Commit 47143b09 authored by Linus Torvalds's avatar Linus Torvalds
Browse files

Merge branch 'drivers' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile

* 'drivers' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
  pci root complex: support for tile architecture
  drivers/net/tile/: on-chip network drivers for the tile architecture
  MAINTAINERS: add drivers/char/hvc_tile.c as maintained by tile
parents 0b9466cc f02cbbe6
......@@ -5839,6 +5839,8 @@ M: Chris Metcalf <cmetcalf@tilera.com>
W: http://www.tilera.com/scm/
S: Supported
F: arch/tile/
F: drivers/char/hvc_tile.c
F: drivers/net/tile/
TLAN NETWORK DRIVER
M: Samuel Chessman <chessman@tux.org>
......
......@@ -329,6 +329,18 @@ endmenu # Tilera-specific configuration
menu "Bus options"
config PCI
bool "PCI support"
default y
select PCI_DOMAINS
---help---
Enable PCI root complex support, so PCIe endpoint devices can
be attached to the Tile chip. Many, but not all, PCI devices
are supported under Tilera's root complex driver.
config PCI_DOMAINS
bool
config NO_IOMEM
def_bool !PCI
......
......@@ -137,4 +137,56 @@ static inline void finv_buffer(void *buffer, size_t size)
mb_incoherent();
}
/*
* Flush & invalidate a VA range that is homed remotely on a single core,
* waiting until the memory controller holds the flushed values.
*/
static inline void finv_buffer_remote(void *buffer, size_t size)
{
char *p;
int i;
/*
* Flush and invalidate the buffer out of the local L1/L2
* and request the home cache to flush and invalidate as well.
*/
__finv_buffer(buffer, size);
/*
* Wait for the home cache to acknowledge that it has processed
* all the flush-and-invalidate requests. This does not mean
* that the flushed data has reached the memory controller yet,
* but it does mean the home cache is processing the flushes.
*/
__insn_mf();
/*
* Issue a load to the last cache line, which can't complete
* until all the previously-issued flushes to the same memory
* controller have also completed. If we weren't striping
* memory, that one load would be sufficient, but since we may
* be, we also need to back up to the last load issued to
* another memory controller, which would be the point where
* we crossed an 8KB boundary (the granularity of striping
* across memory controllers). Keep backing up and doing this
* until we are before the beginning of the buffer, or have
* hit all the controllers.
*/
for (i = 0, p = (char *)buffer + size - 1;
i < (1 << CHIP_LOG_NUM_MSHIMS()) && p >= (char *)buffer;
++i) {
const unsigned long STRIPE_WIDTH = 8192;
/* Force a load instruction to issue. */
*(volatile char *)p;
/* Jump to end of previous stripe. */
p -= STRIPE_WIDTH;
p = (char *)((unsigned long)p | (STRIPE_WIDTH - 1));
}
/* Wait for the loads (and thus flushes) to have completed. */
__insn_mf();
}
#endif /* _ASM_TILE_CACHEFLUSH_H */
......@@ -55,9 +55,6 @@ extern void iounmap(volatile void __iomem *addr);
#define ioremap_writethrough(physaddr, size) ioremap(physaddr, size)
#define ioremap_fullcache(physaddr, size) ioremap(physaddr, size)
void __iomem *ioport_map(unsigned long port, unsigned int len);
extern inline void ioport_unmap(void __iomem *addr) {}
#define mmiowb()
/* Conversion between virtual and physical mappings. */
......@@ -189,12 +186,22 @@ static inline void memcpy_toio(volatile void __iomem *dst, const void *src,
* we never run, uses them unconditionally.
*/
static inline int ioport_panic(void)
static inline long ioport_panic(void)
{
panic("inb/outb and friends do not exist on tile");
return 0;
}
static inline void __iomem *ioport_map(unsigned long port, unsigned int len)
{
return (void __iomem *) ioport_panic();
}
static inline void ioport_unmap(void __iomem *addr)
{
ioport_panic();
}
static inline u8 inb(unsigned long addr)
{
return ioport_panic();
......
/*
* Copyright 2010 Tilera Corporation. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation, version 2.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
* NON INFRINGEMENT. See the GNU General Public License for
* more details.
*/
#ifndef _ASM_TILE_PCI_BRIDGE_H
#define _ASM_TILE_PCI_BRIDGE_H
#include <linux/ioport.h>
#include <linux/pci.h>
struct device_node;
struct pci_controller;
/*
* pci_io_base returns the memory address at which you can access
* the I/O space for PCI bus number `bus' (or NULL on error).
*/
extern void __iomem *pci_bus_io_base(unsigned int bus);
extern unsigned long pci_bus_io_base_phys(unsigned int bus);
extern unsigned long pci_bus_mem_base_phys(unsigned int bus);
/* Allocate a new PCI host bridge structure */
extern struct pci_controller *pcibios_alloc_controller(void);
/* Helper function for setting up resources */
extern void pci_init_resource(struct resource *res, unsigned long start,
unsigned long end, int flags, char *name);
/* Get the PCI host controller for a bus */
extern struct pci_controller *pci_bus_to_hose(int bus);
/*
* Structure of a PCI controller (host bridge)
*/
struct pci_controller {
int index; /* PCI domain number */
struct pci_bus *root_bus;
int first_busno;
int last_busno;
int hv_cfg_fd[2]; /* config{0,1} fds for this PCIe controller */
int hv_mem_fd; /* fd to Hypervisor for MMIO operations */
struct pci_ops *ops;
int irq_base; /* Base IRQ from the Hypervisor */
int plx_gen1; /* flag for PLX Gen 1 configuration */
/* Address ranges that are routed to this controller/bridge. */
struct resource mem_resources[3];
};
static inline struct pci_controller *pci_bus_to_host(struct pci_bus *bus)
{
return bus->sysdata;
}
extern void setup_indirect_pci_nomap(struct pci_controller *hose,
void __iomem *cfg_addr, void __iomem *cfg_data);
extern void setup_indirect_pci(struct pci_controller *hose,
u32 cfg_addr, u32 cfg_data);
extern void setup_grackle(struct pci_controller *hose);
extern unsigned char common_swizzle(struct pci_dev *, unsigned char *);
/*
* The following code swizzles for exactly one bridge. The routine
* common_swizzle below handles multiple bridges. But there are a
* some boards that don't follow the PCI spec's suggestion so we
* break this piece out separately.
*/
static inline unsigned char bridge_swizzle(unsigned char pin,
unsigned char idsel)
{
return (((pin-1) + idsel) % 4) + 1;
}
/*
* The following macro is used to lookup irqs in a standard table
* format for those PPC systems that do not already have PCI
* interrupts properly routed.
*/
/* FIXME - double check this */
#define PCI_IRQ_TABLE_LOOKUP ({ \
long _ctl_ = -1; \
if (idsel >= min_idsel && idsel <= max_idsel && pin <= irqs_per_slot) \
_ctl_ = pci_irq_table[idsel - min_idsel][pin-1]; \
_ctl_; \
})
/*
* Scan the buses below a given PCI host bridge and assign suitable
* resources to all devices found.
*/
extern int pciauto_bus_scan(struct pci_controller *, int);
#ifdef CONFIG_PCI
extern unsigned long pci_address_to_pio(phys_addr_t address);
#else
static inline unsigned long pci_address_to_pio(phys_addr_t address)
{
return (unsigned long)-1;
}
#endif
#endif /* _ASM_TILE_PCI_BRIDGE_H */
......@@ -15,7 +15,29 @@
#ifndef _ASM_TILE_PCI_H
#define _ASM_TILE_PCI_H
#include <asm/pci-bridge.h>
#include <linux/pci.h>
/*
* Structure of a PCI controller (host bridge)
*/
struct pci_controller {
int index; /* PCI domain number */
struct pci_bus *root_bus;
int first_busno;
int last_busno;
int hv_cfg_fd[2]; /* config{0,1} fds for this PCIe controller */
int hv_mem_fd; /* fd to Hypervisor for MMIO operations */
struct pci_ops *ops;
int irq_base; /* Base IRQ from the Hypervisor */
int plx_gen1; /* flag for PLX Gen 1 configuration */
/* Address ranges that are routed to this controller/bridge. */
struct resource mem_resources[3];
};
/*
* The hypervisor maps the entirety of CPA-space as bus addresses, so
......@@ -24,56 +46,12 @@
*/
#define PCI_DMA_BUS_IS_PHYS 1
struct pci_controller *pci_bus_to_hose(int bus);
unsigned char __init common_swizzle(struct pci_dev *dev, unsigned char *pinp);
int __init tile_pci_init(void);
void pci_iounmap(struct pci_dev *dev, void __iomem *addr);
void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long max);
void __devinit pcibios_fixup_bus(struct pci_bus *bus);
int __devinit _tile_cfg_read(struct pci_controller *hose,
int bus,
int slot,
int function,
int offset,
int size,
u32 *val);
int __devinit _tile_cfg_write(struct pci_controller *hose,
int bus,
int slot,
int function,
int offset,
int size,
u32 val);
void __iomem *pci_iomap(struct pci_dev *dev, int bar, unsigned long max);
static inline void pci_iounmap(struct pci_dev *dev, void __iomem *addr) {}
/*
* These are used to to config reads and writes in the early stages of
* setup before the driver infrastructure has been set up enough to be
* able to do config reads and writes.
*/
#define early_cfg_read(where, size, value) \
_tile_cfg_read(controller, \
current_bus, \
pci_slot, \
pci_fn, \
where, \
size, \
value)
#define early_cfg_write(where, size, value) \
_tile_cfg_write(controller, \
current_bus, \
pci_slot, \
pci_fn, \
where, \
size, \
value)
#define PCICFG_BYTE 1
#define PCICFG_WORD 2
#define PCICFG_DWORD 4
void __devinit pcibios_fixup_bus(struct pci_bus *bus);
#define TILE_NUM_PCIE 2
......@@ -88,33 +66,33 @@ static inline int pci_proc_domain(struct pci_bus *bus)
}
/*
* I/O space is currently not supported.
* pcibios_assign_all_busses() tells whether or not the bus numbers
* should be reassigned, in case the BIOS didn't do it correctly, or
* in case we don't have a BIOS and we want to let Linux do it.
*/
static inline int pcibios_assign_all_busses(void)
{
return 1;
}
#define TILE_PCIE_LOWER_IO 0x0
#define TILE_PCIE_UPPER_IO 0x10000
#define TILE_PCIE_PCIE_IO_SIZE 0x0000FFFF
#define _PAGE_NO_CACHE 0
#define _PAGE_GUARDED 0
#define pcibios_assign_all_busses() pci_assign_all_buses
extern int pci_assign_all_buses;
/*
* No special bus mastering setup handling.
*/
static inline void pcibios_set_master(struct pci_dev *dev)
{
/* No special bus mastering setup handling */
}
#define PCIBIOS_MIN_MEM 0
#define PCIBIOS_MIN_IO TILE_PCIE_LOWER_IO
#define PCIBIOS_MIN_IO 0
/*
* This flag tells if the platform is TILEmpower that needs
* special configuration for the PLX switch chip.
*/
extern int blade_pci;
extern int tile_plx_gen1;
/* Use any cpu for PCI. */
#define cpumask_of_pcibus(bus) cpu_online_mask
/* implement the pci_ DMA API in terms of the generic device dma_ one */
#include <asm-generic/pci-dma-compat.h>
......@@ -122,7 +100,4 @@ extern int blade_pci;
/* generic pci stuff */
#include <asm-generic/pci.h>
/* Use any cpu for PCI. */
#define cpumask_of_pcibus(bus) cpu_online_mask
#endif /* _ASM_TILE_PCI_H */
......@@ -292,8 +292,18 @@ extern int kstack_hash;
/* Are we using huge pages in the TLB for kernel data? */
extern int kdata_huge;
/* Support standard Linux prefetching. */
#define ARCH_HAS_PREFETCH
#define prefetch(x) __builtin_prefetch(x)
#define PREFETCH_STRIDE CHIP_L2_LINE_SIZE()
/* Bring a value into the L1D, faulting the TLB if necessary. */
#ifdef __tilegx__
#define prefetch_L1(x) __insn_prefetch_l1_fault((void *)(x))
#else
#define prefetch_L1(x) __insn_prefetch_L1((void *)(x))
#endif
#else /* __ASSEMBLY__ */
/* Do some slow action (e.g. read a slow SPR). */
......
/*
* Copyright 2010 Tilera Corporation. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation, version 2.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
* NON INFRINGEMENT. See the GNU General Public License for
* more details.
*/
/**
* @file drivers/xgbe/impl.h
* Implementation details for the NetIO library.
*/
#ifndef __DRV_XGBE_IMPL_H__
#define __DRV_XGBE_IMPL_H__
#include <hv/netio_errors.h>
#include <hv/netio_intf.h>
#include <hv/drv_xgbe_intf.h>
/** How many groups we have (log2). */
#define LOG2_NUM_GROUPS (12)
/** How many groups we have. */
#define NUM_GROUPS (1 << LOG2_NUM_GROUPS)
/** Number of output requests we'll buffer per tile. */
#define EPP_REQS_PER_TILE (32)
/** Words used in an eDMA command without checksum acceleration. */
#define EDMA_WDS_NO_CSUM 8
/** Words used in an eDMA command with checksum acceleration. */
#define EDMA_WDS_CSUM 10
/** Total available words in the eDMA command FIFO. */
#define EDMA_WDS_TOTAL 128
/*
* FIXME: These definitions are internal and should have underscores!
* NOTE: The actual numeric values here are intentional and allow us to
* optimize the concept "if small ... else if large ... else ...", by
* checking for the low bit being set, and then for non-zero.
* These are used as array indices, so they must have the values (0, 1, 2)
* in some order.
*/
#define SIZE_SMALL (1) /**< Small packet queue. */
#define SIZE_LARGE (2) /**< Large packet queue. */
#define SIZE_JUMBO (0) /**< Jumbo packet queue. */
/** The number of "SIZE_xxx" values. */
#define NETIO_NUM_SIZES 3
/*
* Default numbers of packets for IPP drivers. These values are chosen
* such that CIPP1 will not overflow its L2 cache.
*/
/** The default number of small packets. */
#define NETIO_DEFAULT_SMALL_PACKETS 2750
/** The default number of large packets. */
#define NETIO_DEFAULT_LARGE_PACKETS 2500
/** The default number of jumbo packets. */
#define NETIO_DEFAULT_JUMBO_PACKETS 250
/** Log2 of the size of a memory arena. */
#define NETIO_ARENA_SHIFT 24 /* 16 MB */
/** Size of a memory arena. */
#define NETIO_ARENA_SIZE (1 << NETIO_ARENA_SHIFT)
/** A queue of packets.
*
* This structure partially defines a queue of packets waiting to be
* processed. The queue as a whole is written to by an interrupt handler and
* read by non-interrupt code; this data structure is what's touched by the
* interrupt handler. The other part of the queue state, the read offset, is
* kept in user space, not in hypervisor space, so it is in a separate data
* structure.
*
* The read offset (__packet_receive_read in the user part of the queue
* structure) points to the next packet to be read. When the read offset is
* equal to the write offset, the queue is empty; therefore the queue must
* contain one more slot than the required maximum queue size.
*
* Here's an example of all 3 state variables and what they mean. All
* pointers move left to right.
*
* @code
* I I V V V V I I I I
* 0 1 2 3 4 5 6 7 8 9 10
* ^ ^ ^ ^
* | | |
* | | __last_packet_plus_one
* | __buffer_write
* __packet_receive_read
* @endcode
*
* This queue has 10 slots, and thus can hold 9 packets (_last_packet_plus_one
* = 10). The read pointer is at 2, and the write pointer is at 6; thus,
* there are valid, unread packets in slots 2, 3, 4, and 5. The remaining
* slots are invalid (do not contain a packet).
*/
typedef struct {
/** Byte offset of the next notify packet to be written: zero for the first
* packet on the queue, sizeof (netio_pkt_t) for the second packet on the
* queue, etc. */
volatile uint32_t __packet_write;
/** Offset of the packet after the last valid packet (i.e., when any
* pointer is incremented to this value, it wraps back to zero). */
uint32_t __last_packet_plus_one;
}
__netio_packet_queue_t;
/** A queue of buffers.
*
* This structure partially defines a queue of empty buffers which have been
* obtained via requests to the IPP. (The elements of the queue are packet
* handles, which are transformed into a full netio_pkt_t when the buffer is
* retrieved.) The queue as a whole is written to by an interrupt handler and
* read by non-interrupt code; this data structure is what's touched by the
* interrupt handler. The other parts of the queue state, the read offset and
* requested write offset, are kept in user space, not in hypervisor space, so
* they are in a separate data structure.
*
* The read offset (__buffer_read in the user part of the queue structure)
* points to the next buffer to be read. When the read offset is equal to the
* write offset, the queue is empty; therefore the queue must contain one more
* slot than the required maximum queue size.
*
* The requested write offset (__buffer_requested_write in the user part of
* the queue structure) points to the slot which will hold the next buffer we
* request from the IPP, once we get around to sending such a request. When
* the requested write offset is equal to the write offset, no requests for
* new buffers are outstanding; when the requested write offset is one greater
* than the read offset, no more requests may be sent.
*
* Note that, unlike the packet_queue, the buffer_queue places incoming
* buffers at decreasing addresses. This makes the check for "is it time to
* wrap the buffer pointer" cheaper in the assembly code which receives new
* buffers, and means that the value which defines the queue size,
* __last_buffer, is different than in the packet queue. Also, the offset
* used in the packet_queue is already scaled by the size of a packet; here we
* use unscaled slot indices for the offsets. (These differences are
* historical, and in the future it's possible that the packet_queue will look
* more like this queue.)
*
* @code
* Here's an example of all 4 state variables and what they mean. Remember:
* all pointers move right to left.
*
* V V V I I R R V V V
* 0 1 2 3 4 5 6 7 8 9
* ^ ^ ^ ^
* | | | |
* | | | __last_buffer
* | | __buffer_write
* | __buffer_requested_write
* __buffer_read
* @endcode
*
* This queue has 10 slots, and thus can hold 9 buffers (_last_buffer = 9).
* The read pointer is at 2, and the write pointer is at 6; thus, there are
* valid, unread buffers in slots 2, 1, 0, 9, 8, and 7. The requested write
* pointer is at 4; thus, requests have been made to the IPP for buffers which
* will be placed in slots 6 and 5 when they arrive. Finally, the remaining
* slots are invalid (do not contain a buffer).
*/
typedef struct
{
/** Ordinal number of the next buffer to be written: 0 for the first slot in
* the queue, 1 for the second slot in the queue, etc. */
volatile uint32_t __buffer_write;
/** Ordinal number of the last buffer (i.e., when any pointer is decremented
* below zero, it is reloaded with this value). */
uint32_t __last_buffer;
}
__netio_buffer_queue_t;
/**
* An object for providing Ethernet packets to a process.
*/
typedef struct __netio_queue_impl_t
{
/** The queue of packets waiting to be received. */
__netio_packet_queue_t __packet_receive_queue;
/** The intr bit mask that IDs this device. */
unsigned int __intr_id;
/** Offset to queues of empty buffers, one per size. */
uint32_t __buffer_queue[NETIO_NUM_SIZES];
/** The address of the first EPP tile, or -1 if no EPP. */
/* ISSUE: Actually this is always "0" or "~0". */
uint32_t __epp_location;
/** The queue ID that this queue represents. */
unsigned int __queue_id;
/** Number of acknowledgements received. */
volatile uint32_t __acks_received;
/** Last completion number received for packet_sendv. */
volatile uint32_t __last_completion_rcv;
/** Number of packets allowed to be outstanding. */
uint32_t __max_outstanding;
/** First VA available for packets. */
void* __va_0;
/** First VA in second range available for packets. */
void* __va_1;
/** Padding to align the "__packets" field to the size of a netio_pkt_t. */
uint32_t __padding[3];