Christophe Leroy <christophe.leroy@xxxxxxxxxx> writes: > Le 21/06/2022 à 20:47, Jason A. Donenfeld a écrit : >> On Tue, Jun 21, 2022 at 06:33:11PM +0000, Christophe Leroy wrote: >>> Le 21/06/2022 à 16:08, Jason A. Donenfeld a écrit : >>>> The platform's RNG must be available before random_init() in order to be >>>> useful for initial seeding, which in turn means that it needs to be >>>> called from setup_arch(), rather than from an init call. Fortunately, >>>> each platform already has a setup_arch function pointer, which means we >>>> can wire it up that way. Complicating things, however, is that POWER8 >>>> systems need some per-cpu state and kmalloc, which isn't available at >>>> this stage. So we split things up into an early phase and a later >>>> opportunistic phase. This commit also removes some noisy log messages >>>> that don't add much. >>> >>> Regarding the kmalloc(), I have not looked at it in details, but usually >>> you can use memblock_alloc() when kmalloc is not available yet. >> >> That seems a bit excessive, especially as those allocations are long >> lived. And we don't even *need* it that early, but just before >> random_init(). Michael is running this v5 on the test rig overnight, so >> we'll learn in the Australian morning whether this finally did the trick >> (I hope). > > The fact that they are long lived make them a good candidate for > memblock_alloc(). > > But fair enough, if they are not required that early then just do it later. memblock works but then we trip on ioremap vs early_ioremap. Fixing that is a bit of a pain as we'd have to stop using of_iomap() and we'd also need to switch the mappings to ioremap() later in boot. We'd also have to defer the percpu initialisation. So it's all just a bit of a pain when we actually only need to get the hook ready before random_init() which is called much later in boot when slab/ioremap/percpu are all ready. cheers
