static void __cpuinit
iPTE_SW(u32 **p, struct uasm_reloc **r, unsigned int pte, unsigned int ptr,
unsigned int mode)
{
#ifdef CONFIG_64BIT_PHYS_ADDR
unsigned int hwmode = mode & (_PAGE_VALID | _PAGE_DIRTY);
#endif
uasm_i_ori(p, pte, pte, mode);
#ifdef CONFIG_SMP
# ifdef CONFIG_64BIT_PHYS_ADDR
if (cpu_has_64bits)
uasm_i_scd(p, pte, 0, ptr);
else
# endif
UASM_i_SC(p, pte, 0, ptr);
if (r10000_llsc_war())
uasm_il_beqzl(p, r, pte, label_smp_pgtable_change);
else
uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
# ifdef CONFIG_64BIT_PHYS_ADDR
if (!cpu_has_64bits) {
/* no uasm_i_nop needed */
uasm_i_ll(p, pte, sizeof(pte_t) / 2, ptr);
uasm_i_ori(p, pte, pte, hwmode);
uasm_i_sc(p, pte, sizeof(pte_t) / 2, ptr);
uasm_il_beqz(p, r, pte, label_smp_pgtable_change);
/* no uasm_i_nop needed */
uasm_i_lw(p, pte, 0, ptr);
} else
uasm_i_nop(p);
# else
uasm_i_nop(p);
# endif
#else
# ifdef CONFIG_64BIT_PHYS_ADDR
if (cpu_has_64bits)
uasm_i_sd(p, pte, 0, ptr);
else
# endif
UASM_i_SW(p, pte, 0, ptr);
# ifdef CONFIG_64BIT_PHYS_ADDR
if (!cpu_has_64bits) {
uasm_i_lw(p, pte, sizeof(pte_t) / 2, ptr);
uasm_i_ori(p, pte, pte, hwmode);
uasm_i_sw(p, pte, sizeof(pte_t) / 2, ptr);
uasm_i_lw(p, pte, 0, ptr);
}
# endif
#endif
}
/**
* kvm_mips_build_ret_to_host() - Assemble code to return to the host.
* @addr: Address to start writing code.
*
* Assemble the code to handle return from the guest exit handler
* (kvm_mips_handle_exit()) back to the host, i.e. to the caller of the vcpu_run
* function generated by kvm_mips_build_vcpu_run().
*
* Returns: Next address after end of written function.
*/
static void *kvm_mips_build_ret_to_host(void *addr)
{
u32 *p = addr;
unsigned int i;
/* EBASE is already pointing to Linux */
UASM_i_LW(&p, K1, offsetof(struct kvm_vcpu_arch, host_stack), K1);
UASM_i_ADDIU(&p, K1, K1, -(int)sizeof(struct pt_regs));
/*
* r2/v0 is the return code, shift it down by 2 (arithmetic)
* to recover the err code
*/
uasm_i_sra(&p, K0, V0, 2);
uasm_i_move(&p, V0, K0);
/* Load context saved on the host stack */
for (i = 16; i < 31; ++i) {
if (i == 24)
i = 28;
UASM_i_LW(&p, i, offsetof(struct pt_regs, regs[i]), K1);
}
/* Restore RDHWR access */
UASM_i_LA_mostly(&p, K0, (long)&hwrena);
uasm_i_lw(&p, K0, uasm_rel_lo((long)&hwrena), K0);
uasm_i_mtc0(&p, K0, C0_HWRENA);
/* Restore RA, which is the address we will return to */
UASM_i_LW(&p, RA, offsetof(struct pt_regs, regs[RA]), K1);
uasm_i_jr(&p, RA);
uasm_i_nop(&p);
return p;
}
static inline void build_copy_store_pref(u32 **buf, int off)
{
if (off & cache_line_mask())
return;
if (pref_bias_copy_store) {
uasm_i_pref(buf, pref_dst_mode, pref_bias_copy_store + off,
A0);
} else if (cache_line_size == (half_copy_loop_size << 1)) {
if (cpu_has_cache_cdex_s) {
uasm_i_cache(buf, Create_Dirty_Excl_SD, off, A0);
} else if (cpu_has_cache_cdex_p) {
if (R4600_V1_HIT_CACHEOP_WAR && cpu_is_r4600_v1_x()) {
uasm_i_nop(buf);
uasm_i_nop(buf);
uasm_i_nop(buf);
uasm_i_nop(buf);
}
if (R4600_V2_HIT_CACHEOP_WAR && cpu_is_r4600_v2_x())
uasm_i_lw(buf, ZERO, ZERO, AT);
uasm_i_cache(buf, Create_Dirty_Excl_D, off, A0);
}
}
}
build_get_pgde32(u32 **p, unsigned int tmp, unsigned int ptr)
{
long pgdc = (long)pgd_current;
/* 32 bit SMP has smp_processor_id() stored in CONTEXT. */
#ifdef CONFIG_SMP
#ifdef CONFIG_MIPS_MT_SMTC
/*
* SMTC uses TCBind value as "CPU" index
*/
uasm_i_mfc0(p, ptr, C0_TCBIND);
UASM_i_LA_mostly(p, tmp, pgdc);
uasm_i_srl(p, ptr, ptr, 19);
#else
/*
* smp_processor_id() << 3 is stored in CONTEXT.
*/
uasm_i_mfc0(p, ptr, C0_CONTEXT);
UASM_i_LA_mostly(p, tmp, pgdc);
uasm_i_srl(p, ptr, ptr, 23);
#endif
uasm_i_addu(p, ptr, tmp, ptr);
#else
UASM_i_LA_mostly(p, ptr, pgdc);
#endif
uasm_i_mfc0(p, tmp, C0_BADVADDR); /* get faulting address */
uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
uasm_i_srl(p, tmp, tmp, PGDIR_SHIFT); /* get pgd only bits */
uasm_i_sll(p, tmp, tmp, PGD_T_LOG2);
uasm_i_addu(p, ptr, ptr, tmp); /* add in pgd offset */
}
static void build_copy_load(u32 **buf, int reg, int off)
{
if (cpu_has_64bit_gp_regs) {
uasm_i_ld(buf, reg, off, A1);
} else {
uasm_i_lw(buf, reg, off, A1);
}
}
static void __cpuinit
build_r3000_tlbchange_handler_head(u32 **p, unsigned int pte,
unsigned int ptr)
{
long pgdc = (long)pgd_current;
uasm_i_mfc0(p, pte, C0_BADVADDR);
uasm_i_lui(p, ptr, uasm_rel_hi(pgdc)); /* cp0 delay */
uasm_i_lw(p, ptr, uasm_rel_lo(pgdc), ptr);
uasm_i_srl(p, pte, pte, 22); /* load delay */
uasm_i_sll(p, pte, pte, 2);
uasm_i_addu(p, ptr, ptr, pte);
uasm_i_mfc0(p, pte, C0_CONTEXT);
uasm_i_lw(p, ptr, 0, ptr); /* cp0 delay */
uasm_i_andi(p, pte, pte, 0xffc); /* load delay */
uasm_i_addu(p, ptr, ptr, pte);
uasm_i_lw(p, pte, 0, ptr);
uasm_i_tlbp(p); /* load delay */
}
static __init void build_update_entries(u32 **p, unsigned int tmp,
unsigned int ptep)
{
/*
* 64bit address support (36bit on a 32bit CPU) in a 32bit
* Kernel is a special case. Only a few CPUs use it.
*/
#ifdef CONFIG_64BIT_PHYS_ADDR
if (cpu_has_64bits) {
uasm_i_ld(p, tmp, 0, ptep); /* get even pte */
uasm_i_ld(p, ptep, sizeof(pte_t), ptep); /* get odd pte */
uasm_i_dsrl(p, tmp, tmp, 6); /* convert to entrylo0 */
uasm_i_mtc0(p, tmp, C0_ENTRYLO0); /* load it */
uasm_i_dsrl(p, ptep, ptep, 6); /* convert to entrylo1 */
uasm_i_mtc0(p, ptep, C0_ENTRYLO1); /* load it */
} else {
int pte_off_even = sizeof(pte_t) / 2;
int pte_off_odd = pte_off_even + sizeof(pte_t);
/* The pte entries are pre-shifted */
uasm_i_lw(p, tmp, pte_off_even, ptep); /* get even pte */
uasm_i_mtc0(p, tmp, C0_ENTRYLO0); /* load it */
uasm_i_lw(p, ptep, pte_off_odd, ptep); /* get odd pte */
uasm_i_mtc0(p, ptep, C0_ENTRYLO1); /* load it */
}
#else
UASM_i_LW(p, tmp, 0, ptep); /* get even pte */
UASM_i_LW(p, ptep, sizeof(pte_t), ptep); /* get odd pte */
if (r45k_bvahwbug())
build_tlb_probe_entry(p);
UASM_i_SRL(p, tmp, tmp, 6); /* convert to entrylo0 */
if (r4k_250MHZhwbug())
uasm_i_mtc0(p, 0, C0_ENTRYLO0);
uasm_i_mtc0(p, tmp, C0_ENTRYLO0); /* load it */
UASM_i_SRL(p, ptep, ptep, 6); /* convert to entrylo1 */
if (r45k_bvahwbug())
uasm_i_mfc0(p, tmp, C0_INDEX);
if (r4k_250MHZhwbug())
uasm_i_mtc0(p, 0, C0_ENTRYLO1);
uasm_i_mtc0(p, ptep, C0_ENTRYLO1); /* load it */
#endif
}
/*
* The R3000 TLB handler is simple.
*/
static void __init build_r3000_tlb_refill_handler(void)
{
long pgdc = (long)pgd_current;
u32 *p;
int i;
memset(tlb_handler, 0, sizeof(tlb_handler));
p = tlb_handler;
uasm_i_mfc0(&p, K0, C0_BADVADDR);
uasm_i_lui(&p, K1, uasm_rel_hi(pgdc)); /* cp0 delay */
uasm_i_lw(&p, K1, uasm_rel_lo(pgdc), K1);
uasm_i_srl(&p, K0, K0, 22); /* load delay */
uasm_i_sll(&p, K0, K0, 2);
uasm_i_addu(&p, K1, K1, K0);
uasm_i_mfc0(&p, K0, C0_CONTEXT);
uasm_i_lw(&p, K1, 0, K1); /* cp0 delay */
uasm_i_andi(&p, K0, K0, 0xffc); /* load delay */
uasm_i_addu(&p, K1, K1, K0);
uasm_i_lw(&p, K0, 0, K1);
uasm_i_nop(&p); /* load delay */
uasm_i_mtc0(&p, K0, C0_ENTRYLO0);
uasm_i_mfc0(&p, K1, C0_EPC); /* cp0 delay */
uasm_i_tlbwr(&p); /* cp0 delay */
uasm_i_jr(&p, K1);
uasm_i_rfe(&p); /* branch delay */
if (p > tlb_handler + 32)
panic("TLB refill handler space exceeded");
pr_debug("Wrote TLB refill handler (%u instructions).\n",
(unsigned int)(p - tlb_handler));
pr_debug("\t.set push\n");
pr_debug("\t.set noreorder\n");
for (i = 0; i < (p - tlb_handler); i++)
pr_debug("\t.word 0x%08x\n", tlb_handler[i]);
pr_debug("\t.set pop\n");
memcpy((void *)ebase, tlb_handler, 0x80);
}
/**
* kvm_mips_build_exit() - Assemble common guest exit handler.
* @addr: Address to start writing code.
*
* Assemble the generic guest exit handling code. This is called by the
* exception vectors (generated by kvm_mips_build_exception()), and calls
* kvm_mips_handle_exit(), then either resumes the guest or returns to the host
* depending on the return value.
*
* Returns: Next address after end of written function.
*/
void *kvm_mips_build_exit(void *addr)
{
u32 *p = addr;
unsigned int i;
struct uasm_label labels[3];
struct uasm_reloc relocs[3];
struct uasm_label *l = labels;
struct uasm_reloc *r = relocs;
memset(labels, 0, sizeof(labels));
memset(relocs, 0, sizeof(relocs));
/*
* Generic Guest exception handler. We end up here when the guest
* does something that causes a trap to kernel mode.
*
* Both k0/k1 registers will have already been saved (k0 into the vcpu
* structure, and k1 into the scratch_tmp register).
*
* The k1 register will already contain the kvm_vcpu_arch pointer.
*/
/* Start saving Guest context to VCPU */
for (i = 0; i < 32; ++i) {
/* Guest k0/k1 saved later */
if (i == K0 || i == K1)
continue;
UASM_i_SW(&p, i, offsetof(struct kvm_vcpu_arch, gprs[i]), K1);
}
#ifndef CONFIG_CPU_MIPSR6
/* We need to save hi/lo and restore them on the way out */
uasm_i_mfhi(&p, T0);
UASM_i_SW(&p, T0, offsetof(struct kvm_vcpu_arch, hi), K1);
uasm_i_mflo(&p, T0);
UASM_i_SW(&p, T0, offsetof(struct kvm_vcpu_arch, lo), K1);
#endif
/* Finally save guest k1 to VCPU */
uasm_i_ehb(&p);
UASM_i_MFC0(&p, T0, scratch_tmp[0], scratch_tmp[1]);
UASM_i_SW(&p, T0, offsetof(struct kvm_vcpu_arch, gprs[K1]), K1);
/* Now that context has been saved, we can use other registers */
/* Restore vcpu */
UASM_i_MFC0(&p, A1, scratch_vcpu[0], scratch_vcpu[1]);
uasm_i_move(&p, S1, A1);
/* Restore run (vcpu->run) */
UASM_i_LW(&p, A0, offsetof(struct kvm_vcpu, run), A1);
/* Save pointer to run in s0, will be saved by the compiler */
uasm_i_move(&p, S0, A0);
/*
* Save Host level EPC, BadVaddr and Cause to VCPU, useful to process
* the exception
*/
UASM_i_MFC0(&p, K0, C0_EPC);
UASM_i_SW(&p, K0, offsetof(struct kvm_vcpu_arch, pc), K1);
UASM_i_MFC0(&p, K0, C0_BADVADDR);
UASM_i_SW(&p, K0, offsetof(struct kvm_vcpu_arch, host_cp0_badvaddr),
K1);
uasm_i_mfc0(&p, K0, C0_CAUSE);
uasm_i_sw(&p, K0, offsetof(struct kvm_vcpu_arch, host_cp0_cause), K1);
/* Now restore the host state just enough to run the handlers */
/* Switch EBASE to the one used by Linux */
/* load up the host EBASE */
uasm_i_mfc0(&p, V0, C0_STATUS);
uasm_i_lui(&p, AT, ST0_BEV >> 16);
uasm_i_or(&p, K0, V0, AT);
uasm_i_mtc0(&p, K0, C0_STATUS);
uasm_i_ehb(&p);
UASM_i_LA_mostly(&p, K0, (long)&ebase);
UASM_i_LW(&p, K0, uasm_rel_lo((long)&ebase), K0);
build_set_exc_base(&p, K0);
if (raw_cpu_has_fpu) {
/*
* If FPU is enabled, save FCR31 and clear it so that later
* ctc1's don't trigger FPE for pending exceptions.
*/
uasm_i_lui(&p, AT, ST0_CU1 >> 16);
uasm_i_and(&p, V1, V0, AT);
uasm_il_beqz(&p, &r, V1, label_fpu_1);
uasm_i_nop(&p);
uasm_i_cfc1(&p, T0, 31);
uasm_i_sw(&p, T0, offsetof(struct kvm_vcpu_arch, fpu.fcr31),
K1);
uasm_i_ctc1(&p, ZERO, 31);
uasm_l_fpu_1(&l, p);
}
if (cpu_has_msa) {
/*
* If MSA is enabled, save MSACSR and clear it so that later
* instructions don't trigger MSAFPE for pending exceptions.
*/
uasm_i_mfc0(&p, T0, C0_CONFIG5);
uasm_i_ext(&p, T0, T0, 27, 1); /* MIPS_CONF5_MSAEN */
uasm_il_beqz(&p, &r, T0, label_msa_1);
uasm_i_nop(&p);
uasm_i_cfcmsa(&p, T0, MSA_CSR);
uasm_i_sw(&p, T0, offsetof(struct kvm_vcpu_arch, fpu.msacsr),
K1);
uasm_i_ctcmsa(&p, MSA_CSR, ZERO);
uasm_l_msa_1(&l, p);
}
/* Now that the new EBASE has been loaded, unset BEV and KSU_USER */
uasm_i_addiu(&p, AT, ZERO, ~(ST0_EXL | KSU_USER | ST0_IE));
uasm_i_and(&p, V0, V0, AT);
uasm_i_lui(&p, AT, ST0_CU0 >> 16);
uasm_i_or(&p, V0, V0, AT);
uasm_i_mtc0(&p, V0, C0_STATUS);
uasm_i_ehb(&p);
/* Load up host GP */
UASM_i_LW(&p, GP, offsetof(struct kvm_vcpu_arch, host_gp), K1);
/* Need a stack before we can jump to "C" */
UASM_i_LW(&p, SP, offsetof(struct kvm_vcpu_arch, host_stack), K1);
/* Saved host state */
UASM_i_ADDIU(&p, SP, SP, -(int)sizeof(struct pt_regs));
/*
* XXXKYMA do we need to load the host ASID, maybe not because the
* kernel entries are marked GLOBAL, need to verify
*/
/* Restore host scratch registers, as we'll have clobbered them */
kvm_mips_build_restore_scratch(&p, K0, SP);
/* Restore RDHWR access */
UASM_i_LA_mostly(&p, K0, (long)&hwrena);
uasm_i_lw(&p, K0, uasm_rel_lo((long)&hwrena), K0);
uasm_i_mtc0(&p, K0, C0_HWRENA);
/* Jump to handler */
/*
* XXXKYMA: not sure if this is safe, how large is the stack??
* Now jump to the kvm_mips_handle_exit() to see if we can deal
* with this in the kernel
*/
UASM_i_LA(&p, T9, (unsigned long)kvm_mips_handle_exit);
uasm_i_jalr(&p, RA, T9);
UASM_i_ADDIU(&p, SP, SP, -CALLFRAME_SIZ);
uasm_resolve_relocs(relocs, labels);
p = kvm_mips_build_ret_from_exit(p);
return p;
}
/**
* kvm_mips_build_enter_guest() - Assemble code to resume guest execution.
* @addr: Address to start writing code.
*
* Assemble the code to resume guest execution. This code is common between the
* initial entry into the guest from the host, and returning from the exit
* handler back to the guest.
*
* Returns: Next address after end of written function.
*/
static void *kvm_mips_build_enter_guest(void *addr)
{
u32 *p = addr;
unsigned int i;
struct uasm_label labels[2];
struct uasm_reloc relocs[2];
struct uasm_label *l = labels;
struct uasm_reloc *r = relocs;
memset(labels, 0, sizeof(labels));
memset(relocs, 0, sizeof(relocs));
/* Set Guest EPC */
UASM_i_LW(&p, T0, offsetof(struct kvm_vcpu_arch, pc), K1);
UASM_i_MTC0(&p, T0, C0_EPC);
/* Set the ASID for the Guest Kernel */
UASM_i_LW(&p, T0, offsetof(struct kvm_vcpu_arch, cop0), K1);
UASM_i_LW(&p, T0, offsetof(struct mips_coproc, reg[MIPS_CP0_STATUS][0]),
T0);
uasm_i_andi(&p, T0, T0, KSU_USER | ST0_ERL | ST0_EXL);
uasm_i_xori(&p, T0, T0, KSU_USER);
uasm_il_bnez(&p, &r, T0, label_kernel_asid);
UASM_i_ADDIU(&p, T1, K1,
offsetof(struct kvm_vcpu_arch, guest_kernel_asid));
/* else user */
UASM_i_ADDIU(&p, T1, K1,
offsetof(struct kvm_vcpu_arch, guest_user_asid));
uasm_l_kernel_asid(&l, p);
/* t1: contains the base of the ASID array, need to get the cpu id */
/* smp_processor_id */
uasm_i_lw(&p, T2, offsetof(struct thread_info, cpu), GP);
/* x4 */
uasm_i_sll(&p, T2, T2, 2);
UASM_i_ADDU(&p, T3, T1, T2);
uasm_i_lw(&p, K0, 0, T3);
#ifdef CONFIG_MIPS_ASID_BITS_VARIABLE
/* x sizeof(struct cpuinfo_mips)/4 */
uasm_i_addiu(&p, T3, ZERO, sizeof(struct cpuinfo_mips)/4);
uasm_i_mul(&p, T2, T2, T3);
UASM_i_LA_mostly(&p, AT, (long)&cpu_data[0].asid_mask);
UASM_i_ADDU(&p, AT, AT, T2);
UASM_i_LW(&p, T2, uasm_rel_lo((long)&cpu_data[0].asid_mask), AT);
uasm_i_and(&p, K0, K0, T2);
#else
uasm_i_andi(&p, K0, K0, MIPS_ENTRYHI_ASID);
#endif
uasm_i_mtc0(&p, K0, C0_ENTRYHI);
uasm_i_ehb(&p);
/* Disable RDHWR access */
uasm_i_mtc0(&p, ZERO, C0_HWRENA);
/* load the guest context from VCPU and return */
for (i = 1; i < 32; ++i) {
/* Guest k0/k1 loaded later */
if (i == K0 || i == K1)
continue;
UASM_i_LW(&p, i, offsetof(struct kvm_vcpu_arch, gprs[i]), K1);
}
#ifndef CONFIG_CPU_MIPSR6
/* Restore hi/lo */
UASM_i_LW(&p, K0, offsetof(struct kvm_vcpu_arch, hi), K1);
uasm_i_mthi(&p, K0);
UASM_i_LW(&p, K0, offsetof(struct kvm_vcpu_arch, lo), K1);
uasm_i_mtlo(&p, K0);
#endif
/* Restore the guest's k0/k1 registers */
UASM_i_LW(&p, K0, offsetof(struct kvm_vcpu_arch, gprs[K0]), K1);
UASM_i_LW(&p, K1, offsetof(struct kvm_vcpu_arch, gprs[K1]), K1);
/* Jump to guest */
uasm_i_eret(&p);
uasm_resolve_relocs(relocs, labels);
return p;
}
static void * __init cps_gen_entry_code(unsigned cpu, enum cps_pm_state state)
{
struct uasm_label *l = labels;
struct uasm_reloc *r = relocs;
u32 *buf, *p;
const unsigned r_online = a0;
const unsigned r_nc_count = a1;
const unsigned r_pcohctl = t7;
const unsigned max_instrs = 256;
unsigned cpc_cmd;
enum {
lbl_incready = 1,
lbl_poll_cont,
lbl_secondary_hang,
lbl_disable_coherence,
lbl_flush_fsb,
lbl_invicache,
lbl_flushdcache,
lbl_hang,
lbl_set_cont,
lbl_secondary_cont,
lbl_decready,
};
/* Allocate a buffer to hold the generated code */
p = buf = kcalloc(max_instrs, sizeof(u32), GFP_KERNEL);
if (!buf)
return NULL;
/* Clear labels & relocs ready for (re)use */
memset(labels, 0, sizeof(labels));
memset(relocs, 0, sizeof(relocs));
if (state == CPS_PM_POWER_GATED) {
/* Power gating relies upon CPS SMP */
if (!mips_cps_smp_in_use())
goto out_err;
/*
* Save CPU state. Note the non-standard calling convention
* with the return address placed in v0 to avoid clobbering
* the ra register before it is saved.
*/
UASM_i_LA(&p, t0, (long)mips_cps_pm_save);
uasm_i_jalr(&p, v0, t0);
uasm_i_nop(&p);
}
/*
* Load addresses of required CM & CPC registers. This is done early
* because they're needed in both the enable & disable coherence steps
* but in the coupled case the enable step will only run on one VPE.
*/
UASM_i_LA(&p, r_pcohctl, (long)_gcmp_base + GCMPCLCBOFS(COHCTL));
if (coupled_coherence) {
/* Increment ready_count */
uasm_i_sync(&p, stype_ordering);
uasm_build_label(&l, p, lbl_incready);
uasm_i_ll(&p, t1, 0, r_nc_count);
uasm_i_addiu(&p, t2, t1, 1);
uasm_i_sc(&p, t2, 0, r_nc_count);
uasm_il_beqz(&p, &r, t2, lbl_incready);
uasm_i_addiu(&p, t1, t1, 1);
/* Ordering barrier */
uasm_i_sync(&p, stype_ordering);
/*
* If this is the last VPE to become ready for non-coherence
* then it should branch below.
*/
uasm_il_beq(&p, &r, t1, r_online, lbl_disable_coherence);
uasm_i_nop(&p);
if (state < CPS_PM_POWER_GATED) {
/*
* Otherwise this is not the last VPE to become ready
* for non-coherence. It needs to wait until coherence
* has been disabled before proceeding, which it will do
* by polling for the top bit of ready_count being set.
*/
uasm_i_addiu(&p, t1, zero, -1);
uasm_build_label(&l, p, lbl_poll_cont);
uasm_i_lw(&p, t0, 0, r_nc_count);
uasm_il_bltz(&p, &r, t0, lbl_secondary_cont);
uasm_i_ehb(&p);
uasm_i_yield(&p, zero, t1);
uasm_il_b(&p, &r, lbl_poll_cont);
uasm_i_nop(&p);
} else {
/*
* The core will lose power & this VPE will not continue
* so it can simply halt here.
*/
uasm_i_addiu(&p, t0, zero, TCHALT_H);
uasm_i_mtc0(&p, t0, 2, 4);
uasm_build_label(&l, p, lbl_secondary_hang);
uasm_il_b(&p, &r, lbl_secondary_hang);
uasm_i_nop(&p);
}
}
/*
* This is the point of no return - this VPE will now proceed to
* disable coherence. At this point we *must* be sure that no other
* VPE within the core will interfere with the L1 dcache.
*/
uasm_build_label(&l, p, lbl_disable_coherence);
/* Invalidate the L1 icache */
cps_gen_cache_routine(&p, &l, &r, &cpu_data[cpu].icache,
Index_Invalidate_I, lbl_invicache);
/* Writeback & invalidate the L1 dcache */
cps_gen_cache_routine(&p, &l, &r, &cpu_data[cpu].dcache,
Index_Writeback_Inv_D, lbl_flushdcache);
/* Completion barrier */
uasm_i_sync(&p, stype_memory);
uasm_i_ehb(&p);
/*
* Disable all but self interventions. The load from COHCTL is defined
* by the interAptiv & proAptiv SUMs as ensuring that the operation
* resulting from the preceeding store is complete.
*/
uasm_i_addiu(&p, t0, zero, 1 << cpu_data[cpu].core);
uasm_i_sw(&p, t0, 0, r_pcohctl);
uasm_i_lw(&p, t0, 0, r_pcohctl);
/* Sync to ensure previous interventions are complete */
uasm_i_sync(&p, stype_intervention);
uasm_i_ehb(&p);
/* Disable coherence */
uasm_i_sw(&p, zero, 0, r_pcohctl);
uasm_i_lw(&p, t0, 0, r_pcohctl);
if (state >= CPS_PM_CLOCK_GATED) {
/* TODO: determine whether required based on CPC version */
cps_gen_flush_fsb(&p, &l, &r, &cpu_data[cpu].dcache,
lbl_flush_fsb);
/* Determine the CPC command to issue */
switch (state) {
case CPS_PM_CLOCK_GATED:
cpc_cmd = CPC_Cx_CMD_CLOCKOFF;
break;
case CPS_PM_POWER_GATED:
cpc_cmd = CPC_Cx_CMD_PWRDOWN;
break;
default:
BUG();
goto out_err;
}
/* Issue the CPC command */
UASM_i_LA(&p, t0, (long)addr_cpc_cl_cmd());
uasm_i_addiu(&p, t1, zero, cpc_cmd);
uasm_i_sw(&p, t1, 0, t0);
if (state == CPS_PM_POWER_GATED) {
/* If anything goes wrong just hang */
uasm_build_label(&l, p, lbl_hang);
uasm_il_b(&p, &r, lbl_hang);
uasm_i_nop(&p);
/*
* There's no point generating more code, the core is
* powered down & if powered back up will run from the
* reset vector not from here.
*/
goto gen_done;
}
/* Completion barrier */
uasm_i_sync(&p, stype_memory);
uasm_i_ehb(&p);
}
if (state == CPS_PM_NC_WAIT) {
/*
* At this point it is safe for all VPEs to proceed with
* execution. This VPE will set the top bit of ready_count
* to indicate to the other VPEs that they may continue.
*/
if (coupled_coherence)
cps_gen_set_top_bit(&p, &l, &r, r_nc_count,
lbl_set_cont);
/*
* VPEs which did not disable coherence will continue
* executing, after coherence has been disabled, from this
* point.
*/
uasm_build_label(&l, p, lbl_secondary_cont);
/* Now perform our wait */
uasm_i_wait(&p, 0);
}
/*
* Re-enable coherence. Note that for CPS_PM_NC_WAIT all coupled VPEs
* will run this. The first will actually re-enable coherence & the
* rest will just be performing a rather unusual nop.
*/
uasm_i_addiu(&p, t0, zero, GCMP_CCB_COHCTL_DOMAIN_MSK);
uasm_i_sw(&p, t0, 0, r_pcohctl);
uasm_i_lw(&p, t0, 0, r_pcohctl);
/* Completion barrier */
uasm_i_sync(&p, stype_memory);
uasm_i_ehb(&p);
if (coupled_coherence && (state == CPS_PM_NC_WAIT)) {
/* Decrement ready_count */
uasm_build_label(&l, p, lbl_decready);
uasm_i_sync(&p, stype_ordering);
uasm_i_ll(&p, t1, 0, r_nc_count);
uasm_i_addiu(&p, t2, t1, -1);
uasm_i_sc(&p, t2, 0, r_nc_count);
uasm_il_beqz(&p, &r, t2, lbl_decready);
uasm_i_andi(&p, v0, t1, (1 << fls(smp_num_siblings)) - 1);
/* Ordering barrier */
uasm_i_sync(&p, stype_ordering);
}
if (coupled_coherence && (state == CPS_PM_CLOCK_GATED)) {
/*
* At this point it is safe for all VPEs to proceed with
* execution. This VPE will set the top bit of ready_count
* to indicate to the other VPEs that they may continue.
*/
cps_gen_set_top_bit(&p, &l, &r, r_nc_count, lbl_set_cont);
/*
* This core will be reliant upon another core sending a
* power-up command to the CPC in order to resume operation.
* Thus an arbitrary VPE can't trigger the core leaving the
* idle state and the one that disables coherence might as well
* be the one to re-enable it. The rest will continue from here
* after that has been done.
*/
uasm_build_label(&l, p, lbl_secondary_cont);
/* Ordering barrier */
uasm_i_sync(&p, stype_ordering);
}
/* The core is coherent, time to return to C code */
uasm_i_jr(&p, ra);
uasm_i_nop(&p);
gen_done:
/* Ensure the code didn't exceed the resources allocated for it */
BUG_ON((p - buf) > max_instrs);
BUG_ON((l - labels) > ARRAY_SIZE(labels));
BUG_ON((r - relocs) > ARRAY_SIZE(relocs));
/* Patch branch offsets */
uasm_resolve_relocs(relocs, labels);
/* Flush the icache */
local_flush_icache_range((unsigned long)buf, (unsigned long)p);
return buf;
out_err:
kfree(buf);
return NULL;
}
static void __init cps_gen_flush_fsb(u32 **pp, struct uasm_label **pl,
struct uasm_reloc **pr,
const struct cache_desc *dcache,
int lbl)
{
unsigned i, fsb_size = 8;
unsigned num_loads = (fsb_size * 3) / 2;
unsigned line_stride = 2;
/*
* Ensure that the fill/store buffer (FSB) is not holding the results
* of a prefetch, since if it is then the CPC sequencer may become
* stuck in the D3 (ClrBus) state whilst entering a low power state.
*/
/* TODO: this is interAptiv-specific, generalise it */
/* Preserve perf counter 1 setup */
uasm_i_mfc0(pp, t2, 25, 2); /* PerfCtl1 */
uasm_i_mfc0(pp, t3, 25, 3); /* PerfCnt1 */
/* Setup perf counter 1 to count FSB full pipeline stalls */
uasm_i_addiu(pp, t0, zero, 0x66f);
uasm_i_mtc0(pp, t0, 25, 2); /* PerfCtl1 */
uasm_i_ehb(pp);
uasm_i_mtc0(pp, zero, 25, 3); /* PerfCnt1 */
uasm_i_ehb(pp);
/* Base address for loads */
UASM_i_LA(pp, t0, (long)CKSEG0);
/* Start of clear loop */
uasm_build_label(pl, *pp, lbl);
/* Perform some loads to fill the FSB */
for (i = 0; i < num_loads; i++)
uasm_i_lw(pp, zero, i * dcache->linesz * line_stride, t0);
/*
* Invalidate the new D-cache entries so that the cache will need
* refilling (via the FSB) if the loop is executed again.
*/
for (i = 0; i < num_loads; i++) {
uasm_i_cache(pp, Hit_Invalidate_D,
i * dcache->linesz * line_stride, t0);
uasm_i_cache(pp, Hit_Writeback_Inv_SD,
i * dcache->linesz * line_stride, t0);
}
/* Completion barrier */
uasm_i_sync(pp, stype_memory);
uasm_i_ehb(pp);
/* Check whether the pipeline stalled due to the FSB being full */
uasm_i_mfc0(pp, t1, 25, 3); /* PerfCnt1 */
/* Loop if it didn't */
uasm_il_beqz(pp, pr, t1, lbl);
uasm_i_nop(pp);
/* Restore perf counter 1. The count may well now be wrong... */
uasm_i_mtc0(pp, t2, 25, 2); /* PerfCtl1 */
uasm_i_ehb(pp);
uasm_i_mtc0(pp, t3, 25, 3); /* PerfCnt1 */
uasm_i_ehb(pp);
}
static int __init cps_gen_flush_fsb(u32 **pp, struct uasm_label **pl,
struct uasm_reloc **pr,
const struct cpuinfo_mips *cpu_info,
int lbl)
{
unsigned i, fsb_size = 8;
unsigned num_loads = (fsb_size * 3) / 2;
unsigned line_stride = 2;
unsigned line_size = cpu_info->dcache.linesz;
unsigned perf_counter, perf_event;
unsigned revision = cpu_info->processor_id & PRID_REV_MASK;
/*
* Determine whether this CPU requires an FSB flush, and if so which
* performance counter/event reflect stalls due to a full FSB.
*/
switch (__get_cpu_type(cpu_info->cputype)) {
case CPU_INTERAPTIV:
perf_counter = 1;
perf_event = 51;
break;
case CPU_PROAPTIV:
/* Newer proAptiv cores don't require this workaround */
if (revision >= PRID_REV_ENCODE_332(1, 1, 0))
return 0;
/* On older ones it's unavailable */
return -1;
/* CPUs which do not require the workaround */
case CPU_P5600:
case CPU_I6400:
return 0;
default:
WARN_ONCE(1, "pm-cps: FSB flush unsupported for this CPU\n");
return -1;
}
/*
* Ensure that the fill/store buffer (FSB) is not holding the results
* of a prefetch, since if it is then the CPC sequencer may become
* stuck in the D3 (ClrBus) state whilst entering a low power state.
*/
/* Preserve perf counter setup */
uasm_i_mfc0(pp, t2, 25, (perf_counter * 2) + 0); /* PerfCtlN */
uasm_i_mfc0(pp, t3, 25, (perf_counter * 2) + 1); /* PerfCntN */
/* Setup perf counter to count FSB full pipeline stalls */
uasm_i_addiu(pp, t0, zero, (perf_event << 5) | 0xf);
uasm_i_mtc0(pp, t0, 25, (perf_counter * 2) + 0); /* PerfCtlN */
uasm_i_ehb(pp);
uasm_i_mtc0(pp, zero, 25, (perf_counter * 2) + 1); /* PerfCntN */
uasm_i_ehb(pp);
/* Base address for loads */
UASM_i_LA(pp, t0, (long)CKSEG0);
/* Start of clear loop */
uasm_build_label(pl, *pp, lbl);
/* Perform some loads to fill the FSB */
for (i = 0; i < num_loads; i++)
uasm_i_lw(pp, zero, i * line_size * line_stride, t0);
/*
* Invalidate the new D-cache entries so that the cache will need
* refilling (via the FSB) if the loop is executed again.
*/
for (i = 0; i < num_loads; i++) {
uasm_i_cache(pp, Hit_Invalidate_D,
i * line_size * line_stride, t0);
uasm_i_cache(pp, Hit_Writeback_Inv_SD,
i * line_size * line_stride, t0);
}
/* Completion barrier */
uasm_i_sync(pp, stype_memory);
uasm_i_ehb(pp);
/* Check whether the pipeline stalled due to the FSB being full */
uasm_i_mfc0(pp, t1, 25, (perf_counter * 2) + 1); /* PerfCntN */
/* Loop if it didn't */
uasm_il_beqz(pp, pr, t1, lbl);
uasm_i_nop(pp);
/* Restore perf counter 1. The count may well now be wrong... */
uasm_i_mtc0(pp, t2, 25, (perf_counter * 2) + 0); /* PerfCtlN */
uasm_i_ehb(pp);
uasm_i_mtc0(pp, t3, 25, (perf_counter * 2) + 1); /* PerfCntN */
uasm_i_ehb(pp);
return 0;
}
