Giter Site home page Giter Site logo

ruixiangma / femu Goto Github PK

View Code? Open in Web Editor NEW

This project forked from vtess/femu

0.0 1.0 0.0 79.94 MB

FEMU: Cheap, Accurate, Scalable and Extensible NVMe SSD Emulator

Home Page: https://github.com/ucare-uchicago/femu

License: Other

Makefile 0.24% C 91.56% C++ 2.79% Shell 1.26% Python 2.51% Haxe 0.51% Objective-C 0.28% Assembly 0.50% NSIS 0.02% Perl 0.30% Prolog 0.03% GDB 0.01% GLSL 0.01%

femu's Introduction

FEMU README


Project Description

Briefly speaking, FEMU is a NVMe SSD Emulator. Based upon QEMU/KVM, FEMU is exposed to Guest OS (Linux) as a NVMe block device (e.g. /dev/nvme0nX). It can be used as an emulated whitebox or blackbox SSD: (1). whitebox mode (a.k.a. Software-Defined Flash (SDF), or OpenChannel-SSD) with FTL residing in the host side (e.g. LightNVM) (2). blackbox mode with FTL residing inside the device (most of current commercial SSDs).

FEMU tries to achieve benefits of both SSD Hardware platforms (e.g. CNEX OpenChannel SSD, OpenSSD, etc.) and SSD simulators (e.g. DiskSim+SSD, FlashSim, SSDSim, etc.). Like hardware platforms, FEMU can support running full system stack (Applications + OS + NVMe interface) on top, thus enabling Software-Defined Flash (SDF) alike research with modifications at application, OS, interface or SSD controller architecture level. Like SSD simulators, FEMU can also support internal-SSD/FTL related research. Users can feel free to experiment with new FTL algorithms or SSD performance models to explore new SSD architecture innovations as well as benchmark the new arch changes with real applications, instead of using decade-old disk trace files.

Installation

  1. Make sure you have installed necessary libraries for building QEMU. The dependencies can be installed automatically by

    # Switch to the FEMU building directory
    cd femu/build-femu
    # Copy femu script
    cp ../femu-scripts/femu-copy-scripts.sh .
    ./femu-copy-scripts.sh .
    # only Debian/Ubuntu based distributions supported
    sudo ./pkgdep.sh 
  2. Compile & Install FEMU:

    ./femu-compile.sh

    FEMU binary will appear as x86_64-softmmu/qemu-system-x86_64

Run FEMU

1. Before running

  • FEMU currently uses its own malloc'ed space for data storage, instead of using image-files. However, FEMU still requires a image-file in QEMU command line so as to cheat QEMU to probe correct internal numbers about the backend storage. Thus, if you want to emulate an SSD of 32GB, you need to create an image file of 32GB on your local file system and attach it to QEMU. (This limitation will be remove in near future)

  • FEMU relies on DRAM to provide accurate delay emulation, so make sure you have enough DRAM free space for the emulated SSD.

  • Only Guest Linux version >= 4.14 are supported as FEMU requires the shadow doorbell buffer support in Linux NVMe driver implementation. (Linux 4.12, 4.13 are abandoned due to their wrong implementation in doorbell buffer config support.)

  • To achieve best performance, users need to disable the doorbell write operations in guest Linux NVMe driver since FEMU uses polling. Please see here on how to do this.

2. Run FEMU as an emulated blackbox SSD (device-managed FTL)

Under this mode, each emulated NVMe SSD needs configuration files in the format of vssd1.conf, vssd2.conf, ..., etc. (which should correspond to your virtual NVMe image file names: vssd1.raw, vssd2.raw, etc.) to run.

The key configuration options are explained below:

It configures an emulated SSD with 8 channels and there are 8 chips on each channel. The total SSD size is 1GB.

	PAGE_SIZE           4096            // SSD page size in bytes
	PAGE_NB             256             // # of pages in one block
	SECTOR_SIZE         512             // # sector size in bytes
	FLASH_NB            64              // total # of NAND chips
	BLOCK_NB            16              // # of blocks in one chip

	REG_WRITE_DELAY     40000           // channel transfer time for one page (program) in nanosecond
	CELL_PROGRAM_DELAY  800000          // NAND page program latency in nanosecond
	REG_READ_DELAY      60000           // NAND page read latency in nanosecond
	CELL_READ_DELAY     40000           // channel transfer time for one page (read) in nanosecond
	BLOCK_ERASE_DELAY   3000000         // Block erase latency in nanosecond
	CHANNEL_NB          8               // # of channels
	GC_MODE             2               // GC blocking mode, see hw/block/ssd/common.h for definition

After the FEMU configuration file is ready, boot the VM using the following script:

./run-blackbox.sh

3. Run FEMU as an emulated whitebox SSD (OpenChannel-SSD)

./run-whitebox.sh

Inside the VM, you can play with LightNVM.

Currently FEMU only supports OpenChannel Specification 1.2, the newer 2.0 spec support in work-in-progress and will be added soon.

4. Run FEMU without SSD logic emulation

./run-nossd.sh

In this nossd mode, no SSD emulation logic (either blackbox or whitebox emulation) will be executed. Base NVMe specification is supported, and FEMU in this case handles IOs as fast as possible. It can be used for basic performance benchmarking.

Tuning

To Add ...

Debugging

To Add ...

FEMU Design

Please refer to our FAST paper and design document (to come) ...

Additonal Tweaks

  1. Disable doorbell writes in your guest Linux NVMe driver:

**Note: Linux kernel version less than 4.14 has a wrong implementation over the doorbell buffer config support bit. (Fixed in this commit: 223694b9ae8bfba99f3528d49d07a740af6ff95a). FEMU has been updated to fix this problem accordingly. Thus, in order for FEMU polling to work properly out of box, please use guest Linux >= 4.14.

Otherwise, if you want to stick to 4.12/4.13, please make sure NVME_OACS_DBBUF = 1 << 7 in hw/block/nvme.h as this is what was wrongly implemented in 4.12/4.13**

In Linux 4.14 source code, file drivers/nvme/host/pcie.c, around line 293, you will find below function which is used to indicate whether to perform doorbell write operations.

What we need to do is to add one sentence (return false;) after *dbbuf_db = value;, as shown in the code block below.

After this, recompile your guest Linux kernel.

/* Update dbbuf and return true if an MMIO is required */
static bool nvme_dbbuf_update_and_check_event(u16 value, u32 *dbbuf_db,
					      volatile u32 *dbbuf_ei)
{
	if (dbbuf_db) {
		u16 old_value;

		/*
		 * Ensure that the queue is written before updating
		 * the doorbell in memory
		 */
		wmb();

		old_value = *dbbuf_db;
		*dbbuf_db = value;
			
		/* Disable Doorbell Writes for FEMU: We only need to 
		 * add the following statement */
		return false;
		/* End FEMU modification for NVMe driver */

		if (!nvme_dbbuf_need_event(*dbbuf_ei, value, old_value))
			return false;
	}

	return true;
}

femu's People

Contributors

huaicheng avatar

Watchers

 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.