Linux kernel 2.6.39 + CodeSourcery 2011.03-41 = … Continue.

本文探讨了Linux内核默认使用软件处理ARM处理器未对齐内存访问的问题,并提出两种解决方案:一是通过编译参数调整,二是修改内核配置禁用对齐错误检查,以优化未对齐内存访问。
After some investigation I found that by default Linux kernel uses software handlers (traps) for unaligned memory loads. [1] What does it mean? It means that by default for ARM processors “A bit” in Control Register is set to 1.From “Cortex-A8 Technical Reference Manual” [2]:

[1] | A bit | Banked | Enables strict alignment of data to detect alignment faults in data accesses:
0 = strict alignment fault checking disabled, reset value
1 = strict alignment fault checking enabled.

So at the moment we have default software handling for unaligned memory access faults in kernel and new CodeSourcery 2011.03-41 cross-compiler with “Unaligned access support” feature enabled by default. As I understand, It means that compiler tries to optimize unaligned memory access with assumption that “A bit” in Control Register is unset (reset value) and hardware should handle all unaligned memory loads. But this bit is set by Linux kernel, thus, after memory access optimisation by cross-compiler ARM processor generates alignment faults, which are handled by software traps in kernel.

There are two solutions for this issue. The first one is described in my previous post (“-mno-unaligned-access” parameter). And the second is to disable alignment faults checking in Linux kernel (clear “A bit” in Control Register). I think the second solution is more preferable as hardware unaligned loads support is much faster than software handling. After some investigation I found that alignment traps generation can be controlled using “ALIGNMENT_TRAP” kernel config parameter in “arch/arm/Kconfig”, but this parameter is by default true and it can’t be configured in menuconfig. In order to be able to change it’s value just add “Software alignment trap” text after “bool”:

config ALIGNMENT_TRAP
bool “Software alignment trap”

And unset it in menuconfig:

Kernel Features —>
    [ ] Software alignment trap


After that recompile your kernel. You can check if alignment fault checking is disabled or not in kernel traces:

[ 0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c53c7f

Here “cr=…” is a value of ARM’s Control Register. As you can see “A bit” (bit #1) is set (0x7f = b01111111). After disabling alignment fault checking you should see something like that:

[ 0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c53c7d

Here 0x7d = b01111101 – “A bit” is unset.

P.S. According to kernel documentation disabling alignment fault checking can lead to some unpredictable behavior in some network protocols. At the moment I’m trying to stabilize and test 2.6.39 kernel compiled with CodeSourcery 2011.03-41.

[1] - Linux kernel documentation (“Documentation/unaligned-memory-access.txt”).
[2] – Cortex-A8 Technical Reference Manual.

from:http://blog.galemin.com/tag/codesourcery/

内容概要:本文详细介绍了“秒杀商城”微服务架构的设计与实战全过程,涵盖系统从需求分析、服务拆分、技术选型到核心功能开发、分布式事务处理、容器化部署及监控链路追踪的完整流程。重点解决了高并发场景下的超卖问题,采用Redis预减库存、消息队列削峰、数据库乐观锁等手段保障数据一致性,并通过Nacos实现服务注册发现与配置管理,利用Seata处理跨服务分布式事务,结合RabbitMQ实现异步下单,提升系统吞吐能力。同时,项目支持Docker Compose快速部署和Kubernetes生产级编排,集成Sleuth+Zipkin链路追踪与Prometheus+Grafana监控体系,构建可观测性强的微服务系统。; 适合人群:具备Java基础和Spring Boot开发经验,熟悉微服务基本概念的中高级研发人员,尤其是希望深入理解高并发系统设计、分布式事务、服务治理等核心技术的开发者;适合工作2-5年、有志于转型微服务或提升架构能力的工程师; 使用场景及目标:①学习如何基于Spring Cloud Alibaba构建完整的微服务项目;②掌握秒杀场景下高并发、超卖控制、异步化、削峰填谷等关键技术方案;③实践分布式事务(Seata)、服务熔断降级、链路追踪、统一配置中心等企业级中间件的应用;④完成从本地开发到容器化部署的全流程落地; 阅读建议:建议按照文档提供的七个阶段循序渐进地动手实践,重点关注秒杀流程设计、服务间通信机制、分布式事务实现和系统性能优化部分,结合代码调试与监控工具深入理解各组件协作原理,真正掌握高并发微服务系统的构建能力。
评论
成就一亿技术人!
拼手气红包6.0元
还能输入1000个字符
 
红包 添加红包
表情包 插入表情
 条评论被折叠 查看
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值