[PATCH v7 2/7] hwtracing: Add trace function support for HiSilicon PCIe Tune and Trace device

Yicong Yang yangyicong at huawei.com
Tue Apr 12 05:52:28 PDT 2022


On 2022/4/12 16:39, John Garry wrote:
>>>> +static int hisi_ptt_alloc_trace_buf(struct hisi_ptt *hisi_ptt)
>>>> +{
>>>> +    struct hisi_ptt_trace_ctrl *ctrl = &hisi_ptt->trace_ctrl;
>>>> +    struct device *dev = &hisi_ptt->pdev->dev;
>>>> +    int i;
>>>> +
>>>> +    hisi_ptt->trace_ctrl.buf_index = 0;
>>>> +
>>>> +    /* If the trace buffer has already been allocated, zero it. */
>>>
>>> I am not sure why this is not called from the probe
>>>
>>
>> The buffer allocation is done when necessary as driver will probe the device on booting but
>> the user may never use it. In this condition it's a waste of memory if we allocate the buffers
>> in probe. Currently we'll allocate 16M memory for 4 buffers.
>>
> 
> But that's just not how we do things. We setup the driver fully to be used in the probe. If the user cannot really afford the memory then he/she should not load the driver.
> 

I think in most time user don't need to manually load the driver because of udev.

> In addition, this driver would be used in a machine which will have gigbytes of memory, so I think that the memory mentioned here is relatively insignificant.
> 

ok, yes it'll be used in the server so the memory will be enough in most time. considering this maybe it's proper to move the buffer allocation and the configuration to the driver probe then.

>> So this function is called every time before we start trace. In the first time it will allocate
>> the DMA buffers and it the other times it just zero the buffers to clear the data of last time.
>>
>>>> +    if (ctrl->trace_buf) {
>>>> +        for (i = 0; i < HISI_PTT_TRACE_BUF_CNT; i++)
>>>> +            memset(ctrl->trace_buf[i].addr, 0, HISI_PTT_TRACE_BUF_SIZE);
>>>> +        return 0;
>>>> +    }
>>>> +
>>>> +    ctrl->trace_buf = devm_kcalloc(dev, HISI_PTT_TRACE_BUF_CNT,
>>>> +                       sizeof(struct hisi_ptt_dma_buffer), GFP_KERNEL);
>>>
>>> sizeof(*ctrl->trace_buf) may be better
>>>
>>
>> ok.
>>
>>>> +    if (!ctrl->trace_buf)
>>>> +        return -ENOMEM;
>>>> +
>>>> +    for (i = 0; i < HISI_PTT_TRACE_BUF_CNT; ++i) {
>>>> +        ctrl->trace_buf[i].addr = dmam_alloc_coherent(dev, HISI_PTT_TRACE_BUF_SIZE,
>>>> +                                 &ctrl->trace_buf[i].dma,
>>>> +                                 GFP_KERNEL);
>>>> +        if (!ctrl->trace_buf[i].addr) {
>>>> +            hisi_ptt_free_trace_buf(hisi_ptt);
>>>> +            return -ENOMEM;
>>>> +        }
>>>> +    }
>>>> +
>>>> +    return 0;
>>>> +}
>>>> +
>>>> +static void hisi_ptt_trace_end(struct hisi_ptt *hisi_ptt)
>>>> +{
>>>> +    writel(0, hisi_ptt->iobase + HISI_PTT_TRACE_CTRL);
>>>> +    hisi_ptt->trace_ctrl.started = false;
>>>> +}
>>>> +
>>>> +static int hisi_ptt_trace_start(struct hisi_ptt *hisi_ptt)
>>>> +{
>>>> +    struct hisi_ptt_trace_ctrl *ctrl = &hisi_ptt->trace_ctrl;
>>>> +    u32 val;
>>>> +    int i;
>>>> +
>>>> +    /* Check device idle before start trace */
>>>> +    if (!hisi_ptt_wait_trace_hw_idle(hisi_ptt)) {
>>>> +        pci_err(hisi_ptt->pdev, "Failed to start trace, the device is still busy\n");
>>>> +        return -EBUSY;
>>>> +    }
>>>> +
>>>> +    ctrl->started = true;
>>>> +
>>>> +    /* Reset the DMA before start tracing */
>>>> +    val = readl(hisi_ptt->iobase + HISI_PTT_TRACE_CTRL);
>>>> +    val |= HISI_PTT_TRACE_CTRL_RST;
>>>> +    writel(val, hisi_ptt->iobase + HISI_PTT_TRACE_CTRL);
>>>> +
>>>> +    hisi_ptt_wait_dma_reset_done(hisi_ptt);
>>>> +
>>>> +    val = readl(hisi_ptt->iobase + HISI_PTT_TRACE_CTRL);
>>>> +    val &= ~HISI_PTT_TRACE_CTRL_RST;
>>>> +    writel(val, hisi_ptt->iobase + HISI_PTT_TRACE_CTRL);
>>>> +
>>>> +    /* Clear the interrupt status */
>>>> +    writel(HISI_PTT_TRACE_INT_STAT_MASK, hisi_ptt->iobase + HISI_PTT_TRACE_INT_STAT);
>>>> +    writel(0, hisi_ptt->iobase + HISI_PTT_TRACE_INT_MASK);
>>>> +
>>>> +    /* Configure the trace DMA buffer */
>>>
>>> I am not sure why this sort of thing is done outside probing
>>>
> 
> ...
> 
>>>> +
>>>> +    val = FIELD_GET(HISI_PTT_PMU_DIRECTION_MASK, event->attr.config);
>>>> +    ret = hisi_ptt_trace_valid_config_onehot(val, hisi_ptt_trace_available_direction,
>>>
>>> how about put all those arrays in hisi_ptt_trace_valid_config_onehot() and pass some flag to say which array you want to use? Or something like that. Passing the arrays in this fashion is messy
>>>
>>
>> Since there are 3 configs (type, direction, format) with different available range and setting method (onehot, non-onehot),
>> moving the arrays into the valid checking function means we need to recognize the config types (passed by the caller but need
>> to know the available value array) and the checking method together. That may make the code more complex than now: 1st picking
>> up the right array and judge wich checking method this array applied and 2nd do the checking.
>>
>> Currently it's designed to decouple the checking method and the available value array. The hisi_ptt_trace_valid_config{_onehot}()
>> won't care about which array to use since caller take responsibilty for this. So perhaps current approach is simple and clear
>> enough.
> 
> 
> A couple of points:
> - hisi_ptt_trace_valid_config_type() only has 1x caller so can make it dedicated for that caller
> - there is not much code in hisi_ptt_trace_valid_config_onshot(), so ok to duplicate if makes overall code look better
> 
> So I think dedicated functions make the code simpler, easier to follow, and maintain:
> 

ok then it won't make it more complex. I thought you mean something like hisi_ptt_trace_config(u32 val, enum CONFIG_TYPE type, bool onehot)
as mentioned as "pass some flag to say which array you want to use". The implementation of that will be more complex. Thanks for the great
snippet!

> static int hisi_ptt_trace_valid_config_dir(u32 val)
> {
>     int i;
>     /*
>      * The supported value of the direction parameter. See hisi_ptt.rst
>      * documentation for more details.
>      */
>     static const u32 hisi_ptt_trace_available_direction[] = {
>         0,
>         1,
>         2,
>         3,
>     };
> 
>     for (i = 0; i < ARRAY_SIZE(hisi_ptt_trace_available_direction); i++)
>         if (val == hisi_ptt_trace_available_direction[i])
>             return 0;
> 
>     return -EINVAL;
> }
> 
> static int hisi_ptt_trace_valid_config_format(u32 val)
> {
>     int i;
>     static const u32 hisi_ptt_trace_availble_format[] = {
>         0,    /* 4DW */
>         1,    /* 8DW */
>     };
> 
>     for (i = 0; i < ARRAY_SIZE(hisi_ptt_trace_availble_format); i++)
>         if (val == hisi_ptt_trace_availble_format[i])
>             return 0;
> 
>     return -EINVAL;
> }
> 
> static int hisi_ptt_trace_valid_config_type(u32 val)
> {
>     int i;
>     /* Different types can be set simultaneously */
>     static const u32 hisi_ptt_trace_available_type[] = {
>         1,    /* posted_request */
>         2,    /* non-posted_request */
>         4,    /* completion */
>     };
> 
>     for (i = 0; i < ARRAY_SIZE(hisi_ptt_trace_available_type); i++)
>         val &= ~hisi_ptt_trace_available_type[i];
> 
>     if (val)
>         return -EINVAL;
>     return 0;
> }
> 
> ...
> 
> static int hisi_ptt_pmu_event_init(struct perf_event *event)
> {
> 
> ...
> 
>     val = FIELD_GET(HISI_PTT_PMU_DIRECTION_MASK, event->attr.config);
>     ret = hisi_ptt_trace_valid_config_dir(val);
>     if (ret < 0)
>         goto out;
>     ctrl->direction = val;
> 
>     val = FIELD_GET(HISI_PTT_PMU_TYPE_MASK, event->attr.config);
>     ret = hisi_ptt_trace_valid_config_type(val);
>     if (ret < 0)
>         goto out;
>     ctrl->type = val;
> 
>     val = FIELD_GET(HISI_PTT_PMU_FORMAT_MASK, event->attr.config);
>     ret = hisi_ptt_trace_valid_config_format(val);
>     if (ret < 0)
>         goto out;
>     ctrl->format = val;
> 
>     ...
> }
>>
>>>
>>>> +                         ARRAY_SIZE(hisi_ptt_trace_available_direction));
>>>> +    if (ret < 0)
>>>> +        goto out;
>>>> +    ctrl->direction = val;
>>>> +
>>>> +    val = FIELD_GET(HISI_PTT_PMU_TYPE_MASK, event->attr.config);
>>>> +    ret = hisi_ptt_trace_valid_config(val, hisi_ptt_trace_available_type,
>>>> +                      ARRAY_SIZE(hisi_ptt_trace_available_type));
>>>> +    if (ret < 0)
>>>> +        goto out;
>>>> +    ctrl->type = val;
>>>> +
>>>> +    val = FIELD_GET(HISI_PTT_PMU_FORMAT_MASK, event->attr.config);
>>>> +    ret = hisi_ptt_trace_valid_config_onehot(val, hisi_ptt_trace_availble_format,
>>>> +                         ARRAY_SIZE(hisi_ptt_trace_availble_format));
>>>> +    if (ret < 0)
>>>> +        goto out;
>>>> +    ctrl->format = val;
>>>> +
>>>> +out:
>>>> +    mutex_unlock(&hisi_ptt->mutex);
>>>> +    return ret;
>>>> +}
>>>> +
>>>> +static void *hisi_ptt_pmu_setup_aux(struct perf_event *event, void **pages,
>>>> +                    int nr_pages, bool overwrite)
>>>> +{
>>>> +    struct hisi_ptt_pmu_buf *buf;
>>>> +    struct page **pagelist;
>>>> +    int i;
>>>> +
>>>> +    if (overwrite) {
>>>> +        dev_warn(event->pmu->dev, "Overwrite mode is not supported\n");
>>>> +        return NULL;
>>>> +    }
>>>> +
>>>> +    /* If the pages size less than buffers, we cannot start trace */
>>>> +    if (nr_pages < HISI_PTT_TRACE_TOTAL_BUF_SIZE / PAGE_SIZE)
>>>> +        return NULL;
>>>> +
>>>> +    buf = kzalloc(sizeof(*buf), GFP_KERNEL);
>>>> +    if (!buf)
>>>> +        return NULL;
>>>> +
>>>> +    pagelist = kcalloc(nr_pages, sizeof(*pagelist), GFP_KERNEL);
>>>> +    if (!pagelist) {
>>>> +        kfree(buf);
>>>> +        return NULL;
>>>> +    }
>>>> +
>>>> +    for (i = 0; i < nr_pages; i++)
>>>> +        pagelist[i] = virt_to_page(pages[i]);
>>>> +
>>>> +    buf->base = vmap(pagelist, nr_pages, VM_MAP, PAGE_KERNEL);
>>>> +    if (!buf->base) {
>>>> +        kfree(pagelist);
>>>> +        kfree(buf);
>>>> +        return NULL;
>>>> +    }
>>>> +
>>>> +    buf->nr_pages = nr_pages;
>>>> +    buf->length = nr_pages * PAGE_SIZE;
>>>> +    buf->pos = 0;
>>>> +
>>>> +    kfree(pagelist);
>>>> +    return buf;
>>>> +}
>>>> +
>>>> +static void hisi_ptt_pmu_free_aux(void *aux)
>>>> +{
>>>> +    struct hisi_ptt_pmu_buf *buf = aux;
>>>> +
>>>> +    vunmap(buf->base);
>>>> +    kfree(buf);
>>>> +}
>>>> +
>>>> +static void hisi_ptt_pmu_start(struct perf_event *event, int flags)
>>>> +{
>>>> +    struct hisi_ptt *hisi_ptt = to_hisi_ptt(event->pmu);
>>>> +    struct perf_output_handle *handle = &hisi_ptt->trace_ctrl.handle;
>>>> +    struct hw_perf_event *hwc = &event->hw;
>>>> +    struct hisi_ptt_pmu_buf *buf;
>>>> +    int cpu = event->cpu;
>>>> +    int ret;
>>>> +
>>>> +    hwc->state = 0;
>>>> +    mutex_lock(&hisi_ptt->mutex);
>>>> +    if (hisi_ptt->trace_ctrl.started) {
>>>> +        pci_dbg(hisi_ptt->pdev, "trace has already started\n");
>>>
>>> doesn't perf core guard against this sort of thing?
>>>
>>
>> Maybe not as tested. The perf core will start the events 1)on the cpus user specified or
>> 2)on all the cpus, but the PTT trace is intended to start once on one cpu.
>>
>> For the 2) case, the driver will make default cpu to start the trace and block others
>> in pmu::add(). For the 1) case we'll met the condition here. So the started status is
>> test here to avoid a second start.
> 
> if this is a realistic and sensible usecase then it would be nice to handle in core perf code at some stage
> 

I think at least PTT trace meet this condition but seems we're the only uncore tracing PMU now, so
maybe it's still not common.

>>
>>>> +        goto stop;
>>>> +    }
>>>> +
>>>> +    if (cpu == -1)
>>>> +        cpu = hisi_ptt->trace_ctrl.default_cpu;
>>>> +
>>>> +    /*
>>>> +     * Handle the interrupt on the same cpu which starts the trace to avoid
>>>> +     * context mismatch. Otherwise we'll trigger the WARN from the perf
>>>> +     * core in event_function_local().
>>>> +     */
>>>> +    WARN_ON(irq_set_affinity(pci_irq_vector(hisi_ptt->pdev, HISI_PTT_TRACE_DMA_IRQ),
>>>> +                 cpumask_of(cpu)));
>>>> +
>>>> +    ret = hisi_ptt_alloc_trace_buf(hisi_ptt);
>>>> +    if (ret) {
>>>> +        pci_dbg(hisi_ptt->pdev, "alloc trace buf failed, ret = %d\n", ret);
>>>> +        goto stop;
>>>> +    }
>>>> +
>>>> +    buf = perf_aux_output_begin(handle, event);
>>>> +    if (!buf) {
>>>> +        pci_dbg(hisi_ptt->pdev, "aux output begin failed\n");
>>>> +        goto stop;
>>>> +    }
>>>> +
>>>> +    buf->pos = handle->head % buf->length;
>>>> +
>>>> +    ret = hisi_ptt_trace_start(hisi_ptt);
>>>> +    if (ret) {
>>>> +        pci_dbg(hisi_ptt->pdev, "trace start failed, ret = %d\n", ret);
>>>> +        perf_aux_output_end(handle, 0);
>>>> +        goto stop;
>>>> +    }
>>>> +
>>>> +    mutex_unlock(&hisi_ptt->mutex);
>>>> +    return;
>>>> +stop:
>>>> +    event->hw.state |= PERF_HES_STOPPED;
>>>> +    mutex_unlock(&hisi_ptt->mutex);
>>>> +}
[...]
>>>> +
>>>> +static void hisi_ptt_remove(struct pci_dev *pdev)
>>>> +{
>>>> +    struct hisi_ptt *hisi_ptt = pci_get_drvdata(pdev);
>>>> +
>>>> +    /*
>>>> +     * We have to manually unregister the PMU device rather than make it
>>>> +     * devres managed to keep order that the PMU device's unregistration
>>>> +     * is prior to the release of DMA buffers. As the DMA buffers are
>>>> +     * devm allocated when necessary which is after the registration of
>>>> +     * the PMU device.
>>>> +     */
>>>
>>> do you really need to mention all this?
>>>
>>
>> I think yes. Otherwise people may ask why not register PMU device in managed
>> way as well.
> 
> I expect devres work to be done after hisi_ptt_remove() so I would know this...
> 

yes, but the problem is that the buffers are devm allocated outside probe so the
order is not obvious and maybe a bit hard to understand.

But as suggested to allocate the buffers in the probe stage we can get rid of it
and make PMU registration devres managed as well.

Thanks,
Yicong





More information about the linux-arm-kernel mailing list