/**
* 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
pg_addiu(u32 **buf, unsigned int reg1, unsigned int reg2, unsigned int off)
{
if (cpu_has_64bit_gp_regs && DADDI_WAR && r4k_daddiu_bug()) {
if (off > 0x7fff) {
uasm_i_lui(buf, T9, uasm_rel_hi(off));
uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off));
} else
uasm_i_addiu(buf, T9, ZERO, off);
uasm_i_daddu(buf, reg1, reg2, T9);
} else {
if (off > 0x7fff) {
uasm_i_lui(buf, T9, uasm_rel_hi(off));
uasm_i_addiu(buf, T9, T9, uasm_rel_lo(off));
UASM_i_ADDU(buf, reg1, reg2, T9);
} else
UASM_i_ADDIU(buf, reg1, reg2, off);
}
}
static void __init cps_gen_cache_routine(u32 **pp, struct uasm_label **pl,
struct uasm_reloc **pr,
const struct cache_desc *cache,
unsigned op, int lbl)
{
unsigned cache_size = cache->ways << cache->waybit;
unsigned i;
const unsigned unroll_lines = 32;
/* If the cache isn't present this function has it easy */
if (cache->flags & MIPS_CACHE_NOT_PRESENT)
return;
/* Load base address */
UASM_i_LA(pp, t0, (long)CKSEG0);
/* Calculate end address */
if (cache_size < 0x8000)
uasm_i_addiu(pp, t1, t0, cache_size);
else
UASM_i_LA(pp, t1, (long)(CKSEG0 + cache_size));
/* Start of cache op loop */
uasm_build_label(pl, *pp, lbl);
/* Generate the cache ops */
for (i = 0; i < unroll_lines; i++) {
if (cpu_has_mips_r6) {
uasm_i_cache(pp, op, 0, t0);
uasm_i_addiu(pp, t0, t0, cache->linesz);
} else {
uasm_i_cache(pp, op, i * cache->linesz, t0);
}
}
if (!cpu_has_mips_r6)
/* Update the base address */
uasm_i_addiu(pp, t0, t0, unroll_lines * cache->linesz);
/* Loop if we haven't reached the end address yet */
uasm_il_bne(pp, pr, t0, t1, lbl);
uasm_i_nop(pp);
}
void UASM_i_LA(u32 **buf, unsigned int rs, long addr)
{
UASM_i_LA_mostly(buf, rs, addr);
if (uasm_rel_lo(addr)) {
if (!uasm_in_compat_space_p(addr))
uasm_i_daddiu(buf, rs, rs,
uasm_rel_lo(addr));
else
uasm_i_addiu(buf, rs, rs,
uasm_rel_lo(addr));
}
}
/**
* 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;
}
/**
* kvm_mips_build_vcpu_run() - Assemble function to start running a guest VCPU.
* @addr: Address to start writing code.
*
* Assemble the start of the vcpu_run function to run a guest VCPU. The function
* conforms to the following prototype:
*
* int vcpu_run(struct kvm_run *run, struct kvm_vcpu *vcpu);
*
* The exit from the guest and return to the caller is handled by the code
* generated by kvm_mips_build_ret_to_host().
*
* Returns: Next address after end of written function.
*/
void *kvm_mips_build_vcpu_run(void *addr)
{
u32 *p = addr;
unsigned int i;
/*
* A0: run
* A1: vcpu
*/
/* k0/k1 not being used in host kernel context */
UASM_i_ADDIU(&p, K1, SP, -(int)sizeof(struct pt_regs));
for (i = 16; i < 32; ++i) {
if (i == 24)
i = 28;
UASM_i_SW(&p, i, offsetof(struct pt_regs, regs[i]), K1);
}
/* Save host status */
uasm_i_mfc0(&p, V0, C0_STATUS);
UASM_i_SW(&p, V0, offsetof(struct pt_regs, cp0_status), K1);
/* Save scratch registers, will be used to store pointer to vcpu etc */
kvm_mips_build_save_scratch(&p, V1, K1);
/* VCPU scratch register has pointer to vcpu */
UASM_i_MTC0(&p, A1, scratch_vcpu[0], scratch_vcpu[1]);
/* Offset into vcpu->arch */
UASM_i_ADDIU(&p, K1, A1, offsetof(struct kvm_vcpu, arch));
/*
* Save the host stack to VCPU, used for exception processing
* when we exit from the Guest
*/
UASM_i_SW(&p, SP, offsetof(struct kvm_vcpu_arch, host_stack), K1);
/* Save the kernel gp as well */
UASM_i_SW(&p, GP, offsetof(struct kvm_vcpu_arch, host_gp), K1);
/*
* Setup status register for running the guest in UM, interrupts
* are disabled
*/
UASM_i_LA(&p, K0, ST0_EXL | KSU_USER | ST0_BEV | ST0_KX_IF_64);
uasm_i_mtc0(&p, K0, C0_STATUS);
uasm_i_ehb(&p);
/* load up the new EBASE */
UASM_i_LW(&p, K0, offsetof(struct kvm_vcpu_arch, guest_ebase), K1);
build_set_exc_base(&p, K0);
/*
* Now that the new EBASE has been loaded, unset BEV, set
* interrupt mask as it was but make sure that timer interrupts
* are enabled
*/
uasm_i_addiu(&p, K0, ZERO, ST0_EXL | KSU_USER | ST0_IE | ST0_KX_IF_64);
uasm_i_andi(&p, V0, V0, ST0_IM);
uasm_i_or(&p, K0, K0, V0);
uasm_i_mtc0(&p, K0, C0_STATUS);
uasm_i_ehb(&p);
p = kvm_mips_build_enter_guest(p);
return p;
}
static void __init install_trampoline(u32 *tramp, unsigned int sigreturn)
{
uasm_i_addiu(&tramp, 2, 0, sigreturn);
uasm_i_syscall(&tramp, 0);
}
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);
}
/*
* BVADDR is the faulting address, PTR is scratch.
* PTR will hold the pgd for vmalloc.
*/
static __init void
build_get_pgd_vmalloc64(u32 **p, struct uasm_label **l, struct uasm_reloc **r,
unsigned int bvaddr, unsigned int ptr)
{
long swpd = (long)swapper_pg_dir;
#ifdef MODULE_START
long modd = (long)module_pg_dir;
uasm_l_module_alloc(l, *p);
/*
* Assumption:
* VMALLOC_START >= 0xc000000000000000UL
* MODULE_START >= 0xe000000000000000UL
*/
UASM_i_SLL(p, ptr, bvaddr, 2);
uasm_il_bgez(p, r, ptr, label_vmalloc);
if (uasm_in_compat_space_p(MODULE_START) &&
!uasm_rel_lo(MODULE_START)) {
uasm_i_lui(p, ptr, uasm_rel_hi(MODULE_START)); /* delay slot */
} else {
/* unlikely configuration */
uasm_i_nop(p); /* delay slot */
UASM_i_LA(p, ptr, MODULE_START);
}
uasm_i_dsubu(p, bvaddr, bvaddr, ptr);
if (uasm_in_compat_space_p(modd) && !uasm_rel_lo(modd)) {
uasm_il_b(p, r, label_vmalloc_done);
uasm_i_lui(p, ptr, uasm_rel_hi(modd));
} else {
UASM_i_LA_mostly(p, ptr, modd);
uasm_il_b(p, r, label_vmalloc_done);
if (uasm_in_compat_space_p(modd))
uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(modd));
else
uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(modd));
}
uasm_l_vmalloc(l, *p);
if (uasm_in_compat_space_p(MODULE_START) &&
!uasm_rel_lo(MODULE_START) &&
MODULE_START << 32 == VMALLOC_START)
uasm_i_dsll32(p, ptr, ptr, 0); /* typical case */
else
UASM_i_LA(p, ptr, VMALLOC_START);
#else
uasm_l_vmalloc(l, *p);
UASM_i_LA(p, ptr, VMALLOC_START);
#endif
uasm_i_dsubu(p, bvaddr, bvaddr, ptr);
if (uasm_in_compat_space_p(swpd) && !uasm_rel_lo(swpd)) {
uasm_il_b(p, r, label_vmalloc_done);
uasm_i_lui(p, ptr, uasm_rel_hi(swpd));
} else {
UASM_i_LA_mostly(p, ptr, swpd);
uasm_il_b(p, r, label_vmalloc_done);
if (uasm_in_compat_space_p(swpd))
uasm_i_addiu(p, ptr, ptr, uasm_rel_lo(swpd));
else
uasm_i_daddiu(p, ptr, ptr, uasm_rel_lo(swpd));
}
}
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;
}