PCIe学习笔记（2）错误处理和AER/DPC功能

处理器上错误通常可分为detected和undetected error。Undetected errors可能变得良性(benign)，也可能导致系统故障如silent data corruptions (SDC)。Detected errors则又可分为correctable errors (CE) 和uncorrectable errors (UCE)。
PCIe定义了两种错误报告范式：基线功能和高级错误报告功能（AER）。本文介绍PCIe错误定义及AER/DPC功能。

PCIe Error

PCIe错误分类为Uncorrectable errors和Correctable errors。Uncorrectable错误严重性可以进一步分为fatal和nonfatal。

有3种错误信号机制：

1. CPL status

2. in-band error message:
   Routed-to-RC, reqid标识错误设备BDF，CPL通常返回UR/CA指示uncorrectable error，AER发出error message，RP支持AER记录到Root Error Status寄存器。
   

3. error forwarding (data poisoning)
   EP位指示，non-fatal和fatal error需要启用command.SERR#ENABLE，error message发送由device control bit[3:0]控制。

AER (Advanced Error Reporting)

PCIe AER扩展的功能结构提供更强大的错误报告。可将uncorrectable error编程为fatal或non-fatal。若severity置位，则为faral error，否则为non-fatal error。仅报告最重要错误，遵循优先级列表。

Controller接收TLP detect到错误后的步骤如下：

1. Discard TLP
2. Generate a CA/UR Cpl (for NP)
3. Set status in PCI-compatible status register
4. Set status in AER registers (when enable AER)
5. Generate an error MSG (USP only)
6. For malformed TLPs credit is returned based on the buffer space which has been consumed by the TLP

AER初始化需要启用以下域段：

AER driver用于在发生错误时收集全面的错误信息，报告错误并执行错误恢复操作。

static irqreturn_t aer_irq(int irq, void *context)
{
    struct pcie_device *pdev = (struct pcie_device *)context;
    struct aer_rpc *rpc = get_service_data(pdev);
    struct pci_dev *rp = rpc->rpd;
    int aer = rp->aer_cap;
    struct aer_err_source e_src = {};

    pci_read_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, &e_src.status);
    if (!(e_src.status & AER_ERR_STATUS_MASK))
        return IRQ_NONE;

    pci_read_config_dword(rp, aer + PCI_ERR_ROOT_ERR_SRC, &e_src.id);
    pci_write_config_dword(rp, aer + PCI_ERR_ROOT_STATUS, e_src.status);

    if (!kfifo_put(&rpc->aer_fifo, e_src))
        return IRQ_HANDLED;

    return IRQ_WAKE_THREAD;
}

aer_isr()继续根据severity读取status和mask，并把status和id等信息输出。

static irqreturn_t aer_isr(int irq, void *context)
{
    struct pcie_device *dev = (struct pcie_device *)context;
    struct aer_rpc *rpc = get_service_data(dev);
    struct aer_err_source e_src;

    if (kfifo_is_empty(&rpc->aer_fifo))
        return IRQ_NONE;

    while (kfifo_get(&rpc->aer_fifo, &e_src))
        aer_isr_one_error(rpc, &e_src);
    return IRQ_HANDLED;
}

static void aer_isr_one_error(struct aer_rpc *rpc,
                              struct aer_err_source *e_src)
{
    struct pci_dev *pdev = rpc->rpd;
    struct aer_err_info e_info;

    pci_rootport_aer_stats_incr(pdev, e_src);

    /*
	 * There is a possibility that both correctable error and
	 * uncorrectable error being logged. Report correctable error first.
	 */
    if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
        e_info.id = ERR_COR_ID(e_src->id);
        e_info.severity = AER_CORRECTABLE;

        if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
            e_info.multi_error_valid = 1;
        else
            e_info.multi_error_valid = 0;
        aer_print_port_info(pdev, &e_info);

        if (find_source_device(pdev, &e_info))
            aer_process_err_devices(&e_info);
    }

    if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
        e_info.id = ERR_UNCOR_ID(e_src->id);

        if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
            e_info.severity = AER_FATAL;
        else
            e_info.severity = AER_NONFATAL;

        if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
            e_info.multi_error_valid = 1;
        else
            e_info.multi_error_valid = 0;

        aer_print_port_info(pdev, &e_info);

        if (find_source_device(pdev, &e_info))
            aer_process_err_devices(&e_info);
    }
}

DPC (Downstream Port Containment)

在Downstream Port检测到unmasked uncorrectable error后停止流量，避免数据损坏潜在扩散，并支持CER (Containment Error Recovery)。DPC触发不会被视为错误，但可将其视为correctable error。DPC触发时支持发起中断或ERR_COR消息，流程如下：
DPC trig -> DPC interrupt & DPC trig status/reason -> disable LTSSM
SW clear DPC status -> LTSSM to Detect state -> link retrain

DPC Trigger Enable

• 00b: 默认关闭
• 01b: 启用并在ERR_FATAL触发;
• 10b: 启用并在ERR_NONFATAL/ERR_FATAL触发

DPC Completion Control: 0-CA, 1-UR
DPC Interrupt: DPC触发时发起INTx/MSI/MSI-X
DPC ERR_COR: DPC触发时发起ERR_COR msg，独立于中断
Poisoned TLP Egress Blocking: 不得传输TLP，若未触发DPC返回UR Cpl，否则DPC期间不再接收并丢弃TLP
Software Trigger: 写1触发DPC，读恒为0，可用于disable link，优先于MSI/MSI-X
DL_Active ERR_COR: DL转换为DL_Active时，DSP发起ERR_COR，不会作为错误处理

DPC RC busy: 指示软件RP需保持在DPC揭制状态
AER Uncorrectable Error Status 会根据 DPC Trigger Enable/Status和更新

Root Port Programmed I/O (RP PIO) Error Controls (eDPC)
用于精细管理RP NP遇到的错误 (CFG/IO/Mem UR/CA/CTO)，建议与AER配置同步。若severity置位，作为UCE处理，触发DPC。

static irqreturn_t dpc_irq(int irq, void *context)
{
	struct pci_dev *pdev = context;
	u16 cap = pdev->dpc_cap, status;

	pci_read_config_word(pdev, cap + PCI_EXP_DPC_STATUS, &status);

	if (!(status & PCI_EXP_DPC_STATUS_INTERRUPT) || PCI_POSSIBLE_ERROR(status))
		return IRQ_NONE;

	pci_write_config_word(pdev, cap + PCI_EXP_DPC_STATUS,
			      PCI_EXP_DPC_STATUS_INTERRUPT);
	if (status & PCI_EXP_DPC_STATUS_TRIGGER)
		return IRQ_WAKE_THREAD;
	return IRQ_HANDLED;
}

Linux Reference:
drivers/pci/pcie/aer.c
drivers/pci/pcie/dpc.c