davidlt/kernel-ark
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
bf164c790d
kernel-ark/include/asm-sparc64/spitfire.h
David S. Miller b445e26cbf [SPARC64]: Avoid membar instructions in delay slots. 
In particular, avoid membar instructions in the delay
slot of a jmpl instruction.

UltraSPARC-I, II, IIi, and IIe have a bug, documented in
the UltraSPARC-IIi User's Manual, Appendix K, Erratum 51

The long and short of it is that if the IMU unit misses
on a branch or jmpl, and there is a store buffer synchronizing
membar in the delay slot, the chip can stop fetching instructions.

If interrupts are enabled or some other trap is enabled, the
chip will unwedge itself, but performance will suffer.

We already had a workaround for this bug in a few spots, but
it's better to have the entire tree sanitized for this rule.

Signed-off-by: David S. Miller <davem@davemloft.net>
2005-06-27 15:42:04 -07:00

464 lines
12 KiB
C
Raw Blame History

/* $Id: spitfire.h,v 1.18 2001/11/29 16:42:10 kanoj Exp $
* spitfire.h: SpitFire/BlackBird/Cheetah inline MMU operations.
*
* Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
*/
#ifndef _SPARC64_SPITFIRE_H
#define _SPARC64_SPITFIRE_H
#include <asm/asi.h>
/* The following register addresses are accessible via ASI_DMMU
* and ASI_IMMU, that is there is a distinct and unique copy of
* each these registers for each TLB.
*/
#define TSB_TAG_TARGET 0x0000000000000000 /* All chips */
#define TLB_SFSR 0x0000000000000018 /* All chips */
#define TSB_REG 0x0000000000000028 /* All chips */
#define TLB_TAG_ACCESS 0x0000000000000030 /* All chips */
#define VIRT_WATCHPOINT 0x0000000000000038 /* All chips */
#define PHYS_WATCHPOINT 0x0000000000000040 /* All chips */
#define TSB_EXTENSION_P 0x0000000000000048 /* Ultra-III and later */
#define TSB_EXTENSION_S 0x0000000000000050 /* Ultra-III and later, D-TLB only */
#define TSB_EXTENSION_N 0x0000000000000058 /* Ultra-III and later */
#define TLB_TAG_ACCESS_EXT 0x0000000000000060 /* Ultra-III+ and later */
/* These registers only exist as one entity, and are accessed
* via ASI_DMMU only.
*/
#define PRIMARY_CONTEXT 0x0000000000000008
#define SECONDARY_CONTEXT 0x0000000000000010
#define DMMU_SFAR 0x0000000000000020
#define VIRT_WATCHPOINT 0x0000000000000038
#define PHYS_WATCHPOINT 0x0000000000000040
#define SPITFIRE_HIGHEST_LOCKED_TLBENT (64 - 1)
#define CHEETAH_HIGHEST_LOCKED_TLBENT (16 - 1)
#define L1DCACHE_SIZE 0x4000
#ifndef __ASSEMBLY__
enum ultra_tlb_layout {
spitfire = 0,
cheetah = 1,
cheetah_plus = 2,
};
extern enum ultra_tlb_layout tlb_type;
extern int cheetah_pcache_forced_on;
extern void cheetah_enable_pcache(void);
#define sparc64_highest_locked_tlbent() \
(tlb_type == spitfire ? \
SPITFIRE_HIGHEST_LOCKED_TLBENT : \
CHEETAH_HIGHEST_LOCKED_TLBENT)
static __inline__ unsigned long spitfire_get_isfsr(void)
{
unsigned long ret;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (ret)
: "r" (TLB_SFSR), "i" (ASI_IMMU));
return ret;
}
static __inline__ unsigned long spitfire_get_dsfsr(void)
{
unsigned long ret;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (ret)
: "r" (TLB_SFSR), "i" (ASI_DMMU));
return ret;
}
static __inline__ unsigned long spitfire_get_sfar(void)
{
unsigned long ret;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (ret)
: "r" (DMMU_SFAR), "i" (ASI_DMMU));
return ret;
}
static __inline__ void spitfire_put_isfsr(unsigned long sfsr)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* no outputs */
: "r" (sfsr), "r" (TLB_SFSR), "i" (ASI_IMMU));
}
static __inline__ void spitfire_put_dsfsr(unsigned long sfsr)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* no outputs */
: "r" (sfsr), "r" (TLB_SFSR), "i" (ASI_DMMU));
}
/* The data cache is write through, so this just invalidates the
* specified line.
*/
static __inline__ void spitfire_put_dcache_tag(unsigned long addr, unsigned long tag)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (tag), "r" (addr), "i" (ASI_DCACHE_TAG));
}
/* The instruction cache lines are flushed with this, but note that
* this does not flush the pipeline. It is possible for a line to
* get flushed but stale instructions to still be in the pipeline,
* a flush instruction (to any address) is sufficient to handle
* this issue after the line is invalidated.
*/
static __inline__ void spitfire_put_icache_tag(unsigned long addr, unsigned long tag)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (tag), "r" (addr), "i" (ASI_IC_TAG));
}
static __inline__ unsigned long spitfire_get_dtlb_data(int entry)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (data)
: "r" (entry << 3), "i" (ASI_DTLB_DATA_ACCESS));
/* Clear TTE diag bits. */
data &= ~0x0003fe0000000000UL;
return data;
}
static __inline__ unsigned long spitfire_get_dtlb_tag(int entry)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" (entry << 3), "i" (ASI_DTLB_TAG_READ));
return tag;
}
static __inline__ void spitfire_put_dtlb_data(int entry, unsigned long data)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data), "r" (entry << 3),
"i" (ASI_DTLB_DATA_ACCESS));
}
static __inline__ unsigned long spitfire_get_itlb_data(int entry)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (data)
: "r" (entry << 3), "i" (ASI_ITLB_DATA_ACCESS));
/* Clear TTE diag bits. */
data &= ~0x0003fe0000000000UL;
return data;
}
static __inline__ unsigned long spitfire_get_itlb_tag(int entry)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" (entry << 3), "i" (ASI_ITLB_TAG_READ));
return tag;
}
static __inline__ void spitfire_put_itlb_data(int entry, unsigned long data)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data), "r" (entry << 3),
"i" (ASI_ITLB_DATA_ACCESS));
}
/* Spitfire hardware assisted TLB flushes. */
/* Context level flushes. */
static __inline__ void spitfire_flush_dtlb_primary_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x40), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_primary_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x40), "i" (ASI_IMMU_DEMAP));
}
static __inline__ void spitfire_flush_dtlb_secondary_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x50), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_secondary_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x50), "i" (ASI_IMMU_DEMAP));
}
static __inline__ void spitfire_flush_dtlb_nucleus_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x60), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_nucleus_context(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x60), "i" (ASI_IMMU_DEMAP));
}
/* Page level flushes. */
static __inline__ void spitfire_flush_dtlb_primary_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_primary_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page), "i" (ASI_IMMU_DEMAP));
}
static __inline__ void spitfire_flush_dtlb_secondary_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page | 0x10), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_secondary_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page | 0x10), "i" (ASI_IMMU_DEMAP));
}
static __inline__ void spitfire_flush_dtlb_nucleus_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page | 0x20), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void spitfire_flush_itlb_nucleus_page(unsigned long page)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (page | 0x20), "i" (ASI_IMMU_DEMAP));
}
/* Cheetah has "all non-locked" tlb flushes. */
static __inline__ void cheetah_flush_dtlb_all(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x80), "i" (ASI_DMMU_DEMAP));
}
static __inline__ void cheetah_flush_itlb_all(void)
{
__asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
"membar #Sync"
: /* No outputs */
: "r" (0x80), "i" (ASI_IMMU_DEMAP));
}
/* Cheetah has a 4-tlb layout so direct access is a bit different.
* The first two TLBs are fully assosciative, hold 16 entries, and are
* used only for locked and >8K sized translations. One exists for
* data accesses and one for instruction accesses.
*
* The third TLB is for data accesses to 8K non-locked translations, is
* 2 way assosciative, and holds 512 entries. The fourth TLB is for
* instruction accesses to 8K non-locked translations, is 2 way
* assosciative, and holds 128 entries.
*
* Cheetah has some bug where bogus data can be returned from
* ASI_{D,I}TLB_DATA_ACCESS loads, doing the load twice fixes
* the problem for me. -DaveM
*/
static __inline__ unsigned long cheetah_get_ldtlb_data(int entry)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %%g0\n\t"
"ldxa [%1] %2, %0"
: "=r" (data)
: "r" ((0 << 16) | (entry << 3)),
"i" (ASI_DTLB_DATA_ACCESS));
return data;
}
static __inline__ unsigned long cheetah_get_litlb_data(int entry)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %%g0\n\t"
"ldxa [%1] %2, %0"
: "=r" (data)
: "r" ((0 << 16) | (entry << 3)),
"i" (ASI_ITLB_DATA_ACCESS));
return data;
}
static __inline__ unsigned long cheetah_get_ldtlb_tag(int entry)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" ((0 << 16) | (entry << 3)),
"i" (ASI_DTLB_TAG_READ));
return tag;
}
static __inline__ unsigned long cheetah_get_litlb_tag(int entry)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" ((0 << 16) | (entry << 3)),
"i" (ASI_ITLB_TAG_READ));
return tag;
}
static __inline__ void cheetah_put_ldtlb_data(int entry, unsigned long data)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data),
"r" ((0 << 16) | (entry << 3)),
"i" (ASI_DTLB_DATA_ACCESS));
}
static __inline__ void cheetah_put_litlb_data(int entry, unsigned long data)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data),
"r" ((0 << 16) | (entry << 3)),
"i" (ASI_ITLB_DATA_ACCESS));
}
static __inline__ unsigned long cheetah_get_dtlb_data(int entry, int tlb)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %%g0\n\t"
"ldxa [%1] %2, %0"
: "=r" (data)
: "r" ((tlb << 16) | (entry << 3)), "i" (ASI_DTLB_DATA_ACCESS));
return data;
}
static __inline__ unsigned long cheetah_get_dtlb_tag(int entry, int tlb)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" ((tlb << 16) | (entry << 3)), "i" (ASI_DTLB_TAG_READ));
return tag;
}
static __inline__ void cheetah_put_dtlb_data(int entry, unsigned long data, int tlb)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data),
"r" ((tlb << 16) | (entry << 3)),
"i" (ASI_DTLB_DATA_ACCESS));
}
static __inline__ unsigned long cheetah_get_itlb_data(int entry)
{
unsigned long data;
__asm__ __volatile__("ldxa [%1] %2, %%g0\n\t"
"ldxa [%1] %2, %0"
: "=r" (data)
: "r" ((2 << 16) | (entry << 3)),
"i" (ASI_ITLB_DATA_ACCESS));
return data;
}
static __inline__ unsigned long cheetah_get_itlb_tag(int entry)
{
unsigned long tag;
__asm__ __volatile__("ldxa [%1] %2, %0"
: "=r" (tag)
: "r" ((2 << 16) | (entry << 3)), "i" (ASI_ITLB_TAG_READ));
return tag;
}
static __inline__ void cheetah_put_itlb_data(int entry, unsigned long data)
{
__asm__ __volatile__("stxa %0, [%1] %2\n\t"
"membar #Sync"
: /* No outputs */
: "r" (data), "r" ((2 << 16) | (entry << 3)),
"i" (ASI_ITLB_DATA_ACCESS));
}
#endif /* !(__ASSEMBLY__) */
#endif /* !(_SPARC64_SPITFIRE_H) */
Reference in New Issue View Git Blame Copy Permalink