mirror of
https://github.com/torvalds/linux.git
synced 2024-11-18 01:51:53 +00:00
2bdd1b031b
Previously, one would have to specifically choose CONFIG_OLPC and CONFIG_PCI_GOOLPC in order to enable PCI_OLPC. That doesn't really work for distro kernels, so this patch allows one to choose CONFIG_OLPC and CONFIG_PCI_GOANY in order to build in OLPC support in a generic kernel (as requested by Robert Millan). This also moves GOOLPC before GOANY in the menuconfig list. Finally, make pci_access_init return early if we detect OLPC hardware. There's no need to continue probing stuff, and pci_pcbios_init specifically trashes our settings (we didn't run into that before because PCI_GOANY wasn't supported). Signed-off-by: Andres Salomon <dilinger@debian.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
315 lines
9.4 KiB
C
315 lines
9.4 KiB
C
/*
|
|
* Low-level PCI config space access for OLPC systems who lack the VSA
|
|
* PCI virtualization software.
|
|
*
|
|
* Copyright © 2006 Advanced Micro Devices, Inc.
|
|
*
|
|
* 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; either version 2 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* The AMD Geode chipset (ie: GX2 processor, cs5536 I/O companion device)
|
|
* has some I/O functions (display, southbridge, sound, USB HCIs, etc)
|
|
* that more or less behave like PCI devices, but the hardware doesn't
|
|
* directly implement the PCI configuration space headers. AMD provides
|
|
* "VSA" (Virtual System Architecture) software that emulates PCI config
|
|
* space for these devices, by trapping I/O accesses to PCI config register
|
|
* (CF8/CFC) and running some code in System Management Mode interrupt state.
|
|
* On the OLPC platform, we don't want to use that VSA code because
|
|
* (a) it slows down suspend/resume, and (b) recompiling it requires special
|
|
* compilers that are hard to get. So instead of letting the complex VSA
|
|
* code simulate the PCI config registers for the on-chip devices, we
|
|
* just simulate them the easy way, by inserting the code into the
|
|
* pci_write_config and pci_read_config path. Most of the config registers
|
|
* are read-only anyway, so the bulk of the simulation is just table lookup.
|
|
*/
|
|
|
|
#include <linux/pci.h>
|
|
#include <linux/init.h>
|
|
#include <asm/olpc.h>
|
|
#include <asm/geode.h>
|
|
#include "pci.h"
|
|
|
|
/*
|
|
* In the tables below, the first two line (8 longwords) are the
|
|
* size masks that are used when the higher level PCI code determines
|
|
* the size of the region by writing ~0 to a base address register
|
|
* and reading back the result.
|
|
*
|
|
* The following lines are the values that are read during normal
|
|
* PCI config access cycles, i.e. not after just having written
|
|
* ~0 to a base address register.
|
|
*/
|
|
|
|
static const uint32_t lxnb_hdr[] = { /* dev 1 function 0 - devfn = 8 */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x281022, 0x2200005, 0x6000021, 0x80f808, /* AMD Vendor ID */
|
|
0x0, 0x0, 0x0, 0x0, /* No virtual registers, hence no BAR */
|
|
0x0, 0x0, 0x0, 0x28100b,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t gxnb_hdr[] = { /* dev 1 function 0 - devfn = 8 */
|
|
0xfffffffd, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x28100b, 0x2200005, 0x6000021, 0x80f808, /* NSC Vendor ID */
|
|
0xac1d, 0x0, 0x0, 0x0, /* I/O BAR - base of virtual registers */
|
|
0x0, 0x0, 0x0, 0x28100b,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t lxfb_hdr[] = { /* dev 1 function 1 - devfn = 9 */
|
|
0xff000008, 0xffffc000, 0xffffc000, 0xffffc000,
|
|
0xffffc000, 0x0, 0x0, 0x0,
|
|
|
|
0x20811022, 0x2200003, 0x3000000, 0x0, /* AMD Vendor ID */
|
|
0xfd000000, 0xfe000000, 0xfe004000, 0xfe008000, /* FB, GP, VG, DF */
|
|
0xfe00c000, 0x0, 0x0, 0x30100b, /* VIP */
|
|
0x0, 0x0, 0x0, 0x10e, /* INTA, IRQ14 for graphics accel */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x3d0, 0x3c0, 0xa0000, 0x0, /* VG IO, VG IO, EGA FB, MONO FB */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t gxfb_hdr[] = { /* dev 1 function 1 - devfn = 9 */
|
|
0xff800008, 0xffffc000, 0xffffc000, 0xffffc000,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x30100b, 0x2200003, 0x3000000, 0x0, /* NSC Vendor ID */
|
|
0xfd000000, 0xfe000000, 0xfe004000, 0xfe008000, /* FB, GP, VG, DF */
|
|
0x0, 0x0, 0x0, 0x30100b,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x3d0, 0x3c0, 0xa0000, 0x0, /* VG IO, VG IO, EGA FB, MONO FB */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t aes_hdr[] = { /* dev 1 function 2 - devfn = 0xa */
|
|
0xffffc000, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x20821022, 0x2a00006, 0x10100000, 0x8, /* NSC Vendor ID */
|
|
0xfe010000, 0x0, 0x0, 0x0, /* AES registers */
|
|
0x0, 0x0, 0x0, 0x20821022,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
|
|
static const uint32_t isa_hdr[] = { /* dev f function 0 - devfn = 78 */
|
|
0xfffffff9, 0xffffff01, 0xffffffc1, 0xffffffe1,
|
|
0xffffff81, 0xffffffc1, 0x0, 0x0,
|
|
|
|
0x20901022, 0x2a00049, 0x6010003, 0x802000,
|
|
0x18b1, 0x1001, 0x1801, 0x1881, /* SMB-8 GPIO-256 MFGPT-64 IRQ-32 */
|
|
0x1401, 0x1841, 0x0, 0x20901022, /* PMS-128 ACPI-64 */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0xaa5b, /* IRQ steering */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t ac97_hdr[] = { /* dev f function 3 - devfn = 7b */
|
|
0xffffff81, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x20931022, 0x2a00041, 0x4010001, 0x0,
|
|
0x1481, 0x0, 0x0, 0x0, /* I/O BAR-128 */
|
|
0x0, 0x0, 0x0, 0x20931022,
|
|
0x0, 0x0, 0x0, 0x205, /* IntB, IRQ5 */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t ohci_hdr[] = { /* dev f function 4 - devfn = 7c */
|
|
0xfffff000, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x20941022, 0x2300006, 0xc031002, 0x0,
|
|
0xfe01a000, 0x0, 0x0, 0x0, /* MEMBAR-1000 */
|
|
0x0, 0x0, 0x0, 0x20941022,
|
|
0x0, 0x40, 0x0, 0x40a, /* CapPtr INT-D, IRQA */
|
|
0xc8020001, 0x0, 0x0, 0x0, /* Capabilities - 40 is R/O,
|
|
44 is mask 8103 (power control) */
|
|
0x0, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
};
|
|
|
|
static const uint32_t ehci_hdr[] = { /* dev f function 4 - devfn = 7d */
|
|
0xfffff000, 0x0, 0x0, 0x0,
|
|
0x0, 0x0, 0x0, 0x0,
|
|
|
|
0x20951022, 0x2300006, 0xc032002, 0x0,
|
|
0xfe01b000, 0x0, 0x0, 0x0, /* MEMBAR-1000 */
|
|
0x0, 0x0, 0x0, 0x20951022,
|
|
0x0, 0x40, 0x0, 0x40a, /* CapPtr INT-D, IRQA */
|
|
0xc8020001, 0x0, 0x0, 0x0, /* Capabilities - 40 is R/O, 44 is
|
|
mask 8103 (power control) */
|
|
#if 0
|
|
0x1, 0x40080000, 0x0, 0x0, /* EECP - see EHCI spec section 2.1.7 */
|
|
#endif
|
|
0x01000001, 0x0, 0x0, 0x0, /* EECP - see EHCI spec section 2.1.7 */
|
|
0x2020, 0x0, 0x0, 0x0, /* (EHCI page 8) 60 SBRN (R/O),
|
|
61 FLADJ (R/W), PORTWAKECAP */
|
|
};
|
|
|
|
static uint32_t ff_loc = ~0;
|
|
static uint32_t zero_loc;
|
|
static int bar_probing; /* Set after a write of ~0 to a BAR */
|
|
static int is_lx;
|
|
|
|
#define NB_SLOT 0x1 /* Northbridge - GX chip - Device 1 */
|
|
#define SB_SLOT 0xf /* Southbridge - CS5536 chip - Device F */
|
|
|
|
static int is_simulated(unsigned int bus, unsigned int devfn)
|
|
{
|
|
return (!bus && ((PCI_SLOT(devfn) == NB_SLOT) ||
|
|
(PCI_SLOT(devfn) == SB_SLOT)));
|
|
}
|
|
|
|
static uint32_t *hdr_addr(const uint32_t *hdr, int reg)
|
|
{
|
|
uint32_t addr;
|
|
|
|
/*
|
|
* This is a little bit tricky. The header maps consist of
|
|
* 0x20 bytes of size masks, followed by 0x70 bytes of header data.
|
|
* In the normal case, when not probing a BAR's size, we want
|
|
* to access the header data, so we add 0x20 to the reg offset,
|
|
* thus skipping the size mask area.
|
|
* In the BAR probing case, we want to access the size mask for
|
|
* the BAR, so we subtract 0x10 (the config header offset for
|
|
* BAR0), and don't skip the size mask area.
|
|
*/
|
|
|
|
addr = (uint32_t)hdr + reg + (bar_probing ? -0x10 : 0x20);
|
|
|
|
bar_probing = 0;
|
|
return (uint32_t *)addr;
|
|
}
|
|
|
|
static int pci_olpc_read(unsigned int seg, unsigned int bus,
|
|
unsigned int devfn, int reg, int len, uint32_t *value)
|
|
{
|
|
uint32_t *addr;
|
|
|
|
/* Use the hardware mechanism for non-simulated devices */
|
|
if (!is_simulated(bus, devfn))
|
|
return pci_direct_conf1.read(seg, bus, devfn, reg, len, value);
|
|
|
|
/*
|
|
* No device has config registers past 0x70, so we save table space
|
|
* by not storing entries for the nonexistent registers
|
|
*/
|
|
if (reg >= 0x70)
|
|
addr = &zero_loc;
|
|
else {
|
|
switch (devfn) {
|
|
case 0x8:
|
|
addr = hdr_addr(is_lx ? lxnb_hdr : gxnb_hdr, reg);
|
|
break;
|
|
case 0x9:
|
|
addr = hdr_addr(is_lx ? lxfb_hdr : gxfb_hdr, reg);
|
|
break;
|
|
case 0xa:
|
|
addr = is_lx ? hdr_addr(aes_hdr, reg) : &ff_loc;
|
|
break;
|
|
case 0x78:
|
|
addr = hdr_addr(isa_hdr, reg);
|
|
break;
|
|
case 0x7b:
|
|
addr = hdr_addr(ac97_hdr, reg);
|
|
break;
|
|
case 0x7c:
|
|
addr = hdr_addr(ohci_hdr, reg);
|
|
break;
|
|
case 0x7d:
|
|
addr = hdr_addr(ehci_hdr, reg);
|
|
break;
|
|
default:
|
|
addr = &ff_loc;
|
|
break;
|
|
}
|
|
}
|
|
switch (len) {
|
|
case 1:
|
|
*value = *(uint8_t *)addr;
|
|
break;
|
|
case 2:
|
|
*value = *(uint16_t *)addr;
|
|
break;
|
|
case 4:
|
|
*value = *addr;
|
|
break;
|
|
default:
|
|
BUG();
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static int pci_olpc_write(unsigned int seg, unsigned int bus,
|
|
unsigned int devfn, int reg, int len, uint32_t value)
|
|
{
|
|
/* Use the hardware mechanism for non-simulated devices */
|
|
if (!is_simulated(bus, devfn))
|
|
return pci_direct_conf1.write(seg, bus, devfn, reg, len, value);
|
|
|
|
/* XXX we may want to extend this to simulate EHCI power management */
|
|
|
|
/*
|
|
* Mostly we just discard writes, but if the write is a size probe
|
|
* (i.e. writing ~0 to a BAR), we remember it and arrange to return
|
|
* the appropriate size mask on the next read. This is cheating
|
|
* to some extent, because it depends on the fact that the next
|
|
* access after such a write will always be a read to the same BAR.
|
|
*/
|
|
|
|
if ((reg >= 0x10) && (reg < 0x2c)) {
|
|
/* write is to a BAR */
|
|
if (value == ~0)
|
|
bar_probing = 1;
|
|
} else {
|
|
/*
|
|
* No warning on writes to ROM BAR, CMD, LATENCY_TIMER,
|
|
* CACHE_LINE_SIZE, or PM registers.
|
|
*/
|
|
if ((reg != PCI_ROM_ADDRESS) && (reg != PCI_COMMAND_MASTER) &&
|
|
(reg != PCI_LATENCY_TIMER) &&
|
|
(reg != PCI_CACHE_LINE_SIZE) && (reg != 0x44))
|
|
printk(KERN_WARNING "OLPC PCI: Config write to devfn"
|
|
" %x reg %x value %x\n", devfn, reg, value);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static struct pci_raw_ops pci_olpc_conf = {
|
|
.read = pci_olpc_read,
|
|
.write = pci_olpc_write,
|
|
};
|
|
|
|
int __init pci_olpc_init(void)
|
|
{
|
|
if (!machine_is_olpc() || olpc_has_vsa())
|
|
return -ENODEV;
|
|
|
|
printk(KERN_INFO "PCI: Using configuration type OLPC\n");
|
|
raw_pci_ops = &pci_olpc_conf;
|
|
is_lx = is_geode_lx();
|
|
return 0;
|
|
}
|