PCIe-CXL

KO | EN

gitlite — search

#python #react #ai #typescript #machine-learning #rust #nodejs #java #windows #awesome-list #pytorch #security #mcp #openai

(★ 15)

Note repository for studying Peripheral Component Interconnect Express (PCIe) and Compute Express Link (CXL).

PCIe-CXL 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Table of Contents

- [Table of Contents](#table-of-contents)
- [PCI Express](#pci-express)
  * [PCIe Basics](#pcie-basics)
  * [Structures - PCIe Layer](#structures---pcie-layer)
    + [Software Layer](#software-layer)
    + [Transaction Layer](#transaction-layer)
    + [Data Link Layer](#data-link-layer)
    + [Physical Layer](#physical-layer)
  * [PCIe data transaction protocol](#pcie-data-transaction-protocol)
    + [Properties of TLP (Transaction Layer Packet)](#properties-of-tlp--transaction-layer-packet-)
  * [Link Training and Status State Machine (LTSSM)](#link-training-and-status-state-machine--ltssm-)
    + [States](#states)
      - [Detect](#detect)
      - [Polling](#polling)
      - [Configuration](#configuration)
      - [Recovery](#recovery)
      - [L0](#l0)
      - [L0s](#l0s)
      - [L1](#l1)
      - [L2](#l2)
      - [Disable](#disable)
      - [Loopback](#loopback)
      - [Hot Reset](#hot-reset)
  * [TLP Routing Mechanism](#tlp-routing-mechanism)
    + [Address Spaces](#address-spaces)
    + [Split Transaction Protocol](#split-transaction-protocol)
    + [Three methods of TLP routing](#three-methods-of-tlp-routing)
  * [Root Complex](#root-complex)
    + [PCIe Root Port](#pcie-root-port)
    + [Switch & Bridge](#switch---bridge)
    + [PCIe Endpoint Device 5](#pcie-endpoint-device-5)
    + [Enumeration](#enumeration)
  * [PCI Configuration Space](#pci-configuration-space)
    + [BDF (Bus, Device, Function)](#bdf--bus--device--function-)
    + [Extended configuration space for PCIe](#extended-configuration-space-for-pcie)
    + [Standardized Registers](#standardized-registers)
  * [PCIe Ordered Set](#pcie-ordered-set)

# PCI Express

Since CXL depends on the physical interface of PCIe, it is necessary to review the fundamentals of PCIe. I personally have one summer worth of experience with PCIe during my internship at Samsung Electronics, working with PCIe 5.0 in NAND SSDs.

## PCIe Basics

- `Bi-directional connection` that allows both transmitting and receiving. 
  
  - Each link can either transmit or receive data. 
    
    - `link` : Connection between two devices.
  
  - Hence, combination of 2 simplex makes it `Dual-Simplex Connection`.

- In each `link` there are multiple `lanes`. 
  
  - in forms of : x1, x2, x4, x8, x16, x32

- Uses `Serial Communication` rather than parallel. 
  
  - Parallel comm has issues such as: 
    
    - Parallel bus needs its `flight time` to be lower than a period of `clock`. In high speed communication, this is very difficult. 
      
      - `Flight Time` : Time takes for a signal to reach RX from TX in a bus. 
    
    - `Clock Skew`, as RX side and TX side has different clock arrival.

- Uses `Differential Signal` and sends D+ and D-
  
  - Uses the difference between D+ and D- to recognize a signal.
  
  - This helps lower the impact of noise on signal integrity,

## Structures - PCIe Layer

Packet breakdown:

- `TLP` (Transaction Layer Packet)
  
  - Header
  
  - Data
  
  - `ECRC` (End to end CRC)
    
    - `CRC`: Cycle Redundancy Check

- `DLLP` (Data Link Layer Packet)
  
  - All of `TLP` 
  
  - `LCRC` (Link CRC)
    
    - Used by nearby receiver checking for errors.

- Upon passing `Physical Layer`, `start` and `end` is added to `DLLP`.

### Software Layer

- Start and destination of all `Requests` made in the PCIe.

- Not listed in the official PCIe specification.

### Transaction Layer

- Creates `outbound packets` upon request from software layer.

- `TLP : Transaction layer packets` 
  
  - Non-posted : RX sends `completion packet` upon receiving TLP. 
  
  - Posted : RX does not need to send `completion` upon receiving TLP. 
  
  - Sending completion packets increases accuracy, but creates more burden as time required for waiting for completion.

| Request Type         | Posted?    | Acronym |
| -------------------- | ---------- | ------- |
| Memory Read Request  | Non-Posted | MRd     |
| Memory Write Request | Posted     | MWr     |
| IO Read Request      | Non-Posted | IORd    |
| IO Write Request     | Non-Posted | IOWr    |
| Config Read Request  | Non-Posted | CfgRd   |
| Config Write Request | Non-Posted | CfgWr   |
| Message Request      | Posted     | Msg     |

[More TLP details](#TLP (Transaction Layer Packet))

### Data Link Layer

- Mainly responsible for`TLP error correction`,`Flow control`,`Link Power Management`
  
  - TLP error correction (Ack, Nak Protocol)
    
    - LCRC (Link CRC) & sequence number is added to TLP
    
    - TLP is buffered into Replay Buffer before data is being transmitted. 
    
    - When RX sends TX Ack/Nack `DLLP` as a signal for data receive, next set of data is sent from the buffer. 
  
  - Flow Control 
    
    - TX can only send TLP to RX when RX's buffer is empty. RX uses `DLLP` to report the status of its buffer to TX. 
    
    - When RX processes all its TLP and remove it from the buffer, it sends `Flow Control Update DLLP`. 
  
  - Power Management 
    
    - DLLP can request the status of Link & System Power

### Physical Layer

Physical Layer is divided into two parts: `Logical` and `Electrical` part.

- Logical Part 
  
  - Digital Logic 
  
  - Start and End is added to DLLP 
  
  - `Byte Striping`: packets get divided in each lanes. 
  
  - Link Training and Initialization 
    
    - For smooth exchange of packets in the two links, the following options are checked: 
      
      - Link Width
      
      - Link Data Rate
      
      - Bit Lock per Lane
      
      - Symbol lock per lane 
      
      - Lane Reversal 
      
      - Polarity inversion 
      
      - Lane to Lane de-skew (when using multi-lane)

- Electrical Part
  
  - Analog interface 
  
  - AC coupling: capacitors that filters DC signal from AC/DC signal. 
  
  - `OS: Ordered Set`
    
    - Only used in the Physical Layer. 
    
    - Link Training, Entry or Exit low power state
    
    - Clock tolerance compensation
      
      - Compensates the difference of internal clock between RX and TX side.

## PCIe data transaction protocol

PCIe transaction is divided into 4 categories:

1. Memory

2. IO

3. Configuration

4. Message Transaction

Transactions are divided into `Non-Posted` and `Posted`.

### Properties of TLP (Transaction Layer Packet)

- TLP access 4 different `Address Spaces`.

## Link Training and Status State Machine (LTSSM)

- Sets and resets `links` and `ports` to allow packets to be sent. 
  
  - Physical layer controlled, hardware based processor.

- Diagram of all states that a `link` can have. 
  
  - `LTSSM` has 11 states, including `Detect`, `Polling`. 
  
  - Each state has its substates

![ltssm.png](https://raw.githubusercontent.com/seanhwang10/PCIe-CXL/main/Users/seahwang/Documents/GitHub/Cisco-Internship/figures/ltssm.png)

### States

Related concepts:

- `Ordered Set`: Packets used for communication between devices in the physical layer.

- `Training Sequence` : Ordered sets used for link initialization and training. Acronym : TS. 
  
  - TS1, TS2

- Cold Reset : Power on for system without power.

- Warm Reset : Reset without system losing power.

#### Detect

- State for both ends to detect & identify the link.

- Initial state from Boot or Reset

- `Resets` all logics, ports, registers.

#### Polling

- State to process / establish the following 
  
  - `Bit lock` : Synchronizes the `clock` for both ends of link. 
    
    - we say a `lock` is secured when sync is made. 
  
  - `Symbol lock` : Finding the boundary of continued bits (Gen1, Gen 2)
    
    - 8b/10b (for Gen 1, Gen 2) 
    
    - PCIe 5.0 uses 128b/130b. 
    
    - (Target data size / Actual required size for data send) 
  
  - `Land Polarity` : Checks if lane is `D+` or `D-`
  
  - `Block Assignment` : Founding the boundary of continued bits  using EIEOS (Gen 3)
  
  - `Lane Data Rate` : Speed of data transfer for each lane.

#### Configuration

- State to process / establish the following 
  
  - `Lane reversal` : check if lane connection is reversed. 
    
    - for example, [0:3] can be connected as [3:0] in the other end. 
  
  - `Link & Lane number` : Data displayed as `PAD` in `TS` (Training Sequence) gets changed to a negotiated number. 
    
    - Negotiation is done to utilize maximum performance of both devices at the end of a link. 
  
  - `Lane to Lane de-skew` : Corrects the `skew` 
    
    - `Skew` : Parallel bit streams from multiple lanes arriving the RX device at different times.

#### Recovery

- State that recovers changes made. Processes the following 
  
  - Resets `Bit lock`, `Symbol lock`, `Block alignment`, `Lane to lane de-skew`. 
  
  - Change / Reset the link speed

- State before entering `Loopback`, `Hot reset`, or `Disabled`. 
  
  - Or when state is changed from `L1` to `L0`.

#### L0

- State of **successful transaction of data.**

#### L0s

- Substate of `L0`.

- Saves power while returning to `L0` state can be done fast.

#### L1

- Saves more power than `L0s`, but returning to `L0` state takes longer.

- `L1` state can be entered by: 
  
  - ASPM (Active State Power Management)
    
    - When device connected via PCIe is not in use, this policy changes `links` to low power state. 
  
  - PMSW (Power Management Software)
    
    - Can change the device to low power state. 
    
    - D1, D2, D3hot, D3cold

#### L2

- Maximum power saving mode. 
  
  - Turns the device **OFF** to achieve that.

#### Disable

- Makes the `TX` ***Electrically Idle*** when `RX` is at ***low impedance state.***

- TS1's `TC (Training Control) disable bit` -> 1 
  
  - Sends 16 of `TC disable bit` set to 1

#### Loopback

- State for testing.

- Enters the state when TS1's `Loopback bit` is set to 1.

- Receiver `echos` the received packet back. 
  
  - Sends the same packet back to the TX.

#### Hot Reset

- Link can be reset by setting `Secondary Bus Reset Bit` as 1. 
  
  - Located in `Bridge Control Register`

- TS1's TC Hot reset bit -> 1 
  
  - Sends a `TC Hot reset bit` set to 1.

## TLP Routing Mechanism

Transaction Layer Packets gets transmitted from one link to the other while following TLP routing mechanism. There are 4 different `Address Spaces` that TLP access.

### Address Spaces

- Memory (Read, Write)
  
  - Transfer data to or from a location in the system memory map.

- IO (Read, Write)
  
  - Transfer data to or from a location in the system IO map.

- Configuration (Read, Write)
  
  - Transfer data to or from a location in the configuration space of a PCI-compatible device.

- Message (Baseline, Vendor-Specific)
  
  - General in-band messaging and event reporting (without consuming memory or IO address resources)

### Split Transaction Protocol

Access to the address spaces in PCIe are accomplished with split-transaction `requests` and `completions`.

- Improvement over earlier bus protocols like PCI.

- Splits and generates two seperate TLP with independent routing for the `Request TLP` and `Completion TLP`.

- Higher performance, but more overhead
  
  - A link is available for activity in the time between request and completion = BETTER PERFORMANCE
  
  - Requires and generates additional TLP for a single transaction = MORE OVERHEAD

### Three methods of TLP routing

There are three methods of TLP routing. TLP Variants that targets one of the 4 `address spaces` will use one of the following routing method.

1. Address Routing

2. ID Routing

3. Implicit Routing

| TLP Type                                                                        | Routing Method                                |
| ------------------------------------------------------------------------------- | --------------------------------------------- |
| Memory read (MRd), Memory Read Lock (MRdLk), Memory write (MWr)                 | Address Routing                               |
| IO Read (IORd), IO Write (IOWr)                                                 | Address Routing                               |
| Configuration Read Type 0-1 (CfgRd0-1), Configuration Write Type 0-1 (CfgWr0-1) | ID Routing                                    |
| Message (Msg), Message with Data (MsgD)                                         | Address Routing, ID Routing, Implicit Routing |
| Completion (Cpl), Completion with Data (CplD)                                   | ID Routing                                    |

## Root Complex

- Interface between CPU and PCIe busses. For example:
  
  - Processor interface
  
  - DRAM interface 
  
  - Configurations / chips

- Logically connects PCIe's hierarchy domain into a single PCIe hierarchy.

- A single fabric instance is consisted of the following: 
  
  - Rc 
  
  - Multiple Endpoints (I/O devices)
  
  - Switch
  
  - PCIe to PCI/PCI-X Bridge

![](https://raw.githubusercontent.com/seanhwang10/PCIe-CXL/main/Users/seahwang/Documents/GitHub/Cisco-Internship/figures/root%20complex.png)

PCIe Timeout : Transaction 사이에 커맨드 주고 기다리는 시간.

Root complex 에서 핸들링 함.

### PCIe Root Port

- PCIe port inside a root complex.

- Each Root Port has a divided hierarchy

- Each hierarchy can be configured with one endpoint, and sub-hierarchy with one or more switch and endpoints.

![3.png](https://raw.githubusercontent.com/seanhwang10/PCIe-CXL/main/Users/seahwang/Documents/GitHub/Cisco-Internship/figures/3.png)

Fig 3.

### Switch & Bridge

- `Switch` is a device that allows a single `Root Port` to connect with multiple devices. It orks as a packet router, and uses the packet address and other routing information to identify the direction of packets.

- `Switch` can have multiple `Downstream Ports`, and one `Upstream Port`.

- A `Bridge` provides an interface with other busses. 
  
  - Interface includes PCI, PCI-X or other PCIe bus.

![4.jpeg](https://raw.githubusercontent.com/seanhwang10/PCIe-CXL/main/Users/seahwang/Documents/GitHub/Cisco-Internship/figures/4.jpeg)

Fig 4

### PCIe Endpoint Device 5

- Endpoint Device is at the bottom or tree topology. 
  
  - And has a single `Upstream Port`

- Has `Type 0 Configuration Space Header`

- `Switch` or `Bridge` does NOT serve as a initiator or completer in a bus with PCIe transaction. 
  
  - Endpoint device can act as a `Requester` or `Completer` for Non-PCIe devices.

- Three types of Endpoint Devices 
  
  - Legacy (PCI Endpoint)
    
    - I/O request and complete possible.
    
    - Can't perform memory transaction over 4G address (32 bit Address Register)
  
  - PCIe Endpoint
    
    - I/O request should not be done. 
    
    - Can perform memory transaction over 4G address. 
  
  - Root Complex Integrated Endpoint 
    
    - Root port inside the `Root Complex` logic. 
    
    - I/O request should not be done 
    
    - No `Power management` and `Hot Plug`

Fig 5

### Enumeration

-  Enumerates through BDF of a PCI device (explained below)

- Process of allocating bus numbers and system resources by configuration software. 
  
  - Done after finding system topology

- in x86 PCIe architecture, `enumeration` takes place on BIOS hardware initialization process. 
  
  - All registers are set before boot loader.

- System software can re-allocate enumeration.

## PCI Configuration Space

- PCI bus uses `configuration space` to get the basic information of the connected device when a connection is made. 
  
  - Identifies what type of device is connected 
  
  - What protocol should be used

- Configuration space includes set of registers such as `Device ID`, `Vendor ID`, `Class Code` etc that includes the device information. 
  
  - In PCIe, `extended configuration space` 
  
  - The registers in `configuration space` are allocated in memory.

- All PCI devices have `BDF (Bus:Device:Fuction)` number. 
  
  - Scanning all BDFs will allow system to identify the device connected to it.

- Configuration spae is 256 bytes long 
  
  - Can allocate address for 
    
    - 8 bit PCI bus 
    
    - 5 bit device
    
    - 3 bit function number (BDF)

![6.png](https://raw.githubusercontent.com/seanhwang10/PCIe-CXL/main/Users/seahwang/Documents/GitHub/Cisco-Internship/figures/6.png)

Fig 6 BDF

### BDF (Bus, Device, Function)

There are

- 256 `Bus` 
  
  - 32 `Device` on each `Bus` 
    
    - 8 `Functions` on each `Devices`

`Bus`: Number of the bus.

`Device` : Number of a certain device connected to the bus.

`Function` : Number of functions that each devices can have.

### Extended configuration space for PCIe

- PCI-X supports up to 4096 bytes.

- Standard : First 4 bytes = 0x100. 
  
  - This implies for extended configuration space.

### Standardized Registers

The following is the standard registers in `PCI Type 0 Configuration Space Heder`

- `Device ID, Vendor ID` Register
  
  - Identifies the Device. 
  
  - 16bit `Vendor ID` is allocated by PCI-SIG. 
  
  - 16bit `Device ID` is set by the Vendor.

- `Status` Register 
  
  - Used to notify the status such as: 
    
    - Operation of specific functions 
    
    - Errors

- `Command` Register 
  
  - Has `bitmask` that can activate/deactivate options such as: 
    
    - Interrupt Disable 
    
    - Memory Write and Invalidate 
    
    - Bus Master Enable

- `Class Code` Register
  
  - Shows the class (type) of a PCI device. 
  
  - For example: storage, USB controller, Network Card etc.

- `Header Type` Register
  
  - Sets the layout of the header for remaining 48 bytes (64-16). 
    
    - 10h(16) onwrds in the figure below. 
  
  - Type 1: Root Complex, Switches, Bridges 
  
  - Type 0: Endpoint

- `Cache Line Size` Register
  
  - Sets the cache line size 
  
  - Not applicable for PCIe

- `Base Address` Register
  
  - Stores address of memory shared by the Device and OS. 
  
  - For Device Driver which is stored in memory.

- `Subsystem ID (SSID) & Subsystem Vendor ID (SVID)` 
  
  - Used to distinguish a specific model of device. 
    
    - `Vendor ID` : Chipset Manufacturer 
    
    - `Subsystem Vendor ID` : Card Manufacturer

# Personal Note

공부하고싶은것:

- Hetrogeneous Computing

- Server disaggregation

- UCS - Unified Computing System

Compute Express Link (CXL)

- Leverages PCIe, and targets high performance computational workloads.
  
  - Example: AI, Machine learning
  
  - 이유: 최근 데이터센터에 빅데이터, ML 등 고집적 워크로드 연산 demand 높음.
    
    - 이런 High intensity연산 보조를 위해 GPU, FPGA등 가속기 도입 확대중.
    
    - 고성능 컴퓨팅 환경에서 프로세서 <-> 가속기 간 메모리 공유를 통해 초저지연 컴퓨팅 구현하여 서버 내 메모리 리소스 범용성, 확장성 UP.

- 디바이스가 저지연 고 bandwidth로 시스템과 통신하고, 시스템이 디바이스의 메모리에 접근할수 있게 해주는 쌍방으로 이로운 interconnect.

## CXL Basics

- Interconnect for processors, with high bandwidth and low latency.
  
  - Aims to establish a memory sharing between host processor and CXL devices.

- L1 L2 L3 캐시 메모리

- CXL 사용하는 이유가 더 많은 메모리를 사용 가능하게 하기 위해
  
  - FPGA로 메모리 추가 기능만 사용하는것같음
  
  - 더 가격이 쌈. 컴퓨터에 가까워질수록 (캐시, DRAM등) 가격은 비싸짐.

- 각 서버 유닛당 BMC라는 칩이 있음
  
  - BMC는 Out of Band 임. CPU랑 스토리지 등 서버 유닛의 주 목적에 직접적으로는 아무 영향을 끼치지 않음.
  
  - 네트워크를 통해서 각 유닛의 BMC를 접근할수 있고 BMC를 통해 유닛의 각종 정보를 얻을 수 있음. 현재 온도, 팬 스피드, 등. 나중에 더 추가하자

- BMC로 접근하는 방법은 Hai 가 알려준대로 터미널 이용해서 들어가면 됨.

Rank Margin Test

Intel MLC - Memory latency checker

PMEM2 - in house memory testing tool

STREAM - benchmark tool

Mt Rainier C-Series Servers

Go through both My Rainer 1U and 2U, focus more on 2U.

- CXL cards go on riser
  
  - 2U : up to 8 slots for standard add-ons

## Supporting Memory

R-DIMM, LR-DIMM DDR5 288 pin DIMM and 16 NVDIMM modules

DRAM 종류들임 :

- DIMM : 아무 기능 없는 메모리

- U-DIMM : ECC (Unbuffered or Unregistered DIMM)
  
  - 가격 저렴, 저전력 사용, 빠른 응답률 (버퍼가 없기 때문)
  
  - 최대 2 Rank 까지 확장 가능.. 채널당 2개 DIMM만 사용 가능.

- R-DIMM : ECC + REG (Registered DIMM)
  
  - 메모리에 버퍼를 추가하여 DIMM의 주소와 명령을 기다리고 ECC가 강력함.
  
  - DIMMs당 4개 RANK로 확장 가능. 레지스터 칩 때문에 원가 더 비쌈

- LR-DIMM : ECC + REG + 데이터신호 제어 (Load Reduced DIMM)
  
  - RDIMM 에 isolation memory buffer 을 적용해서

- NVDIMM : ECC + REG + 데이터신호 제어 + SSD (Non Volatile DIMM)

Feature Types:

- ECC : 데이터 에러 정정 기능

- REG : 다수의 메모리 구성 시, RAM 슬롯 거리 차이로 모듈간 데이터 이동속도 차이로 발생하는 신호왜곡을 방지하는 기능.
  
  - Registered DIMM 추가 정보

- 데이터신호 제어 : 버퍼를 추가하여 데이터의 신호를 제어

## CIMC, BMC

/opt/homebrew/Cellar/telnet/64/bin/telnet

- CIMC 는 KVM, BIOS update등 기능으로 쓰임.

- 웹 브라우져에 CIMC IP 그대로 치면 됨.

### Boot:

```bash
  # BIOS > UEFI built in EFI Shell:
  Shell> FS3:
  FS3:\> efi\redhat\grubx64.efi
```

Press E on linux boot menu >> LINUX Boot edit mode.

`iomem=relaxed` kernel parameter.

Kernel parameter view: `cat /proc/cmdline`

#### BCM Log: `/var/log/sel/log`

리셋:

CIMC -> Chassis -> Faults and Logs ->
-> System Event Log
-> Cisco IMC log

## MEMTESTER 돌리기

```shell
sudo su # ROOT MODE로 만들기
sudo setpci -s 98:0.0 0x4.L=0x7

sudo daxctl reconfigure-device --mode=system-ram 
--no-online --force dax0.0

sudo echo "base=0x2080 000000 size=0x80 000000 
type=uncachable" > /proc/mtrr #숫자 띄우기 없이

./memtester -p 0x2080000000 1M 10
# 로그 남기기 
./memtester -p 0x2080000000 1M 10 > log.txt  
```

로그 남기기 꿀팁

```bash
<memtester command> | tee output.txt
```

FCS : Frame check sum error

### PATH 추가: `sudo nano /etc/paths`

seahwang@SEAHWANG-M-V32G ~ % ssh [admin@172.28.94.55](mailto:admin@172.28.94.55)

C240-WZP24430AC8# connect debug-shell

[ help ]# sldp l

`ssh ip: 172.28.30.195`

telnet ip port

cat /etc/bashrc

change TMOUT to 0 with vi.

`RACK IP: 172.28.30.6 , PORT 2008`

`$rack_fan_control  -c get_sensor_readings;`

Get all sensor reading from BMC

`$ipmi-query sensors all | grep -v "discrete";`

BMC more inclusive; longer

```shell
ipmi-query sel
ipmitool -I bmc sel elist
echo BEGIN THE OEM SEL
ipmi-query sel oem
```

generate logs

```
ipmitool -I bmc sel clear
ipmi-query sel clear
```

clear the logs

```
rack_fan_control -c set_log_level 7
tail -F /nv/etc/log/eng-repo/messages
```

debug

`pwmtest getdutycycle all`

get duty cycle of all fans.

### Fan Control

```shell
$echo 100 > /tmp/fanspeed.txt
$pwmtest getdutycycle all
```

콘솔세팅값

console=ttyS0,115200

filter : sed

cxv: awk

## LSPCI

1) 경로 : /sbin/lspci

2) 요약 : 시스템에 있는 PCI 디바이스 정보를 출력

3) 사용 방법 : lspci [옵션]

4) 옵션

-b : 커널을 대신하여 카드를 이용한 IRQ와 주소를 출력

-m : 디바이스 스크립트에서 이용하기 알맞도록 디바이스 정보를 문자 형태로 출력

-n : 벤더와 디바이스 코드를 출력

-s domain bus:slot.func : 지정된 디바이스의 정보만을 출력. PCI domain(0~~ffff), bus(0~~ff), slot(0~~1f), function(0~~7)로 구성되어 있음

-t : 디바이스 사이에 연결을 트리 형식으로 출력

-v : 상세한 디바이스 정보를 출력

-vv : -v 옵션보다 상세한 정보를 출력

5) 추가 설명

lspci 명령어는 시스템 관리 명령어로서 시스템에 있는 모든 PCI 디바이스 목록을 출력합니다. 이 명령어는 시스템에 디바이스 드라이버가 정상적으로 동작하는지 확인하거나 드라이버를 디버깅 하는데 사용합니다. lspci를 이용하여 시스템에 설치된 이더넷 디바이스를 검색하고, 자세한 장치 정보를 얻어봅니다. lspci 명령어만 실행하면 시스템의 모든 PCI 디바이스 목록을 출력합니다. grep 명령어를 이용하여 ethernet 스트링이 포함된 라인만 필터링해 봅니다.

PCIe-CXL

PCIe-CXL 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

PCIe-CXL

PCIe-CXL 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README