diff options
Diffstat (limited to 'target/linux/ubicom32/files/arch/ubicom32/kernel/ubicom32_context_switch.S')
-rw-r--r-- | target/linux/ubicom32/files/arch/ubicom32/kernel/ubicom32_context_switch.S | 359 |
1 files changed, 359 insertions, 0 deletions
diff --git a/target/linux/ubicom32/files/arch/ubicom32/kernel/ubicom32_context_switch.S b/target/linux/ubicom32/files/arch/ubicom32/kernel/ubicom32_context_switch.S new file mode 100644 index 0000000000..08db4c057a --- /dev/null +++ b/target/linux/ubicom32/files/arch/ubicom32/kernel/ubicom32_context_switch.S @@ -0,0 +1,359 @@ +/* + * arch/ubicom32/kernel/ubicom32_context_switch.S + * Implements context switch and return functions. + * + * (C) Copyright 2009, Ubicom, Inc. + * + * This file is part of the Ubicom32 Linux Kernel Port. + * + * The Ubicom32 Linux Kernel Port is free software: you can redistribute + * it and/or modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation, either version 2 of the + * License, or (at your option) any later version. + * + * The Ubicom32 Linux Kernel Port is distributed in the hope that it + * will be useful, but WITHOUT ANY WARRANTY; without even the implied + * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See + * the GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with the Ubicom32 Linux Kernel Port. If not, + * see <http://www.gnu.org/licenses/>. + * + * Ubicom32 implementation derived from (with many thanks): + * arch/m68knommu + * arch/blackfin + * arch/parisc + */ +#include <linux/sys.h> +#include <linux/linkage.h> +#include <asm/asm-offsets.h> +#include <asm/ubicom32-common.h> +#include <asm/ip5000.h> +#include <asm/range-protect.h> + +/* + * begin_restore_context() + * Restore most of the context from sp (struct pt_reg *) + * + * This *can* be called without the global atomic lock. (because sp is + * not restored!) Only d15 and a3 are allowed to be used after this + * before calling complete_restore_context + */ +.macro begin_restore_context + move.4 d0, PT_D0(sp) + move.4 d1, PT_D1(sp) + move.4 d2, PT_D2(sp) + move.4 d3, PT_D3(sp) + move.4 d4, PT_D4(sp) + move.4 d5, PT_D5(sp) + move.4 d6, PT_D6(sp) + move.4 d7, PT_D7(sp) + move.4 d8, PT_D8(sp) + move.4 d9, PT_D9(sp) + move.4 d10, PT_D10(sp) + move.4 d11, PT_D11(sp) + move.4 d12, PT_D12(sp) + move.4 d13, PT_D13(sp) + move.4 d14, PT_D14(sp) +;; move.4 d15, PT_D15(sp) + move.4 a0, PT_A0(sp) + move.4 a1, PT_A1(sp) + move.4 a2, PT_A2(sp) +;; move.4 a3, PT_A3(sp) + move.4 a4, PT_A4(sp) + move.4 a5, PT_A5(sp) + move.4 a6, PT_A6(sp) + move.4 acc0_hi, PT_ACC0HI(sp) + move.4 acc0_lo, PT_ACC0LO(sp) + move.4 mac_rc16, PT_MAC_RC16(sp) + move.4 acc1_hi, PT_ACC1HI(sp) + move.4 acc1_lo, PT_ACC1LO(sp) + move.4 source3, PT_SOURCE3(sp) + move.4 int_mask0, PT_INT_MASK0(sp) + move.4 int_mask1, PT_INT_MASK1(sp) +.endm + +/* + * complete_restore_context() + * Completely restore the context from sp (struct pt_reg *) + * + * Note: Recovered PC and CSR are saved on the stack and are to be + * popped off before returning. + */ +.macro complete_restore_context + move.4 a3, sp + move.4 d15, PT_D15(sp) + move.4 sp, PT_SP(a3) ; Recover Stack pointer from save area + move.4 -4(sp)++, PT_PC(a3) ; Recover saved PC and save to stack + move.4 -4(sp)++, PT_CSR(a3) ; Recover saved csr and save to stack + move.4 a3, PT_A3(a3) +.endm + +/* + * old restore_context macro + */ +.macro restore_context + begin_restore_context + complete_restore_context +.endm + +/* + * ldsr_thread_enable_interrupts() + * An assembly version of the enable interrupts function. + * + * The stack is fair game but all registers MUST be preserved. + * + */ +.macro ldsr_thread_enable_interrupts + move.4 -4(sp)++, d3 ; Push d3 + move.4 -4(sp)++, a3 ; Push a3 + + /* + * Read the ROSR and obtain ~(1 << tid) + */ + lsr.4 d3, rosr, #0x2 ; Move the thread portion of ROSR into d3 + lsl.4 d3, #1, d3 ; perform a (1 << tid) + not.4 d3, d3 ; Negate the value of d3 == ~(1 << threadid) + + /* + * Get the value of the ldsr_soft_irq_mask + */ + moveai a3, #%hi(ldsr_soft_irq_mask) + move.4 a3, %lo(ldsr_soft_irq_mask)(a3) + + /* + * Now re-enable interrupts for this thread and then + * wakeup the LDSR. + */ + and.4 scratchpad1, scratchpad1, d3 + move.4 int_set0, a3 + + /* + * Restore the registers. + */ + move.4 a3, (sp)4++ + move.4 d3, (sp)4++ +.endm + +/* + * ret_from_interrupt_to_kernel() + * RFI function that is where do_IRQ() returns to if the thread was + * in kernel space. + */ + .section .text.ret_from_interrupt_to_kernel, "ax", @progbits + .global ret_from_interrupt_to_kernel +ret_from_interrupt_to_kernel: + begin_restore_context ; Restore the thread context + atomic_lock_acquire ; Enter critical section + complete_restore_context ; Restore the thread context + atomic_lock_release ; Leave critical section + ldsr_thread_enable_interrupts ; enable the threads interrupts + move.4 csr, (sp)4++ ; Restore csr from the stack + ret (sp)4++ + +/* + * ret_from_interrupt_to_user() + * RFI function that is where do_IRQ() returns to if the thread was + * in user space. + * + * TODO: Do we really need the critical section handling in this code? + * + */ + .section .text.ret_from_interrupt_to_user, "ax", @progbits + .global ret_from_interrupt_to_user +ret_from_interrupt_to_user: + ldsr_thread_enable_interrupts ; enable the threads interrupts + /* + * Set a1 to the thread info pointer, no need to save it as we are + * restoring userspace and will never return + */ + movei d0, #(~(ASM_THREAD_SIZE-1)) + and.4 a1, sp, d0 + + /* + * Test if the scheduler needs to be called. + */ + btst TI_FLAGS(a1), #ASM_TIF_NEED_RESCHED + jmpeq.t 2f + call a5, schedule ; Call the scheduler. I will come back here. + + /* + * See if we have pending signals and call do_signal + * if needed. + */ +2: + btst TI_FLAGS(a1), #ASM_TIF_SIGPENDING ; Any signals needed? + jmpeq.t 1f + + /* + * Now call do_signal() + */ + move.4 d0, #0 ; oldset pointer is NULL + move.4 d1, sp ; d1 is the regs pointer + call a5, do_signal ; Call do_signal() + + /* + * Back from do_signal(), re-enter critical section. + */ +1: + begin_restore_context ; Restore the thread context + atomic_lock_acquire ; Enter critical section + call a3, __complete_and_return_to_userspace ; jump to unprotected section + +/* + * restore_all_registers() + * + * restore_all_registers will be the alternate exit route for + * preempted processes that have called a signal handler + * and are returning back to user space. + */ + .section .text.restore_all_registers, "ax", @progbits + .global restore_all_registers +restore_all_registers: + begin_restore_context ; Restore the thread context + atomic_lock_acquire ; Enter critical section + call a3, __complete_and_return_to_userspace + +/* + * __complete_and_return_to_userspace + * + * restores the second half of the context and returns + * You must have the atomic lock when you call this function + */ + .section .kernel_unprotected, "ax", @progbits +__complete_and_return_to_userspace: + disable_kernel_ranges_for_current d15 ; disable kernel ranges + complete_restore_context ; restore previous context + atomic_lock_release ; Leave critical section + move.4 csr, (sp)4++ ; Restore csr from the stack + ret (sp)4++ + +/* + * ret_from_fork() + * Called on the child's return from fork system call. + */ + .section .text.ret_from_fork, "ax", @progbits + .global ret_from_fork +ret_from_fork: + ;;; d0 contains the arg for schedule_tail + ;;; the others we don't care about as they are in PT_REGS (sp) + call a5, schedule_tail + + atomic_lock_acquire ; Enter critical section + + move.4 a3, sp + move.4 d0, PT_D0(a3) ; Restore D0 + move.4 d1, PT_D1(a3) ; Restore D1 + move.4 d2, PT_D2(a3) ; Restore D2 + move.4 d3, PT_D3(a3) ; Restore D3 + move.4 d10, PT_D10(a3) ; Restore D10 + move.4 d11, PT_D11(a3) ; Restore D11 + move.4 d12, PT_D12(a3) ; Restore D12 + move.4 d13, PT_D13(a3) ; Restore D13 + move.4 a1, PT_A1(a3) ; Restore A1 + move.4 a2, PT_A2(a3) ; Restore A2 + move.4 a5, PT_A5(a3) ; Restore A5 + move.4 a6, PT_A6(a3) ; Restore A6 + ;; I think atomic_lock_acquire could be moved here.. + move.4 sp, PT_SP(a3) ; Restore sp + move.4 a4, PT_PC(a3) ; Restore pc in register a4 + move.4 PT_FRAME_TYPE(a3), #0 ; Clear frame_type to indicate it is invalid. + +#ifdef CONFIG_PROTECT_KERNEL + call a3, __ret_from_fork_bottom_half + .section .kernel_unprotected, "ax", @progbits +__ret_from_fork_bottom_half: + disable_kernel_ranges_for_current d15 +#endif + atomic_lock_release ; Leave critical section + calli a4, 0(a4) ; Return. + +/* + * __switch_to() + * + * Call with: + * void *__switch_to(struct task_struct *prev, struct thread_struct *prev_switch, + * struct thread_struct *next_switch) + */ + .section .text.__switch_to, "ax", @progbits + .global __switch_to +__switch_to: + + /* + * Set up register a3 to point to save area. + */ + movea a3, d1 ; a3 now holds prev_switch + move.4 (a3)4++, d10 + move.4 (a3)4++, d11 + move.4 (a3)4++, d12 + move.4 (a3)4++, d13 + move.4 (a3)4++, a1 + move.4 (a3)4++, a2 + move.4 (a3)4++, a5 + move.4 (a3)4++, a6 + move.4 (a3)4++, a7 + + /* + * Set up register a3 to point to restore area. + */ + movea a3, d2 ; a3 now holds next_switch + move.4 d10 , (a3)4++ + move.4 d11 , (a3)4++ + move.4 d12 , (a3)4++ + move.4 d13 , (a3)4++ + move.4 a1 , (a3)4++ + move.4 a2 , (a3)4++ + move.4 a5 , (a3)4++ + move.4 a6 , (a3)4++ + move.4 a7 , (a3)4++ + + /* + * Load the sw_ksp with the proper thread_info pointer. + */ + movei d15, #(~(ASM_THREAD_SIZE-1)) + and.4 a3, sp, d15 ; a3 now has the thread info pointer + moveai a4, #%hi(sw_ksp) + lea.1 a4, %lo(sw_ksp)(a4) ; a4 now has the base address of sw_ksp array + lsr.4 d15, ROSR, #2 ; Thread number - bit's 6 through 31 are zeroes anyway. + move.4 (a4, d15), a3 ; Load the thread info pointer into the hw_ksp array.. + + /* + * We are done with context switch. Time to return.. + */ + calli a5, 0(a5) + .size __switch_to, . - __switch_to + +/* + * ubicom32_emulate_insn() + * Emulates the instruction. + * + * Call with: + * unsigned int ubicom32_emulate_insn(int source1, int source2, int source3, int *save_acc, int *save_csr); + */ + .section .text.ubicom32_emulate_insn, "ax", @progbits + .global ubicom32_emulate_insn + .global trap_emulate +ubicom32_emulate_insn: + movea a3, d3 ; a3 holds save_acc pointer + movea a4, d4 ; a4 hods save_csr pointer + move.4 source3, d2 + move.4 acc0_lo, (a3) + move.4 acc0_hi, 4(a3) + move.4 acc1_lo, 8(a3) + move.4 acc1_hi, 12(a3) + move.4 mac_rc16, 16(a3) + move.4 CSR, (a4) + setcsr_flush 0 + +trap_emulate: + move.4 d0, d1 + setcsr_flush 0 + move.4 (a4), CSR ; Save csr + move.4 (a3), acc0_lo + move.4 4(a3), acc0_hi + move.4 8(a3), acc1_lo + move.4 12(a3), acc1_hi + move.4 16(a3), mac_rc16 + ret a5 + .size ubicom32_emulate_insn, . - ubicom32_emulate_insn |