freebsd-nq/sys/arm/mv/mv_machdep.c
Rafal Jaworowski 71f00bea66 Auto-size kernel page tables allocation on Marvell systems.
This allows mini dumps to fully work for these platforms.

Obtained from:	Juniper Networks, Semihalf
2008-11-06 16:25:12 +00:00

642 lines
18 KiB
C

/*-
* Copyright (c) 1994-1998 Mark Brinicombe.
* Copyright (c) 1994 Brini.
* All rights reserved.
*
* This code is derived from software written for Brini by Mark Brinicombe
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. All advertising materials mentioning features or use of this software
* must display the following acknowledgement:
* This product includes software developed by Brini.
* 4. The name of the company nor the name of the author may be used to
* endorse or promote products derived from this software without specific
* prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY BRINI ``AS IS'' AND ANY EXPRESS OR IMPLIED
* WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL BRINI OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
* INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* from: FreeBSD: //depot/projects/arm/src/sys/arm/at91/kb920x_machdep.c, rev 45
*/
#include "opt_msgbuf.h"
#include "opt_ddb.h"
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#define _ARM32_BUS_DMA_PRIVATE
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/signalvar.h>
#include <sys/imgact.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/linker.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/pcpu.h>
#include <sys/proc.h>
#include <sys/ptrace.h>
#include <sys/cons.h>
#include <sys/bio.h>
#include <sys/bus.h>
#include <sys/buf.h>
#include <sys/exec.h>
#include <sys/kdb.h>
#include <sys/msgbuf.h>
#include <machine/reg.h>
#include <machine/cpu.h>
#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pager.h>
#include <vm/vm_map.h>
#include <vm/vnode_pager.h>
#include <machine/pte.h>
#include <machine/pmap.h>
#include <machine/vmparam.h>
#include <machine/pcb.h>
#include <machine/undefined.h>
#include <machine/machdep.h>
#include <machine/metadata.h>
#include <machine/armreg.h>
#include <machine/bus.h>
#include <sys/reboot.h>
#include <machine/bootinfo.h>
#include <arm/mv/mvvar.h> /* XXX eventually this should be eliminated */
#ifdef DEBUG
#define debugf(fmt, args...) printf(fmt, ##args)
#else
#define debugf(fmt, args...)
#endif
/*
* This is the number of L2 page tables required for covering max
* (hypothetical) memsize of 4GB and all kernel mappings (vectors, msgbuf,
* stacks etc.), uprounded to be divisible by 4.
*/
#define KERNEL_PT_MAX 78
/* Define various stack sizes in pages */
#define IRQ_STACK_SIZE 1
#define ABT_STACK_SIZE 1
#define UND_STACK_SIZE 1
/* Maximum number of memory regions */
#define MEM_REGIONS 8
extern unsigned char kernbase[];
extern unsigned char _etext[];
extern unsigned char _edata[];
extern unsigned char __bss_start[];
extern unsigned char _end[];
extern u_int data_abort_handler_address;
extern u_int prefetch_abort_handler_address;
extern u_int undefined_handler_address;
extern const struct pmap_devmap *pmap_devmap_bootstrap_table;
extern vm_offset_t pmap_bootstrap_lastaddr;
struct pv_addr kernel_pt_table[KERNEL_PT_MAX];
extern int *end;
struct pcpu __pcpu;
struct pcpu *pcpup = &__pcpu;
/* Physical and virtual addresses for some global pages */
vm_paddr_t phys_avail[10];
vm_paddr_t dump_avail[4];
vm_offset_t physical_pages;
struct pv_addr systempage;
struct pv_addr msgbufpv;
struct pv_addr irqstack;
struct pv_addr undstack;
struct pv_addr abtstack;
struct pv_addr kernelstack;
static struct trapframe proc0_tf;
struct mem_region {
vm_offset_t mr_start;
vm_size_t mr_size;
};
static struct mem_region availmem_regions[MEM_REGIONS];
static int availmem_regions_sz;
struct bootinfo *bootinfo;
static void print_kenv(void);
static void print_kernel_section_addr(void);
static void print_bootinfo(void);
static void physmap_init(int);
static char *
kenv_next(char *cp)
{
if (cp != NULL) {
while (*cp != 0)
cp++;
cp++;
if (*cp == 0)
cp = NULL;
}
return (cp);
}
static void
print_kenv(void)
{
int len;
char *cp;
debugf("loader passed (static) kenv:\n");
if (kern_envp == NULL) {
debugf(" no env, null ptr\n");
return;
}
debugf(" kern_envp = 0x%08x\n", (uint32_t)kern_envp);
len = 0;
for (cp = kern_envp; cp != NULL; cp = kenv_next(cp))
debugf(" %x %s\n", (uint32_t)cp, cp);
}
static void
print_bootinfo(void)
{
struct bi_mem_region *mr;
struct bi_eth_addr *eth;
int i, j;
debugf("bootinfo:\n");
if (bootinfo == NULL) {
debugf(" no bootinfo, null ptr\n");
return;
}
debugf(" version = 0x%08x\n", bootinfo->bi_version);
debugf(" ccsrbar = 0x%08x\n", bootinfo->bi_bar_base);
debugf(" cpu_clk = 0x%08x\n", bootinfo->bi_cpu_clk);
debugf(" bus_clk = 0x%08x\n", bootinfo->bi_bus_clk);
debugf(" mem regions:\n");
mr = (struct bi_mem_region *)bootinfo->bi_data;
for (i = 0; i < bootinfo->bi_mem_reg_no; i++, mr++)
debugf(" #%d, base = 0x%08x, size = 0x%08x\n", i,
mr->mem_base, mr->mem_size);
debugf(" eth addresses:\n");
eth = (struct bi_eth_addr *)mr;
for (i = 0; i < bootinfo->bi_eth_addr_no; i++, eth++) {
debugf(" #%d, addr = ", i);
for (j = 0; j < 6; j++)
debugf("%02x ", eth->mac_addr[j]);
debugf("\n");
}
}
static void
print_kernel_section_addr(void)
{
debugf("kernel image addresses:\n");
debugf(" kernbase = 0x%08x\n", (uint32_t)kernbase);
debugf(" _etext (sdata) = 0x%08x\n", (uint32_t)_etext);
debugf(" _edata = 0x%08x\n", (uint32_t)_edata);
debugf(" __bss_start = 0x%08x\n", (uint32_t)__bss_start);
debugf(" _end = 0x%08x\n", (uint32_t)_end);
}
struct bi_mem_region *
bootinfo_mr(void)
{
return ((struct bi_mem_region *)bootinfo->bi_data);
}
static void
physmap_init(int hardcoded)
{
int i, j, cnt;
vm_offset_t phys_kernelend, kernload;
uint32_t s, e, sz;
struct mem_region *mp, *mp1;
phys_kernelend = KERNPHYSADDR + (virtual_avail - KERNVIRTADDR);
kernload = KERNPHYSADDR;
/*
* Use hardcoded physical addresses if we don't use memory regions
* from metadata.
*/
if (hardcoded) {
phys_avail[0] = 0;
phys_avail[1] = kernload;
phys_avail[2] = phys_kernelend;
phys_avail[3] = PHYSMEM_SIZE;
phys_avail[4] = 0;
phys_avail[5] = 0;
return;
}
/*
* Remove kernel physical address range from avail
* regions list. Page align all regions.
* Non-page aligned memory isn't very interesting to us.
* Also, sort the entries for ascending addresses.
*/
sz = 0;
cnt = availmem_regions_sz;
debugf("processing avail regions:\n");
for (mp = availmem_regions; mp->mr_size; mp++) {
s = mp->mr_start;
e = mp->mr_start + mp->mr_size;
debugf(" %08x-%08x -> ", s, e);
/* Check whether this region holds all of the kernel. */
if (s < kernload && e > phys_kernelend) {
availmem_regions[cnt].mr_start = phys_kernelend;
availmem_regions[cnt++].mr_size = e - phys_kernelend;
e = kernload;
}
/* Look whether this regions starts within the kernel. */
if (s >= kernload && s < phys_kernelend) {
if (e <= phys_kernelend)
goto empty;
s = phys_kernelend;
}
/* Now look whether this region ends within the kernel. */
if (e > kernload && e <= phys_kernelend) {
if (s >= kernload) {
goto empty;
}
e = kernload;
}
/* Now page align the start and size of the region. */
s = round_page(s);
e = trunc_page(e);
if (e < s)
e = s;
sz = e - s;
debugf("%08x-%08x = %x\n", s, e, sz);
/* Check whether some memory is left here. */
if (sz == 0) {
empty:
printf("skipping\n");
bcopy(mp + 1, mp,
(cnt - (mp - availmem_regions)) * sizeof(*mp));
cnt--;
mp--;
continue;
}
/* Do an insertion sort. */
for (mp1 = availmem_regions; mp1 < mp; mp1++)
if (s < mp1->mr_start)
break;
if (mp1 < mp) {
bcopy(mp1, mp1 + 1, (char *)mp - (char *)mp1);
mp1->mr_start = s;
mp1->mr_size = sz;
} else {
mp->mr_start = s;
mp->mr_size = sz;
}
}
availmem_regions_sz = cnt;
/* Fill in phys_avail table, based on availmem_regions */
debugf("fill in phys_avail:\n");
for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
availmem_regions[i].mr_start,
availmem_regions[i].mr_start + availmem_regions[i].mr_size,
availmem_regions[i].mr_size);
phys_avail[j] = availmem_regions[i].mr_start;
phys_avail[j + 1] = availmem_regions[i].mr_start +
availmem_regions[i].mr_size;
}
phys_avail[j] = 0;
phys_avail[j + 1] = 0;
}
void *
initarm(void *mdp, void *unused __unused)
{
struct pv_addr kernel_l1pt;
vm_offset_t freemempos, l2_start, lastaddr;
uint32_t memsize, l2size;
struct bi_mem_region *mr;
void *kmdp;
u_int l1pagetable;
int i = 0, j = 0;
kmdp = NULL;
lastaddr = 0;
memsize = 0;
set_cpufuncs();
/*
* Mask metadata pointer: it is supposed to be on page boundary. If
* the first argument (mdp) doesn't point to a valid address the
* bootloader must have passed us something else than the metadata
* ptr... In this case we want to fall back to some built-in settings.
*/
mdp = (void *)((uint32_t)mdp & ~PAGE_MASK);
/* Parse metadata and fetch parameters */
if (mdp != NULL) {
preload_metadata = mdp;
kmdp = preload_search_by_type("elf kernel");
if (kmdp != NULL) {
bootinfo = (struct bootinfo *)preload_search_info(kmdp,
MODINFO_METADATA|MODINFOMD_BOOTINFO);
boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int);
kern_envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *);
lastaddr = MD_FETCH(kmdp, MODINFOMD_KERNEND, vm_offset_t);
}
/* Initialize memory regions table */
mr = bootinfo_mr();
for (i = 0; i < bootinfo->bi_mem_reg_no; i++, mr++) {
if (i == MEM_REGIONS)
break;
availmem_regions[i].mr_start = mr->mem_base;
availmem_regions[i].mr_size = mr->mem_size;
memsize += mr->mem_size;
}
availmem_regions_sz = i;
} else {
/* Fall back to hardcoded boothowto flags and metadata. */
boothowto = RB_VERBOSE | RB_SINGLE;
lastaddr = fake_preload_metadata();
/*
* Assume a single memory region of size specified in board
* configuration file.
*/
memsize = PHYSMEM_SIZE;
}
/*
* If memsize is invalid, we can neither proceed nor panic (too
* early for console output).
*/
if (memsize == 0)
while (1);
/* Platform-specific initialisation */
if (platform_pmap_init() != 0)
return (NULL);
pcpu_init(pcpup, 0, sizeof(struct pcpu));
PCPU_SET(curthread, &thread0);
/* Calculate number of L2 tables needed for mapping vm_page_array */
l2size = (memsize / PAGE_SIZE) * sizeof(struct vm_page);
l2size = (l2size >> L1_S_SHIFT) + 1;
/*
* Add one table for end of kernel map, one for stacks, msgbuf and
* L1 and L2 tables map and one for vectors map.
*/
l2size += 3;
/* Make it divisible by 4 */
l2size = (l2size + 3) & ~3;
#define KERNEL_TEXT_BASE (KERNBASE)
freemempos = (lastaddr + PAGE_MASK) & ~PAGE_MASK;
/* Define a macro to simplify memory allocation */
#define valloc_pages(var, np) \
alloc_pages((var).pv_va, (np)); \
(var).pv_pa = (var).pv_va + (KERNPHYSADDR - KERNVIRTADDR);
#define alloc_pages(var, np) \
(var) = freemempos; \
freemempos += (np * PAGE_SIZE); \
memset((char *)(var), 0, ((np) * PAGE_SIZE));
while (((freemempos - L1_TABLE_SIZE) & (L1_TABLE_SIZE - 1)) != 0)
freemempos += PAGE_SIZE;
valloc_pages(kernel_l1pt, L1_TABLE_SIZE / PAGE_SIZE);
for (i = 0; i < l2size; ++i) {
if (!(i % (PAGE_SIZE / L2_TABLE_SIZE_REAL))) {
valloc_pages(kernel_pt_table[i],
L2_TABLE_SIZE / PAGE_SIZE);
j = i;
} else {
kernel_pt_table[i].pv_va = kernel_pt_table[j].pv_va +
L2_TABLE_SIZE_REAL * (i - j);
kernel_pt_table[i].pv_pa =
kernel_pt_table[i].pv_va - KERNVIRTADDR +
KERNPHYSADDR;
}
}
/*
* Allocate a page for the system page mapped to 0x00000000
* or 0xffff0000. This page will just contain the system vectors
* and can be shared by all processes.
*/
valloc_pages(systempage, 1);
/* Allocate stacks for all modes */
valloc_pages(irqstack, IRQ_STACK_SIZE);
valloc_pages(abtstack, ABT_STACK_SIZE);
valloc_pages(undstack, UND_STACK_SIZE);
valloc_pages(kernelstack, KSTACK_PAGES);
valloc_pages(msgbufpv, round_page(MSGBUF_SIZE) / PAGE_SIZE);
/*
* Now we start construction of the L1 page table
* We start by mapping the L2 page tables into the L1.
* This means that we can replace L1 mappings later on if necessary
*/
l1pagetable = kernel_l1pt.pv_va;
/*
* Try to map as much as possible of kernel text and data using
* 1MB section mapping and for the rest of initial kernel address
* space use L2 coarse tables.
*
* Link L2 tables for mapping remainder of kernel (modulo 1MB)
* and kernel structures
*/
l2_start = lastaddr & ~(L1_S_OFFSET);
for (i = 0 ; i < l2size - 1; i++)
pmap_link_l2pt(l1pagetable, l2_start + i * L1_S_SIZE,
&kernel_pt_table[i]);
pmap_curmaxkvaddr = l2_start + (l2size - 1) * L1_S_SIZE;
/* Map kernel code and data */
pmap_map_chunk(l1pagetable, KERNVIRTADDR, KERNPHYSADDR,
(((uint32_t)(lastaddr) - KERNVIRTADDR) + PAGE_MASK) & ~PAGE_MASK,
VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
/* Map L1 directory and allocated L2 page tables */
pmap_map_chunk(l1pagetable, kernel_l1pt.pv_va, kernel_l1pt.pv_pa,
L1_TABLE_SIZE, VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
pmap_map_chunk(l1pagetable, kernel_pt_table[0].pv_va,
kernel_pt_table[0].pv_pa,
L2_TABLE_SIZE_REAL * l2size,
VM_PROT_READ|VM_PROT_WRITE, PTE_PAGETABLE);
/* Map allocated stacks and msgbuf */
pmap_map_chunk(l1pagetable, irqstack.pv_va, irqstack.pv_pa,
freemempos - irqstack.pv_va,
VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
/* Link and map the vector page */
pmap_link_l2pt(l1pagetable, ARM_VECTORS_HIGH,
&kernel_pt_table[l2size - 1]);
pmap_map_entry(l1pagetable, ARM_VECTORS_HIGH, systempage.pv_pa,
VM_PROT_READ|VM_PROT_WRITE, PTE_CACHE);
pmap_devmap_bootstrap(l1pagetable, pmap_devmap_bootstrap_table);
cpu_domains((DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2)) |
DOMAIN_CLIENT);
setttb(kernel_l1pt.pv_pa);
cpu_tlb_flushID();
cpu_domains(DOMAIN_CLIENT << (PMAP_DOMAIN_KERNEL * 2));
cninit();
physmem = memsize / PAGE_SIZE;
debugf("initarm: console initialized\n");
debugf(" arg1 mdp = 0x%08x\n", (uint32_t)mdp);
debugf(" boothowto = 0x%08x\n", boothowto);
print_bootinfo();
print_kernel_section_addr();
print_kenv();
/*
* Re-initialise decode windows
*/
if (soc_decode_win() != 0)
printf("WARNING: could not re-initialise decode windows! "
"Running with existing settings...\n");
/*
* Pages were allocated during the secondary bootstrap for the
* stacks for different CPU modes.
* We must now set the r13 registers in the different CPU modes to
* point to these stacks.
* Since the ARM stacks use STMFD etc. we must set r13 to the top end
* of the stack memory.
*/
cpu_control(CPU_CONTROL_MMU_ENABLE, CPU_CONTROL_MMU_ENABLE);
set_stackptr(PSR_IRQ32_MODE,
irqstack.pv_va + IRQ_STACK_SIZE * PAGE_SIZE);
set_stackptr(PSR_ABT32_MODE,
abtstack.pv_va + ABT_STACK_SIZE * PAGE_SIZE);
set_stackptr(PSR_UND32_MODE,
undstack.pv_va + UND_STACK_SIZE * PAGE_SIZE);
/*
* We must now clean the cache again....
* Cleaning may be done by reading new data to displace any
* dirty data in the cache. This will have happened in setttb()
* but since we are boot strapping the addresses used for the read
* may have just been remapped and thus the cache could be out
* of sync. A re-clean after the switch will cure this.
* After booting there are no gross reloations of the kernel thus
* this problem will not occur after initarm().
*/
cpu_idcache_wbinv_all();
/* Set stack for exception handlers */
data_abort_handler_address = (u_int)data_abort_handler;
prefetch_abort_handler_address = (u_int)prefetch_abort_handler;
undefined_handler_address = (u_int)undefinedinstruction_bounce;
undefined_init();
proc_linkup0(&proc0, &thread0);
thread0.td_kstack = kernelstack.pv_va;
thread0.td_kstack_pages = KSTACK_PAGES;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
thread0.td_pcb->pcb_flags = 0;
thread0.td_frame = &proc0_tf;
pcpup->pc_curpcb = thread0.td_pcb;
arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
dump_avail[0] = 0;
dump_avail[1] = memsize;
dump_avail[2] = 0;
dump_avail[3] = 0;
pmap_bootstrap(freemempos, pmap_bootstrap_lastaddr, &kernel_l1pt);
msgbufp = (void *)msgbufpv.pv_va;
msgbufinit(msgbufp, MSGBUF_SIZE);
mutex_init();
/*
* Prepare map of physical memory regions available to vm subsystem.
* If metadata pointer doesn't point to a valid address, use hardcoded
* values.
*/
physmap_init((mdp != NULL) ? 0 : 1);
/* Do basic tuning, hz etc */
init_param1();
init_param2(physmem);
kdb_init();
return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
sizeof(struct pcb)));
}
struct arm32_dma_range *
bus_dma_get_range(void)
{
return (NULL);
}
int
bus_dma_get_range_nb(void)
{
return (0);
}