From f3156851616b4f3cb1d6c567fe46adcfd43ad8f8 Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Fri, 10 Nov 2017 12:40:00 -0800 Subject: ARCv2: boot log: updates for HS48: dual-issue, ECC, Loop Buffer Print the hardware support for ECC, Loop Buffer as well as the runtime enabled status Note that unlike the existing boot printing, this one is not read from pre-decoded hardware capabilty info cached in cpuinfo[] struct. Instead we read the AUX regs on the spot and print it, without botherign to save anywhere. There is no point in saving static hardware capabilites in memory when its use is very sporadic and non-performance critical, mainly for /proc/cpuinfo. This gets worse in SMP, given it is per-cpu, and pretty much exactly same across all cpus. So only info needed at runtime (e.g. TLB geometry) needs to be cached in cpuinfo[]. So going fwd we will start converting code to this paradigm. Signed-off-by: Vineet Gupta --- arch/arc/kernel/setup.c | 43 +++++++++++++++++++++++++++++++++++++++---- 1 file changed, 39 insertions(+), 4 deletions(-) (limited to 'arch/arc/kernel') diff --git a/arch/arc/kernel/setup.c b/arch/arc/kernel/setup.c index fb83844daeea..7ef7d9a8ff89 100644 --- a/arch/arc/kernel/setup.c +++ b/arch/arc/kernel/setup.c @@ -199,8 +199,10 @@ static void read_arc_build_cfg_regs(void) unsigned int exec_ctrl; READ_BCR(AUX_EXEC_CTRL, exec_ctrl); - cpu->extn.dual_iss_exist = 1; - cpu->extn.dual_iss_enb = exec_ctrl & 1; + cpu->extn.dual_enb = exec_ctrl & 1; + + /* dual issue always present for this core */ + cpu->extn.dual = 1; } } @@ -253,7 +255,7 @@ static char *arc_cpu_mumbojumbo(int cpu_id, char *buf, int len) cpu_id, cpu->name, cpu->details, is_isa_arcompact() ? "ARCompact" : "ARCv2", IS_AVAIL1(cpu->isa.be, "[Big-Endian]"), - IS_AVAIL3(cpu->extn.dual_iss_exist, cpu->extn.dual_iss_enb, " Dual-Issue")); + IS_AVAIL3(cpu->extn.dual, cpu->extn.dual_enb, " Dual-Issue ")); n += scnprintf(buf + n, len - n, "Timers\t\t: %s%s%s%s%s%s\nISA Extn\t: ", IS_AVAIL1(cpu->extn.timer0, "Timer0 "), @@ -293,11 +295,26 @@ static char *arc_cpu_mumbojumbo(int cpu_id, char *buf, int len) if (cpu->bpu.ver) n += scnprintf(buf + n, len - n, - "BPU\t\t: %s%s match, cache:%d, Predict Table:%d\n", + "BPU\t\t: %s%s match, cache:%d, Predict Table:%d", IS_AVAIL1(cpu->bpu.full, "full"), IS_AVAIL1(!cpu->bpu.full, "partial"), cpu->bpu.num_cache, cpu->bpu.num_pred); + if (is_isa_arcv2()) { + struct bcr_lpb lpb; + + READ_BCR(ARC_REG_LPB_BUILD, lpb); + if (lpb.ver) { + unsigned int ctl; + ctl = read_aux_reg(ARC_REG_LPB_CTRL); + + n += scnprintf(buf + n, len - n, " Loop Buffer:%d %s", + lpb.entries, + IS_DISABLED_RUN(!ctl)); + } + } + + n += scnprintf(buf + n, len - n, "\n"); return buf; } @@ -326,6 +343,24 @@ static char *arc_extn_mumbojumbo(int cpu_id, char *buf, int len) cpu->dccm.base_addr, TO_KB(cpu->dccm.sz), cpu->iccm.base_addr, TO_KB(cpu->iccm.sz)); + if (is_isa_arcv2()) { + + /* Error Protection: ECC/Parity */ + struct bcr_erp erp; + READ_BCR(ARC_REG_ERP_BUILD, erp); + + if (erp.ver) { + struct ctl_erp ctl; + READ_BCR(ARC_REG_ERP_CTRL, ctl); + + /* inverted bits: 0 means enabled */ + n += scnprintf(buf + n, len - n, "Extn [ECC]\t: %s%s%s%s%s%s\n", + IS_AVAIL3(erp.ic, !ctl.dpi, "IC "), + IS_AVAIL3(erp.dc, !ctl.dpd, "DC "), + IS_AVAIL3(erp.mmu, !ctl.mpd, "MMU ")); + } + } + n += scnprintf(buf + n, len - n, "OS ABI [v%d]\t: %s\n", EF_ARC_OSABI_CURRENT >> 8, EF_ARC_OSABI_CURRENT == EF_ARC_OSABI_V3 ? -- cgit v1.3-18-gd494 From 4d431290402c8d867af7ba45ee75407d68748c4a Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Sat, 9 May 2015 18:27:30 +0530 Subject: ARCv2: perf: tweak overflow interrupt Current perf ISR loops thru all 32 counters, checking for each if it caused the interrupt. Instead only loop thru counters which actually interrupted (typically 1). Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/kernel/perf_event.c | 24 ++++++++++++++---------- 1 file changed, 14 insertions(+), 10 deletions(-) (limited to 'arch/arc/kernel') diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c index 2ce24e74f879..0eaa132a2c90 100644 --- a/arch/arc/kernel/perf_event.c +++ b/arch/arc/kernel/perf_event.c @@ -377,21 +377,22 @@ static irqreturn_t arc_pmu_intr(int irq, void *dev) struct perf_sample_data data; struct arc_pmu_cpu *pmu_cpu = this_cpu_ptr(&arc_pmu_cpu); struct pt_regs *regs; - int active_ints; + unsigned int active_ints; int idx; arc_pmu_disable(&arc_pmu->pmu); active_ints = read_aux_reg(ARC_REG_PCT_INT_ACT); + if (!active_ints) + goto done; regs = get_irq_regs(); - for (idx = 0; idx < arc_pmu->n_counters; idx++) { - struct perf_event *event = pmu_cpu->act_counter[idx]; + do { + struct perf_event *event; struct hw_perf_event *hwc; - if (!(active_ints & (1 << idx))) - continue; + idx = __ffs(active_ints); /* Reset interrupt flag by writing of 1 */ write_aux_reg(ARC_REG_PCT_INT_ACT, 1 << idx); @@ -404,19 +405,22 @@ static irqreturn_t arc_pmu_intr(int irq, void *dev) write_aux_reg(ARC_REG_PCT_INT_CTRL, read_aux_reg(ARC_REG_PCT_INT_CTRL) | (1 << idx)); + event = pmu_cpu->act_counter[idx]; hwc = &event->hw; WARN_ON_ONCE(hwc->idx != idx); arc_perf_event_update(event, &event->hw, event->hw.idx); perf_sample_data_init(&data, 0, hwc->last_period); - if (!arc_pmu_event_set_period(event)) - continue; + if (arc_pmu_event_set_period(event)) { + if (perf_event_overflow(event, &data, regs)) + arc_pmu_stop(event, 0); + } - if (perf_event_overflow(event, &data, regs)) - arc_pmu_stop(event, 0); - } + active_ints &= ~(1U << idx); + } while (active_ints); +done: arc_pmu_enable(&arc_pmu->pmu); return IRQ_HANDLED; -- cgit v1.3-18-gd494 From 5b9027d6d044d4917992119d184ab0bb616489cc Mon Sep 17 00:00:00 2001 From: Vineet Gupta Date: Thu, 8 Oct 2015 22:17:48 +0530 Subject: ARCv2: perf: optimize given that num counters <= 32 use ffz primitive which maps to ARCv2 instruction, vs. non atomic __test_and_set_bit It is unlikely if we will even have more than 32 counters, but still add a BUILD_BUG to catch that Acked-by: Peter Zijlstra (Intel) Signed-off-by: Vineet Gupta --- arch/arc/kernel/perf_event.c | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) (limited to 'arch/arc/kernel') diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c index 0eaa132a2c90..8aec462d90fb 100644 --- a/arch/arc/kernel/perf_event.c +++ b/arch/arc/kernel/perf_event.c @@ -336,15 +336,12 @@ static int arc_pmu_add(struct perf_event *event, int flags) struct hw_perf_event *hwc = &event->hw; int idx = hwc->idx; - if (__test_and_set_bit(idx, pmu_cpu->used_mask)) { - idx = find_first_zero_bit(pmu_cpu->used_mask, - arc_pmu->n_counters); - if (idx == arc_pmu->n_counters) - return -EAGAIN; - - __set_bit(idx, pmu_cpu->used_mask); - hwc->idx = idx; - } + idx = ffz(pmu_cpu->used_mask[0]); + if (idx == arc_pmu->n_counters) + return -EAGAIN; + + __set_bit(idx, pmu_cpu->used_mask); + hwc->idx = idx; write_aux_reg(ARC_REG_PCT_INDEX, idx); @@ -465,6 +462,7 @@ static int arc_pmu_device_probe(struct platform_device *pdev) pr_err("This core does not have performance counters!\n"); return -ENODEV; } + BUILD_BUG_ON(ARC_PERF_MAX_COUNTERS > 32); BUG_ON(pct_bcr.c > ARC_PERF_MAX_COUNTERS); READ_BCR(ARC_REG_CC_BUILD, cc_bcr); -- cgit v1.3-18-gd494