Linux 调频策略引发讨论 华为建议默认开启 CPU Boost Arm 工程师提出异议Linux FM strategy sparks discussion Huawei suggests turning on CPU Boost Arm by default Engineer raises objection


AI free token/AI免费积分

IT之家 10 月 16 日消息,Phoronix 发现,华为工程师张鹏杰(Pengjie Zhang)昨日向 Linux 内核提交补丁,提议在兼容相关能力的前提下默认启用 ACPI CPPC CPUFreq 驱动中的 CPU Boost 模式。为此,该补丁修改了 cppc_cpufreq.c 文件,新增了 20 行代码用于初始化 Boost 默认状态。

ACPI CPPC(Collaborative Processor Performance Control,协作式处理器表现控制)是一套由 ACPI 定义的处理器表现控制机制,操作平台可借助它与平台固件交互,调节处理器表现。

在当前通用 ACPI CPPC 驱动下,当处理器的最高表现等级(Highest Performance)高于标称表现等级(Nominal Performance)时,Boost 模式默认处于禁用状态,使用者必须借助 sysfs 手动启动。

Phoronix 指出,ACPI CPPC 驱动广泛用于 ARM64 Linux 服务器平台,而对于使用这些平台的使用者而言更希望保持峰值表现。

Linux 调频策略引发讨论 华为建议默认开启 CPU Boost Arm 工程师提出异议

张鹏杰表示,只要有至少一个 CPU 兼容 Boost,就应该自动为所有兼容该能力的策略启用 Boost。

不过,Arm 工程师、Linux 开发者 Christian Loehle 对此提出了不同看法。他认为,现代处理器的功耗和热外观约束太过于复杂,Boost 模式本身并不能保证处理器不会降频。

我时常在想笔者是否应该完全保留 Boost,如今芯片的功耗和散热边界已经足够复杂,以至于 Boost 的意义不大,它既不能保证不会降频,也不一定能取得请求的表现。

他指出,懂行的使用者可以自行借助 scaling_max_freq 单独设置表现上限,使其适应不同平台、工作负载和运行环境。

Loehle 还提及,华为也曾引入过另一个相关 commit(54e74df5d76d),他对此的使用情形感到好奇,并认为在 Linux 上具备一个标称表现不等于最高表现的平台当下只会带来更多麻烦。

IT之家提醒:此次补丁针对的是通用 ACPI CPPC CPUFreq 驱动,并非面向所有 Linux 处理器平台统一启动 Boost。最终是否合并,仍取决于 Linux 内核维护者后续的讨论与审查。

本文来源:https://www.ithome.com/1/014/175.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

IT House October 16 news, Phoronix found that Huawei engineer Pengjie Zhang submitted a patch to the Linux kernel yesterday, proposing to enable the CPU Boost mode in the ACPI CPPC CPUFreq driver by default under the premise of compatibility with related capabilities. To this end, the patch modifies the cppc_cpufreq.c file and adds 20 lines of code to initialize the Boost default state.

ACPI CPPC (Collaborative Processor Performance Control) is a set of processor performance control mechanisms defined by ACPI. The operating platform can interact with the platform firmware to adjust the processor performance.

Under the current generic ACPI CPPC driver, the Boost mode is disabled by default when the Highest Performance of the processor is higher than the Nominal Performance, and the user must start it manually with sysfs.

Phoronix noted that the ACPI CPPC driver is widely used in ARM64 Linux server platforms, and for users using these platforms, it is more desirable to maintain peak performance.

Zhang Pengjie said that as long as at least one CPU is compatible with Boost, Boost should be automatically enabled for all policies that are compatible with that capability.

Linux FM strategy sparks discussion Huawei suggests turning on CPU Boost Arm by default Engineer raises objection

However, Arm engineer and Linux developer Christian Loehle disagrees. He believes that the power consumption and thermal appearance constraints of modern processors are too complicated, and the Boost mode itself does not guarantee that the processor will not degrade.

I often wonder if I should keep Boost completely. Nowadays, the power consumption and heat dissipation boundaries of the chip are complicated enough that Boost is of little significance. It can neither guarantee that it will not reduce the frequency, nor necessarily achieve the requested performance.

He pointed out that knowledgeable users can set their own performance caps with scaling_max_freq to adapt to different platforms, workloads, and operating environments.

Loehle also mentioned that Huawei has also introduced another related commit (54e74df5d76d), which he is curious about the use case and believes that having a platform with a nominal performance that is not equal to the highest performance on Linux will only bring more trouble.

IT Home Reminder: This patch is for the generic ACPI CPPC CPUFreq driver and does not launch Boost uniformly for all Linux processor platforms. The eventual merge remains subject to subsequent discussion and review by Linux kernel maintainers.

This article is from: https://www.ithome.com/1/014/175.htm For copyright issues, please write to: 13951431913#139.com or submit a review on the right.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry