From dbdf20db537a5369c65330f878ad4905020a8bfa Mon Sep 17 00:00:00 2001 From: Bernd Schmidt Date: Wed, 7 Jan 2009 23:14:38 +0800 Subject: Blackfin arch: Faster C implementation of no-MPU CPLB handler This is a mixture ofcMichael McTernan's patch and the existing cplb-mpu code. We ditch the old cplb-nompu implementation, which is a good example of why a good algorithm in a HLL is preferrable to a bad algorithm written in assembly. Rather than try to construct a table of all posible CPLBs and search it, we just create a (smaller) table of memory regions and their attributes. Some of the data structures are now unified for both the mpu and nompu cases. A lot of needless complexity in cplbinit.c is removed. Further optimizations: * compile cplbmgr.c with a lot of -ffixed-reg options, and omit saving these registers on the stack when entering a CPLB exception. * lose cli/nop/nop/sti sequences for some workarounds - these don't * make sense in an exception context Additional code unification should be possible after this. [Mike Frysinger : - convert CPP if statements to C if statements - remove redundant statements - use a do...while loop rather than a for loop to get slightly better optimization and to avoid gcc "may be used uninitialized" warnings ... we know that the [id]cplb_nr_bounds variables will never be 0, so this is OK - the no-mpu code was the last user of MAX_MEM_SIZE and with that rewritten, we can punt it - add some BUG_ON() checks to make sure we dont overflow the small cplb_bounds array - add i/d cplb entries for the bootrom because there is functions/data in there we want to access - we do not need a NULL trailing entry as any time we access the bounds arrays, we use the nr_bounds variable ] Signed-off-by: Michael McTernan Signed-off-by: Mike Frysinger Signed-off-by: Bernd Schmidt Signed-off-by: Bryan Wu --- arch/blackfin/kernel/cplb-nompu/cplbmgr.c | 283 ++++++++++++++++++++++++++++++ 1 file changed, 283 insertions(+) create mode 100644 arch/blackfin/kernel/cplb-nompu/cplbmgr.c (limited to 'arch/blackfin/kernel/cplb-nompu/cplbmgr.c') diff --git a/arch/blackfin/kernel/cplb-nompu/cplbmgr.c b/arch/blackfin/kernel/cplb-nompu/cplbmgr.c new file mode 100644 index 00000000000..376249ab269 --- /dev/null +++ b/arch/blackfin/kernel/cplb-nompu/cplbmgr.c @@ -0,0 +1,283 @@ +/* + * File: arch/blackfin/kernel/cplb-nompu-c/cplbmgr.c + * Based on: arch/blackfin/kernel/cplb-mpu/cplbmgr.c + * Author: Michael McTernan + * + * Created: 01Nov2008 + * Description: CPLB miss handler. + * + * Modified: + * Copyright 2008 Airvana Inc. + * Copyright 2004-2007 Analog Devices Inc. + * + * Bugs: Enter bugs at http://blackfin.uclinux.org/ + * + * 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. + */ + +#include +#include +#include +#include +#include + +/* + * WARNING + * + * This file is compiled with certain -ffixed-reg options. We have to + * make sure not to call any functions here that could clobber these + * registers. + */ + +int nr_dcplb_miss[NR_CPUS], nr_icplb_miss[NR_CPUS]; +int nr_dcplb_supv_miss[NR_CPUS], nr_icplb_supv_miss[NR_CPUS]; +int nr_cplb_flush[NR_CPUS], nr_dcplb_prot[NR_CPUS]; + +#ifdef CONFIG_EXCPT_IRQ_SYSC_L1 +#define MGR_ATTR __attribute__((l1_text)) +#else +#define MGR_ATTR +#endif + +/* + * We're in an exception handler. The normal cli nop nop workaround + * isn't going to do very much, as the only thing that can interrupt + * us is an NMI, and the cli isn't going to stop that. + */ +#define NOWA_SSYNC __asm__ __volatile__ ("ssync;") + +/* Anomaly handlers provide SSYNCs, so avoid extra if anomaly is present */ +#if ANOMALY_05000125 + +#define bfin_write_DMEM_CONTROL_SSYNC(v) bfin_write_DMEM_CONTROL(v) +#define bfin_write_IMEM_CONTROL_SSYNC(v) bfin_write_IMEM_CONTROL(v) + +#else + +#define bfin_write_DMEM_CONTROL_SSYNC(v) \ + do { NOWA_SSYNC; bfin_write_DMEM_CONTROL(v); NOWA_SSYNC; } while (0) +#define bfin_write_IMEM_CONTROL_SSYNC(v) \ + do { NOWA_SSYNC; bfin_write_IMEM_CONTROL(v); NOWA_SSYNC; } while (0) + +#endif + +static inline void write_dcplb_data(int cpu, int idx, unsigned long data, + unsigned long addr) +{ + unsigned long ctrl = bfin_read_DMEM_CONTROL(); + bfin_write_DMEM_CONTROL_SSYNC(ctrl & ~ENDCPLB); + bfin_write32(DCPLB_DATA0 + idx * 4, data); + bfin_write32(DCPLB_ADDR0 + idx * 4, addr); + bfin_write_DMEM_CONTROL_SSYNC(ctrl); + +#ifdef CONFIG_CPLB_INFO + dcplb_tbl[cpu][idx].addr = addr; + dcplb_tbl[cpu][idx].data = data; +#endif +} + +static inline void write_icplb_data(int cpu, int idx, unsigned long data, + unsigned long addr) +{ + unsigned long ctrl = bfin_read_IMEM_CONTROL(); + + bfin_write_IMEM_CONTROL_SSYNC(ctrl & ~ENICPLB); + bfin_write32(ICPLB_DATA0 + idx * 4, data); + bfin_write32(ICPLB_ADDR0 + idx * 4, addr); + bfin_write_IMEM_CONTROL_SSYNC(ctrl); + +#ifdef CONFIG_CPLB_INFO + icplb_tbl[cpu][idx].addr = addr; + icplb_tbl[cpu][idx].data = data; +#endif +} + +/* + * Given the contents of the status register, return the index of the + * CPLB that caused the fault. + */ +static inline int faulting_cplb_index(int status) +{ + int signbits = __builtin_bfin_norm_fr1x32(status & 0xFFFF); + return 30 - signbits; +} + +/* + * Given the contents of the status register and the DCPLB_DATA contents, + * return true if a write access should be permitted. + */ +static inline int write_permitted(int status, unsigned long data) +{ + if (status & FAULT_USERSUPV) + return !!(data & CPLB_SUPV_WR); + else + return !!(data & CPLB_USER_WR); +} + +/* Counters to implement round-robin replacement. */ +static int icplb_rr_index[NR_CPUS] PDT_ATTR; +static int dcplb_rr_index[NR_CPUS] PDT_ATTR; + +/* + * Find an ICPLB entry to be evicted and return its index. + */ +static int evict_one_icplb(int cpu) +{ + int i = first_switched_icplb + icplb_rr_index[cpu]; + if (i >= MAX_CPLBS) { + i -= MAX_CPLBS - first_switched_icplb; + icplb_rr_index[cpu] -= MAX_CPLBS - first_switched_icplb; + } + icplb_rr_index[cpu]++; + return i; +} + +static int evict_one_dcplb(int cpu) +{ + int i = first_switched_dcplb + dcplb_rr_index[cpu]; + if (i >= MAX_CPLBS) { + i -= MAX_CPLBS - first_switched_dcplb; + dcplb_rr_index[cpu] -= MAX_CPLBS - first_switched_dcplb; + } + dcplb_rr_index[cpu]++; + return i; +} + +MGR_ATTR static int icplb_miss(int cpu) +{ + unsigned long addr = bfin_read_ICPLB_FAULT_ADDR(); + int status = bfin_read_ICPLB_STATUS(); + int idx; + unsigned long i_data, base, addr1, eaddr; + + nr_icplb_miss[cpu]++; + if (unlikely(status & FAULT_USERSUPV)) + nr_icplb_supv_miss[cpu]++; + + base = 0; + for (idx = 0; idx < icplb_nr_bounds; idx++) { + eaddr = icplb_bounds[idx].eaddr; + if (addr < eaddr) + break; + base = eaddr; + } + if (unlikely(idx == icplb_nr_bounds)) + return CPLB_NO_ADDR_MATCH; + + i_data = icplb_bounds[idx].data; + if (unlikely(i_data == 0)) + return CPLB_NO_ADDR_MATCH; + + addr1 = addr & ~(SIZE_4M - 1); + addr &= ~(SIZE_1M - 1); + i_data |= PAGE_SIZE_1MB; + if (addr1 >= base && (addr1 + SIZE_4M) <= eaddr) { + /* + * This works because + * (PAGE_SIZE_4MB & PAGE_SIZE_1MB) == PAGE_SIZE_1MB. + */ + i_data |= PAGE_SIZE_4MB; + addr = addr1; + } + + /* Pick entry to evict */ + idx = evict_one_icplb(cpu); + + write_icplb_data(cpu, idx, i_data, addr); + + return CPLB_RELOADED; +} + +MGR_ATTR static int dcplb_miss(int cpu) +{ + unsigned long addr = bfin_read_DCPLB_FAULT_ADDR(); + int status = bfin_read_DCPLB_STATUS(); + int idx; + unsigned long d_data, base, addr1, eaddr; + + nr_dcplb_miss[cpu]++; + if (unlikely(status & FAULT_USERSUPV)) + nr_dcplb_supv_miss[cpu]++; + + base = 0; + for (idx = 0; idx < dcplb_nr_bounds; idx++) { + eaddr = dcplb_bounds[idx].eaddr; + if (addr < eaddr) + break; + base = eaddr; + } + if (unlikely(idx == dcplb_nr_bounds)) + return CPLB_NO_ADDR_MATCH; + + d_data = dcplb_bounds[idx].data; + if (unlikely(d_data == 0)) + return CPLB_NO_ADDR_MATCH; + + addr1 = addr & ~(SIZE_4M - 1); + addr &= ~(SIZE_1M - 1); + d_data |= PAGE_SIZE_1MB; + if (addr1 >= base && (addr1 + SIZE_4M) <= eaddr) { + /* + * This works because + * (PAGE_SIZE_4MB & PAGE_SIZE_1MB) == PAGE_SIZE_1MB. + */ + d_data |= PAGE_SIZE_4MB; + addr = addr1; + } + + /* Pick entry to evict */ + idx = evict_one_dcplb(cpu); + + write_dcplb_data(cpu, idx, d_data, addr); + + return CPLB_RELOADED; +} + +MGR_ATTR static noinline int dcplb_protection_fault(int cpu) +{ + int status = bfin_read_DCPLB_STATUS(); + + nr_dcplb_prot[cpu]++; + + if (likely(status & FAULT_RW)) { + int idx = faulting_cplb_index(status); + unsigned long regaddr = DCPLB_DATA0 + idx * 4; + unsigned long data = bfin_read32(regaddr); + + /* Check if fault is to dirty a clean page */ + if (!(data & CPLB_WT) && !(data & CPLB_DIRTY) && + write_permitted(status, data)) { + + dcplb_tbl[cpu][idx].data = data; + bfin_write32(regaddr, data); + return CPLB_RELOADED; + } + } + + return CPLB_PROT_VIOL; +} + +MGR_ATTR int cplb_hdr(int seqstat, struct pt_regs *regs) +{ + int cause = seqstat & 0x3f; + unsigned int cpu = smp_processor_id(); + switch (cause) { + case 0x2C: + return icplb_miss(cpu); + case 0x26: + return dcplb_miss(cpu); + default: + if (unlikely(cause == 0x23)) + return dcplb_protection_fault(cpu); + + return CPLB_UNKNOWN_ERR; + } +} -- cgit v1.2.3