On Wed, Jul 08, 2026 at 02:50:04PM +0200, Thierry Reding wrote:

On Tue, Jul 07, 2026 at 12:27:13PM +0100, Will Deacon wrote:

quoted

On Mon, Jul 06, 2026 at 03:49:24PM +0200, Thierry Reding wrote:

quoted

On Fri, Jul 03, 2026 at 06:13:31PM +0100, Will Deacon wrote:

quoted

On Thu, Jul 02, 2026 at 06:41:23PM +0200, Thierry Reding wrote:

quoted

On Thu, Jul 02, 2026 at 03:46:44PM +0200, Thierry Reding wrote:

quoted

On Thu, Jul 02, 2026 at 10:18:47AM +0100, Will Deacon wrote:

quoted

On Wed, Jul 01, 2026 at 06:08:15PM +0200, Thierry Reding wrote:

quoted

From: Chun Ng <redacted>

Add helpers to swap PROT_NORMAL and PROT_DEVICE_nGnRnE protection bits
on a kernel-linear-map range.

That sounds like a really terrible idea. Why is this necessary and how
does it interact with things like load_unaligned_zeropad()?

This is necessary because once the memory controller has walled off the
new memory region the CPU must not access it under any circumstances or
it'll cause the CPU to lock up (I think technically it'll hit an SError
but in practice that just means it'll freeze, as far as I can tell).

Probably doesn't interact well at all with load_unaligned_zeropad().

quoted

I think you should unmap the memory from the linear map and memremap()
it instead.

Given that the memory can never be accessed by the CPU after the memory
controller locks it down, I don't think we'll even need memremap(). The
only thing we really need is the sg_table we hand out via the DMA BUFs
so that they can be used by device drivers to program their DMA engines
internally.

Looking through some of the architecture code around this, shouldn't we
simply be using set_memory_encrypted() and set_memory_decrypted() for
this? While they might've been created for slightly other use-cases,
they seem to be doing exactly what we want (i.e. remove the page range
from the linear mapping and flushing it, or restoring the valid bit and
standard permissions, respectively).

Ah... I guess we can't do it because we're not in a realm world and so
the early checks in __set_memory_enc_dec() would return early and turn
it into a no-op.

How about if I extract a common helper and provide set_memory_p() and
set_memory_np() in terms of those. Those are available on x86 and
PowerPC as well, so fairly standard. I suppose at that point we're
closer to set_memory_valid().

Why not just call set_direct_map_invalid_noflush() +
flush_tlb_kernel_range() for each page? We already have APIs for this.

Having a "standard" helper with a fixed and documented purposed seemed
like a preferable approach for this particular case. We also may want to
make the driver that uses this buildable as a module, in which case we'd
need to export these rather low-level APIs. And then there's also the
fact that we typically call this on a rather large region of memory
(usually something like 512 MiB), so doing it page-by-page is rather
suboptimal.

quoted

The big challenge I see with any linear map manipulation, however, is
that it will rely on can_set_direct_map() which likely means you need to
give up some performance and/or security to make this work. Does memory
become inaccesible dynamically at runtime? If not, the best bet would
be to describe it as a carveout in the DT and mark it as "no-map" so
we avoid mapping it in the first place.

VPR exists in two modes: static and resizable. For static VPR we do
exactly that: describe it as carveout in DT with no-map and deal with it
accordingly in the driver. Resizable VPR is for device that have small
amounts of RAM. Content-protected video playback will in the worst case
consume around 1.8 GiB of RAM, so we want to be able to reuse for other
purposes when VPR is unused on those devices. In that case, the memory
is also described as a reserved-memory region in DT, but it is marked as
reusable so that it can be managed by CMA.

The resize operation is fairly slow to begin with because we need to
stall the GPU and put it into reset before the operation, then take it
out of reset and resume it afterwards.

What kind of performance impact do you expect?

You'll need to measure it, but we've seen reports of double-digit
percentage regressions in performance and power. As I said, the problem
is that you need to split the linear map to 4k page at runtime to unmap
the dynamic carveout, but that isn't something that can be done on most
CPUs. Therefore you end up having to use page-granular mappings for the
entire thing, similarly to how 'rodata_full' drives can_set_direct_map()
and the perf/power hit affects everything.

It's hard to know what to suggest... I wonder if any of the memory
hotplug logic could help here?

I've read up on memory hotplug a bit and it sounds like it could fit
this really nicely. Given that we only use CMA (along with the extra
patches to it) to make sure that any buffers are reclaimed for VPR use,
we should be able to get rid of the CMA usage altogether and replace it
with online_pages() and offline_pages() instead. Rather than using a
fixed set of CMA areas like we currently do, each "chunk" in the VPR
driver could represent a memory block instead (which looks like it will
be 128 MiB for 4 KiB pages and 512 MiB for 64 KiB pages). We currently
use 512 MiB as the chunk size, so it should be relatively similar and
easy to adjust.

One issue that we would absolutely need this memory to be ZONE_MOVABLE
from the start. Using no-map in DT and then online_pages() probably will
not work because there's no struct page for the memory. So we're left
with keeping the memory onlined by default, in which case we'd need some
way for DT to instruct the memory to be put into ZONE_MOVABLE always.

There's a "hotpluggable" property for "memory" nodes, maybe that can be
extended to apply to reserved-memory nodes as well?

I haven't been having much success with this. memblock_mark_hotplug()
doesn't have much of an effect because the kernel clears this flag
automatically at some point, so by the time the movable zone is created
there's no memory left that's marked hotpluggable. I don't know if it's
a good idea to modify the code to keep the flag.

Another thing I briefly tried was to use add_memory_driver_managed()
together with the no-map flag in an attempt to get the memory explicitly
added as movable, but that fails because __request_resource() notices
that the reserved memory is actually part of the system RAM that was
registered earlier.

I think in order for this to work the bindings would probably need to
change, such that reserved-memory nodes aren't used but it's described
using the memory nodes instead. That way a piece of system RAM could be
carved out and added by the VPR driver. I don't know if the kernel would
like this kind of splicing of the system RAM, though.

It's all very close to what I need for this, but doesn't quite fit. Any
ideas which of the options is best? Right now it sounds like finding a
way to make this region explicitly ZONE_MOVABLE would be the best. That
should allow offline_pages() and online_pages() to be used, which seems
like the cleanest approach.

Thierry