Merge tag 'csky-for-linus-5.3-rc1' of git://github.com/c-sky/csky-linux

Pull arch/csky pupdates from Guo Ren:
 "This round of csky subsystem gives two features (ASID algorithm
  update, Perf pmu record support) and some fixups.

  ASID updates:
   - Revert mmu ASID mechanism
   - Add new asid lib code from arm
   - Use generic asid algorithm to implement switch_mm
   - Improve tlb operation with help of asid

  Perf pmu record support:
   - Init pmu as a device
   - Add count-width property for csky pmu
   - Add pmu interrupt support
   - Fix perf record in kernel/user space
   - dt-bindings: Add csky PMU bindings

  Fixes:
   - Fixup no panic in kernel for some traps
   - Fixup some error count in 810 & 860.
   - Fixup abiv1 memset error"

* tag 'csky-for-linus-5.3-rc1' of git://github.com/c-sky/csky-linux:
  csky: Fixup abiv1 memset error
  csky: Improve tlb operation with help of asid
  csky: Use generic asid algorithm to implement switch_mm
  csky: Add new asid lib code from arm
  csky: Revert mmu ASID mechanism
  dt-bindings: csky: Add csky PMU bindings
  dt-bindings: interrupt-controller: Update csky mpintc
  csky: Fixup some error count in 810 & 860.
  csky: Fix perf record in kernel/user space
  csky: Add pmu interrupt support
  csky: Add count-width property for csky pmu
  csky: Init pmu as a device
  csky: Fixup no panic in kernel for some traps
  csky: Select intc & timer drivers
This commit is contained in:
Linus Torvalds
2019-07-19 12:15:33 -07:00
20 changed files with 890 additions and 358 deletions

View File

@@ -9,17 +9,44 @@
#include <linux/platform_device.h>
#define CSKY_PMU_MAX_EVENTS 32
#define DEFAULT_COUNT_WIDTH 48
#define HPCR "<0, 0x0>" /* PMU Control reg */
#define HPCNTENR "<0, 0x4>" /* Count Enable reg */
#define HPCR "<0, 0x0>" /* PMU Control reg */
#define HPSPR "<0, 0x1>" /* Start PC reg */
#define HPEPR "<0, 0x2>" /* End PC reg */
#define HPSIR "<0, 0x3>" /* Soft Counter reg */
#define HPCNTENR "<0, 0x4>" /* Count Enable reg */
#define HPINTENR "<0, 0x5>" /* Interrupt Enable reg */
#define HPOFSR "<0, 0x6>" /* Interrupt Status reg */
/* The events for a given PMU register set. */
struct pmu_hw_events {
/*
* The events that are active on the PMU for the given index.
*/
struct perf_event *events[CSKY_PMU_MAX_EVENTS];
/*
* A 1 bit for an index indicates that the counter is being used for
* an event. A 0 means that the counter can be used.
*/
unsigned long used_mask[BITS_TO_LONGS(CSKY_PMU_MAX_EVENTS)];
};
static uint64_t (*hw_raw_read_mapping[CSKY_PMU_MAX_EVENTS])(void);
static void (*hw_raw_write_mapping[CSKY_PMU_MAX_EVENTS])(uint64_t val);
struct csky_pmu_t {
struct pmu pmu;
uint32_t hpcr;
static struct csky_pmu_t {
struct pmu pmu;
struct pmu_hw_events __percpu *hw_events;
struct platform_device *plat_device;
uint32_t count_width;
uint32_t hpcr;
u64 max_period;
} csky_pmu;
static int csky_pmu_irq;
#define to_csky_pmu(p) (container_of(p, struct csky_pmu, pmu))
#define cprgr(reg) \
({ \
@@ -701,6 +728,20 @@ static const int csky_pmu_hw_map[PERF_COUNT_HW_MAX] = {
#define CACHE_OP_UNSUPPORTED 0xffff
static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
[C(L1D)] = {
#ifdef CONFIG_CPU_CK810
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = 0x5,
[C(RESULT_MISS)] = 0x6,
},
#else
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = 0x14,
[C(RESULT_MISS)] = 0x15,
@@ -710,9 +751,10 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
[C(RESULT_MISS)] = 0x17,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = 0x5,
[C(RESULT_MISS)] = 0x6,
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
#endif
},
[C(L1I)] = {
[C(OP_READ)] = {
@@ -729,6 +771,20 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
},
},
[C(LL)] = {
#ifdef CONFIG_CPU_CK810
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = 0x7,
[C(RESULT_MISS)] = 0x8,
},
#else
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = 0x18,
[C(RESULT_MISS)] = 0x19,
@@ -738,29 +794,48 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
[C(RESULT_MISS)] = 0x1b,
},
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = 0x7,
[C(RESULT_MISS)] = 0x8,
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
#endif
},
[C(DTLB)] = {
#ifdef CONFIG_CPU_CK810
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = 0x5,
[C(RESULT_MISS)] = 0xb,
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
#else
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = 0x14,
[C(RESULT_MISS)] = 0xb,
},
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = 0x16,
[C(RESULT_MISS)] = 0xb,
},
#endif
[C(OP_PREFETCH)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
},
[C(ITLB)] = {
#ifdef CONFIG_CPU_CK810
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
},
#else
[C(OP_READ)] = {
[C(RESULT_ACCESS)] = 0x3,
[C(RESULT_MISS)] = 0xa,
},
#endif
[C(OP_WRITE)] = {
[C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
[C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
@@ -800,11 +875,57 @@ static const int csky_pmu_cache_map[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
},
};
int csky_pmu_event_set_period(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
s64 left = local64_read(&hwc->period_left);
s64 period = hwc->sample_period;
int ret = 0;
if (unlikely(left <= -period)) {
left = period;
local64_set(&hwc->period_left, left);
hwc->last_period = period;
ret = 1;
}
if (unlikely(left <= 0)) {
left += period;
local64_set(&hwc->period_left, left);
hwc->last_period = period;
ret = 1;
}
if (left > (s64)csky_pmu.max_period)
left = csky_pmu.max_period;
/*
* The hw event starts counting from this event offset,
* mark it to be able to extract future "deltas":
*/
local64_set(&hwc->prev_count, (u64)(-left));
if (hw_raw_write_mapping[hwc->idx] != NULL)
hw_raw_write_mapping[hwc->idx]((u64)(-left) &
csky_pmu.max_period);
cpwcr(HPOFSR, ~BIT(hwc->idx) & cprcr(HPOFSR));
perf_event_update_userpage(event);
return ret;
}
static void csky_perf_event_update(struct perf_event *event,
struct hw_perf_event *hwc)
{
uint64_t prev_raw_count = local64_read(&hwc->prev_count);
uint64_t new_raw_count = hw_raw_read_mapping[hwc->idx]();
/*
* Sign extend count value to 64bit, otherwise delta calculation
* would be incorrect when overflow occurs.
*/
uint64_t new_raw_count = sign_extend64(
hw_raw_read_mapping[hwc->idx](), csky_pmu.count_width - 1);
int64_t delta = new_raw_count - prev_raw_count;
/*
@@ -816,6 +937,11 @@ static void csky_perf_event_update(struct perf_event *event,
local64_sub(delta, &hwc->period_left);
}
static void csky_pmu_reset(void *info)
{
cpwcr(HPCR, BIT(31) | BIT(30) | BIT(1));
}
static void csky_pmu_read(struct perf_event *event)
{
csky_perf_event_update(event, &event->hw);
@@ -844,6 +970,30 @@ static int csky_pmu_event_init(struct perf_event *event)
struct hw_perf_event *hwc = &event->hw;
int ret;
switch (event->attr.type) {
case PERF_TYPE_HARDWARE:
if (event->attr.config >= PERF_COUNT_HW_MAX)
return -ENOENT;
ret = csky_pmu_hw_map[event->attr.config];
if (ret == HW_OP_UNSUPPORTED)
return -ENOENT;
hwc->idx = ret;
break;
case PERF_TYPE_HW_CACHE:
ret = csky_pmu_cache_event(event->attr.config);
if (ret == CACHE_OP_UNSUPPORTED)
return -ENOENT;
hwc->idx = ret;
break;
case PERF_TYPE_RAW:
if (hw_raw_read_mapping[event->attr.config] == NULL)
return -ENOENT;
hwc->idx = event->attr.config;
break;
default:
return -ENOENT;
}
if (event->attr.exclude_user)
csky_pmu.hpcr = BIT(2);
else if (event->attr.exclude_kernel)
@@ -853,29 +1003,7 @@ static int csky_pmu_event_init(struct perf_event *event)
csky_pmu.hpcr |= BIT(1) | BIT(0);
switch (event->attr.type) {
case PERF_TYPE_HARDWARE:
if (event->attr.config >= PERF_COUNT_HW_MAX)
return -ENOENT;
ret = csky_pmu_hw_map[event->attr.config];
if (ret == HW_OP_UNSUPPORTED)
return -ENOENT;
hwc->idx = ret;
return 0;
case PERF_TYPE_HW_CACHE:
ret = csky_pmu_cache_event(event->attr.config);
if (ret == CACHE_OP_UNSUPPORTED)
return -ENOENT;
hwc->idx = ret;
return 0;
case PERF_TYPE_RAW:
if (hw_raw_read_mapping[event->attr.config] == NULL)
return -ENOENT;
hwc->idx = event->attr.config;
return 0;
default:
return -ENOENT;
}
return 0;
}
/* starts all counters */
@@ -892,6 +1020,7 @@ static void csky_pmu_disable(struct pmu *pmu)
static void csky_pmu_start(struct perf_event *event, int flags)
{
unsigned long flg;
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
@@ -903,16 +1032,34 @@ static void csky_pmu_start(struct perf_event *event, int flags)
hwc->state = 0;
csky_pmu_event_set_period(event);
local_irq_save(flg);
cpwcr(HPINTENR, BIT(idx) | cprcr(HPINTENR));
cpwcr(HPCNTENR, BIT(idx) | cprcr(HPCNTENR));
local_irq_restore(flg);
}
static void csky_pmu_stop_event(struct perf_event *event)
{
unsigned long flg;
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
local_irq_save(flg);
cpwcr(HPINTENR, ~BIT(idx) & cprcr(HPINTENR));
cpwcr(HPCNTENR, ~BIT(idx) & cprcr(HPCNTENR));
local_irq_restore(flg);
}
static void csky_pmu_stop(struct perf_event *event, int flags)
{
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
if (!(event->hw.state & PERF_HES_STOPPED)) {
cpwcr(HPCNTENR, ~BIT(idx) & cprcr(HPCNTENR));
csky_pmu_stop_event(event);
event->hw.state |= PERF_HES_STOPPED;
}
@@ -925,22 +1072,26 @@ static void csky_pmu_stop(struct perf_event *event, int flags)
static void csky_pmu_del(struct perf_event *event, int flags)
{
struct pmu_hw_events *hw_events = this_cpu_ptr(csky_pmu.hw_events);
struct hw_perf_event *hwc = &event->hw;
csky_pmu_stop(event, PERF_EF_UPDATE);
hw_events->events[hwc->idx] = NULL;
perf_event_update_userpage(event);
}
/* allocate hardware counter and optionally start counting */
static int csky_pmu_add(struct perf_event *event, int flags)
{
struct pmu_hw_events *hw_events = this_cpu_ptr(csky_pmu.hw_events);
struct hw_perf_event *hwc = &event->hw;
local64_set(&hwc->prev_count, 0);
if (hw_raw_write_mapping[hwc->idx] != NULL)
hw_raw_write_mapping[hwc->idx](0);
hw_events->events[hwc->idx] = event;
hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
if (flags & PERF_EF_START)
csky_pmu_start(event, PERF_EF_RELOAD);
@@ -949,8 +1100,110 @@ static int csky_pmu_add(struct perf_event *event, int flags)
return 0;
}
int __init init_hw_perf_events(void)
static irqreturn_t csky_pmu_handle_irq(int irq_num, void *dev)
{
struct perf_sample_data data;
struct pmu_hw_events *cpuc = this_cpu_ptr(csky_pmu.hw_events);
struct pt_regs *regs;
int idx;
/*
* Did an overflow occur?
*/
if (!cprcr(HPOFSR))
return IRQ_NONE;
/*
* Handle the counter(s) overflow(s)
*/
regs = get_irq_regs();
csky_pmu_disable(&csky_pmu.pmu);
for (idx = 0; idx < CSKY_PMU_MAX_EVENTS; ++idx) {
struct perf_event *event = cpuc->events[idx];
struct hw_perf_event *hwc;
/* Ignore if we don't have an event. */
if (!event)
continue;
/*
* We have a single interrupt for all counters. Check that
* each counter has overflowed before we process it.
*/
if (!(cprcr(HPOFSR) & BIT(idx)))
continue;
hwc = &event->hw;
csky_perf_event_update(event, &event->hw);
perf_sample_data_init(&data, 0, hwc->last_period);
csky_pmu_event_set_period(event);
if (perf_event_overflow(event, &data, regs))
csky_pmu_stop_event(event);
}
csky_pmu_enable(&csky_pmu.pmu);
/*
* Handle the pending perf events.
*
* Note: this call *must* be run with interrupts disabled. For
* platforms that can have the PMU interrupts raised as an NMI, this
* will not work.
*/
irq_work_run();
return IRQ_HANDLED;
}
static int csky_pmu_request_irq(irq_handler_t handler)
{
int err, irqs;
struct platform_device *pmu_device = csky_pmu.plat_device;
if (!pmu_device)
return -ENODEV;
irqs = min(pmu_device->num_resources, num_possible_cpus());
if (irqs < 1) {
pr_err("no irqs for PMUs defined\n");
return -ENODEV;
}
csky_pmu_irq = platform_get_irq(pmu_device, 0);
if (csky_pmu_irq < 0)
return -ENODEV;
err = request_percpu_irq(csky_pmu_irq, handler, "csky-pmu",
this_cpu_ptr(csky_pmu.hw_events));
if (err) {
pr_err("unable to request IRQ%d for CSKY PMU counters\n",
csky_pmu_irq);
return err;
}
return 0;
}
static void csky_pmu_free_irq(void)
{
int irq;
struct platform_device *pmu_device = csky_pmu.plat_device;
irq = platform_get_irq(pmu_device, 0);
if (irq >= 0)
free_percpu_irq(irq, this_cpu_ptr(csky_pmu.hw_events));
}
int init_hw_perf_events(void)
{
csky_pmu.hw_events = alloc_percpu_gfp(struct pmu_hw_events,
GFP_KERNEL);
if (!csky_pmu.hw_events) {
pr_info("failed to allocate per-cpu PMU data.\n");
return -ENOMEM;
}
csky_pmu.pmu = (struct pmu) {
.pmu_enable = csky_pmu_enable,
.pmu_disable = csky_pmu_disable,
@@ -1022,10 +1275,97 @@ int __init init_hw_perf_events(void)
hw_raw_write_mapping[0x1a] = csky_pmu_write_l2wac;
hw_raw_write_mapping[0x1b] = csky_pmu_write_l2wmc;
csky_pmu.pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
cpwcr(HPCR, BIT(31) | BIT(30) | BIT(1));
return perf_pmu_register(&csky_pmu.pmu, "cpu", PERF_TYPE_RAW);
return 0;
}
arch_initcall(init_hw_perf_events);
static int csky_pmu_starting_cpu(unsigned int cpu)
{
enable_percpu_irq(csky_pmu_irq, 0);
return 0;
}
static int csky_pmu_dying_cpu(unsigned int cpu)
{
disable_percpu_irq(csky_pmu_irq);
return 0;
}
int csky_pmu_device_probe(struct platform_device *pdev,
const struct of_device_id *of_table)
{
struct device_node *node = pdev->dev.of_node;
int ret;
ret = init_hw_perf_events();
if (ret) {
pr_notice("[perf] failed to probe PMU!\n");
return ret;
}
if (of_property_read_u32(node, "count-width",
&csky_pmu.count_width)) {
csky_pmu.count_width = DEFAULT_COUNT_WIDTH;
}
csky_pmu.max_period = BIT(csky_pmu.count_width) - 1;
csky_pmu.plat_device = pdev;
/* Ensure the PMU has sane values out of reset. */
on_each_cpu(csky_pmu_reset, &csky_pmu, 1);
ret = csky_pmu_request_irq(csky_pmu_handle_irq);
if (ret) {
csky_pmu.pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
pr_notice("[perf] PMU request irq fail!\n");
}
ret = cpuhp_setup_state(CPUHP_AP_PERF_ONLINE, "AP_PERF_ONLINE",
csky_pmu_starting_cpu,
csky_pmu_dying_cpu);
if (ret) {
csky_pmu_free_irq();
free_percpu(csky_pmu.hw_events);
return ret;
}
ret = perf_pmu_register(&csky_pmu.pmu, "cpu", PERF_TYPE_RAW);
if (ret) {
csky_pmu_free_irq();
free_percpu(csky_pmu.hw_events);
}
return ret;
}
const static struct of_device_id csky_pmu_of_device_ids[] = {
{.compatible = "csky,csky-pmu"},
{},
};
static int csky_pmu_dev_probe(struct platform_device *pdev)
{
return csky_pmu_device_probe(pdev, csky_pmu_of_device_ids);
}
static struct platform_driver csky_pmu_driver = {
.driver = {
.name = "csky-pmu",
.of_match_table = csky_pmu_of_device_ids,
},
.probe = csky_pmu_dev_probe,
};
static int __init csky_pmu_probe(void)
{
int ret;
ret = platform_driver_register(&csky_pmu_driver);
if (ret)
pr_notice("[perf] PMU initialization failed\n");
else
pr_notice("[perf] PMU initialization done\n");
return ret;
}
device_initcall(csky_pmu_probe);

View File

@@ -212,8 +212,6 @@ void csky_start_secondary(void)
TLBMISS_HANDLER_SETUP_PGD(swapper_pg_dir);
TLBMISS_HANDLER_SETUP_PGD_KERNEL(swapper_pg_dir);
asid_cache(smp_processor_id()) = ASID_FIRST_VERSION;
#ifdef CONFIG_CPU_HAS_FPU
init_fpu();
#endif

View File

@@ -120,6 +120,7 @@ asmlinkage void trap_c(struct pt_regs *regs)
switch (vector) {
case VEC_ZERODIV:
die_if_kernel("Kernel mode ZERO DIV", regs, vector);
sig = SIGFPE;
break;
/* ptrace */
@@ -128,6 +129,7 @@ asmlinkage void trap_c(struct pt_regs *regs)
sig = SIGTRAP;
break;
case VEC_ILLEGAL:
die_if_kernel("Kernel mode ILLEGAL", regs, vector);
#ifndef CONFIG_CPU_NO_USER_BKPT
if (*(uint16_t *)instruction_pointer(regs) != USR_BKPT)
#endif
@@ -139,6 +141,7 @@ asmlinkage void trap_c(struct pt_regs *regs)
case VEC_TRAP1:
/* jtagserver breakpoint */
case VEC_BREAKPOINT:
die_if_kernel("Kernel mode BKPT", regs, vector);
info.si_code = TRAP_BRKPT;
sig = SIGTRAP;
break;
@@ -150,8 +153,10 @@ asmlinkage void trap_c(struct pt_regs *regs)
#endif
#ifdef CONFIG_CPU_HAS_FPU
case VEC_FPE:
die_if_kernel("Kernel mode FPE", regs, vector);
return fpu_fpe(regs);
case VEC_PRIV:
die_if_kernel("Kernel mode PRIV", regs, vector);
if (fpu_libc_helper(regs))
return;
#endif