* [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map
@ 2026-10-01 8:52 Benjamin Kelenyi
2026-10-13 10:42 ` Konrad Dybcio
2026-10-13 16:35 ` Christopher Obbard
0 siblings, 2 replies; 3+ messages in thread
From: Benjamin Kelenyi @ 2026-10-01 8:52 UTC (permalink / raw)
To: Bjorn Andersson, Konrad Dybcio, Rob Herring, Krzysztof Kozlowski,
Conor Dooley
Cc: Vinayak Katoch, Ekansh Gupta, Dmitry Baryshkov,
Christopher Obbard, Abel Vesa, linux-arm-msm, devicetree,
linux-kernel, Benjamin Kelenyi
The ADSP remote heap is declared as a reusable CMA pool. The ADSP has no
IOMMU, so fastrpc hands the whole region to the DSP with
qcom_scm_assign_mem() when it probes, while CMA keeps lending the pool's
pages to movable allocations. Once memory runs low the kernel touches a
page the DSP now owns, and the board resets without an oops.
On an Arduino UNO Q (QRB2210, 4 GB), writing and reading back 3.5 GB
through the page cache resets the board within seconds. With fastrpc
blacklisted, or with the heap declared no-map, the same test passes
repeatedly.
Declare the heap no-map, so the kernel never uses it, as the other
platforms' remote heaps are.
Fixes: 3fbf3b85bf86 ("arm64: dts: qcom: agatti: Add memory-region for audio PD")
Link: https://lore.kernel.org/all/20260908070331.2189A1F00A3A@smtp.kernel.org/
Signed-off-by: Benjamin Kelenyi <benjamin@wendy.sh>
---
arch/arm64/boot/dts/qcom/agatti.dtsi | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/arm64/boot/dts/qcom/agatti.dtsi b/arch/arm64/boot/dts/qcom/agatti.dtsi
index d9f4cbf..3ff97ea 100644
--- a/arch/arm64/boot/dts/qcom/agatti.dtsi
+++ b/arch/arm64/boot/dts/qcom/agatti.dtsi
@@ -385,7 +385,7 @@
adsp_rpc_remote_heap_mem: adsp-rpc-remote-heap {
compatible = "shared-dma-pool";
alloc-ranges = <0x0 0x0 0x1 0x0>;
- reusable;
+ no-map;
alignment = <0x0 0x400000>;
size = <0x0 0x800000>;
};
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map
2026-10-01 8:52 [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map Benjamin Kelenyi
@ 2026-10-13 10:42 ` Konrad Dybcio
2026-10-13 16:35 ` Christopher Obbard
1 sibling, 0 replies; 3+ messages in thread
From: Konrad Dybcio @ 2026-10-13 10:42 UTC (permalink / raw)
To: Benjamin Kelenyi, Bjorn Andersson, Konrad Dybcio, Rob Herring,
Krzysztof Kozlowski, Conor Dooley
Cc: Vinayak Katoch, Ekansh Gupta, Dmitry Baryshkov,
Christopher Obbard, Abel Vesa, linux-arm-msm, devicetree,
linux-kernel
On 10/1/26 10:52 AM, Benjamin Kelenyi wrote:
> The ADSP remote heap is declared as a reusable CMA pool. The ADSP has no
> IOMMU, so fastrpc hands the whole region to the DSP with
> qcom_scm_assign_mem() when it probes, while CMA keeps lending the pool's
> pages to movable allocations. Once memory runs low the kernel touches a
> page the DSP now owns, and the board resets without an oops.
>
> On an Arduino UNO Q (QRB2210, 4 GB), writing and reading back 3.5 GB
> through the page cache resets the board within seconds. With fastrpc
> blacklisted, or with the heap declared no-map, the same test passes
> repeatedly.
>
> Declare the heap no-map, so the kernel never uses it, as the other
> platforms' remote heaps are.
>
> Fixes: 3fbf3b85bf86 ("arm64: dts: qcom: agatti: Add memory-region for audio PD")
> Link: https://lore.kernel.org/all/20260908070331.2189A1F00A3A@smtp.kernel.org/
> Signed-off-by: Benjamin Kelenyi <benjamin@wendy.sh>
> ---
Reviewed-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
Konrad
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map
2026-10-01 8:52 [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map Benjamin Kelenyi
2026-10-13 10:42 ` Konrad Dybcio
@ 2026-10-13 16:35 ` Christopher Obbard
1 sibling, 0 replies; 3+ messages in thread
From: Christopher Obbard @ 2026-10-13 16:35 UTC (permalink / raw)
To: Benjamin Kelenyi
Cc: Bjorn Andersson, Konrad Dybcio, Rob Herring, Krzysztof Kozlowski,
Conor Dooley, Vinayak Katoch, Ekansh Gupta, Dmitry Baryshkov,
Christopher Obbard, Abel Vesa, linux-arm-msm, devicetree,
linux-kernel
On Thu, 01 Oct 2026 11:52:47 +0300, Benjamin Kelenyi <benjamin@wendy.sh> wrote:
> The ADSP remote heap is declared as a reusable CMA pool. The ADSP has no
> IOMMU, so fastrpc hands the whole region to the DSP with
> qcom_scm_assign_mem() when it probes, while CMA keeps lending the pool's
> pages to movable allocations. Once memory runs low the kernel touches a
> page the DSP now owns, and the board resets without an oops.
>
> On an Arduino UNO Q (QRB2210, 4 GB), writing and reading back 3.5 GB
> through the page cache resets the board within seconds. With fastrpc
> blacklisted, or with the heap declared no-map, the same test passes
> repeatedly.
>
> [...]
Reviewed-by: Christopher Obbard <chris.obbard@oss.qualcomm.com>
--
Christopher Obbard
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-10-13 16:35 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-01 8:52 [PATCH] arm64: dts: qcom: agatti: Make the ADSP remote heap no-map Benjamin Kelenyi
2026-10-13 10:42 ` Konrad Dybcio
2026-10-13 16:35 ` Christopher Obbard
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®