From c51b4488cd5bff08ed5690a8f303ff7f0894da2a Mon Sep 17 00:00:00 2001 From: Graf Yang Date: Wed, 7 Jan 2009 23:14:39 +0800 Subject: [PATCH] Blackfin arch: SMP supporting patchset: BF561 related code Blackfin dual core BF561 processor can support SMP like features. https://docs.blackfin.uclinux.org/doku.php?id=linux-kernel:smp-like In this patch, we provide SMP extend to BF561 kernel code Signed-off-by: Graf Yang Signed-off-by: Mike Frysinger Signed-off-by: Bryan Wu --- arch/blackfin/mach-bf561/Kconfig | 6 +- arch/blackfin/mach-bf561/Makefile | 1 + arch/blackfin/mach-bf561/atomic.S | 919 ++++++++++++++++++ .../mach-bf561/include/mach/blackfin.h | 4 + .../mach-bf561/include/mach/defBF561.h | 3 + .../mach-bf561/include/mach/mem_map.h | 120 +++ arch/blackfin/mach-bf561/include/mach/smp.h | 22 + arch/blackfin/mach-bf561/secondary.S | 215 ++++ arch/blackfin/mach-bf561/smp.c | 172 ++++ 9 files changed, 1459 insertions(+), 3 deletions(-) create mode 100644 arch/blackfin/mach-bf561/atomic.S create mode 100644 arch/blackfin/mach-bf561/include/mach/smp.h create mode 100644 arch/blackfin/mach-bf561/secondary.S create mode 100644 arch/blackfin/mach-bf561/smp.c diff --git a/arch/blackfin/mach-bf561/Kconfig b/arch/blackfin/mach-bf561/Kconfig index 3f4895450be..5d56438cad2 100644 --- a/arch/blackfin/mach-bf561/Kconfig +++ b/arch/blackfin/mach-bf561/Kconfig @@ -4,9 +4,9 @@ source "arch/blackfin/mach-bf561/boards/Kconfig" menu "BF561 Specific Configuration" -comment "Core B Support" +if (!SMP) -menu "Core B Support" +comment "Core B Support" config BF561_COREB bool "Enable Core B support" @@ -25,7 +25,7 @@ config BF561_COREB_RESET 0 is set, and will reset PC to 0xff600000 when COREB_SRAM_INIT is cleared. -endmenu +endif comment "Interrupt Priority Assignment" diff --git a/arch/blackfin/mach-bf561/Makefile b/arch/blackfin/mach-bf561/Makefile index f39235a5578..c37f00c0edb 100644 --- a/arch/blackfin/mach-bf561/Makefile +++ b/arch/blackfin/mach-bf561/Makefile @@ -7,3 +7,4 @@ extra-y := head.o obj-y := ints-priority.o dma.o obj-$(CONFIG_BF561_COREB) += coreb.o +obj-$(CONFIG_SMP) += smp.o secondary.o atomic.o diff --git a/arch/blackfin/mach-bf561/atomic.S b/arch/blackfin/mach-bf561/atomic.S new file mode 100644 index 00000000000..9439bc6bd01 --- /dev/null +++ b/arch/blackfin/mach-bf561/atomic.S @@ -0,0 +1,919 @@ +/* + * File: arch/blackfin/mach-bf561/atomic.S + * Author: Philippe Gerum + * + * Copyright 2007 Analog Devices Inc. + * + * This program 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. + * + * This program 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 this program; if not, see the file COPYING, or write + * to the Free Software Foundation, Inc., + * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include +#include +#include +#include + +.text + +.macro coreslot_loadaddr reg:req + \reg\().l = _corelock; + \reg\().h = _corelock; +.endm + +/* + * r0 = address of atomic data to flush and invalidate (32bit). + * + * Clear interrupts and return the old mask. + * We assume that no atomic data can span cachelines. + * + * Clobbers: r2:0, p0 + */ +ENTRY(_get_core_lock) + r1 = -L1_CACHE_BYTES; + r1 = r0 & r1; + cli r0; + coreslot_loadaddr p0; +.Lretry_corelock: + testset (p0); + if cc jump .Ldone_corelock; + SSYNC(r2); + jump .Lretry_corelock +.Ldone_corelock: + p0 = r1; + CSYNC(r2); + flushinv[p0]; + SSYNC(r2); + rts; +ENDPROC(_get_core_lock) + +/* + * r0 = address of atomic data in uncacheable memory region (32bit). + * + * Clear interrupts and return the old mask. + * + * Clobbers: r0, p0 + */ +ENTRY(_get_core_lock_noflush) + cli r0; + coreslot_loadaddr p0; +.Lretry_corelock_noflush: + testset (p0); + if cc jump .Ldone_corelock_noflush; + SSYNC(r2); + jump .Lretry_corelock_noflush +.Ldone_corelock_noflush: + rts; +ENDPROC(_get_core_lock_noflush) + +/* + * r0 = interrupt mask to restore. + * r1 = address of atomic data to flush and invalidate (32bit). + * + * Interrupts are masked on entry (see _get_core_lock). + * Clobbers: r2:0, p0 + */ +ENTRY(_put_core_lock) + /* Write-through cache assumed, so no flush needed here. */ + coreslot_loadaddr p0; + r1 = 0; + [p0] = r1; + SSYNC(r2); + sti r0; + rts; +ENDPROC(_put_core_lock) + +#ifdef __ARCH_SYNC_CORE_DCACHE + +ENTRY(___raw_smp_mark_barrier_asm) + [--sp] = rets; + [--sp] = ( r7:5 ); + [--sp] = r0; + [--sp] = p1; + [--sp] = p0; + call _get_core_lock_noflush; + + /* + * Calculate current core mask + */ + GET_CPUID(p1, r7); + r6 = 1; + r6 <<= r7; + + /* + * Set bit of other cores in barrier mask. Don't change current core bit. + */ + p1.l = _barrier_mask; + p1.h = _barrier_mask; + r7 = [p1]; + r5 = r7 & r6; + r7 = ~r6; + cc = r5 == 0; + if cc jump 1f; + r7 = r7 | r6; +1: + [p1] = r7; + SSYNC(r2); + + call _put_core_lock; + p0 = [sp++]; + p1 = [sp++]; + r0 = [sp++]; + ( r7:5 ) = [sp++]; + rets = [sp++]; + rts; +ENDPROC(___raw_smp_mark_barrier_asm) + +ENTRY(___raw_smp_check_barrier_asm) + [--sp] = rets; + [--sp] = ( r7:5 ); + [--sp] = r0; + [--sp] = p1; + [--sp] = p0; + call _get_core_lock_noflush; + + /* + * Calculate current core mask + */ + GET_CPUID(p1, r7); + r6 = 1; + r6 <<= r7; + + /* + * Clear current core bit in barrier mask if it is set. + */ + p1.l = _barrier_mask; + p1.h = _barrier_mask; + r7 = [p1]; + r5 = r7 & r6; + cc = r5 == 0; + if cc jump 1f; + r6 = ~r6; + r7 = r7 & r6; + [p1] = r7; + SSYNC(r2); + + call _put_core_lock; + + /* + * Invalidate the entire D-cache of current core. + */ + sp += -12; + call _resync_core_dcache + sp += 12; + jump 2f; +1: + call _put_core_lock; +2: + p0 = [sp++]; + p1 = [sp++]; + r0 = [sp++]; + ( r7:5 ) = [sp++]; + rets = [sp++]; + rts; +ENDPROC(___raw_smp_check_barrier_asm) + +/* + * r0 = irqflags + * r1 = address of atomic data + * + * Clobbers: r2:0, p1:0 + */ +_start_lock_coherent: + + [--sp] = rets; + [--sp] = ( r7:6 ); + r7 = r0; + p1 = r1; + + /* + * Determine whether the atomic data was previously + * owned by another CPU (=r6). + */ + GET_CPUID(p0, r2); + r1 = 1; + r1 <<= r2; + r2 = ~r1; + + r1 = [p1]; + r1 >>= 28; /* CPU fingerprints are stored in the high nibble. */ + r6 = r1 & r2; + r1 = [p1]; + r1 <<= 4; + r1 >>= 4; + [p1] = r1; + + /* + * Release the core lock now, but keep IRQs disabled while we are + * performing the remaining housekeeping chores for the current CPU. + */ + coreslot_loadaddr p0; + r1 = 0; + [p0] = r1; + + /* + * If another CPU has owned the same atomic section before us, + * then our D-cached copy of the shared data protected by the + * current spin/write_lock may be obsolete. + */ + cc = r6 == 0; + if cc jump .Lcache_synced + + /* + * Invalidate the entire D-cache of the current core. + */ + sp += -12; + call _resync_core_dcache + sp += 12; + +.Lcache_synced: + SSYNC(r2); + sti r7; + ( r7:6 ) = [sp++]; + rets = [sp++]; + rts + +/* + * r0 = irqflags + * r1 = address of atomic data + * + * Clobbers: r2:0, p1:0 + */ +_end_lock_coherent: + + p1 = r1; + GET_CPUID(p0, r2); + r2 += 28; + r1 = 1; + r1 <<= r2; + r2 = [p1]; + r2 = r1 | r2; + [p1] = r2; + r1 = p1; + jump _put_core_lock; + +#endif /* __ARCH_SYNC_CORE_DCACHE */ + +/* + * r0 = &spinlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_spin_is_locked_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r3 = [p1]; + cc = bittst( r3, 0 ); + r3 = cc; + r1 = p1; + call _put_core_lock; + rets = [sp++]; + r0 = r3; + rts; +ENDPROC(___raw_spin_is_locked_asm) + +/* + * r0 = &spinlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_spin_lock_asm) + p1 = r0; + [--sp] = rets; +.Lretry_spinlock: + call _get_core_lock; + r1 = p1; + r2 = [p1]; + cc = bittst( r2, 0 ); + if cc jump .Lbusy_spinlock +#ifdef __ARCH_SYNC_CORE_DCACHE + r3 = p1; + bitset ( r2, 0 ); /* Raise the lock bit. */ + [p1] = r2; + call _start_lock_coherent +#else + r2 = 1; + [p1] = r2; + call _put_core_lock; +#endif + rets = [sp++]; + rts; + +.Lbusy_spinlock: + /* We don't touch the atomic area if busy, so that flush + will behave like nop in _put_core_lock. */ + call _put_core_lock; + SSYNC(r2); + r0 = p1; + jump .Lretry_spinlock +ENDPROC(___raw_spin_lock_asm) + +/* + * r0 = &spinlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_spin_trylock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r1 = p1; + r3 = [p1]; + cc = bittst( r3, 0 ); + if cc jump .Lfailed_trylock +#ifdef __ARCH_SYNC_CORE_DCACHE + bitset ( r3, 0 ); /* Raise the lock bit. */ + [p1] = r3; + call _start_lock_coherent +#else + r2 = 1; + [p1] = r2; + call _put_core_lock; +#endif + r0 = 1; + rets = [sp++]; + rts; +.Lfailed_trylock: + call _put_core_lock; + r0 = 0; + rets = [sp++]; + rts; +ENDPROC(___raw_spin_trylock_asm) + +/* + * r0 = &spinlock->lock + * + * Clobbers: r2:0, p1:0 + */ +ENTRY(___raw_spin_unlock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r2 = [p1]; + bitclr ( r2, 0 ); + [p1] = r2; + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _end_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + rts; +ENDPROC(___raw_spin_unlock_asm) + +/* + * r0 = &rwlock->lock + * + * Clobbers: r2:0, p1:0 + */ +ENTRY(___raw_read_lock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; +.Lrdlock_try: + r1 = [p1]; + r1 += -1; + [p1] = r1; + cc = r1 < 0; + if cc jump .Lrdlock_failed + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _start_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + rts; + +.Lrdlock_failed: + r1 += 1; + [p1] = r1; +.Lrdlock_wait: + r1 = p1; + call _put_core_lock; + SSYNC(r2); + r0 = p1; + call _get_core_lock; + r1 = [p1]; + cc = r1 < 2; + if cc jump .Lrdlock_wait; + jump .Lrdlock_try +ENDPROC(___raw_read_lock_asm) + +/* + * r0 = &rwlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_read_trylock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r1 = [p1]; + cc = r1 <= 0; + if cc jump .Lfailed_tryrdlock; + r1 += -1; + [p1] = r1; + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _start_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + r0 = 1; + rts; +.Lfailed_tryrdlock: + r1 = p1; + call _put_core_lock; + rets = [sp++]; + r0 = 0; + rts; +ENDPROC(___raw_read_trylock_asm) + +/* + * r0 = &rwlock->lock + * + * Note: Processing controlled by a reader lock should not have + * any side-effect on cache issues with the other core, so we + * just release the core lock and exit (no _end_lock_coherent). + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_read_unlock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r1 = [p1]; + r1 += 1; + [p1] = r1; + r1 = p1; + call _put_core_lock; + rets = [sp++]; + rts; +ENDPROC(___raw_read_unlock_asm) + +/* + * r0 = &rwlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_write_lock_asm) + p1 = r0; + r3.l = lo(RW_LOCK_BIAS); + r3.h = hi(RW_LOCK_BIAS); + [--sp] = rets; + call _get_core_lock; +.Lwrlock_try: + r1 = [p1]; + r1 = r1 - r3; +#ifdef __ARCH_SYNC_CORE_DCACHE + r2 = r1; + r2 <<= 4; + r2 >>= 4; + cc = r2 == 0; +#else + cc = r1 == 0; +#endif + if !cc jump .Lwrlock_wait + [p1] = r1; + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _start_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + rts; + +.Lwrlock_wait: + r1 = p1; + call _put_core_lock; + SSYNC(r2); + r0 = p1; + call _get_core_lock; + r1 = [p1]; +#ifdef __ARCH_SYNC_CORE_DCACHE + r1 <<= 4; + r1 >>= 4; +#endif + cc = r1 == r3; + if !cc jump .Lwrlock_wait; + jump .Lwrlock_try +ENDPROC(___raw_write_lock_asm) + +/* + * r0 = &rwlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_write_trylock_asm) + p1 = r0; + [--sp] = rets; + call _get_core_lock; + r1 = [p1]; + r2.l = lo(RW_LOCK_BIAS); + r2.h = hi(RW_LOCK_BIAS); + cc = r1 == r2; + if !cc jump .Lfailed_trywrlock; +#ifdef __ARCH_SYNC_CORE_DCACHE + r1 >>= 28; + r1 <<= 28; +#else + r1 = 0; +#endif + [p1] = r1; + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _start_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + r0 = 1; + rts; + +.Lfailed_trywrlock: + r1 = p1; + call _put_core_lock; + rets = [sp++]; + r0 = 0; + rts; +ENDPROC(___raw_write_trylock_asm) + +/* + * r0 = &rwlock->lock + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_write_unlock_asm) + p1 = r0; + r3.l = lo(RW_LOCK_BIAS); + r3.h = hi(RW_LOCK_BIAS); + [--sp] = rets; + call _get_core_lock; + r1 = [p1]; + r1 = r1 + r3; + [p1] = r1; + r1 = p1; +#ifdef __ARCH_SYNC_CORE_DCACHE + call _end_lock_coherent +#else + call _put_core_lock; +#endif + rets = [sp++]; + rts; +ENDPROC(___raw_write_unlock_asm) + +/* + * r0 = ptr + * r1 = value + * + * Add a signed value to a 32bit word and return the new value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_atomic_update_asm) + p1 = r0; + r3 = r1; + [--sp] = rets; + call _get_core_lock; + r2 = [p1]; + r3 = r3 + r2; + [p1] = r3; + r1 = p1; + call _put_core_lock; + r0 = r3; + rets = [sp++]; + rts; +ENDPROC(___raw_atomic_update_asm) + +/* + * r0 = ptr + * r1 = mask + * + * Clear the mask bits from a 32bit word and return the old 32bit value + * atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_atomic_clear_asm) + p1 = r0; + r3 = ~r1; + [--sp] = rets; + call _get_core_lock; + r2 = [p1]; + r3 = r2 & r3; + [p1] = r3; + r3 = r2; + r1 = p1; + call _put_core_lock; + r0 = r3; + rets = [sp++]; + rts; +ENDPROC(___raw_atomic_clear_asm) + +/* + * r0 = ptr + * r1 = mask + * + * Set the mask bits into a 32bit word and return the old 32bit value + * atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_atomic_set_asm) + p1 = r0; + r3 = r1; + [--sp] = rets; + call _get_core_lock; + r2 = [p1]; + r3 = r2 | r3; + [p1] = r3; + r3 = r2; + r1 = p1; + call _put_core_lock; + r0 = r3; + rets = [sp++]; + rts; +ENDPROC(___raw_atomic_set_asm) + +/* + * r0 = ptr + * r1 = mask + * + * XOR the mask bits with a 32bit word and return the old 32bit value + * atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_atomic_xor_asm) + p1 = r0; + r3 = r1; + [--sp] = rets; + call _get_core_lock; + r2 = [p1]; + r3 = r2 ^ r3; + [p1] = r3; + r3 = r2; + r1 = p1; + call _put_core_lock; + r0 = r3; + rets = [sp++]; + rts; +ENDPROC(___raw_atomic_xor_asm) + +/* + * r0 = ptr + * r1 = mask + * + * Perform a logical AND between the mask bits and a 32bit word, and + * return the masked value. We need this on this architecture in + * order to invalidate the local cache before testing. + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_atomic_test_asm) + p1 = r0; + r3 = r1; + r1 = -L1_CACHE_BYTES; + r1 = r0 & r1; + p0 = r1; + flushinv[p0]; + SSYNC(r2); + r0 = [p1]; + r0 = r0 & r3; + rts; +ENDPROC(___raw_atomic_test_asm) + +/* + * r0 = ptr + * r1 = value + * + * Swap *ptr with value and return the old 32bit value atomically. + * Clobbers: r3:0, p1:0 + */ +#define __do_xchg(src, dst) \ + p1 = r0; \ + r3 = r1; \ + [--sp] = rets; \ + call _get_core_lock; \ + r2 = src; \ + dst = r3; \ + r3 = r2; \ + r1 = p1; \ + call _put_core_lock; \ + r0 = r3; \ + rets = [sp++]; \ + rts; + +ENTRY(___raw_xchg_1_asm) + __do_xchg(b[p1] (z), b[p1]) +ENDPROC(___raw_xchg_1_asm) + +ENTRY(___raw_xchg_2_asm) + __do_xchg(w[p1] (z), w[p1]) +ENDPROC(___raw_xchg_2_asm) + +ENTRY(___raw_xchg_4_asm) + __do_xchg([p1], [p1]) +ENDPROC(___raw_xchg_4_asm) + +/* + * r0 = ptr + * r1 = new + * r2 = old + * + * Swap *ptr with new if *ptr == old and return the previous *ptr + * value atomically. + * + * Clobbers: r3:0, p1:0 + */ +#define __do_cmpxchg(src, dst) \ + [--sp] = rets; \ + [--sp] = r4; \ + p1 = r0; \ + r3 = r1; \ + r4 = r2; \ + call _get_core_lock; \ + r2 = src; \ + cc = r2 == r4; \ + if !cc jump 1f; \ + dst = r3; \ + 1: r3 = r2; \ + r1 = p1; \ + call _put_core_lock; \ + r0 = r3; \ + r4 = [sp++]; \ + rets = [sp++]; \ + rts; + +ENTRY(___raw_cmpxchg_1_asm) + __do_cmpxchg(b[p1] (z), b[p1]) +ENDPROC(___raw_cmpxchg_1_asm) + +ENTRY(___raw_cmpxchg_2_asm) + __do_cmpxchg(w[p1] (z), w[p1]) +ENDPROC(___raw_cmpxchg_2_asm) + +ENTRY(___raw_cmpxchg_4_asm) + __do_cmpxchg([p1], [p1]) +ENDPROC(___raw_cmpxchg_4_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Set a bit in a 32bit word and return the old 32bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_set_asm) + r2 = r1; + r1 = 1; + r1 <<= r2; + jump ___raw_atomic_set_asm +ENDPROC(___raw_bit_set_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Clear a bit in a 32bit word and return the old 32bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_clear_asm) + r2 = r1; + r1 = 1; + r1 <<= r2; + jump ___raw_atomic_clear_asm +ENDPROC(___raw_bit_clear_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Toggle a bit in a 32bit word and return the old 32bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_toggle_asm) + r2 = r1; + r1 = 1; + r1 <<= r2; + jump ___raw_atomic_xor_asm +ENDPROC(___raw_bit_toggle_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Test-and-set a bit in a 32bit word and return the old bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_test_set_asm) + [--sp] = rets; + [--sp] = r1; + call ___raw_bit_set_asm + r1 = [sp++]; + r2 = 1; + r2 <<= r1; + r0 = r0 & r2; + cc = r0 == 0; + if cc jump 1f + r0 = 1; +1: + rets = [sp++]; + rts; +ENDPROC(___raw_bit_test_set_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Test-and-clear a bit in a 32bit word and return the old bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_test_clear_asm) + [--sp] = rets; + [--sp] = r1; + call ___raw_bit_clear_asm + r1 = [sp++]; + r2 = 1; + r2 <<= r1; + r0 = r0 & r2; + cc = r0 == 0; + if cc jump 1f + r0 = 1; +1: + rets = [sp++]; + rts; +ENDPROC(___raw_bit_test_clear_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Test-and-toggle a bit in a 32bit word, + * and return the old bit value atomically. + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_test_toggle_asm) + [--sp] = rets; + [--sp] = r1; + call ___raw_bit_toggle_asm + r1 = [sp++]; + r2 = 1; + r2 <<= r1; + r0 = r0 & r2; + cc = r0 == 0; + if cc jump 1f + r0 = 1; +1: + rets = [sp++]; + rts; +ENDPROC(___raw_bit_test_toggle_asm) + +/* + * r0 = ptr + * r1 = bitnr + * + * Test a bit in a 32bit word and return its value. + * We need this on this architecture in order to invalidate + * the local cache before testing. + * + * Clobbers: r3:0, p1:0 + */ +ENTRY(___raw_bit_test_asm) + r2 = r1; + r1 = 1; + r1 <<= r2; + jump ___raw_atomic_test_asm +ENDPROC(___raw_bit_test_asm) + +/* + * r0 = ptr + * + * Fetch and return an uncached 32bit value. + * + * Clobbers: r2:0, p1:0 + */ +ENTRY(___raw_uncached_fetch_asm) + p1 = r0; + r1 = -L1_CACHE_BYTES; + r1 = r0 & r1; + p0 = r1; + flushinv[p0]; + SSYNC(r2); + r0 = [p1]; + rts; +ENDPROC(___raw_uncached_fetch_asm) diff --git a/arch/blackfin/mach-bf561/include/mach/blackfin.h b/arch/blackfin/mach-bf561/include/mach/blackfin.h index 0ea8666e676..f79f6626b7e 100644 --- a/arch/blackfin/mach-bf561/include/mach/blackfin.h +++ b/arch/blackfin/mach-bf561/include/mach/blackfin.h @@ -66,8 +66,12 @@ #define bfin_read_SIC_IMASK(x) bfin_read32(SICA_IMASK0 + (x << 2)) #define bfin_write_SIC_IMASK(x, val) bfin_write32((SICA_IMASK0 + (x << 2)), val) +#define bfin_read_SICB_IMASK(x) bfin_read32(SICB_IMASK0 + (x << 2)) +#define bfin_write_SICB_IMASK(x, val) bfin_write32((SICB_IMASK0 + (x << 2)), val) #define bfin_read_SIC_ISR(x) bfin_read32(SICA_ISR0 + (x << 2)) #define bfin_write_SIC_ISR(x, val) bfin_write32((SICA_ISR0 + (x << 2)), val) +#define bfin_read_SICB_ISR(x) bfin_read32(SICB_ISR0 + (x << 2)) +#define bfin_write_SICB_ISR(x, val) bfin_write32((SICB_ISR0 + (x << 2)), val) #define BFIN_UART_NR_PORTS 1 diff --git a/arch/blackfin/mach-bf561/include/mach/defBF561.h b/arch/blackfin/mach-bf561/include/mach/defBF561.h index 4eca2026bb9..d7c50975965 100644 --- a/arch/blackfin/mach-bf561/include/mach/defBF561.h +++ b/arch/blackfin/mach-bf561/include/mach/defBF561.h @@ -912,6 +912,9 @@ #define ACTIVE_PLLDISABLED 0x0004 /* Processor In Active Mode With PLL Disabled */ #define PLL_LOCKED 0x0020 /* PLL_LOCKCNT Has Been Reached */ +/* SICA_SYSCR Masks */ +#define COREB_SRAM_INIT 0x0020 + /* SWRST Mask */ #define SYSTEM_RESET 0x0007 /* Initiates a system software reset */ #define DOUBLE_FAULT_A 0x0008 /* Core A Double Fault Causes Reset */ diff --git a/arch/blackfin/mach-bf561/include/mach/mem_map.h b/arch/blackfin/mach-bf561/include/mach/mem_map.h index f1d4c0637bd..488c3bda65b 100644 --- a/arch/blackfin/mach-bf561/include/mach/mem_map.h +++ b/arch/blackfin/mach-bf561/include/mach/mem_map.h @@ -85,4 +85,124 @@ #define L1_SCRATCH_START COREA_L1_SCRATCH_START #define L1_SCRATCH_LENGTH 0x1000 +#ifndef __ASSEMBLY__ + +#ifdef CONFIG_SMP + +#define get_l1_scratch_start_cpu(cpu) \ + ({ unsigned long __addr; \ + __addr = (cpu) ? COREB_L1_SCRATCH_START : COREA_L1_SCRATCH_START;\ + __addr; }) + +#define get_l1_code_start_cpu(cpu) \ + ({ unsigned long __addr; \ + __addr = (cpu) ? COREB_L1_CODE_START : COREA_L1_CODE_START; \ + __addr; }) + +#define get_l1_data_a_start_cpu(cpu) \ + ({ unsigned long __addr; \ + __addr = (cpu) ? COREB_L1_DATA_A_START : COREA_L1_DATA_A_START;\ + __addr; }) + +#define get_l1_data_b_start_cpu(cpu) \ + ({ unsigned long __addr; \ + __addr = (cpu) ? COREB_L1_DATA_B_START : COREA_L1_DATA_B_START;\ + __addr; }) + +#define get_l1_scratch_start() get_l1_scratch_start_cpu(blackfin_core_id()) +#define get_l1_code_start() get_l1_code_start_cpu(blackfin_core_id()) +#define get_l1_data_a_start() get_l1_data_a_start_cpu(blackfin_core_id()) +#define get_l1_data_b_start() get_l1_data_b_start_cpu(blackfin_core_id()) + +#else /* !CONFIG_SMP */ +#define get_l1_scratch_start_cpu(cpu) L1_SCRATCH_START +#define get_l1_code_start_cpu(cpu) L1_CODE_START +#define get_l1_data_a_start_cpu(cpu) L1_DATA_A_START +#define get_l1_data_b_start_cpu(cpu) L1_DATA_B_START +#define get_l1_scratch_start() L1_SCRATCH_START +#define get_l1_code_start() L1_CODE_START +#define get_l1_data_a_start() L1_DATA_A_START +#define get_l1_data_b_start() L1_DATA_B_START +#endif /* !CONFIG_SMP */ + +#else /* __ASSEMBLY__ */ + +/* + * The following macros both return the address of the PDA for the + * current core. + * + * In its first safe (and hairy) form, the macro neither clobbers any + * register aside of the output Preg, nor uses the stack, since it + * could be called with an invalid stack pointer, or the current stack + * space being uncovered by any CPLB (e.g. early exception handling). + * + * The constraints on the second form are a bit relaxed, and the code + * is allowed to use the specified Dreg for determining the PDA + * address to be returned into Preg. + */ +#ifdef CONFIG_SMP +#define GET_PDA_SAFE(preg) \ + preg.l = lo(DSPID); \ + preg.h = hi(DSPID); \ + preg = [preg]; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + preg = preg << 2; \ + if cc jump 2f; \ + cc = preg == 0x0; \ + preg.l = _cpu_pda; \ + preg.h = _cpu_pda; \ + if !cc jump 3f; \ +1: \ + /* preg = 0x0; */ \ + cc = !cc; /* restore cc to 0 */ \ + jump 4f; \ +2: \ + cc = preg == 0x0; \ + preg.l = _cpu_pda; \ + preg.h = _cpu_pda; \ + if cc jump 4f; \ + /* preg = 0x1000000; */ \ + cc = !cc; /* restore cc to 1 */ \ +3: \ + preg = [preg]; \ +4: + +#define GET_PDA(preg, dreg) \ + preg.l = lo(DSPID); \ + preg.h = hi(DSPID); \ + dreg = [preg]; \ + preg.l = _cpu_pda; \ + preg.h = _cpu_pda; \ + cc = bittst(dreg, 0); \ + if !cc jump 1f; \ + preg = [preg]; \ +1: \ + +#define GET_CPUID(preg, dreg) \ + preg.l = lo(DSPID); \ + preg.h = hi(DSPID); \ + dreg = [preg]; \ + dreg = ROT dreg BY -1; \ + dreg = CC; + +#else +#define GET_PDA_SAFE(preg) \ + preg.l = _cpu_pda; \ + preg.h = _cpu_pda; + +#define GET_PDA(preg, dreg) GET_PDA_SAFE(preg) +#endif /* CONFIG_SMP */ + +#endif /* __ASSEMBLY__ */ + #endif /* _MEM_MAP_533_H_ */ diff --git a/arch/blackfin/mach-bf561/include/mach/smp.h b/arch/blackfin/mach-bf561/include/mach/smp.h new file mode 100644 index 00000000000..f9e65ebe81b --- /dev/null +++ b/arch/blackfin/mach-bf561/include/mach/smp.h @@ -0,0 +1,22 @@ +#ifndef _MACH_BF561_SMP +#define _MACH_BF561_SMP + +struct task_struct; + +void platform_init_cpus(void); + +void platform_prepare_cpus(unsigned int max_cpus); + +int platform_boot_secondary(unsigned int cpu, struct task_struct *idle); + +void platform_secondary_init(unsigned int cpu); + +void platform_request_ipi(int (*handler)(int, void *)); + +void platform_send_ipi(cpumask_t callmap); + +void platform_send_ipi_cpu(unsigned int cpu); + +void platform_clear_ipi(unsigned int cpu); + +#endif /* !_MACH_BF561_SMP */ diff --git a/arch/blackfin/mach-bf561/secondary.S b/arch/blackfin/mach-bf561/secondary.S new file mode 100644 index 00000000000..35280f06b7b --- /dev/null +++ b/arch/blackfin/mach-bf561/secondary.S @@ -0,0 +1,215 @@ +/* + * File: arch/blackfin/mach-bf561/secondary.S + * Based on: arch/blackfin/mach-bf561/head.S + * Author: Philippe Gerum + * + * Copyright 2007 Analog Devices Inc. + * + * Description: BF561 coreB bootstrap file + * + * This program 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. + * + * This program 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 this program; if not, see the file COPYING, or write + * to the Free Software Foundation, Inc., + * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include +#include + +__INIT + +/* Lay the initial stack into the L1 scratch area of Core B */ +#define INITIAL_STACK (COREB_L1_SCRATCH_START + L1_SCRATCH_LENGTH - 12) + +ENTRY(_coreb_trampoline_start) + /* Set the SYSCFG register */ + R0 = 0x36; + SYSCFG = R0; /*Enable Cycle Counter and Nesting Of Interrupts(3rd Bit)*/ + R0 = 0; + + /*Clear Out All the data and pointer Registers*/ + R1 = R0; + R2 = R0; + R3 = R0; + R4 = R0; + R5 = R0; + R6 = R0; + R7 = R0; + + P0 = R0; + P1 = R0; + P2 = R0; + P3 = R0; + P4 = R0; + P5 = R0; + + LC0 = r0; + LC1 = r0; + L0 = r0; + L1 = r0; + L2 = r0; + L3 = r0; + + /* Clear Out All the DAG Registers*/ + B0 = r0; + B1 = r0; + B2 = r0; + B3 = r0; + + I0 = r0; + I1 = r0; + I2 = r0; + I3 = r0; + + M0 = r0; + M1 = r0; + M2 = r0; + M3 = r0; + + /* Turn off the icache */ + p0.l = LO(IMEM_CONTROL); + p0.h = HI(IMEM_CONTROL); + R1 = [p0]; + R0 = ~ENICPLB; + R0 = R0 & R1; + + /* Anomaly 05000125 */ +#ifdef ANOMALY_05000125 + CLI R2; + SSYNC; +#endif + [p0] = R0; + SSYNC; +#ifdef ANOMALY_05000125 + STI R2; +#endif + + /* Turn off the dcache */ + p0.l = LO(DMEM_CONTROL); + p0.h = HI(DMEM_CONTROL); + R1 = [p0]; + R0 = ~ENDCPLB; + R0 = R0 & R1; + + /* Anomaly 05000125 */ +#ifdef ANOMALY_05000125 + CLI R2; + SSYNC; +#endif + [p0] = R0; + SSYNC; +#ifdef ANOMALY_05000125 + STI R2; +#endif + + /* in case of double faults, save a few things */ + p0.l = _init_retx_coreb; + p0.h = _init_retx_coreb; + R0 = RETX; + [P0] = R0; + +#ifdef CONFIG_DEBUG_DOUBLEFAULT + /* Only save these if we are storing them, + * This happens here, since L1 gets clobbered + * below + */ + GET_PDA(p0, r0); + r7 = [p0 + PDA_RETX]; + p1.l = _init_saved_retx_coreb; + p1.h = _init_saved_retx_coreb; + [p1] = r7; + + r7 = [p0 + PDA_DCPLB]; + p1.l = _init_saved_dcplb_fault_addr_coreb; + p1.h = _init_saved_dcplb_fault_addr_coreb; + [p1] = r7; + + r7 = [p0 + PDA_ICPLB]; + p1.l = _init_saved_icplb_fault_addr_coreb; + p1.h = _init_saved_icplb_fault_addr_coreb; + [p1] = r7; + + r7 = [p0 + PDA_SEQSTAT]; + p1.l = _init_saved_seqstat_coreb; + p1.h = _init_saved_seqstat_coreb; + [p1] = r7; +#endif + + /* Initialize stack pointer */ + sp.l = lo(INITIAL_STACK); + sp.h = hi(INITIAL_STACK); + fp = sp; + usp = sp; + + /* This section keeps the processor in supervisor mode + * during core B startup. Branches to the idle task. + */ + + /* EVT15 = _real_start */ + + p0.l = lo(EVT15); + p0.h = hi(EVT15); + p1.l = _coreb_start; + p1.h = _coreb_start; + [p0] = p1; + csync; + + p0.l = lo(IMASK); + p0.h = hi(IMASK); + p1.l = IMASK_IVG15; + p1.h = 0x0; + [p0] = p1; + csync; + + raise 15; + p0.l = .LWAIT_HERE; + p0.h = .LWAIT_HERE; + reti = p0; +#if defined(ANOMALY_05000281) + nop; nop; nop; +#endif + rti; + +.LWAIT_HERE: + jump .LWAIT_HERE; +ENDPROC(_coreb_trampoline_start) +ENTRY(_coreb_trampoline_end) + +ENTRY(_coreb_start) + [--sp] = reti; + + p0.l = lo(WDOGB_CTL); + p0.h = hi(WDOGB_CTL); + r0 = 0xAD6(z); + w[p0] = r0; /* Clear the watchdog. */ + ssync; + + /* + * switch to IDLE stack. + */ + p0.l = _secondary_stack; + p0.h = _secondary_stack; + sp = [p0]; + usp = sp; + fp = sp; + sp += -12; + call _init_pda + sp += 12; + call _secondary_start_kernel; +.L_exit: + jump.s .L_exit; +ENDPROC(_coreb_start) + +__FINIT diff --git a/arch/blackfin/mach-bf561/smp.c b/arch/blackfin/mach-bf561/smp.c new file mode 100644 index 00000000000..23fd4c13897 --- /dev/null +++ b/arch/blackfin/mach-bf561/smp.c @@ -0,0 +1,172 @@ +/* + * File: arch/blackfin/mach-bf561/smp.c + * Author: Philippe Gerum + * + * Copyright 2007 Analog Devices Inc. + * + * This program 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. + * + * This program 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 this program; if not, see the file COPYING, or write + * to the Free Software Foundation, Inc., + * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include +#include +#include +#include +#include +#include + +#define COREB_SRAM_BASE 0xff600000 +#define COREB_SRAM_SIZE 0x4000 + +extern char coreb_trampoline_start, coreb_trampoline_end; + +static DEFINE_SPINLOCK(boot_lock); + +static cpumask_t cpu_callin_map; + +/* + * platform_init_cpus() - Tell the world about how many cores we + * have. This is called while setting up the architecture support + * (setup_arch()), so don't be too demanding here with respect to + * available kernel services. + */ + +void __init platform_init_cpus(void) +{ + cpu_set(0, cpu_possible_map); /* CoreA */ + cpu_set(1, cpu_possible_map); /* CoreB */ +} + +void __init platform_prepare_cpus(unsigned int max_cpus) +{ + int len; + + len = &coreb_trampoline_end - &coreb_trampoline_start + 1; + BUG_ON(len > COREB_SRAM_SIZE); + + dma_memcpy((void *)COREB_SRAM_BASE, &coreb_trampoline_start, len); + + /* Both cores ought to be present on a bf561! */ + cpu_set(0, cpu_present_map); /* CoreA */ + cpu_set(1, cpu_present_map); /* CoreB */ + + printk(KERN_INFO "CoreB bootstrap code to SRAM %p via DMA.\n", (void *)COREB_SRAM_BASE); +} + +int __init setup_profiling_timer(unsigned int multiplier) /* not supported */ +{ + return -EINVAL; +} + +void __cpuinit platform_secondary_init(unsigned int cpu) +{ + local_irq_disable(); + + /* Clone setup for peripheral interrupt sources from CoreA. */ + bfin_write_SICB_IMASK0(bfin_read_SICA_IMASK0()); + bfin_write_SICB_IMASK1(bfin_read_SICA_IMASK1()); + SSYNC(); + + /* Clone setup for IARs from CoreA. */ + bfin_write_SICB_IAR0(bfin_read_SICA_IAR0()); + bfin_write_SICB_IAR1(bfin_read_SICA_IAR1()); + bfin_write_SICB_IAR2(bfin_read_SICA_IAR2()); + bfin_write_SICB_IAR3(bfin_read_SICA_IAR3()); + bfin_write_SICB_IAR4(bfin_read_SICA_IAR4()); + bfin_write_SICB_IAR5(bfin_read_SICA_IAR5()); + bfin_write_SICB_IAR6(bfin_read_SICA_IAR6()); + bfin_write_SICB_IAR7(bfin_read_SICA_IAR7()); + SSYNC(); + + local_irq_enable(); + + /* Calibrate loops per jiffy value. */ + calibrate_delay(); + + /* Store CPU-private information to the cpu_data array. */ + bfin_setup_cpudata(cpu); + + /* We are done with local CPU inits, unblock the boot CPU. */ + cpu_set(cpu, cpu_callin_map); + spin_lock(&boot_lock); + spin_unlock(&boot_lock); +} + +int __cpuinit platform_boot_secondary(unsigned int cpu, struct task_struct *idle) +{ + unsigned long timeout; + + /* CoreB already running?! */ + BUG_ON((bfin_read_SICA_SYSCR() & COREB_SRAM_INIT) == 0); + + printk(KERN_INFO "Booting Core B.\n"); + + spin_lock(&boot_lock); + + /* Kick CoreB, which should start execution from CORE_SRAM_BASE. */ + SSYNC(); + bfin_write_SICA_SYSCR(bfin_read_SICA_SYSCR() & ~COREB_SRAM_INIT); + SSYNC(); + + timeout = jiffies + 1 * HZ; + while (time_before(jiffies, timeout)) { + if (cpu_isset(cpu, cpu_callin_map)) + break; + udelay(100); + barrier(); + } + + spin_unlock(&boot_lock); + + return cpu_isset(cpu, cpu_callin_map) ? 0 : -ENOSYS; +} + +void __init platform_request_ipi(irq_handler_t handler) +{ + int ret; + + ret = request_irq(IRQ_SUPPLE_0, handler, IRQF_DISABLED, + "SMP interrupt", handler); + if (ret) + panic("Cannot request supplemental interrupt 0 for IPI service\n"); +} + +void platform_send_ipi(cpumask_t callmap) +{ + unsigned int cpu; + + for_each_cpu_mask(cpu, callmap) { + BUG_ON(cpu >= 2); + SSYNC(); + bfin_write_SICB_SYSCR(bfin_read_SICB_SYSCR() | (1 << (6 + cpu))); + SSYNC(); + } +} + +void platform_send_ipi_cpu(unsigned int cpu) +{ + BUG_ON(cpu >= 2); + SSYNC(); + bfin_write_SICB_SYSCR(bfin_read_SICB_SYSCR() | (1 << (6 + cpu))); + SSYNC(); +} + +void platform_clear_ipi(unsigned int cpu) +{ + BUG_ON(cpu >= 2); + SSYNC(); + bfin_write_SICB_SYSCR(bfin_read_SICB_SYSCR() | (1 << (10 + cpu))); + SSYNC(); +} -- 2.20.1