_device_context_array)); free(ctrl->dcbaa); memset(ctrl, '\0', sizeof(struct xhci_ctrl)); } @@ -218,15 +233,13 @@ static void xhci_link_segments(struct xhci_ctrl *ctrl, struct xhci_segment *prev struct xhci_segment *next, bool link_trbs) { u32 val; - u64 val_64 = 0;
if (!prev || !next) return; prev->next = next; if (link_trbs) { - val_64 = xhci_virt_to_bus(ctrl, next->trbs); prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr = - cpu_to_le64(val_64); + cpu_to_le64(next->dma);
/* * Set the last TRB in the segment to @@ -273,7 +286,7 @@ static void xhci_initialize_ring_info(struct xhci_ring *ring) * @param none * Return: pointer to the newly allocated SEGMENT */ -static struct xhci_segment *xhci_segment_alloc(void) +static struct xhci_segment *xhci_segment_alloc(struct xhci_ctrl *ctrl) { struct xhci_segment *seg;
@@ -281,6 +294,7 @@ static struct xhci_segment *xhci_segment_alloc(void) BUG_ON(!seg);
seg->trbs = xhci_malloc(SEGMENT_SIZE); + seg->dma = xhci_dma_map(ctrl, seg->trbs, SEGMENT_SIZE);
seg->next = NULL;
@@ -314,7 +328,7 @@ struct xhci_ring *xhci_ring_alloc(struct xhci_ctrl *ctrl, unsigned int num_segs, if (num_segs == 0) return ring;
- ring->first_seg = xhci_segment_alloc(); + ring->first_seg = xhci_segment_alloc(ctrl); BUG_ON(!ring->first_seg);
num_segs--; @@ -323,7 +337,7 @@ struct xhci_ring *xhci_ring_alloc(struct xhci_ctrl *ctrl, unsigned int num_segs, while (num_segs > 0) { struct xhci_segment *next;
- next = xhci_segment_alloc(); + next = xhci_segment_alloc(ctrl); BUG_ON(!next);
xhci_link_segments(ctrl, prev, next, link_trbs); @@ -372,7 +386,8 @@ static int xhci_scratchpad_alloc(struct xhci_ctrl *ctrl) if (!scratchpad->sp_array) goto fail_sp2;
- val_64 = xhci_virt_to_bus(ctrl, scratchpad->sp_array); + val_64 = xhci_dma_map(ctrl, scratchpad->sp_array, + num_sp * sizeof(u64)); ctrl->dcbaa->dev_context_ptrs[0] = cpu_to_le64(val_64);
xhci_flush_cache((uintptr_t)&ctrl->dcbaa->dev_context_ptrs[0], @@ -386,16 +401,18 @@ static int xhci_scratchpad_alloc(struct xhci_ctrl *ctrl) } BUG_ON(i == 16);
- page_size = 1 << (i + 12); - buf = memalign(page_size, num_sp * page_size); + ctrl->page_size = 1 << (i + 12); + buf = memalign(ctrl->page_size, num_sp * ctrl->page_size); if (!buf) goto fail_sp3; - memset(buf, '\0', num_sp * page_size); - xhci_flush_cache((uintptr_t)buf, num_sp * page_size); + memset(buf, '\0', num_sp * ctrl->page_size); + xhci_flush_cache((uintptr_t)buf, num_sp * ctrl->page_size);
+ scratchpad->scratchpad = buf; + val_64 = xhci_dma_map(ctrl, buf, num_sp * ctrl->page_size); for (i = 0; i < num_sp; i++) { - val_64 = xhci_virt_to_bus(ctrl, buf + i * page_size); scratchpad->sp_array[i] = cpu_to_le64(val_64); + val_64 += ctrl->page_size; }
xhci_flush_cache((uintptr_t)scratchpad->sp_array, @@ -437,6 +454,7 @@ static struct xhci_container_ctx ctx->size += CTX_SIZE(xhci_readl(&ctrl->hccr->cr_hccparams));
ctx->bytes = xhci_malloc(ctx->size); + ctx->dma = xhci_dma_map(ctrl, ctx->bytes, ctx->size);
return ctx; } @@ -487,7 +505,7 @@ int xhci_alloc_virt_device(struct xhci_ctrl *ctrl, unsigned int slot_id) /* Allocate endpoint 0 ring */ virt_dev->eps[0].ring = xhci_ring_alloc(ctrl, 1, true);
- byte_64 = xhci_virt_to_bus(ctrl, virt_dev->out_ctx->bytes); + byte_64 = virt_dev->out_ctx->dma;
/* Point to output device context in dcbaa. */ ctrl->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(byte_64); @@ -523,15 +541,16 @@ int xhci_mem_init(struct xhci_ctrl *ctrl, struct xhci_hccr *hccr, return -ENOMEM; }
- val_64 = xhci_virt_to_bus(ctrl, ctrl->dcbaa); + ctrl->dcbaa->dma = xhci_dma_map(ctrl, ctrl->dcbaa, + sizeof(struct xhci_device_context_array)); /* Set the pointer in DCBAA register */ - xhci_writeq(&hcor->or_dcbaap, val_64); + xhci_writeq(&hcor->or_dcbaap, ctrl->dcbaa->dma);
/* Command ring control pointer register initialization */ ctrl->cmd_ring = xhci_ring_alloc(ctrl, 1, true);
/* Set the address in the Command Ring Control register */ - trb_64 = xhci_virt_to_bus(ctrl, ctrl->cmd_ring->first_seg->trbs); + trb_64 = ctrl->cmd_ring->first_seg->dma; val_64 = xhci_readq(&hcor->or_crcr); val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) | (trb_64 & (u64) ~CMD_RING_RSVD_BITS) | @@ -555,6 +574,8 @@ int xhci_mem_init(struct xhci_ctrl *ctrl, struct xhci_hccr *hccr, ctrl->event_ring = xhci_ring_alloc(ctrl, ERST_NUM_SEGS, false); ctrl->erst.entries = xhci_malloc(sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS); + ctrl->erst.erst_dma_addr = xhci_dma_map(ctrl, ctrl->erst.entries, + sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS);
ctrl->erst.num_entries = ERST_NUM_SEGS;
@@ -562,7 +583,7 @@ int xhci_mem_init(struct xhci_ctrl *ctrl, struct xhci_hccr *hccr, val < ERST_NUM_SEGS; val++) { struct xhci_erst_entry *entry = &ctrl->erst.entries[val]; - trb_64 = xhci_virt_to_bus(ctrl, seg->trbs); + trb_64 = seg->dma; entry->seg_addr = cpu_to_le64(trb_64); entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT); entry->rsvd = 0; @@ -571,7 +592,8 @@ int xhci_mem_init(struct xhci_ctrl *ctrl, struct xhci_hccr *hccr, xhci_flush_cache((uintptr_t)ctrl->erst.entries, ERST_NUM_SEGS * sizeof(struct xhci_erst_entry));
- deq = xhci_virt_to_bus(ctrl, ctrl->event_ring->dequeue); + deq = xhci_trb_virt_to_dma(ctrl->event_ring->deq_seg, + ctrl->event_ring->dequeue);
/* Update HC event ring dequeue pointer */ xhci_writeq(&ctrl->ir_set->erst_dequeue, @@ -586,7 +608,7 @@ int xhci_mem_init(struct xhci_ctrl *ctrl, struct xhci_hccr *hccr, /* this is the event ring segment table pointer */ val_64 = xhci_readq(&ctrl->ir_set->erst_base); val_64 &= ERST_PTR_MASK; - val_64 |= xhci_virt_to_bus(ctrl, ctrl->erst.entries) & ~ERST_PTR_MASK; + val_64 |= ctrl->erst.erst_dma_addr & ~ERST_PTR_MASK;
xhci_writeq(&ctrl->ir_set->erst_base, val_64);
@@ -849,7 +871,7 @@ void xhci_setup_addressable_virt_dev(struct xhci_ctrl *ctrl, /* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */ ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3));
- trb_64 = xhci_virt_to_bus(ctrl, virt_dev->eps[0].ring->first_seg->trbs); + trb_64 = virt_dev->eps[0].ring->first_seg->dma; ep0_ctx->deq = cpu_to_le64(trb_64 | virt_dev->eps[0].ring->cycle_state);
/* diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c index eb6dfcdb09..c8260cbdf9 100644 --- a/drivers/usb/host/xhci-ring.c +++ b/drivers/usb/host/xhci-ring.c @@ -24,6 +24,24 @@
#include <usb/xhci.h>
+/* + * Returns zero if the TRB isn't in this segment, otherwise it returns the DMA + * address of the TRB. + */ +dma_addr_t xhci_trb_virt_to_dma(struct xhci_segment *seg, + union xhci_trb *trb) +{ + unsigned long segment_offset; + + if (!seg || !trb || trb < seg->trbs) + return 0; + /* offset in TRBs */ + segment_offset = trb - seg->trbs; + if (segment_offset >= TRBS_PER_SEGMENT) + return 0; + return seg->dma + (segment_offset * sizeof(*trb)); +} + /** * Is this TRB a link TRB or was the last TRB the last TRB in this event ring * segment? I.e. would the updated event TRB pointer step off the end of the @@ -180,10 +198,8 @@ static void inc_deq(struct xhci_ctrl *ctrl, struct xhci_ring *ring) * @param trb_fields pointer to trb field array containing TRB contents * Return: pointer to the enqueued trb */ -static struct xhci_generic_trb *queue_trb(struct xhci_ctrl *ctrl, - struct xhci_ring *ring, - bool more_trbs_coming, - unsigned int *trb_fields) +static dma_addr_t queue_trb(struct xhci_ctrl *ctrl, struct xhci_ring *ring, + bool more_trbs_coming, unsigned int *trb_fields) { struct xhci_generic_trb *trb; int i; @@ -197,7 +213,7 @@ static struct xhci_generic_trb *queue_trb(struct xhci_ctrl *ctrl,
inc_enq(ctrl, ring, more_trbs_coming);
- return trb; + return xhci_trb_virt_to_dma(ring->enq_seg, (union xhci_trb *)trb); }
/** @@ -271,19 +287,15 @@ static int prepare_ring(struct xhci_ctrl *ctrl, struct xhci_ring *ep_ring, * @param cmd Command type to enqueue * Return: none */ -void xhci_queue_command(struct xhci_ctrl *ctrl, u8 *ptr, u32 slot_id, +void xhci_queue_command(struct xhci_ctrl *ctrl, dma_addr_t addr, u32 slot_id, u32 ep_index, trb_type cmd) { u32 fields[4]; - u64 val_64 = 0;
BUG_ON(prepare_ring(ctrl, ctrl->cmd_ring, EP_STATE_RUNNING));
- if (ptr) - val_64 = xhci_virt_to_bus(ctrl, ptr); - - fields[0] = lower_32_bits(val_64); - fields[1] = upper_32_bits(val_64); + fields[0] = lower_32_bits(addr); + fields[1] = upper_32_bits(addr); fields[2] = 0; fields[3] = TRB_TYPE(cmd) | SLOT_ID_FOR_TRB(slot_id) | ctrl->cmd_ring->cycle_state; @@ -399,12 +411,15 @@ static void giveback_first_trb(struct usb_device *udev, int ep_index, */ void xhci_acknowledge_event(struct xhci_ctrl *ctrl) { + dma_addr_t deq; + /* Advance our dequeue pointer to the next event */ inc_deq(ctrl, ctrl->event_ring);
/* Inform the hardware */ - xhci_writeq(&ctrl->ir_set->erst_dequeue, - xhci_virt_to_bus(ctrl, ctrl->event_ring->dequeue) | ERST_EHB); + deq = xhci_trb_virt_to_dma(ctrl->event_ring->deq_seg, + ctrl->event_ring->dequeue); + xhci_writeq(&ctrl->ir_set->erst_dequeue, deq | ERST_EHB); }
/** @@ -490,17 +505,19 @@ static void reset_ep(struct usb_device *udev, int ep_index) struct xhci_ctrl *ctrl = xhci_get_ctrl(udev); struct xhci_ring *ring = ctrl->devs[udev->slot_id]->eps[ep_index].ring; union xhci_trb *event; + u64 addr; u32 field;
printf("Resetting EP %d...\n", ep_index); - xhci_queue_command(ctrl, NULL, udev->slot_id, ep_index, TRB_RESET_EP); + xhci_queue_command(ctrl, 0, udev->slot_id, ep_index, TRB_RESET_EP); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); field = le32_to_cpu(event->trans_event.flags); BUG_ON(TRB_TO_SLOT_ID(field) != udev->slot_id); xhci_acknowledge_event(ctrl);
- xhci_queue_command(ctrl, (void *)((uintptr_t)ring->enqueue | - ring->cycle_state), udev->slot_id, ep_index, TRB_SET_DEQ); + addr = xhci_trb_virt_to_dma(ring->enq_seg, + (void *)((uintptr_t)ring->enqueue | ring->cycle_state)); + xhci_queue_command(ctrl, addr, udev->slot_id, ep_index, TRB_SET_DEQ); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); BUG_ON(TRB_TO_SLOT_ID(le32_to_cpu(event->event_cmd.flags)) != udev->slot_id || GET_COMP_CODE(le32_to_cpu( @@ -521,9 +538,10 @@ static void abort_td(struct usb_device *udev, int ep_index) struct xhci_ctrl *ctrl = xhci_get_ctrl(udev); struct xhci_ring *ring = ctrl->devs[udev->slot_id]->eps[ep_index].ring; union xhci_trb *event; + u64 addr; u32 field;
- xhci_queue_command(ctrl, NULL, udev->slot_id, ep_index, TRB_STOP_RING); + xhci_queue_command(ctrl, 0, udev->slot_id, ep_index, TRB_STOP_RING);
event = xhci_wait_for_event(ctrl, TRB_TRANSFER); field = le32_to_cpu(event->trans_event.flags); @@ -539,8 +557,9 @@ static void abort_td(struct usb_device *udev, int ep_index) event->event_cmd.status)) != COMP_SUCCESS); xhci_acknowledge_event(ctrl);
- xhci_queue_command(ctrl, (void *)((uintptr_t)ring->enqueue | - ring->cycle_state), udev->slot_id, ep_index, TRB_SET_DEQ); + addr = xhci_trb_virt_to_dma(ring->enq_seg, + (void *)((uintptr_t)ring->enqueue | ring->cycle_state)); + xhci_queue_command(ctrl, addr, udev->slot_id, ep_index, TRB_SET_DEQ); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); BUG_ON(TRB_TO_SLOT_ID(le32_to_cpu(event->event_cmd.flags)) != udev->slot_id || GET_COMP_CODE(le32_to_cpu( @@ -609,8 +628,8 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe, u64 addr; int ret; u32 trb_fields[4]; - u64 val_64 = xhci_virt_to_bus(ctrl, buffer); - void *last_transfer_trb_addr; + u64 buf_64 = xhci_dma_map(ctrl, buffer, length); + dma_addr_t last_transfer_trb_addr; int available_length;
debug("dev=%p, pipe=%lx, buffer=%p, length=%d\n", @@ -633,7 +652,7 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe, * we send request in more than 1 TRB by chaining them. */ running_total = TRB_MAX_BUFF_SIZE - - (lower_32_bits(val_64) & (TRB_MAX_BUFF_SIZE - 1)); + (lower_32_bits(buf_64) & (TRB_MAX_BUFF_SIZE - 1)); trb_buff_len = running_total; running_total &= TRB_MAX_BUFF_SIZE - 1;
@@ -678,7 +697,7 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe, * that the buffer should not span 64KB boundary. if so * we send request in more than 1 TRB by chaining them. */ - addr = val_64; + addr = buf_64;
if (trb_buff_len > length) trb_buff_len = length; @@ -754,7 +773,7 @@ again: }
if ((uintptr_t)(le64_to_cpu(event->trans_event.buffer)) != - (uintptr_t)xhci_virt_to_bus(ctrl, last_transfer_trb_addr)) { + (uintptr_t)last_transfer_trb_addr) { available_length -= (int)EVENT_TRB_LEN(le32_to_cpu(event->trans_event.transfer_len)); xhci_acknowledge_event(ctrl); @@ -768,6 +787,7 @@ again: record_transfer_result(udev, event, available_length); xhci_acknowledge_event(ctrl); xhci_inval_cache((uintptr_t)buffer, length); + xhci_dma_unmap(ctrl, buf_64, length);
return (udev->status != USB_ST_NOT_PROC) ? 0 : -1; } @@ -911,7 +931,7 @@ int xhci_ctrl_tx(struct usb_device *udev, unsigned long pipe, if (length > 0) { if (req->requesttype & USB_DIR_IN) field |= TRB_DIR_IN; - buf_64 = xhci_virt_to_bus(ctrl, buffer); + buf_64 = xhci_dma_map(ctrl, buffer, length);
trb_fields[0] = lower_32_bits(buf_64); trb_fields[1] = upper_32_bits(buf_64); @@ -961,8 +981,10 @@ int xhci_ctrl_tx(struct usb_device *udev, unsigned long pipe, }
/* Invalidate buffer to make it available to usb-core */ - if (length > 0) + if (length > 0) { xhci_inval_cache((uintptr_t)buffer, length); + xhci_dma_unmap(ctrl, buf_64, length); + }
if (GET_COMP_CODE(le32_to_cpu(event->trans_event.transfer_len)) == COMP_SHORT_TX) { diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c index dbeb88afe3..440b0224b1 100644 --- a/drivers/usb/host/xhci.c +++ b/drivers/usb/host/xhci.c @@ -448,7 +448,7 @@ static int xhci_configure_endpoints(struct usb_device *udev, bool ctx_change) in_ctx = virt_dev->in_ctx;
xhci_flush_cache((uintptr_t)in_ctx->bytes, in_ctx->size); - xhci_queue_command(ctrl, in_ctx->bytes, udev->slot_id, 0, + xhci_queue_command(ctrl, in_ctx->dma, udev->slot_id, 0, ctx_change ? TRB_EVAL_CONTEXT : TRB_CONFIG_EP); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); BUG_ON(TRB_TO_SLOT_ID(le32_to_cpu(event->event_cmd.flags)) @@ -585,7 +585,8 @@ static int xhci_set_configuration(struct usb_device *udev) cpu_to_le32(MAX_BURST(max_burst) | ERROR_COUNT(err_count));
- trb_64 = xhci_virt_to_bus(ctrl, virt_dev->eps[ep_index].ring->enqueue); + trb_64 = xhci_trb_virt_to_dma(virt_dev->eps[ep_index].ring->enq_seg, + virt_dev->eps[ep_index].ring->enqueue); ep_ctx[ep_index]->deq = cpu_to_le64(trb_64 | virt_dev->eps[ep_index].ring->cycle_state);
@@ -643,7 +644,8 @@ static int xhci_address_device(struct usb_device *udev, int root_portnr) ctrl_ctx->add_flags = cpu_to_le32(SLOT_FLAG | EP0_FLAG); ctrl_ctx->drop_flags = 0;
- xhci_queue_command(ctrl, (void *)ctrl_ctx, slot_id, 0, TRB_ADDR_DEV); + xhci_queue_command(ctrl, virt_dev->in_ctx->dma, + slot_id, 0, TRB_ADDR_DEV); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); BUG_ON(TRB_TO_SLOT_ID(le32_to_cpu(event->event_cmd.flags)) != slot_id);
@@ -718,7 +720,7 @@ static int _xhci_alloc_device(struct usb_device *udev) return 0; }
- xhci_queue_command(ctrl, NULL, 0, 0, TRB_ENABLE_SLOT); + xhci_queue_command(ctrl, 0, 0, 0, TRB_ENABLE_SLOT); event = xhci_wait_for_event(ctrl, TRB_COMPLETION); BUG_ON(GET_COMP_CODE(le32_to_cpu(event->event_cmd.status)) != COMP_SUCCESS); diff --git a/include/usb/xhci.h b/include/usb/xhci.h index ea4cf3f52b..85c359fa1b 100644 --- a/include/usb/xhci.h +++ b/include/usb/xhci.h @@ -16,6 +16,7 @@ #ifndef HOST_XHCI_H_ #define HOST_XHCI_H_
+#include <iommu.h> #include <phys2bus.h> #include <asm/types.h> #include <asm/cache.h> @@ -490,6 +491,7 @@ struct xhci_container_ctx {
int size; u8 *bytes; + dma_addr_t dma; };
/** @@ -688,6 +690,8 @@ struct xhci_input_control_ctx { struct xhci_device_context_array { /* 64-bit device addresses; we only write 32-bit addresses */ __le64 dev_context_ptrs[MAX_HC_SLOTS]; + /* private xHCD pointers */ + dma_addr_t dma; }; /* TODO: write function to set the 64-bit device DMA address */ /* @@ -997,6 +1001,7 @@ struct xhci_segment { union xhci_trb *trbs; /* private to HCD */ struct xhci_segment *next; + dma_addr_t dma; };
struct xhci_ring { @@ -1025,11 +1030,14 @@ struct xhci_erst_entry { struct xhci_erst { struct xhci_erst_entry *entries; unsigned int num_entries; + /* xhci->event_ring keeps track of segment dma addresses */ + dma_addr_t erst_dma_addr; /* Num entries the ERST can contain */ unsigned int erst_size; };
struct xhci_scratchpad { + void *scratchpad; u64 *sp_array; };
@@ -1216,6 +1224,7 @@ struct xhci_ctrl { struct xhci_virt_device *devs[MAX_HC_SLOTS]; int rootdev; u16 hci_version; + int page_size; u32 quirks; #define XHCI_MTK_HOST BIT(0) }; @@ -1226,7 +1235,7 @@ struct xhci_ctrl { #define xhci_to_dev(_ctrl) NULL #endif
-unsigned long trb_addr(struct xhci_segment *seg, union xhci_trb *trb); +dma_addr_t xhci_trb_virt_to_dma(struct xhci_segment *seg, union xhci_trb *trb); struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_container_ctx *ctx); struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_ctrl *ctrl, @@ -1243,7 +1252,7 @@ void xhci_slot_copy(struct xhci_ctrl *ctrl, struct xhci_container_ctx *out_ctx); void xhci_setup_addressable_virt_dev(struct xhci_ctrl *ctrl, struct usb_device *udev, int hop_portnr); -void xhci_queue_command(struct xhci_ctrl *ctrl, u8 *ptr, +void xhci_queue_command(struct xhci_ctrl *ctrl, dma_addr_t addr, u32 slot_id, u32 ep_index, trb_type cmd); void xhci_acknowledge_event(struct xhci_ctrl *ctrl); union xhci_trb *xhci_wait_for_event(struct xhci_ctrl *ctrl, trb_type expected); @@ -1284,14 +1293,22 @@ extern struct dm_usb_ops xhci_usb_ops;
struct xhci_ctrl *xhci_get_ctrl(struct usb_device *udev);
-static inline dma_addr_t xhci_virt_to_bus(struct xhci_ctrl *ctrl, void *addr) +static inline dma_addr_t xhci_dma_map(struct xhci_ctrl *ctrl, void *addr, + size_t size) { +#if CONFIG_IS_ENABLED(IOMMU) + return dev_iommu_dma_map(xhci_to_dev(ctrl), addr, size); +#else return dev_phys_to_bus(xhci_to_dev(ctrl), virt_to_phys(addr)); +#endif }
-static inline void *xhci_bus_to_virt(struct xhci_ctrl *ctrl, dma_addr_t addr) +static inline void xhci_dma_unmap(struct xhci_ctrl *ctrl, dma_addr_t addr, + size_t size) { - return phys_to_virt(dev_bus_to_phys(xhci_to_dev(ctrl), addr)); +#if CONFIG_IS_ENABLED(IOMMU) + dev_iommu_dma_unmap(xhci_to_dev(ctrl), addr, size); +#endif }
#endif /* HOST_XHCI_H_ */
Systems such as Apple's M1 and M2 SoCs may have separate IOMMUs for each PCIe root port. In this case the right IOMMU for a PCI device behind a particular root port is described by an "iommu-map" property in the device tree. Parse this property and use it to find the right IOMMU device for PCI devices.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- drivers/iommu/iommu-uclass.c | 65 ++++++++++++++++++++++++++++++++++++ 1 file changed, 65 insertions(+)
diff --git a/drivers/iommu/iommu-uclass.c b/drivers/iommu/iommu-uclass.c index f6b1457736..72f123df55 100644 --- a/drivers/iommu/iommu-uclass.c +++ b/drivers/iommu/iommu-uclass.c @@ -8,10 +8,71 @@ #include <common.h> #include <dm.h> #include <iommu.h> +#include <malloc.h> #include <phys2bus.h> #include <asm/io.h>
#if (CONFIG_IS_ENABLED(OF_CONTROL) && !CONFIG_IS_ENABLED(OF_PLATDATA)) + +#if CONFIG_IS_ENABLED(PCI) +static int dev_pci_iommu_enable(struct udevice *dev) +{ + struct udevice *parent = dev->parent; + struct udevice *dev_iommu; + u32 *iommu_map; + u32 iommu_map_mask, length, phandle, rid, rid_base; + int i, count, len, ret; + + while (parent) { + len = dev_read_size(parent, "iommu-map"); + if (len > 0) + break; + parent = parent->parent; + } + + if (len <= 0) + return 0; + + iommu_map = malloc(len); + if (!iommu_map) + return -ENOMEM; + + count = len / sizeof(u32); + ret = dev_read_u32_array(parent, "iommu-map", iommu_map, count); + if (ret < 0) { + free(iommu_map); + return 0; + } + + iommu_map_mask = dev_read_u32_default(parent, "iommu-map-mask", ~0); + rid = (dm_pci_get_bdf(dev) >> 8) & iommu_map_mask; + + /* Loop over entries until mapping is found. */ + for (i = 0; i < count; i += 4) { + rid_base = iommu_map[i]; + phandle = iommu_map[i + 1]; + length = iommu_map[i + 3]; + + if (rid < rid_base || rid >= rid_base + length) + continue; + + ret = uclass_get_device_by_phandle_id(UCLASS_IOMMU, phandle, + &dev_iommu); + if (ret) { + debug("%s: uclass_get_device_by_ofnode failed: %d\n", + __func__, ret); + free(iommu_map); + return ret; + } + dev->iommu = dev_iommu; + break; + } + + free(iommu_map); + return 0; +} +#endif + int dev_iommu_enable(struct udevice *dev) { struct ofnode_phandle_args args; @@ -39,6 +100,10 @@ int dev_iommu_enable(struct udevice *dev) dev->iommu = dev_iommu; }
+ if (CONFIG_IS_ENABLED(PCI) && count < 0 && + device_is_on_pci_bus(dev)) + return dev_pci_iommu_enable(dev); + return 0; } #endif
Test that we correctly probe an IOMMU that is mapped by an "iommu-map" device tree property of a PCIe controller node.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- arch/sandbox/dts/test.dts | 2 ++ test/dm/iommu.c | 30 ++++++++++++++++++++++++++++++ 2 files changed, 32 insertions(+)
diff --git a/arch/sandbox/dts/test.dts b/arch/sandbox/dts/test.dts index dffe10adbf..b576b61a8c 100644 --- a/arch/sandbox/dts/test.dts +++ b/arch/sandbox/dts/test.dts @@ -1025,6 +1025,8 @@ #size-cells = <2>; ranges = <0x02000000 0 0x10000000 0x10000000 0 0x2000000 0x01000000 0 0x20000000 0x20000000 0 0x2000>; + iommu-map = <0x0010 &iommu 0 1>; + iommu-map-mask = <0xfffffff8>; pci@0,0 { compatible = "pci-generic"; reg = <0x0000 0 0 0 0>; diff --git a/test/dm/iommu.c b/test/dm/iommu.c index 6f932ef233..62d38f1214 100644 --- a/test/dm/iommu.c +++ b/test/dm/iommu.c @@ -68,3 +68,33 @@ static int dm_test_iommu_noiommu(struct unit_test_state *uts) return 0; } DM_TEST(dm_test_iommu_noiommu, UT_TESTF_SCAN_FDT); + +static int dm_test_iommu_pci(struct unit_test_state *uts) +{ + struct udevice *dev; + + ut_assertok(uclass_find_device(UCLASS_IOMMU, 0, &dev)); + ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); + + /* Probing P2SB probes the IOMMU through the "iommu-map" property */ + ut_assertok(uclass_probe_all(UCLASS_P2SB)); + ut_assert(dev_get_flags(dev) & DM_FLAG_ACTIVATED); + + return 0; +} +DM_TEST(dm_test_iommu_pci, UT_TESTF_SCAN_FDT); + +static int dm_test_iommu_pci_noiommu(struct unit_test_state *uts) +{ + struct udevice *dev; + + ut_assertok(uclass_find_device(UCLASS_IOMMU, 0, &dev)); + ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); + + /* Probing PMC should not probe the IOMMU */ + ut_assertok(uclass_probe_all(UCLASS_ACPI_PMC)); + ut_assert(!(dev_get_flags(dev) & DM_FLAG_ACTIVATED)); + + return 0; +} +DM_TEST(dm_test_iommu_pci_noiommu, UT_TESTF_SCAN_FDT);
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
Test that we correctly probe an IOMMU that is mapped by an "iommu-map" device tree property of a PCIe controller node.
Signed-off-by: Mark Kettenis kettenis@openbsd.org
arch/sandbox/dts/test.dts | 2 ++ test/dm/iommu.c | 30 ++++++++++++++++++++++++++++++ 2 files changed, 32 insertions(+)
Reviewed-by: Simon Glass sjg@chromium.org
This driver supports the PCIe controller on the Apple M1 and M2 SoCs. The code is adapted from the Linux driver.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- MAINTAINERS | 1 + arch/arm/Kconfig | 2 + drivers/pci/Kconfig | 9 + drivers/pci/Makefile | 1 + drivers/pci/pcie_apple.c | 354 +++++++++++++++++++++++++++++++++++++++ 5 files changed, 367 insertions(+) create mode 100644 drivers/pci/pcie_apple.c
diff --git a/MAINTAINERS b/MAINTAINERS index 3fc4cd0f12..b8a947f9d3 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -122,6 +122,7 @@ F: arch/arm/mach-apple/ F: configs/apple_m1_defconfig F: drivers/iommu/apple_dart.c F: drivers/nvme/nvme_apple.c +F: drivers/pci/pcie_apple.c F: drivers/pinctrl/pinctrl-apple.c F: drivers/watchdog/apple_wdt.c F: include/configs/apple.h diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index cac4fa09fd..780815269b 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -963,6 +963,7 @@ config ARCH_APPLE bool "Apple SoCs" select ARM64 select CLK + select CMD_PCI select CMD_USB select DM select DM_GPIO @@ -977,6 +978,7 @@ config ARCH_APPLE select LINUX_KERNEL_IMAGE_HEADER select OF_BOARD_SETUP select OF_CONTROL + select PCI select PINCTRL select POSITION_INDEPENDENT select POWER_DOMAIN diff --git a/drivers/pci/Kconfig b/drivers/pci/Kconfig index 22f4995453..d61596cd7c 100644 --- a/drivers/pci/Kconfig +++ b/drivers/pci/Kconfig @@ -105,6 +105,15 @@ config PCIE_ECAM_SYNQUACER Note that this must be configured when boot because Linux driver expects the PCIe RC has been configured in the bootloader.
+config PCIE_APPLE + bool "Enable Apple PCIe driver" + depends on ARCH_APPLE + imply PCI_INIT_R + default y + help + Say Y here if you want to enable PCIe controller support on + Apple SoCs. + config PCI_GT64120 bool "GT64120 PCI support" depends on MIPS diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile index dd1ad91ced..d393f1ba03 100644 --- a/drivers/pci/Makefile +++ b/drivers/pci/Makefile @@ -13,6 +13,7 @@ obj-$(CONFIG_PCI) += pci_auto_common.o pci_common.o
obj-$(CONFIG_PCIE_ECAM_GENERIC) += pcie_ecam_generic.o obj-$(CONFIG_PCIE_ECAM_SYNQUACER) += pcie_ecam_synquacer.o +obj-$(CONFIG_PCIE_APPLE) += pcie_apple.o obj-$(CONFIG_PCI_GT64120) += pci_gt64120.o obj-$(CONFIG_PCI_MPC85XX) += pci_mpc85xx.o obj-$(CONFIG_PCI_MSC01) += pci_msc01.o diff --git a/drivers/pci/pcie_apple.c b/drivers/pci/pcie_apple.c new file mode 100644 index 0000000000..9b08e1e5da --- /dev/null +++ b/drivers/pci/pcie_apple.c @@ -0,0 +1,354 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * PCIe host bridge driver for Apple system-on-chips. + * + * The HW is ECAM compliant. + * + * Initialization requires enabling power and clocks, along with a + * number of register pokes. + * + * Copyright (C) 2021 Alyssa Rosenzweig alyssa@rosenzweig.io + * Copyright (C) 2021 Google LLC + * Copyright (C) 2021 Corellium LLC + * Copyright (C) 2021 Mark Kettenis kettenis@openbsd.org + * + * Author: Alyssa Rosenzweig alyssa@rosenzweig.io + * Author: Marc Zyngier maz@kernel.org + */ + +#include <common.h> +#include <dm.h> +#include <dm/device_compat.h> +#include <dm/devres.h> +#include <mapmem.h> +#include <pci.h> +#include <asm/io.h> +#include <asm-generic/gpio.h> +#include <linux/delay.h> +#include <linux/iopoll.h> + +#define CORE_RC_PHYIF_CTL 0x00024 +#define CORE_RC_PHYIF_CTL_RUN BIT(0) +#define CORE_RC_PHYIF_STAT 0x00028 +#define CORE_RC_PHYIF_STAT_REFCLK BIT(4) +#define CORE_RC_CTL 0x00050 +#define CORE_RC_CTL_RUN BIT(0) +#define CORE_RC_STAT 0x00058 +#define CORE_RC_STAT_READY BIT(0) +#define CORE_FABRIC_STAT 0x04000 +#define CORE_FABRIC_STAT_MASK 0x001F001F +#define CORE_LANE_CFG(port) (0x84000 + 0x4000 * (port)) +#define CORE_LANE_CFG_REFCLK0REQ BIT(0) +#define CORE_LANE_CFG_REFCLK1REQ BIT(1) +#define CORE_LANE_CFG_REFCLK0ACK BIT(2) +#define CORE_LANE_CFG_REFCLK1ACK BIT(3) +#define CORE_LANE_CFG_REFCLKEN (BIT(9) | BIT(10)) +#define CORE_LANE_CTL(port) (0x84004 + 0x4000 * (port)) +#define CORE_LANE_CTL_CFGACC BIT(15) + +#define PORT_LTSSMCTL 0x00080 +#define PORT_LTSSMCTL_START BIT(0) +#define PORT_INTSTAT 0x00100 +#define PORT_INT_TUNNEL_ERR 31 +#define PORT_INT_CPL_TIMEOUT 23 +#define PORT_INT_RID2SID_MAPERR 22 +#define PORT_INT_CPL_ABORT 21 +#define PORT_INT_MSI_BAD_DATA 19 +#define PORT_INT_MSI_ERR 18 +#define PORT_INT_REQADDR_GT32 17 +#define PORT_INT_AF_TIMEOUT 15 +#define PORT_INT_LINK_DOWN 14 +#define PORT_INT_LINK_UP 12 +#define PORT_INT_LINK_BWMGMT 11 +#define PORT_INT_AER_MASK (15 << 4) +#define PORT_INT_PORT_ERR 4 +#define PORT_INT_INTx(i) i +#define PORT_INT_INTx_MASK 15 +#define PORT_INTMSK 0x00104 +#define PORT_INTMSKSET 0x00108 +#define PORT_INTMSKCLR 0x0010c +#define PORT_MSICFG 0x00124 +#define PORT_MSICFG_EN BIT(0) +#define PORT_MSICFG_L2MSINUM_SHIFT 4 +#define PORT_MSIBASE 0x00128 +#define PORT_MSIBASE_1_SHIFT 16 +#define PORT_MSIADDR 0x00168 +#define PORT_LINKSTS 0x00208 +#define PORT_LINKSTS_UP BIT(0) +#define PORT_LINKSTS_BUSY BIT(2) +#define PORT_LINKCMDSTS 0x00210 +#define PORT_OUTS_NPREQS 0x00284 +#define PORT_OUTS_NPREQS_REQ BIT(24) +#define PORT_OUTS_NPREQS_CPL BIT(16) +#define PORT_RXWR_FIFO 0x00288 +#define PORT_RXWR_FIFO_HDR GENMASK(15, 10) +#define PORT_RXWR_FIFO_DATA GENMASK(9, 0) +#define PORT_RXRD_FIFO 0x0028C +#define PORT_RXRD_FIFO_REQ GENMASK(6, 0) +#define PORT_OUTS_CPLS 0x00290 +#define PORT_OUTS_CPLS_SHRD GENMASK(14, 8) +#define PORT_OUTS_CPLS_WAIT GENMASK(6, 0) +#define PORT_APPCLK 0x00800 +#define PORT_APPCLK_EN BIT(0) +#define PORT_APPCLK_CGDIS BIT(8) +#define PORT_STATUS 0x00804 +#define PORT_STATUS_READY BIT(0) +#define PORT_REFCLK 0x00810 +#define PORT_REFCLK_EN BIT(0) +#define PORT_REFCLK_CGDIS BIT(8) +#define PORT_PERST 0x00814 +#define PORT_PERST_OFF BIT(0) +#define PORT_RID2SID(i16) (0x00828 + 4 * (i16)) +#define PORT_RID2SID_VALID BIT(31) +#define PORT_RID2SID_SID_SHIFT 16 +#define PORT_RID2SID_BUS_SHIFT 8 +#define PORT_RID2SID_DEV_SHIFT 3 +#define PORT_RID2SID_FUNC_SHIFT 0 +#define PORT_OUTS_PREQS_HDR 0x00980 +#define PORT_OUTS_PREQS_HDR_MASK GENMASK(9, 0) +#define PORT_OUTS_PREQS_DATA 0x00984 +#define PORT_OUTS_PREQS_DATA_MASK GENMASK(15, 0) +#define PORT_TUNCTRL 0x00988 +#define PORT_TUNCTRL_PERST_ON BIT(0) +#define PORT_TUNCTRL_PERST_ACK_REQ BIT(1) +#define PORT_TUNSTAT 0x0098c +#define PORT_TUNSTAT_PERST_ON BIT(0) +#define PORT_TUNSTAT_PERST_ACK_PEND BIT(1) +#define PORT_PREFMEM_ENABLE 0x00994 + +struct apple_pcie_priv { + struct udevice *dev; + void __iomem *base; + void __iomem *cfg_base; + struct list_head ports; +}; + +struct apple_pcie_port { + struct apple_pcie_priv *pcie; + struct gpio_desc reset; + ofnode np; + void __iomem *base; + struct list_head entry; + int idx; +}; + +static void rmw_set(u32 set, void __iomem *addr) +{ + writel_relaxed(readl_relaxed(addr) | set, addr); +} + +static void rmw_clear(u32 clr, void __iomem *addr) +{ + writel_relaxed(readl_relaxed(addr) & ~clr, addr); +} + +static int apple_pcie_config_address(const struct udevice *bus, + pci_dev_t bdf, uint offset, + void **paddress) +{ + struct apple_pcie_priv *pcie = dev_get_priv(bus); + void *addr; + + addr = pcie->cfg_base; + addr += PCIE_ECAM_OFFSET(PCI_BUS(bdf), PCI_DEV(bdf), + PCI_FUNC(bdf), offset); + *paddress = addr; + + return 0; +} + +static int apple_pcie_read_config(const struct udevice *bus, pci_dev_t bdf, + uint offset, ulong *valuep, + enum pci_size_t size) +{ + int ret; + + ret = pci_generic_mmap_read_config(bus, apple_pcie_config_address, + bdf, offset, valuep, size); + return ret; +} + +static int apple_pcie_write_config(struct udevice *bus, pci_dev_t bdf, + uint offset, ulong value, + enum pci_size_t size) +{ + return pci_generic_mmap_write_config(bus, apple_pcie_config_address, + bdf, offset, value, size); +} + +static const struct dm_pci_ops apple_pcie_ops = { + .read_config = apple_pcie_read_config, + .write_config = apple_pcie_write_config, +}; + +static int apple_pcie_setup_refclk(struct apple_pcie_priv *pcie, + struct apple_pcie_port *port) +{ + u32 stat; + int res; + + res = readl_poll_sleep_timeout(pcie->base + CORE_RC_PHYIF_STAT, stat, + stat & CORE_RC_PHYIF_STAT_REFCLK, + 100, 50000); + if (res < 0) + return res; + + rmw_set(CORE_LANE_CTL_CFGACC, pcie->base + CORE_LANE_CTL(port->idx)); + rmw_set(CORE_LANE_CFG_REFCLK0REQ, pcie->base + CORE_LANE_CFG(port->idx)); + + res = readl_poll_sleep_timeout(pcie->base + CORE_LANE_CFG(port->idx), + stat, stat & CORE_LANE_CFG_REFCLK0ACK, + 100, 50000); + if (res < 0) + return res; + + rmw_set(CORE_LANE_CFG_REFCLK1REQ, pcie->base + CORE_LANE_CFG(port->idx)); + res = readl_poll_sleep_timeout(pcie->base + CORE_LANE_CFG(port->idx), + stat, stat & CORE_LANE_CFG_REFCLK1ACK, + 100, 50000); + + if (res < 0) + return res; + + rmw_clear(CORE_LANE_CTL_CFGACC, pcie->base + CORE_LANE_CTL(port->idx)); + + rmw_set(CORE_LANE_CFG_REFCLKEN, pcie->base + CORE_LANE_CFG(port->idx)); + rmw_set(PORT_REFCLK_EN, port->base + PORT_REFCLK); + + return 0; +} + +static int apple_pcie_setup_port(struct apple_pcie_priv *pcie, ofnode np) +{ + struct apple_pcie_port *port; + struct gpio_desc reset; + fdt_addr_t addr; + u32 stat, idx; + int ret; + + ret = gpio_request_by_name_nodev(np, "reset-gpios", 0, &reset, 0); + if (ret) + return ret; + + port = devm_kzalloc(pcie->dev, sizeof(*port), GFP_KERNEL); + if (!port) + return -ENOMEM; + + ret = ofnode_read_u32_index(np, "reg", 0, &idx); + if (ret) + return ret; + + /* Use the first reg entry to work out the port index */ + port->idx = idx >> 11; + port->pcie = pcie; + port->reset = reset; + port->np = np; + + addr = dev_read_addr_index(pcie->dev, port->idx + 2); + if (addr == FDT_ADDR_T_NONE) + return -EINVAL; + port->base = map_sysmem(addr, 0); + + rmw_set(PORT_APPCLK_EN, port->base + PORT_APPCLK); + + /* Assert PERST# before setting up the clock */ + dm_gpio_set_value(&reset, 1); + + ret = apple_pcie_setup_refclk(pcie, port); + if (ret < 0) + return ret; + + /* The minimal Tperst-clk value is 100us (PCIe CEM r5.0, 2.9.2) */ + udelay(100); + + /* Deassert PERST# */ + rmw_set(PORT_PERST_OFF, port->base + PORT_PERST); + dm_gpio_set_value(&reset, 0); + + /* Wait for 100ms after PERST# deassertion (PCIe r5.0, 6.6.1) */ + udelay(100 * 1000); + + ret = readl_poll_sleep_timeout(port->base + PORT_STATUS, stat, + stat & PORT_STATUS_READY, 100, 250000); + if (ret < 0) { + dev_err(pcie->dev, "port %d ready wait timeout\n", port->idx); + return ret; + } + + rmw_clear(PORT_REFCLK_CGDIS, port->base + PORT_REFCLK); + rmw_clear(PORT_APPCLK_CGDIS, port->base + PORT_APPCLK); + + list_add_tail(&port->entry, &pcie->ports); + + writel_relaxed(PORT_LTSSMCTL_START, port->base + PORT_LTSSMCTL); + + /* + * Deliberately ignore the link not coming up as connected + * devices (e.g. the WiFi controller) may not be powerd up. + */ + readl_poll_sleep_timeout(port->base + PORT_LINKSTS, stat, + (stat & PORT_LINKSTS_UP), 100, 100000); + + return 0; +} + +static int apple_pcie_probe(struct udevice *dev) +{ + struct apple_pcie_priv *pcie = dev_get_priv(dev); + fdt_addr_t addr; + ofnode of_port; + int i, ret; + + pcie->dev = dev; + addr = dev_read_addr_index(dev, 0); + if (addr == FDT_ADDR_T_NONE) + return -EINVAL; + pcie->cfg_base = map_sysmem(addr, 0); + + addr = dev_read_addr_index(dev, 1); + if (addr == FDT_ADDR_T_NONE) + return -EINVAL; + pcie->base = map_sysmem(addr, 0); + + INIT_LIST_HEAD(&pcie->ports); + + for (of_port = ofnode_first_subnode(dev_ofnode(dev)); + ofnode_valid(of_port); + of_port = ofnode_next_subnode(of_port)) { + ret = apple_pcie_setup_port(pcie, of_port); + if (ret) { + dev_err(pcie->dev, "Port %d setup fail: %d\n", i, ret); + return ret; + } + } + + return 0; +} + +static int apple_pcie_remove(struct udevice *dev) +{ + struct apple_pcie_priv *pcie = dev_get_priv(dev); + struct apple_pcie_port *port, *tmp; + + list_for_each_entry_safe(port, tmp, &pcie->ports, entry) { + gpio_free_list_nodev(&port->reset, 1); + free(port); + } + + return 0; +} + +static const struct udevice_id apple_pcie_of_match[] = { + { .compatible = "apple,pcie" }, + { /* sentinel */ } +}; + +U_BOOT_DRIVER(apple_pcie) = { + .name = "apple_pcie", + .id = UCLASS_PCI, + .of_match = apple_pcie_of_match, + .probe = apple_pcie_probe, + .remove = apple_pcie_remove, + .priv_auto = sizeof(struct apple_pcie_priv), + .ops = &apple_pcie_ops, +};
Some Apple Silicon machines have a PCIe XHCI controller in additon to the DWC3 controllers integrated on the SoC. On the Mac mini the Type-A ports are handled by this PCIe controller. Enabling it allows the use of these ports in U-Boot.
Signed-off-by: Mark Kettenis kettenis@openbsd.org --- configs/apple_m1_defconfig | 1 + 1 file changed, 1 insertion(+)
diff --git a/configs/apple_m1_defconfig b/configs/apple_m1_defconfig index 4ba09334d4..b4ecf73cbc 100644 --- a/configs/apple_m1_defconfig +++ b/configs/apple_m1_defconfig @@ -15,6 +15,7 @@ CONFIG_APPLE_SPI_KEYB=y CONFIG_NVME_APPLE=y CONFIG_USB_XHCI_HCD=y CONFIG_USB_XHCI_DWC3=y +CONFIG_USB_XHCI_PCI=y CONFIG_USB_KEYBOARD=y CONFIG_SYS_WHITE_ON_BLACK=y CONFIG_NO_FB_CLEAR=y
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de --- drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c index 440b0224b1..9e33c5d855 100644 --- a/drivers/usb/host/xhci.c +++ b/drivers/usb/host/xhci.c @@ -944,7 +944,7 @@ static int xhci_submit_root(struct usb_device *udev, unsigned long pipe, case USB_DT_HUB: case USB_DT_SS_HUB: debug("USB_DT_HUB config\n"); - srcptr = &descriptor.hub; + srcptr = &ctrl->hub_desc; srclen = 0x8; break; default: @@ -1203,21 +1203,22 @@ static int xhci_lowlevel_init(struct xhci_ctrl *ctrl) /* initializing xhci data structures */ if (xhci_mem_init(ctrl, hccr, hcor) < 0) return -ENOMEM; + ctrl->hub_desc = descriptor.hub;
reg = xhci_readl(&hccr->cr_hcsparams1); - descriptor.hub.bNbrPorts = HCS_MAX_PORTS(reg); - printf("Register %x NbrPorts %d\n", reg, descriptor.hub.bNbrPorts); + ctrl->hub_desc.bNbrPorts = HCS_MAX_PORTS(reg); + printf("Register %x NbrPorts %d\n", reg, ctrl->hub_desc.bNbrPorts);
/* Port Indicators */ reg = xhci_readl(&hccr->cr_hccparams); if (HCS_INDICATOR(reg)) - put_unaligned(get_unaligned(&descriptor.hub.wHubCharacteristics) - | 0x80, &descriptor.hub.wHubCharacteristics); + put_unaligned(get_unaligned(&ctrl->hub_desc.wHubCharacteristics) + | 0x80, &ctrl->hub_desc.wHubCharacteristics);
/* Port Power Control */ if (HCC_PPC(reg)) - put_unaligned(get_unaligned(&descriptor.hub.wHubCharacteristics) - | 0x01, &descriptor.hub.wHubCharacteristics); + put_unaligned(get_unaligned(&ctrl->hub_desc.wHubCharacteristics) + | 0x01, &ctrl->hub_desc.wHubCharacteristics);
if (xhci_start(hcor)) { xhci_reset(hcor); diff --git a/include/usb/xhci.h b/include/usb/xhci.h index 85c359fa1b..4a4ac10229 100644 --- a/include/usb/xhci.h +++ b/include/usb/xhci.h @@ -1222,6 +1222,7 @@ struct xhci_ctrl { struct xhci_erst_entry entry[ERST_NUM_SEGS]; struct xhci_scratchpad *scratchpad; struct xhci_virt_device *devs[MAX_HC_SLOTS]; + struct usb_hub_descriptor hub_desc; int rootdev; u16 hci_version; int page_size;
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de
drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
Tested on: Chromebook Brya (Intel Alder Lake): Tested-by: Simon Glass sjg@chromium.org
On 2/21/23 00:45, Simon Glass wrote:
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de
drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
Tested on: Chromebook Brya (Intel Alder Lake): Tested-by: Simon Glass sjg@chromium.org
Mark, do you plan to pick the entire series via some apple tree or shall I pick the USB parts via USB tree ? If the later, then please collect the AB/RB tags and send me just the two xhci patches separately. If the former, no action is needed.
Date: Tue, 21 Feb 2023 01:46:22 +0100 From: Marek Vasut marex@denx.de
Hi Marek,
On 2/21/23 00:45, Simon Glass wrote:
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de
drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
Tested on: Chromebook Brya (Intel Alder Lake): Tested-by: Simon Glass sjg@chromium.org
Mark, do you plan to pick the entire series via some apple tree or shall I pick the USB parts via USB tree ? If the later, then please collect the AB/RB tags and send me just the two xhci patches separately. If the former, no action is needed.
Tom already merged the whole series into master a few weeks ago. Not sure why Simon thought it made sense to add his Tested-by tags when the patch in question is already on master.
Cheers,
Mark
On 2/21/23 10:28, Mark Kettenis wrote:
Date: Tue, 21 Feb 2023 01:46:22 +0100 From: Marek Vasut marex@denx.de
Hi Marek,
On 2/21/23 00:45, Simon Glass wrote:
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de
drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
Tested on: Chromebook Brya (Intel Alder Lake): Tested-by: Simon Glass sjg@chromium.org
Mark, do you plan to pick the entire series via some apple tree or shall I pick the USB parts via USB tree ? If the later, then please collect the AB/RB tags and send me just the two xhci patches separately. If the former, no action is needed.
Tom already merged the whole series into master a few weeks ago. Not sure why Simon thought it made sense to add his Tested-by tags when the patch in question is already on master.
Uh, well, then all is good, thanks !
Hi Mark,
On Tue, 21 Feb 2023 at 02:28, Mark Kettenis mark.kettenis@xs4all.nl wrote:
Date: Tue, 21 Feb 2023 01:46:22 +0100 From: Marek Vasut marex@denx.de
Hi Marek,
On 2/21/23 00:45, Simon Glass wrote:
On Sat, 21 Jan 2023 at 12:28, Mark Kettenis kettenis@openbsd.org wrote:
When a system has multiple XHCI controllers, some of the properties described in the descriptor of the root hub (such as the number of ports) might differ between controllers. Fix this by switching from a single global hub descriptor to a hub descriptor per controller.
Signed-off-by: Mark Kettenis kettenis@openbsd.org Reviewed-by: Marek Vasut marex@denx.de
drivers/usb/host/xhci.c | 15 ++++++++------- include/usb/xhci.h | 1 + 2 files changed, 9 insertions(+), 7 deletions(-)
Tested on: Chromebook Brya (Intel Alder Lake): Tested-by: Simon Glass sjg@chromium.org
Mark, do you plan to pick the entire series via some apple tree or shall I pick the USB parts via USB tree ? If the later, then please collect the AB/RB tags and send me just the two xhci patches separately. If the former, no action is needed.
Tom already merged the whole series into master a few weeks ago. Not sure why Simon thought it made sense to add his Tested-by tags when the patch in question is already on master.
Because sometimes these things get reverted and I didn't want to have to remember that it fixed my problem.
Regards, Simon
participants (5)
-
Marek Vasut -
Mark Kettenis -
Mark Kettenis -
Simon Glass -
Tom Rini