Hello,

On Tue, Dec 14, 2021 at 05:29:22PM +0100, Geert Uytterhoeven wrote:

Hi Vladimir and Dennis,

On Wed, Dec 1, 2021 at 12:53 PM Vladimir Murzin [off-list ref] wrote:

quoted

On 11/30/21 5:41 PM, Dennis Zhou wrote:

quoted

On Tue, Nov 30, 2021 at 05:29:54PM +0000, Vladimir Murzin wrote:

quoted

Currently, NOMMU pull km allocator via !SMP dependency because most of
them are UP, yet for SMP+NOMMU vm allocator gets pulled which:

* may lead to broken build [1]
* ...or not working runtime due to [2]

It looks like SMP+NOMMU case was overlooked in bbddff054587 ("percpu:
use percpu allocator on UP too") so restore that.

[1]
For ARM SMP+NOMMU (R-class cores)

arm-none-linux-gnueabihf-ld: mm/percpu.o: in function `pcpu_post_unmap_tlb_flush':
mm/percpu-vm.c:188: undefined reference to `flush_tlb_kernel_range'

[2]
static inline
int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
                pgprot_t prot, struct page **pages, unsigned int page_shift)
{
       return -EINVAL;
}

Signed-off-by: Vladimir Murzin <redacted>
---
 mm/Kconfig | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mm/Kconfig b/mm/Kconfig
index d16ba92..66331e0 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig

@@ -425,9 +425,8 @@ config THP_SWAP
 # UP and nommu archs use km based percpu allocator
 #
 config NEED_PER_CPU_KM
-    depends on !SMP
     bool
-    default y
+    default !SMP || !MMU

Should this be `depends on !SMP || !MMU` with default yes? Because with
SMP && MMU, it shouldn't be an option to run with percpu-km.

IIUC these are equivalent, truth table would not change if is under "depends"
or "default"

SMP    MMU   NEED_PER_CPU_KM
 y      y    !y || !y => n || n => n
 y      n    !y || !n => n || y => y
 n      y    !n || !y => y || n => y
 n      n    !n || !n => y || y => y

quoted

quoted

 config CLEANCACHE
     bool "Enable cleancache driver to cache clean pages if tmem is present"
--
2.7.4

It's interesting to me that this is all coming up at once. Earlier this
month I had the same conversation with people involved with sh [1].

[1] https://lore.kernel.org/linux-sh/YY7tp5attRyK42Zk@fedora/ (local)

I can pull this shortly once I see whatever happened to linux-sh.

Ahh, good to know! Adding SH folks here (start of discussion [0]). I see you came
to the same conclusion, right?

IIRC, RISC-V also have SMP+NOMMU, so adding them as well.

I had seen the j-Core thread, but completely forgot about
Canaan K210 (RV64 SMP+NOMMU).

This became commit 3583521aabac76e5 ("percpu: km: ensure it is used
with NOMMU (either UP or SMP)").  And now booting K210 prints:

    percpu: wasting 10 pages per chunk

a) Is this bad?

It's not great.. Can you share the line on boot with the following
prefix: pcpu-alloc [1].

I'm a little surprised here because this means it's allocating not
against the right atomic size. I don't necesarily think it's an issue of
switching from percpu-vm to percpu-km.

b) What exactly was this fixing, and how would I trigger the bad case
   on K210 before, if it was affected at all?

In v5.14, I merged Roman's request for percpu depopulation [2]. This
required calls to flush the tlb. There is an abstraction layer:
percpu-vm vs percpu-km. So if an architecture is using percpu-vm but
doesn't have an MMU AND doesn't map out appropriately the tlb flush
call, then it fails. This happened on arm + sh architectures. Now RV
might be mapping it out appropriately so they never saw the issue.

Now, there is also a bigger caveat with using percpu-vm without an MMU.
In percpu-vm, we allocate pages on demand and map them in into
pre-allocated vmas. This means there are 2 scenarios that I haven't
looked into deeper. 1, the vma alloc maps to allocating physical pages.
2, we're lucky percpu allocates backwards in the vma so we haven't had a
collision problem yet.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/dennis/percpu.git/tree/mm/percpu.c#n2507
[2] https://lore.kernel.org/lkml/20210419225047.3415425-1-dennis@kernel.org/ (local)

Thanks,
Dennis

quoted

[0] https://lore.kernel.org/linux-mm/20211130172954.129587-1-vladimir.murzin@arm.com/T/ (local)

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds