Commit 98c5584c authored by David S. Miller's avatar David S. Miller Committed by David S. Miller

[SPARC64]: Add infrastructure for dynamic TSB sizing.

This also cleans up tsb_context_switch().  The assembler
routine is now __tsb_context_switch() and the former is
an inline function that picks out the bits from the mm_struct
and passes it into the assembler code as arguments.

setup_tsb_parms() computes the locked TLB entry to map the
TSB.  Later when we support using the physical address quad
load instructions of Cheetah+ and later, we'll simply use
the physical address for the TSB register value and set
the map virtual and PTE both to zero.
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 09f94287
......@@ -330,8 +330,7 @@ beyond_if:
current->mm->start_stack =
(unsigned long) create_aout32_tables((char __user *)bprm->p, bprm);
tsb_context_switch(__pa(current->mm->pgd),
current->mm->context.sparc64_tsb);
tsb_context_switch(mm);
start_thread32(regs, ex.a_entry, current->mm->start_stack);
if (current->ptrace & PT_PTRACED)
......
......@@ -441,8 +441,7 @@ void flush_thread(void)
mm = t->task->mm;
if (mm)
tsb_context_switch(__pa(mm->pgd),
mm->context.sparc64_tsb);
tsb_context_switch(mm);
set_thread_wsaved(0);
......
......@@ -130,48 +130,36 @@ winfix_trampoline:
* schedule() time.
*
* %o0: page table physical address
* %o1: TSB address
* %o1: TSB register value
* %o2: TSB virtual address
* %o3: TSB mapping locked PTE
*
* We have to run this whole thing with interrupts
* disabled so that the current cpu doesn't change
* due to preemption.
*/
.align 32
.globl tsb_context_switch
tsb_context_switch:
.globl __tsb_context_switch
__tsb_context_switch:
rdpr %pstate, %o5
wrpr %o5, PSTATE_IE, %pstate
ldub [%g6 + TI_CPU], %o3
sethi %hi(trap_block), %o4
sllx %o3, TRAP_BLOCK_SZ_SHIFT, %o3
or %o4, %lo(trap_block), %o4
add %o4, %o3, %o4
stx %o0, [%o4 + TRAP_PER_CPU_PGD_PADDR]
brgez %o1, 9f
nop
ldub [%g6 + TI_CPU], %g1
sethi %hi(trap_block), %g2
sllx %g1, TRAP_BLOCK_SZ_SHIFT, %g1
or %g2, %lo(trap_block), %g2
add %g2, %g1, %g2
stx %o0, [%g2 + TRAP_PER_CPU_PGD_PADDR]
/* Lock TSB into D-TLB. */
sethi %hi(PAGE_SIZE), %o3
and %o3, %o1, %o3
sethi %hi(TSBMAP_BASE), %o2
add %o2, %o3, %o2
/* XXX handle PAGE_SIZE != 8K correctly... */
mov TSB_REG, %g1
stxa %o2, [%g1] ASI_DMMU
stxa %o1, [%g1] ASI_DMMU
membar #Sync
stxa %o2, [%g1] ASI_IMMU
stxa %o1, [%g1] ASI_IMMU
membar #Sync
#define KERN_HIGHBITS ((_PAGE_VALID|_PAGE_SZBITS)^0xfffff80000000000)
#define KERN_LOWBITS (_PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_W | _PAGE_L)
sethi %uhi(KERN_HIGHBITS), %g2
or %g2, %ulo(KERN_HIGHBITS), %g2
sllx %g2, 32, %g2
or %g2, KERN_LOWBITS, %g2
#undef KERN_HIGHBITS
#undef KERN_LOWBITS
xor %o1, %g2, %o1
brz %o2, 9f
nop
/* We use entry 61 for this locked entry. This is the spitfire
* TLB entry number, and luckily cheetah masks the value with
......@@ -184,11 +172,10 @@ tsb_context_switch:
stxa %o2, [%g1] ASI_DMMU
membar #Sync
mov (61 << 3), %g1
stxa %o1, [%g1] ASI_DTLB_DATA_ACCESS
stxa %o3, [%g1] ASI_DTLB_DATA_ACCESS
membar #Sync
9:
wrpr %o5, %pstate
retl
mov %o2, %o0
nop
......@@ -9,13 +9,7 @@
#include <asm/tlbflush.h>
#include <asm/tlb.h>
#include <asm/mmu_context.h>
#define TSB_ENTRY_ALIGNMENT 16
struct tsb {
unsigned long tag;
unsigned long pte;
} __attribute__((aligned(TSB_ENTRY_ALIGNMENT)));
#include <asm/pgtable.h>
/* We use an 8K TSB for the whole kernel, this allows to
* handle about 4MB of modules and vmalloc mappings without
......@@ -27,10 +21,10 @@ struct tsb {
extern struct tsb swapper_tsb[KERNEL_TSB_NENTRIES];
static inline unsigned long tsb_hash(unsigned long vaddr)
static inline unsigned long tsb_hash(unsigned long vaddr, unsigned long nentries)
{
vaddr >>= PAGE_SHIFT;
return vaddr & (KERNEL_TSB_NENTRIES - 1);
return vaddr & (nentries - 1);
}
static inline int tag_compare(struct tsb *entry, unsigned long vaddr, unsigned long context)
......@@ -51,7 +45,8 @@ void flush_tsb_kernel_range(unsigned long start, unsigned long end)
unsigned long v;
for (v = start; v < end; v += PAGE_SIZE) {
struct tsb *ent = &swapper_tsb[tsb_hash(v)];
unsigned long hash = tsb_hash(v, KERNEL_TSB_NENTRIES);
struct tsb *ent = &swapper_tsb[hash];
if (tag_compare(ent, v, 0)) {
ent->tag = 0UL;
......@@ -63,8 +58,9 @@ void flush_tsb_kernel_range(unsigned long start, unsigned long end)
void flush_tsb_user(struct mmu_gather *mp)
{
struct mm_struct *mm = mp->mm;
struct tsb *tsb = (struct tsb *) mm->context.sparc64_tsb;
struct tsb *tsb = mm->context.tsb;
unsigned long ctx = ~0UL;
unsigned long nentries = mm->context.tsb_nentries;
int i;
if (CTX_VALID(mm->context))
......@@ -76,7 +72,7 @@ void flush_tsb_user(struct mmu_gather *mp)
v &= ~0x1UL;
ent = &tsb[tsb_hash(v)];
ent = &tsb[tsb_hash(v, nentries)];
if (tag_compare(ent, v, ctx)) {
ent->tag = 0UL;
membar_storeload_storestore();
......@@ -84,6 +80,83 @@ void flush_tsb_user(struct mmu_gather *mp)
}
}
static void setup_tsb_params(struct mm_struct *mm, unsigned long tsb_bytes)
{
unsigned long tsb_reg, base, tsb_paddr;
unsigned long page_sz, tte;
mm->context.tsb_nentries = tsb_bytes / sizeof(struct tsb);
base = TSBMAP_BASE;
tte = (_PAGE_VALID | _PAGE_L | _PAGE_CP |
_PAGE_CV | _PAGE_P | _PAGE_W);
tsb_paddr = __pa(mm->context.tsb);
/* Use the smallest page size that can map the whole TSB
* in one TLB entry.
*/
switch (tsb_bytes) {
case 8192 << 0:
tsb_reg = 0x0UL;
#ifdef DCACHE_ALIASING_POSSIBLE
base += (tsb_paddr & 8192);
#endif
tte |= _PAGE_SZ8K;
page_sz = 8192;
break;
case 8192 << 1:
tsb_reg = 0x1UL;
tte |= _PAGE_SZ64K;
page_sz = 64 * 1024;
break;
case 8192 << 2:
tsb_reg = 0x2UL;
tte |= _PAGE_SZ64K;
page_sz = 64 * 1024;
break;
case 8192 << 3:
tsb_reg = 0x3UL;
tte |= _PAGE_SZ64K;
page_sz = 64 * 1024;
break;
case 8192 << 4:
tsb_reg = 0x4UL;
tte |= _PAGE_SZ512K;
page_sz = 512 * 1024;
break;
case 8192 << 5:
tsb_reg = 0x5UL;
tte |= _PAGE_SZ512K;
page_sz = 512 * 1024;
break;
case 8192 << 6:
tsb_reg = 0x6UL;
tte |= _PAGE_SZ512K;
page_sz = 512 * 1024;
break;
case 8192 << 7:
tsb_reg = 0x7UL;
tte |= _PAGE_SZ4MB;
page_sz = 4 * 1024 * 1024;
break;
};
tsb_reg |= base;
tsb_reg |= (tsb_paddr & (page_sz - 1UL));
tte |= (tsb_paddr & ~(page_sz - 1UL));
mm->context.tsb_reg_val = tsb_reg;
mm->context.tsb_map_vaddr = base;
mm->context.tsb_map_pte = tte;
}
int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
{
unsigned long page = get_zeroed_page(GFP_KERNEL);
......@@ -92,14 +165,22 @@ int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
if (unlikely(!page))
return -ENOMEM;
mm->context.sparc64_tsb = (unsigned long *) page;
mm->context.tsb = (struct tsb *) page;
setup_tsb_params(mm, PAGE_SIZE);
return 0;
}
void destroy_context(struct mm_struct *mm)
{
free_page((unsigned long) mm->context.sparc64_tsb);
free_page((unsigned long) mm->context.tsb);
/* We can remove these later, but for now it's useful
* to catch any bogus post-destroy_context() references
* to the TSB.
*/
mm->context.tsb = NULL;
mm->context.tsb_reg_val = 0UL;
spin_lock(&ctx_alloc_lock);
......
......@@ -90,9 +90,20 @@
#ifndef __ASSEMBLY__
#define TSB_ENTRY_ALIGNMENT 16
struct tsb {
unsigned long tag;
unsigned long pte;
} __attribute__((aligned(TSB_ENTRY_ALIGNMENT)));
typedef struct {
unsigned long sparc64_ctx_val;
unsigned long *sparc64_tsb;
struct tsb *tsb;
unsigned long tsb_nentries;
unsigned long tsb_reg_val;
unsigned long tsb_map_vaddr;
unsigned long tsb_map_pte;
} mm_context_t;
#endif /* !__ASSEMBLY__ */
......
......@@ -22,7 +22,15 @@ extern void get_new_mmu_context(struct mm_struct *mm);
extern int init_new_context(struct task_struct *tsk, struct mm_struct *mm);
extern void destroy_context(struct mm_struct *mm);
extern unsigned long tsb_context_switch(unsigned long pgd_pa, unsigned long *tsb);
extern void __tsb_context_switch(unsigned long pgd_pa, unsigned long tsb_reg,
unsigned long tsb_vaddr, unsigned long tsb_pte);
static inline void tsb_context_switch(struct mm_struct *mm)
{
__tsb_context_switch(__pa(mm->pgd), mm->context.tsb_reg_val,
mm->context.tsb_map_vaddr,
mm->context.tsb_map_pte);
}
/* Set MMU context in the actual hardware. */
#define load_secondary_context(__mm) \
......@@ -52,8 +60,7 @@ static inline void switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, str
if (!ctx_valid || (old_mm != mm)) {
load_secondary_context(mm);
tsb_context_switch(__pa(mm->pgd),
mm->context.sparc64_tsb);
tsb_context_switch(mm);
}
/* Even if (mm == old_mm) we _must_ check
......@@ -91,7 +98,7 @@ static inline void activate_mm(struct mm_struct *active_mm, struct mm_struct *mm
load_secondary_context(mm);
__flush_tlb_mm(CTX_HWBITS(mm->context), SECONDARY_CONTEXT);
tsb_context_switch(__pa(mm->pgd), mm->context.sparc64_tsb);
tsb_context_switch(mm);
}
#endif /* !(__ASSEMBLY__) */
......
......@@ -19,7 +19,7 @@
* stxa %g5, [%g0] ASI_{D,I}TLB_DATA_IN
* retry
*
*
* Each 16-byte slot of the TSB is the 8-byte tag and then the 8-byte
* PTE. The TAG is of the same layout as the TLB TAG TARGET mmu
* register which is:
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment