> On 26 Sep 2023, at 11:04, Michal Orzel <[email protected]> wrote: > > Hi Luca, > > On 26/09/2023 11:52, Luca Fancellu wrote: >> >> >>> On 26 Sep 2023, at 06:33, Leo Yan <[email protected]> wrote: >>> >>> During the Linux kernel booting, an error is reported by the Xen >>> hypervisor: >>> >>> (XEN) arch/arm/p2m.c:2202: d0v0: Failing to acquire the MFN 0x1a02dc >>> >>> The kernel attempts to use an invalid memory frame number, which can be >>> converted to: 0x1a02dc << PAGE_SHIFT, resulting in 0x1_a02d_c000. >>> >>> The invalid memory frame falls into a reserved memory node, which is >>> defined in the device tree as below: >>> >>> reserved-memory { >>> #address-cells = <0x02>; >>> #size-cells = <0x02>; >>> ranges; >>> >>> ... >>> >>> ethosn_reserved { >>> compatible = "shared-dma-pool"; >>> reg = <0x01 0xa0000000 0x00 0x20000000>; >>> status = "disabled"; >>> no-map; >>> }; >>> >>> ... >>> }; >>> >>> Xen excludes all reserved memory regions from the frame management >>> through the dt_unreserved_regions() function. On the other hand, the >>> reserved memory nodes are passed to the Linux kernel. However, the Linux >>> kernel ignores the 'ethosn_reserved' node since its status is >>> "disabled". This leads to the Linux kernel to allocate pages from the >>> reserved memory range. >> >> I might be wrong, but reading the specifications seems that “status” is not >> a property >> of the child nodes of /reserved-memory, so I’m not sure Xen should do >> something about it. > Please take a look at dt documentation (v0.4) for /memory and > /reserved-memory. > Under the tables listing possible properties, there is a statement: > Note: All other standard properties (Section 2.3) are allowed but are > optional.
Thanks for pointing that out, I missed that bit. > > "status" is part of standard properties so it is perfectly fine for /memory > or /reserved-memory > nodes to have it defined. > > ~Michal
