Re: Future of armv7
- In reply to: Minsoo Choo : "Re: Future of armv7"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Wed, 19 Nov 2025 14:56:48 UTC
On Tue, 18 Nov 2025 01:56:08 +0000 Minsoo Choo <minsoochoo0122@proton.me> wrote: > - Maintaining at least one 32-bit platform makes porting FreeBSD to new > 32-bit platforms easier in future. > > Following is why I decided to vote towards the removal. > - Usage of armv7 hardwares have been quickly dwindling in the past few > years. Since I started studying in univeristy, the only armv7 I've seen is > ARM Cortex-M for old stm32 courses. Even including Cortex-M, I don't see > anyone around me who uses armv7 in their project or work. stable/15 will > give 3 more years of support for armv7, and I expect that people will use > armv7 dramatically less by then. > - Warner mentioned OOM issues. 32-bit architectures usually have maximum > memory capacity of 4GB, and introducing another 32-bit architecture won't > solve this problem (unless they come up with something like PAE, but why > would anyone come up with PAE if there are 64-bit architectures?). Thus, > claims like "maintain at least one 32-bit architecture for future 32-bit > architecture" don't really make sense to me. Even if we assume there will be > a revolutionary 32-bit platform that can replace all existing 64-bit > platforms under special circumstances, how do we know when the day will > come? People in this list have said that there might be a future 32-bit > platform, but I haven't heard what the plan is if there is NO such > architecture in future. How can we tell? When is the time to remove armv7 if > we really don't see another 32-bit alternative for decades? > - Cost of maintaining, testing, and releasing FreeBSD on these platforms. As > FreeBSD build system encompasses different architectures and operating > systems should abstract different hardwares, armv6 might affect code other > than its own architecture-specific code, such as jemalloc that Warner have > mentioned. OOM and compiling is not an issue when the system is just target. For example, in VM/clouds, even i386 images are used because they use less memory so you can buy more of them for same price. Another target is embedded/IoT world: while vast majority of constrained devices have less than dozen Kb of RAM (see IEEE 802.15.4/6LoWPAN world, C2 class by RFC 7228), situation may change towards 32-bits addressing in future when 32-bit may become more cheap than current 32-bit architectures. -- WBR, @nuclight