Re: [PATCH v8 3/4] drivers/perf: add DesignWare PCIe PMU driver

[Yicong Yang]

On 2023/10/18 11:33, Shuai Xue wrote:
> 
> 
> On 2023/10/17 17:39, Jonathan Cameron wrote:
>> On Tue, 17 Oct 2023 09:32:34 +0800
>> Shuai Xue <xueshuai@xxxxxxxxxxxxxxxxx> wrote:
>>
>>> This commit adds the PCIe Performance Monitoring Unit (PMU) driver support
>>> for T-Head Yitian SoC chip. Yitian is based on the Synopsys PCI Express
>>> Core controller IP which provides statistics feature. The PMU is a PCIe
>>> configuration space register block provided by each PCIe Root Port in a
>>> Vendor-Specific Extended Capability named RAS D.E.S (Debug, Error
>>> injection, and Statistics).
>>>
>>> To facilitate collection of statistics the controller provides the
>>> following two features for each Root Port:
>>>
>>> - one 64-bit counter for Time Based Analysis (RX/TX data throughput and
>>>   time spent in each low-power LTSSM state) and
>>> - one 32-bit counter for Event Counting (error and non-error events for
>>>   a specified lane)
>>>
>>> Note: There is no interrupt for counter overflow.
>>>
>>> This driver adds PMU devices for each PCIe Root Port. And the PMU device is
>>> named based the BDF of Root Port. For example,
>>>
>>>     30:03.0 PCI bridge: Device 1ded:8000 (rev 01)
>>>
>>> the PMU device name for this Root Port is dwc_rootport_3018.
>>>
>>> Example usage of counting PCIe RX TLP data payload (Units of 16 bytes)::
>>
>> Question follow through from previous patch comment, why not just
>> multiply it by 16 when you read it?  Is something in perf going to
>> overflow?
> 
> As we discussed in Path 1/4, the unit 16 is not general for all groups of
> Time Based Analysis, I prefer to leave unit part to end perf users.
> 
>>
>>>
>>>     $# perf stat -a -e dwc_rootport_3018/Rx_PCIe_TLP_Data_Payload/
>>>
>>> average RX bandwidth can be calculated like this:
>>>
>>>     PCIe TX Bandwidth = PCIE_TX_DATA * 16B / Measure_Time_Window
>>>
>>> Signed-off-by: Shuai Xue <xueshuai@xxxxxxxxxxxxxxxxx>
>>
>> Most of the comments inline aren't perf driver specific.  To me that
>> part of it looks fine but I'm only an intermittent reviewer of perf
>> drivers, so that needs more eyes!
>>
>> Anyhow, to my eyes this is coming together well but there are a few things
>> that don't look quite right yet.
> 
> Thank you for valuable comments, I will try my best to address them :)
> 
>>
>> ...
>>
>>> +#define DWC_PCIE_LANE_EVENT_MAX_PERIOD		GENMASK_ULL(31, 0)
>>> +#define DWC_PCIE_TIME_BASED_EVENT_MAX_PERIOD	GENMASK_ULL(63, 0)
>>> +
>>> +struct dwc_pcie_pmu {
>>> +	struct pmu		pmu;
>>> +	struct pci_dev		*pdev;		/* Root Port device */
>>> +	u16			ras_des;	/* RAS DES capability offset */
>>
>> Could call it ras_des_offset as then the comment wouldn't be needed...
> 
> Agreed, will fix it.
> 
>>
>>> +	u32			nr_lanes;
>>> +
>>> +	struct list_head	pmu_node;
>>> +	struct hlist_node	cpuhp_node;
>>> +	struct perf_event	*event[DWC_PCIE_EVENT_TYPE_MAX];
>>> +	int			on_cpu;
>>> +	bool			registered;
>>> +};
>>>
>>
>>
>>> +static void dwc_pcie_pmu_unregister_pmu(void *data)
>>> +{
>>> +	struct dwc_pcie_pmu *pcie_pmu = data;
>>> +
>>> +	if (!pcie_pmu->registered)
>>> +		return;
>>> +
>>> +	perf_pmu_unregister(&pcie_pmu->pmu);
>>> +	pcie_pmu->registered = false;
>>> +	list_del(&pcie_pmu->pmu_node);
>> For simplicity or reviewing I'd expect either:
>> a) Exact reverse order of what happened in probe.
>> b) A comment on why not.
>>
>> That probably jus means that the perf_pmu_unregister
>> should be first.
> 
> I guess you mean perf_pmu_unregister should be last?
> Bellow is the exact reverse order of what happened in probe.
> 
> 	pcie_pmu->registered = false;
> 	list_del(&pcie_pmu->pmu_node);
> 	perf_pmu_unregister(&pcie_pmu->pmu);
> 
>>
>>> +}
>>
>>
>> ...
>>
>>> +static int dwc_pcie_pmu_probe(struct platform_device *plat_dev)
>>> +{
>>> +	struct pci_dev *pdev = NULL;
>>> +	struct dwc_pcie_pmu *pcie_pmu;
>>> +	bool notify = false;
>>> +	char *name;
>>> +	u32 bdf;
>>> +	int ret;
>>> +
>>> +	/* Match the rootport with VSEC_RAS_DES_ID, and register a PMU for it */
>>> +	for_each_pci_dev(pdev) {
>>> +		u16 vsec;
>>> +		u32 val;
>>> +
>>> +		if (!(pci_is_pcie(pdev) &&
>>> +		      pci_pcie_type(pdev) == PCI_EXP_TYPE_ROOT_PORT))
>>> +			continue;
>>> +
>>> +		vsec = pci_find_vsec_capability(pdev, PCI_VENDOR_ID_ALIBABA,
>>> +						DWC_PCIE_VSEC_RAS_DES_ID);
>>> +		if (!vsec)
>>> +			continue;
>>> +
>>> +		pci_read_config_dword(pdev, vsec + PCI_VNDR_HEADER, &val);
>>> +		if (PCI_VNDR_HEADER_REV(val) != 0x04 ||
>>> +		    PCI_VNDR_HEADER_LEN(val) != 0x100)
>>
>> I'm curious - why check the header length?  Paranoia / defensive coding or does it vary?
>> I'd expect it to be fixed for a given revision. Ideally it should be backwards compatible
>> so that revs above 4 will always work (possibly with missing features) with rev 4 targeting
>> software but I guess we can't guaranteed that...
> 
> Kind of defensive coding. Agreed, I will remove the hender length check to make the driver more
> compatible.
>>
>>> +			continue;
>>> +		pci_dbg(pdev,
>>> +			"Detected PCIe Vendor-Specific Extended Capability RAS DES\n");
>>> +
>>> +		bdf = PCI_DEVID(pdev->bus->number, pdev->devfn);
>>> +		name = devm_kasprintf(&plat_dev->dev, GFP_KERNEL, "dwc_rootport_%x",
>>> +				      bdf);
>>> +		if (!name) {
>>> +			ret = -ENOMEM;
>>> +			goto out;
>>> +		}
>>> +
>>> +		/* All checks passed, go go go */
>>> +		pcie_pmu = devm_kzalloc(&plat_dev->dev, sizeof(*pcie_pmu), GFP_KERNEL);
>>> +		if (!pcie_pmu) {
>>> +			ret = -ENOMEM;
>>> +			goto out;
>>> +		}
>>> +
>>> +		pcie_pmu->pdev = pdev;
>>> +		pcie_pmu->ras_des = vsec;
>>> +		pcie_pmu->nr_lanes = pcie_get_width_cap(pdev);
>>> +		pcie_pmu->on_cpu = -1;
>>> +		pcie_pmu->pmu = (struct pmu){
>>> +			.module		= THIS_MODULE,
>>> +			.attr_groups	= dwc_pcie_attr_groups,
>>> +			.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
>>> +			.task_ctx_nr	= perf_invalid_context,
>>> +			.event_init	= dwc_pcie_pmu_event_init,
>>> +			.add		= dwc_pcie_pmu_event_add,
>>> +			.del		= dwc_pcie_pmu_event_del,
>>> +			.start		= dwc_pcie_pmu_event_start,
>>> +			.stop		= dwc_pcie_pmu_event_stop,
>>> +			.read		= dwc_pcie_pmu_event_update,
>>> +		};
>>> +
>>> +		/* Add this instance to the list used by the offline callback */
>>> +		ret = cpuhp_state_add_instance(dwc_pcie_pmu_hp_state,
>>> +					       &pcie_pmu->cpuhp_node);
>>> +		if (ret) {
>>> +			pci_err(pcie_pmu->pdev,
>>> +				"Error %d registering hotplug @%x\n", ret, bdf);
>>> +			goto out;
>>> +		}
>>> +
>>> +		/* Unwind when platform driver removes */
>>> +		ret = devm_add_action_or_reset(
>>> +			&plat_dev->dev, dwc_pcie_pmu_remove_cpuhp_instance,
>>> +			&pcie_pmu->cpuhp_node);
>>> +		if (ret)
>>> +			goto out;
>>> +
>>> +		ret = perf_pmu_register(&pcie_pmu->pmu, name, -1);
>>> +		if (ret) {
>>> +			pci_err(pcie_pmu->pdev,
>>> +				"Error %d registering PMU @%x\n", ret, bdf);
>>> +			goto out;
>>> +		}
>>> +		ret = devm_add_action_or_reset(
>>> +			&plat_dev->dev, dwc_pcie_pmu_unregister_pmu, pcie_pmu);
>>> +		if (ret)
>>> +			goto out;
>> This is messy because your devm callback also deals with the bit below.
>> So if the _or_reset here happens because this call fails, the list_del will
>> happen on something that was never added.  Simple fix is move this down to after
>> pcie_pmu->registered given none of the next few line of code can fail anyway.
> 
> Goot point. I missed the fact that if devm_add_action_or_reset() fails, the action
> dwc_pcie_pmu_unregister_pmu() will be called right away.
> 
> Will fix it in next version.
> 
> 
>>
>>> +
>>> +		/* Cache PMU to handle pci device hotplug */
>>> +		list_add(&pcie_pmu->pmu_node, &dwc_pcie_pmu_head);
>>> +		pcie_pmu->registered = true;
>>> +		notify = true;
>>> +	}
>>> +
>>> +	if (notify && bus_register_notifier(&pci_bus_type, &dwc_pcie_pmu_nb))
>>> +		notify = false;
>> As mentioned below, I'd expect the bus_unregister_notifier to be in a remove()
>> callback, or you could use a devm_add_action_or_reset() an a simple callback.
>> You can register that as
>>
>> 	if (notify) {
>> 		if (bus_register_notifier() == 0)
>> 			ret = devm_add_action_or_reset(unreg_notifier,
>> 						       &dwc_pcie_pmu_nb);
>> 	}
>> and then you don't need to track if it was registered or not as the
>> cleanup only happens if it was.
> 
> Good suggestion. Will also use devm_add_action_or_reset() to unwind.
> 
>>
>>
>>> +
>>> +	if (notify)
>>> +		dwc_pcie_pmu_notify = true;
>>> +
>>> +	return 0;
>>> +
>>> +out:
>>> +	pci_dev_put(pdev);
>>> +
>>> +	return ret;
>>> +}
>>> +
>>
>>
>> ...
>>
>>> +
>>> +static int dwc_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *cpuhp_node)
>>> +{
>>> +	struct dwc_pcie_pmu *pcie_pmu;
>>> +	struct pci_dev *pdev;
>>> +	int node;
>>> +	cpumask_t mask;
>>> +	unsigned int target;
>>> +
>>> +	pcie_pmu = hlist_entry_safe(cpuhp_node, struct dwc_pcie_pmu, cpuhp_node);
>>> +	/* Nothing to do if this CPU doesn't own the PMU */
>>> +	if (cpu != pcie_pmu->on_cpu)
>>> +		return 0;
>>> +
>>> +	pcie_pmu->on_cpu = -1;
>>> +	pdev = pcie_pmu->pdev;
>>> +	node = dev_to_node(&pdev->dev);
>>> +	if (cpumask_and(&mask, cpumask_of_node(node), cpu_online_mask) &&
>>> +	    cpumask_andnot(&mask, &mask, cpumask_of(cpu)))
>>> +		target = cpumask_any(&mask);
>>> +	else
>>> +		target = cpumask_any_but(cpu_online_mask, cpu);
>>> +
>>> +	if (target >= nr_cpu_ids) {
>>> +		pci_err(pcie_pmu->pdev, "There is no CPU to set\n");
>>
>> You have a local variable for pdev, use it here as well.
> 
> Will use local pdev directly.
> 
>>
>>> +		return 0;
>>> +	}
>>> +
>>> +	/* This PMU does NOT support interrupt, just migrate context. */
>>> +	perf_pmu_migrate_context(&pcie_pmu->pmu, cpu, target);
>>> +	pcie_pmu->on_cpu = target;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static struct platform_driver dwc_pcie_pmu_driver = {
>>> +	.probe = dwc_pcie_pmu_probe,
>>> +	.driver = {.name = "dwc_pcie_pmu",},
>>> +};
>>> +
>>> +static int __init dwc_pcie_pmu_init(void)
>>> +{
>>> +	int ret;
>>> +
>>> +	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
>>> +				      "perf/dwc_pcie_pmu:online",
>>> +				      dwc_pcie_pmu_online_cpu,
>>> +				      dwc_pcie_pmu_offline_cpu);
>>> +	if (ret < 0)
>>> +		return ret;
>>> +
>>> +	dwc_pcie_pmu_hp_state = ret;
>>> +
>>> +	ret = platform_driver_register(&dwc_pcie_pmu_driver);
>>> +	if (ret) {
>>> +		cpuhp_remove_multi_state(dwc_pcie_pmu_hp_state);
>>> +		return ret;
>>> +	}
>>> +
>>> +	dwc_pcie_pmu_dev = platform_device_register_simple(
>>> +				"dwc_pcie_pmu", PLATFORM_DEVID_NONE, NULL, 0);
>>> +	if (IS_ERR(dwc_pcie_pmu_dev)) {
>>> +		platform_driver_unregister(&dwc_pcie_pmu_driver);
>>
>> Why no cpuhp_remove_multi_state() in this error path?
>>
>> I'd move to the approach of a gotos and a single error handling block as
>> that makes this sort of thing easier to spot.
> 
> Agreed, will use the approach of a gotos to handle errors.
> 
>>
>>> +		return PTR_ERR(dwc_pcie_pmu_dev);
>>> +	}
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static void __exit dwc_pcie_pmu_exit(void)
>>> +{
>>> +	platform_device_unregister(dwc_pcie_pmu_dev);
>>> +	platform_driver_unregister(&dwc_pcie_pmu_driver);
>>> +	cpuhp_remove_multi_state(dwc_pcie_pmu_hp_state);
>>> +
>>> +	if (dwc_pcie_pmu_notify)
>>
>> If you have something unusual like this a driver module_exit() it definitely
>> deserves a comment on why.  I'm surprised by this as I'd expect the notifier
>> to be unregistered in the driver remove so not sure why this is here.
>> I've lost track of earlier discussions so if this was addressed then all
>> we need is a comment here for the next person to run into it!
> 
> All replied above, I will unregistered the notifier by devm_add_action_or_reset().
> 
> I am curious about that what the difference between unregistered in module_exit()
> and remove()?
> 

>From my understanding, if you register it in probe() then should undo it in remove().
Otherwise you should register it in module_init(). Just make them coupled to make
sure cleanup the resources correctly.

This driver is a bit different since device and driver are created in module_init()
so will works fine in most cases, because the device/driver removal will happens the
same time when unloading the module. However if manually unbind the driver and device
without unloading the module, we'll miss to unregister the notifier in the currently
implementation.

>>
>>> +		bus_unregister_notifier(&pci_bus_type, &dwc_pcie_pmu_nb);
>>> +}
> 
> .
>