From patchwork Tue Dec 2 00:18:34 2008 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Carl Love X-Patchwork-Id: 11686 Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from ozlabs.org (localhost [127.0.0.1]) by ozlabs.org (Postfix) with ESMTP id A6B33DE262 for ; Tue, 2 Dec 2008 11:20:53 +1100 (EST) X-Original-To: linuxppc-dev@ozlabs.org Delivered-To: linuxppc-dev@ozlabs.org Received: from e32.co.us.ibm.com (e32.co.us.ibm.com [32.97.110.150]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client CN "e32.co.us.ibm.com", Issuer "Equifax" (verified OK)) by ozlabs.org (Postfix) with ESMTPS id 2158ADDEE3; Tue, 2 Dec 2008 11:19:33 +1100 (EST) Received: from d03relay04.boulder.ibm.com (d03relay04.boulder.ibm.com [9.17.195.106]) by e32.co.us.ibm.com (8.13.1/8.13.1) with ESMTP id mB20I97G024040; Mon, 1 Dec 2008 17:18:09 -0700 Received: from d03av01.boulder.ibm.com (d03av01.boulder.ibm.com [9.17.195.167]) by d03relay04.boulder.ibm.com (8.13.8/8.13.8/NCO v9.1) with ESMTP id mB20JVHr214888; Mon, 1 Dec 2008 17:19:31 -0700 Received: from d03av01.boulder.ibm.com (loopback [127.0.0.1]) by d03av01.boulder.ibm.com (8.12.11.20060308/8.13.3) with ESMTP id mB20JUDx026767; Mon, 1 Dec 2008 17:19:30 -0700 Received: from [9.47.21.78] (carll-linux-desktop.beaverton.ibm.com [9.47.21.78]) by d03av01.boulder.ibm.com (8.12.11.20060308/8.12.11) with ESMTP id mB20JSI9026739 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Mon, 1 Dec 2008 17:19:29 -0700 Subject: [Patch 2/3] Kernel patch, IBM CELL processor OProfile cleanup and restructuring From: Carl Love To: linux-kernel , linuxppc-dev@ozlabs.org, oprofile-list@lists.sourceforge.net, cel , cbe-oss-dev Date: Mon, 01 Dec 2008 16:18:34 -0800 Message-Id: <1228177114.6509.287.camel@carll-linux-desktop> Mime-Version: 1.0 X-Mailer: Evolution 2.12.1 X-BeenThere: linuxppc-dev@ozlabs.org X-Mailman-Version: 2.1.11 Precedence: list List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: linuxppc-dev-bounces+patchwork-incoming=ozlabs.org@ozlabs.org Errors-To: linuxppc-dev-bounces+patchwork-incoming=ozlabs.org@ozlabs.org This patch restructures and cleans up the code a bit to make it easier to add new functionality later. The patch makes no functional changes to the existing code. Signed-off-by: Carl Love Index: Cell_kernel_11_10_2008/arch/powerpc/oprofile/op_model_cell.c =================================================================== --- Cell_kernel_11_10_2008.orig/arch/powerpc/oprofile/op_model_cell.c +++ Cell_kernel_11_10_2008/arch/powerpc/oprofile/op_model_cell.c @@ -40,14 +40,9 @@ #include "../platforms/cell/interrupt.h" #include "cell/pr_util.h" -static void cell_global_stop_spu(void); - -/* - * spu_cycle_reset is the number of cycles between samples. - * This variable is used for SPU profiling and should ONLY be set - * at the beginning of cell_reg_setup; otherwise, it's read-only. - */ -static unsigned int spu_cycle_reset; +#define PPU_PROFILING 0 +#define SPU_PROFILING_CYCLES 1 +#define SPU_PROFILING_EVENTS 2 #define NUM_SPUS_PER_NODE 8 #define SPU_CYCLES_EVENT_NUM 2 /* event number for SPU_CYCLES */ @@ -66,6 +61,14 @@ static unsigned int spu_cycle_reset; #define MAX_SPU_COUNT 0xFFFFFF /* maximum 24 bit LFSR value */ +/* + * spu_cycle_reset is the number of cycles between samples. + * This variable is used for SPU profiling and should ONLY be set + * at the beginning of cell_reg_setup; otherwise, it's read-only. + */ +static unsigned int spu_cycle_reset; +static unsigned int profiling_mode; + struct pmc_cntrl_data { unsigned long vcntr; unsigned long evnts; @@ -122,7 +125,6 @@ static struct { #define GET_INPUT_CONTROL(x) ((x & 0x00000004) >> 2) static DEFINE_PER_CPU(unsigned long[NR_PHYS_CTRS], pmc_values); - static struct pmc_cntrl_data pmc_cntrl[NUM_THREADS][NR_PHYS_CTRS]; /* @@ -165,7 +167,7 @@ static int spu_rtas_token; /* token fo static u32 reset_value[NR_PHYS_CTRS]; static int num_counters; static int oprofile_running; -static DEFINE_SPINLOCK(virt_cntr_lock); +static DEFINE_SPINLOCK(cntr_lock); static u32 ctr_enabled; @@ -367,7 +369,7 @@ static void write_pm_cntrl(int cpu) if (pm_regs.pm_cntrl.stop_at_max == 1) val |= CBE_PM_STOP_AT_MAX; - if (pm_regs.pm_cntrl.trace_mode == 1) + if (pm_regs.pm_cntrl.trace_mode != 0) val |= CBE_PM_TRACE_MODE_SET(pm_regs.pm_cntrl.trace_mode); if (pm_regs.pm_cntrl.freeze == 1) @@ -441,7 +443,7 @@ static void cell_virtual_cntr(unsigned l * not both playing with the counters on the same node. */ - spin_lock_irqsave(&virt_cntr_lock, flags); + spin_lock_irqsave(&cntr_lock, flags); prev_hdw_thread = hdw_thread; @@ -527,7 +529,7 @@ static void cell_virtual_cntr(unsigned l cbe_enable_pm(cpu); } - spin_unlock_irqrestore(&virt_cntr_lock, flags); + spin_unlock_irqrestore(&cntr_lock, flags); mod_timer(&timer_virt_cntr, jiffies + HZ / 10); } @@ -541,44 +543,30 @@ static void start_virt_cntrs(void) add_timer(&timer_virt_cntr); } -/* This function is called once for all cpus combined */ -static int cell_reg_setup(struct op_counter_config *ctr, +static int cell_reg_setup_spu_cycles(struct op_counter_config *ctr, struct op_system_config *sys, int num_ctrs) { - int i, j, cpu; - spu_cycle_reset = 0; - - if (ctr[0].event == SPU_CYCLES_EVENT_NUM) { - spu_cycle_reset = ctr[0].count; - - /* - * Each node will need to make the rtas call to start - * and stop SPU profiling. Get the token once and store it. - */ - spu_rtas_token = rtas_token("ibm,cbe-spu-perftools"); - - if (unlikely(spu_rtas_token == RTAS_UNKNOWN_SERVICE)) { - printk(KERN_ERR - "%s: rtas token ibm,cbe-spu-perftools unknown\n", - __func__); - return -EIO; - } - } - - pm_rtas_token = rtas_token("ibm,cbe-perftools"); + spu_cycle_reset = ctr[0].count; /* - * For all events excetp PPU CYCLEs, each node will need to make - * the rtas cbe-perftools call to setup and reset the debug bus. - * Make the token lookup call once and store it in the global - * variable pm_rtas_token. + * Each node will need to make the rtas call to start + * and stop SPU profiling. Get the token once and store it. */ - if (unlikely(pm_rtas_token == RTAS_UNKNOWN_SERVICE)) { + spu_rtas_token = rtas_token("ibm,cbe-spu-perftools"); + + if (unlikely(spu_rtas_token == RTAS_UNKNOWN_SERVICE)) { printk(KERN_ERR - "%s: rtas token ibm,cbe-perftools unknown\n", + "%s: rtas token ibm,cbe-spu-perftools unknown\n", __func__); return -EIO; } + return 0; +} + +static int cell_reg_setup_ppu(struct op_counter_config *ctr, + struct op_system_config *sys, int num_ctrs) +{ + int i, j, cpu; num_counters = num_ctrs; @@ -665,6 +653,41 @@ static int cell_reg_setup(struct op_coun } +/* This function is called once for all cpus combined */ +static int cell_reg_setup(struct op_counter_config *ctr, + struct op_system_config *sys, int num_ctrs) +{ + int ret; + + spu_cycle_reset = 0; + + /* + * For all events except PPU CYCLEs, each node will need to make + * the rtas cbe-perftools call to setup and reset the debug bus. + * Make the token lookup call once and store it in the global + * variable pm_rtas_token. + */ + pm_rtas_token = rtas_token("ibm,cbe-perftools"); + + if (unlikely(pm_rtas_token == RTAS_UNKNOWN_SERVICE)) { + printk(KERN_ERR + "%s: rtas token ibm,cbe-perftools unknown\n", + __func__); + return -EIO; + } + + if (ctr[0].event == SPU_CYCLES_EVENT_NUM) { + profiling_mode = SPU_PROFILING_CYCLES; + ret = cell_reg_setup_spu_cycles(ctr, sys, num_ctrs); + } else { + profiling_mode = PPU_PROFILING; + ret = cell_reg_setup_ppu(ctr, sys, num_ctrs); + } + + return ret; +} + + /* This function is called once for each cpu */ static int cell_cpu_setup(struct op_counter_config *cntr) @@ -673,7 +696,11 @@ static int cell_cpu_setup(struct op_coun u32 num_enabled = 0; int i; - if (spu_cycle_reset) + /* Cycle based SPU profiling does not use the performance + * counters. The trace array is configured to collect + * the data. + */ + if (profiling_mode == SPU_PROFILING_CYCLES) return 0; /* There is one performance monitor per processor chip (i.e. node), @@ -686,7 +713,6 @@ static int cell_cpu_setup(struct op_coun cbe_disable_pm(cpu); cbe_disable_pm_interrupts(cpu); - cbe_write_pm(cpu, pm_interval, 0); cbe_write_pm(cpu, pm_start_stop, 0); cbe_write_pm(cpu, group_control, pm_regs.group_control); cbe_write_pm(cpu, debug_bus_control, pm_regs.debug_bus_control); @@ -885,7 +911,94 @@ static struct notifier_block cpu_freq_no }; #endif -static int cell_global_start_spu(struct op_counter_config *ctr) +/* + * Note the generic OProfile stop calls do not support returning + * an error on stop. Hence, will not return an error if the FW + * calls fail on stop. Failure to reset the debug bus is not an issue. + * Failure to disable the SPU profiling is not an issue. The FW calls + * to enable the performance counters and debug bus will work even if + * the hardware was not cleanly reset. + */ +static void cell_global_stop_spu_cycles(void) +{ + int subfunc, rtn_value; + unsigned int lfsr_value; + int cpu; + + oprofile_running = 0; + +#ifdef CONFIG_CPU_FREQ + cpufreq_unregister_notifier(&cpu_freq_notifier_block, + CPUFREQ_TRANSITION_NOTIFIER); +#endif + + for_each_online_cpu(cpu) { + if (cbe_get_hw_thread_id(cpu)) + continue; + + subfunc = 3; /* + * 2 - activate SPU tracing, + * 3 - deactivate + */ + lfsr_value = 0x8f100000; + + rtn_value = rtas_call(spu_rtas_token, 3, 1, NULL, + subfunc, cbe_cpu_to_node(cpu), + lfsr_value); + + if (unlikely(rtn_value != 0)) { + printk(KERN_ERR + "%s: rtas call ibm,cbe-spu-perftools " \ + "failed, return = %d\n", + __func__, rtn_value); + } + + /* Deactivate the signals */ + pm_rtas_reset_signals(cbe_cpu_to_node(cpu)); + } + + if (profiling_mode == SPU_PROFILING_CYCLES) + stop_spu_profiling_cycles(); +} + +static void cell_global_stop_ppu(void) +{ + int cpu; + + /* + * This routine will be called once for the system. + * There is one performance monitor per node, so we + * only need to perform this function once per node. + */ + del_timer_sync(&timer_virt_cntr); + oprofile_running = 0; + smp_wmb(); + + for_each_online_cpu(cpu) { + if (cbe_get_hw_thread_id(cpu)) + continue; + + cbe_sync_irq(cbe_cpu_to_node(cpu)); + /* Stop the counters */ + cbe_disable_pm(cpu); + + /* Deactivate the signals */ + pm_rtas_reset_signals(cbe_cpu_to_node(cpu)); + + /* Deactivate interrupts */ + cbe_disable_pm_interrupts(cpu); + } +} + +static void cell_global_stop(void) +{ + if (profiling_mode == PPU_PROFILING) + cell_global_stop_ppu(); + else + cell_global_stop_spu_cycles(); +} + +static int cell_global_start_spu_cycles(struct op_counter_config *ctr) { int subfunc; unsigned int lfsr_value; @@ -955,14 +1068,14 @@ static int cell_global_start_spu(struct if (unlikely(ret != 0)) { printk(KERN_ERR - "%s: rtas call ibm,cbe-spu-perftools failed, return = %d\n", - __func__, ret); + "%s: rtas call ibm,cbe-spu-perftools failed, " \ + "return = %d\n", __func__, ret); rtas_error = -EIO; goto out; } } - rtas_error = start_spu_profiling(spu_cycle_reset); + rtas_error = start_spu_profiling_cycles(spu_cycle_reset); if (rtas_error) goto out_stop; @@ -970,7 +1083,7 @@ static int cell_global_start_spu(struct return 0; out_stop: - cell_global_stop_spu(); /* clean up the PMU/debug bus */ + cell_global_stop_spu_cycles(); /* clean up the PMU/debug bus */ out: return rtas_error; } @@ -1024,99 +1137,15 @@ static int cell_global_start_ppu(struct static int cell_global_start(struct op_counter_config *ctr) { - if (spu_cycle_reset) - return cell_global_start_spu(ctr); + if (profiling_mode == SPU_PROFILING_CYCLES) + return cell_global_start_spu_cycles(ctr); else return cell_global_start_ppu(ctr); } -/* - * Note the generic OProfile stop calls do not support returning - * an error on stop. Hence, will not return an error if the FW - * calls fail on stop. Failure to reset the debug bus is not an issue. - * Failure to disable the SPU profiling is not an issue. The FW calls - * to enable the performance counters and debug bus will work even if - * the hardware was not cleanly reset. - */ -static void cell_global_stop_spu(void) -{ - int subfunc, rtn_value; - unsigned int lfsr_value; - int cpu; - - oprofile_running = 0; -#ifdef CONFIG_CPU_FREQ - cpufreq_unregister_notifier(&cpu_freq_notifier_block, - CPUFREQ_TRANSITION_NOTIFIER); -#endif - - for_each_online_cpu(cpu) { - if (cbe_get_hw_thread_id(cpu)) - continue; - - subfunc = 3; /* - * 2 - activate SPU tracing, - * 3 - deactivate - */ - lfsr_value = 0x8f100000; - - rtn_value = rtas_call(spu_rtas_token, 3, 1, NULL, - subfunc, cbe_cpu_to_node(cpu), - lfsr_value); - - if (unlikely(rtn_value != 0)) { - printk(KERN_ERR - "%s: rtas call ibm,cbe-spu-perftools failed, return = %d\n", - __func__, rtn_value); - } - - /* Deactivate the signals */ - pm_rtas_reset_signals(cbe_cpu_to_node(cpu)); - } - - stop_spu_profiling(); -} - -static void cell_global_stop_ppu(void) -{ - int cpu; - - /* - * This routine will be called once for the system. - * There is one performance monitor per node, so we - * only need to perform this function once per node. - */ - del_timer_sync(&timer_virt_cntr); - oprofile_running = 0; - smp_wmb(); - - for_each_online_cpu(cpu) { - if (cbe_get_hw_thread_id(cpu)) - continue; - - cbe_sync_irq(cbe_cpu_to_node(cpu)); - /* Stop the counters */ - cbe_disable_pm(cpu); - - /* Deactivate the signals */ - pm_rtas_reset_signals(cbe_cpu_to_node(cpu)); - - /* Deactivate interrupts */ - cbe_disable_pm_interrupts(cpu); - } -} - -static void cell_global_stop(void) -{ - if (spu_cycle_reset) - cell_global_stop_spu(); - else - cell_global_stop_ppu(); -} - -static void cell_handle_interrupt(struct pt_regs *regs, - struct op_counter_config *ctr) +static void cell_handle_interrupt_ppu(struct pt_regs *regs, + struct op_counter_config *ctr) { u32 cpu; u64 pc; @@ -1132,7 +1161,7 @@ static void cell_handle_interrupt(struct * routine are not running at the same time. See the * cell_virtual_cntr() routine for additional comments. */ - spin_lock_irqsave(&virt_cntr_lock, flags); + spin_lock_irqsave(&cntr_lock, flags); /* * Need to disable and reenable the performance counters @@ -1185,7 +1214,14 @@ static void cell_handle_interrupt(struct */ cbe_enable_pm(cpu); } - spin_unlock_irqrestore(&virt_cntr_lock, flags); + spin_unlock_irqrestore(&cntr_lock, flags); +} + +static void cell_handle_interrupt(struct pt_regs *regs, + struct op_counter_config *ctr) +{ + if (profiling_mode == PPU_PROFILING) + cell_handle_interrupt_ppu(regs, ctr); } /* @@ -1195,7 +1231,8 @@ static void cell_handle_interrupt(struct */ static int cell_sync_start(void) { - if (spu_cycle_reset) + if ((profiling_mode == SPU_PROFILING_CYCLES) || + (profiling_mode == SPU_PROFILING_EVENTS)) return spu_sync_start(); else return DO_GENERIC_SYNC; @@ -1203,7 +1240,8 @@ static int cell_sync_start(void) static int cell_sync_stop(void) { - if (spu_cycle_reset) + if ((profiling_mode == SPU_PROFILING_CYCLES) || + (profiling_mode == SPU_PROFILING_EVENTS)) return spu_sync_stop(); else return 1; Index: Cell_kernel_11_10_2008/arch/powerpc/oprofile/cell/spu_profiler.c =================================================================== --- Cell_kernel_11_10_2008.orig/arch/powerpc/oprofile/cell/spu_profiler.c +++ Cell_kernel_11_10_2008/arch/powerpc/oprofile/cell/spu_profiler.c @@ -31,8 +31,8 @@ static unsigned int profiling_interval; #define SPU_PC_MASK 0xFFFF -static DEFINE_SPINLOCK(sample_array_lock); -unsigned long sample_array_lock_flags; +static DEFINE_SPINLOCK(oprof_spu_smpl_arry_lck); +unsigned long oprof_spu_smpl_arry_lck_flags; void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset) { @@ -145,13 +145,13 @@ static enum hrtimer_restart profile_spus * sample array must be loaded and then processed for a given * cpu. The sample array is not per cpu. */ - spin_lock_irqsave(&sample_array_lock, - sample_array_lock_flags); + spin_lock_irqsave(&oprof_spu_smpl_arry_lck, + oprof_spu_smpl_arry_lck_flags); num_samples = cell_spu_pc_collection(cpu); if (num_samples == 0) { - spin_unlock_irqrestore(&sample_array_lock, - sample_array_lock_flags); + spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck, + oprof_spu_smpl_arry_lck_flags); continue; } @@ -162,8 +162,8 @@ static enum hrtimer_restart profile_spus num_samples); } - spin_unlock_irqrestore(&sample_array_lock, - sample_array_lock_flags); + spin_unlock_irqrestore(&oprof_spu_smpl_arry_lck, + oprof_spu_smpl_arry_lck_flags); } smp_wmb(); /* insure spu event buffer updates are written */ @@ -182,13 +182,13 @@ static enum hrtimer_restart profile_spus static struct hrtimer timer; /* - * Entry point for SPU profiling. + * Entry point for SPU cycle profiling. * NOTE: SPU profiling is done system-wide, not per-CPU. * * cycles_reset is the count value specified by the user when * setting up OProfile to count SPU_CYCLES. */ -int start_spu_profiling(unsigned int cycles_reset) +int start_spu_profiling_cycles(unsigned int cycles_reset) { ktime_t kt; @@ -212,10 +212,10 @@ int start_spu_profiling(unsigned int cyc return 0; } -void stop_spu_profiling(void) +void stop_spu_profiling_cycles(void) { spu_prof_running = 0; hrtimer_cancel(&timer); kfree(samples); - pr_debug("SPU_PROF: stop_spu_profiling issued\n"); + pr_debug("SPU_PROF: stop_spu_profiling_cycles issued\n"); }