Provide access to pcpu structures for SMP kernels.
The basic idea is to use a the same virtual address as a window onto distinct physical memory locations - one per processor. The physical address that you access through this mapping depends on which cpu you are currently executing on. We can now use the same virtual address on any processor to access its per-cpu area. The details are: - The virtual address for 'struct pcpu *pcpup' is obtained by stealing 2 pages worth of KVA in pmap_bootstrap(). - The mapping from the constant virtual address to a distinct physical page is done in cpu_pcpu_init() through a wired TLB entry. - A side-effect of this is that we reserve 2 pages worth of memory for the pcpu but in reality it needs much less than that. The unused memory is now used as the boot stack for the BSP and APs. Remove SMP-specific bits from locore.S. The plan is to use a separate mpboot.S for AP bootstrap. Discussed on: freebsd-mips Approved by: imp (mentor)
This commit is contained in:
neel
parent
248442cb37
commit
f0bf9d2db5
@ -309,8 +309,16 @@
|
||||
|
||||
/*
|
||||
* The first TLB entry that write random hits.
|
||||
* TLB entry 0 maps the kernel stack of the currently running thread
|
||||
* TLB entry 1 maps the pcpu area of processor (only for SMP builds)
|
||||
*/
|
||||
#define KSTACK_TLB_ENTRY 0
|
||||
#ifdef SMP
|
||||
#define PCPU_TLB_ENTRY 1
|
||||
#define VMWIRED_ENTRIES 2
|
||||
#else
|
||||
#define VMWIRED_ENTRIES 1
|
||||
#endif /* SMP */
|
||||
|
||||
/*
|
||||
* The number of process id entries.
|
||||
|
||||
@ -38,35 +38,15 @@
|
||||
struct pmap *pc_curpmap; /* pmap of curthread */ \
|
||||
u_int32_t pc_next_asid; /* next ASID to alloc */ \
|
||||
u_int32_t pc_asid_generation; /* current ASID generation */ \
|
||||
u_int pc_pending_ipis; /* the IPIs pending to this CPU */ \
|
||||
void *pc_boot_stack;
|
||||
u_int pc_pending_ipis; /* IPIs pending to this CPU */
|
||||
|
||||
#ifdef _KERNEL
|
||||
|
||||
#ifdef SMP
|
||||
static __inline struct pcpu*
|
||||
get_pcpup(void)
|
||||
{
|
||||
/*
|
||||
* FREEBSD_DEVELOPERS_FIXME
|
||||
* In multiprocessor case, store/retrieve the pcpu structure
|
||||
* address for current CPU in scratch register for fast access.
|
||||
*
|
||||
* In this routine, read the scratch register to retrieve the PCPU
|
||||
* structure for this CPU
|
||||
*/
|
||||
struct pcpu *ret;
|
||||
extern char pcpu_space[MAXCPU][PAGE_SIZE * 2];
|
||||
#define PCPU_ADDR(cpu) (struct pcpu *)(pcpu_space[(cpu)])
|
||||
|
||||
/* ret should contain the pointer to the PCPU structure for this CPU */
|
||||
return(ret);
|
||||
}
|
||||
|
||||
#define PCPUP ((struct pcpu *)get_pcpup())
|
||||
#else
|
||||
/* Uni processor systems */
|
||||
extern struct pcpu *pcpup;
|
||||
#define PCPUP pcpup
|
||||
#endif /* SMP */
|
||||
|
||||
#define PCPU_ADD(member, value) (PCPUP->pc_ ## member += (value))
|
||||
#define PCPU_GET(member) (PCPUP->pc_ ## member)
|
||||
|
||||
@ -82,7 +82,6 @@ ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
|
||||
ASSYM(PC_SEGBASE, offsetof(struct pcpu, pc_segbase));
|
||||
ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
|
||||
ASSYM(PC_FPCURTHREAD, offsetof(struct pcpu, pc_fpcurthread));
|
||||
ASSYM(PC_BOOT_STACK, offsetof(struct pcpu, pc_boot_stack));
|
||||
ASSYM(PC_CPUID, offsetof(struct pcpu, pc_cpuid));
|
||||
ASSYM(PC_CURPMAP, offsetof(struct pcpu, pc_curpmap));
|
||||
|
||||
|
||||
@ -77,14 +77,9 @@
|
||||
GLOBAL(fenvp)
|
||||
.space 4 # Assumes mips32? Is that OK?
|
||||
#endif
|
||||
GLOBAL(stackspace)
|
||||
.space NBPG /* Smaller than it should be since it's temp. */
|
||||
.align 8
|
||||
GLOBAL(topstack)
|
||||
|
||||
|
||||
.set noreorder
|
||||
|
||||
|
||||
.text
|
||||
|
||||
GLOBAL(btext)
|
||||
@ -133,6 +128,7 @@ VECTOR(_locore, unknown)
|
||||
or t2, t1
|
||||
mtc0 t2, COP_0_STATUS_REG
|
||||
COP0_SYNC
|
||||
|
||||
/* Make sure KSEG0 is cached */
|
||||
li t0, CFG_K0_CACHED
|
||||
mtc0 t0, MIPS_COP_0_CONFIG
|
||||
@ -157,13 +153,6 @@ VECTOR(_locore, unknown)
|
||||
sw t0, _C_LABEL(cpu_id)
|
||||
sw t1, _C_LABEL(fpu_id)
|
||||
|
||||
/*
|
||||
* Initialize stack and call machine startup.
|
||||
*/
|
||||
PTR_LA sp, _C_LABEL(topstack) - START_FRAME
|
||||
PTR_LA gp, _C_LABEL(_gp)
|
||||
sw zero, START_FRAME - 4(sp) # Zero out old ra for debugger
|
||||
|
||||
/*xxximp
|
||||
* now that we pass a0...a3 to the platform_init routine, do we need
|
||||
* to stash this stuff here?
|
||||
@ -174,58 +163,19 @@ VECTOR(_locore, unknown)
|
||||
#endif
|
||||
|
||||
/*
|
||||
* The following needs to be done differently for each platform and
|
||||
* there needs to be a good way to plug this in.
|
||||
* Initialize stack and call machine startup.
|
||||
*/
|
||||
#if defined(SMP) && defined(CPU_XLR)
|
||||
/*
|
||||
* Block all the slave CPUs
|
||||
*/
|
||||
/* XXX a0, a1, a2 shouldn't be used here */
|
||||
/*
|
||||
* Read the cpu id from the cp0 config register
|
||||
* cpuid[9:4], thrid[3: 0]
|
||||
*/
|
||||
mfc0 a0, COP_0_CONFIG, 7
|
||||
srl a1, a0, 4
|
||||
andi a1, a1, 0x3f
|
||||
andi a0, a0, 0xf
|
||||
PTR_LA sp, _C_LABEL(pcpu_space)
|
||||
addiu sp, (NBPG * 2) - START_FRAME
|
||||
|
||||
/* calculate linear cpuid */
|
||||
sll t0, a1, 2
|
||||
addu a2, t0, a0
|
||||
/* Initially, disable all hardware threads on each core except thread0 */
|
||||
li t1, VCPU_ID_0
|
||||
li t2, XLR_THREAD_ENABLE_IND
|
||||
mtcr t1, t2
|
||||
#endif
|
||||
sw zero, START_FRAME - 4(sp) # Zero out old ra for debugger
|
||||
sw zero, START_FRAME - 8(sp) # Zero out old fp for debugger
|
||||
|
||||
|
||||
#if defined(TARGET_OCTEON) /* Maybe this is mips32/64 generic? */
|
||||
.set push
|
||||
.set mips32r2
|
||||
rdhwr t0, $0
|
||||
.set pop
|
||||
#else
|
||||
move t0, zero
|
||||
#endif
|
||||
|
||||
/* Stage the secondary cpu start until later */
|
||||
bne t0, zero, start_secondary
|
||||
nop
|
||||
|
||||
#ifdef SMP
|
||||
PTR_LA t0, _C_LABEL(__pcpu)
|
||||
SET_CPU_PCPU(t0)
|
||||
/* If not master cpu, jump... */
|
||||
/*XXX this assumes the above #if 0'd code runs */
|
||||
bne a2, zero, start_secondary
|
||||
nop
|
||||
#endif
|
||||
PTR_LA gp, _C_LABEL(_gp)
|
||||
|
||||
/* Call the platform-specific startup code. */
|
||||
jal _C_LABEL(platform_start)
|
||||
sw zero, START_FRAME - 8(sp) # Zero out old fp for debugger
|
||||
nop
|
||||
|
||||
PTR_LA sp, _C_LABEL(thread0)
|
||||
lw a0, TD_PCB(sp)
|
||||
@ -238,25 +188,4 @@ VECTOR(_locore, unknown)
|
||||
|
||||
PANIC("Startup failed!")
|
||||
|
||||
#ifdef SMP
|
||||
start_secondary:
|
||||
move a0, a1
|
||||
2:
|
||||
addiu t0, PCPU_SIZE
|
||||
subu a1, 1
|
||||
bne a1, zero, 2b
|
||||
nop
|
||||
SET_CPU_PCPU(t0)
|
||||
smp_wait:
|
||||
lw sp, PC_BOOT_STACK(t0)
|
||||
beqz sp, smp_wait
|
||||
nop
|
||||
jal _C_LABEL(smp_init_secondary)
|
||||
nop
|
||||
#else
|
||||
start_secondary:
|
||||
b start_secondary
|
||||
nop
|
||||
#endif
|
||||
|
||||
VECTOR_END(_locore)
|
||||
|
||||
@ -115,12 +115,28 @@ int clocks_running = 0;
|
||||
|
||||
vm_offset_t kstack0;
|
||||
|
||||
/*
|
||||
* Each entry in the pcpu_space[] array is laid out in the following manner:
|
||||
* struct pcpu for cpu 'n' pcpu_space[n]
|
||||
* boot stack for cpu 'n' pcpu_space[n] + PAGE_SIZE * 2 - START_FRAME
|
||||
*
|
||||
* Note that the boot stack grows downwards and we assume that we never
|
||||
* use enough stack space to trample over the 'struct pcpu' that is at
|
||||
* the beginning of the array.
|
||||
*
|
||||
* The array is aligned on a (PAGE_SIZE * 2) boundary so that the 'struct pcpu'
|
||||
* is always in the even page frame of the wired TLB entry on SMP kernels.
|
||||
*
|
||||
* The array is in the .data section so that the stack does not get zeroed out
|
||||
* when the .bss section is zeroed.
|
||||
*/
|
||||
char pcpu_space[MAXCPU][PAGE_SIZE * 2] \
|
||||
__aligned(PAGE_SIZE * 2) __section(".data");
|
||||
|
||||
#ifdef SMP
|
||||
struct pcpu __pcpu[MAXCPU];
|
||||
char pcpu_boot_stack[KSTACK_PAGES * PAGE_SIZE * MAXCPU];
|
||||
struct pcpu *pcpup = 0; /* initialized in pmap_bootstrap() */
|
||||
#else
|
||||
struct pcpu pcpu;
|
||||
struct pcpu *pcpup = &pcpu;
|
||||
struct pcpu *pcpup = (struct pcpu *)pcpu_space;
|
||||
#endif
|
||||
|
||||
vm_offset_t phys_avail[PHYS_AVAIL_ENTRIES + 2];
|
||||
@ -269,11 +285,7 @@ void
|
||||
mips_pcpu0_init()
|
||||
{
|
||||
/* Initialize pcpu info of cpu-zero */
|
||||
#ifdef SMP
|
||||
pcpu_init(&__pcpu[0], 0, sizeof(struct pcpu));
|
||||
#else
|
||||
pcpu_init(pcpup, 0, sizeof(struct pcpu));
|
||||
#endif
|
||||
pcpu_init(PCPU_ADDR(0), 0, sizeof(struct pcpu));
|
||||
PCPU_SET(curthread, &thread0);
|
||||
}
|
||||
|
||||
@ -283,6 +295,10 @@ mips_pcpu0_init()
|
||||
void
|
||||
mips_proc0_init(void)
|
||||
{
|
||||
#ifdef SMP
|
||||
if (platform_processor_id() != 0)
|
||||
panic("BSP must be processor number 0");
|
||||
#endif
|
||||
proc_linkup0(&proc0, &thread0);
|
||||
|
||||
KASSERT((kstack0 & PAGE_MASK) == 0,
|
||||
@ -410,12 +426,27 @@ void
|
||||
cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
|
||||
{
|
||||
#ifdef SMP
|
||||
if (cpuid != 0)
|
||||
pcpu->pc_boot_stack = (void *)(pcpu_boot_stack + cpuid *
|
||||
(KSTACK_PAGES * PAGE_SIZE));
|
||||
vm_paddr_t pa;
|
||||
struct tlb tlb;
|
||||
int lobits;
|
||||
#endif
|
||||
|
||||
pcpu->pc_next_asid = 1;
|
||||
pcpu->pc_asid_generation = 1;
|
||||
|
||||
#ifdef SMP
|
||||
/*
|
||||
* Map the pcpu structure at the virtual address 'pcpup'.
|
||||
* We use a wired tlb index to do this one-time mapping.
|
||||
*/
|
||||
memset(&tlb, 0, sizeof(tlb));
|
||||
pa = vtophys(pcpu);
|
||||
lobits = PTE_RW | PTE_V | PTE_G | PTE_CACHE;
|
||||
tlb.tlb_hi = (vm_offset_t)pcpup;
|
||||
tlb.tlb_lo0 = mips_paddr_to_tlbpfn(pa) | lobits;
|
||||
tlb.tlb_lo1 = mips_paddr_to_tlbpfn(pa + PAGE_SIZE) | lobits;
|
||||
Mips_TLBWriteIndexed(PCPU_TLB_ENTRY, &tlb);
|
||||
#endif
|
||||
}
|
||||
|
||||
int
|
||||
|
||||
@ -355,6 +355,17 @@ pmap_bootstrap(void)
|
||||
virtual_avail = VM_MIN_KERNEL_ADDRESS + VM_KERNEL_ALLOC_OFFSET;
|
||||
virtual_end = VM_MAX_KERNEL_ADDRESS;
|
||||
|
||||
#ifdef SMP
|
||||
/*
|
||||
* Steal some virtual address space to map the pcpu area.
|
||||
*/
|
||||
virtual_avail = roundup2(virtual_avail, PAGE_SIZE * 2);
|
||||
pcpup = (struct pcpu *)virtual_avail;
|
||||
virtual_avail += PAGE_SIZE * 2;
|
||||
if (bootverbose)
|
||||
printf("pcpu is available at virtual address %p.\n", pcpup);
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Steal some virtual space that will not be in kernel_segmap. This
|
||||
* va memory space will be used to map in kernel pages that are
|
||||
@ -428,8 +439,8 @@ pmap_bootstrap(void)
|
||||
kernel_pmap->pm_segtab = kernel_segmap;
|
||||
kernel_pmap->pm_active = ~0;
|
||||
TAILQ_INIT(&kernel_pmap->pm_pvlist);
|
||||
kernel_pmap->pm_asid[PCPU_GET(cpuid)].asid = PMAP_ASID_RESERVED;
|
||||
kernel_pmap->pm_asid[PCPU_GET(cpuid)].gen = 0;
|
||||
kernel_pmap->pm_asid[0].asid = PMAP_ASID_RESERVED;
|
||||
kernel_pmap->pm_asid[0].gen = 0;
|
||||
pmap_max_asid = VMNUM_PIDS;
|
||||
MachSetPID(0);
|
||||
}
|
||||
|
||||
Loading…
Reference in New Issue
Block a user