Date: Mon, 16 Apr 2018 21:24:24 +0000 From: bugzilla-noreply@freebsd.org To: ppc@FreeBSD.org Subject: [Bug 226974] kernel DSI read trap at boot Message-ID: <bug-226974-21-ewvYjNFRjm@https.bugs.freebsd.org/bugzilla/> In-Reply-To: <bug-226974-21@https.bugs.freebsd.org/bugzilla/> References: <bug-226974-21@https.bugs.freebsd.org/bugzilla/>
next in thread | previous in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D226974 --- Comment #12 from Leandro Lupori <leandro.lupori@gmail.com> --- I have an update on this. I think I've found what is causing this issue. When ofw_mem_regions() is called during initialization, it looks for /chosen/fdtmemreserv property in the device tree. When FDT is enabled, this property will always exist, as it's a "fake" property returned by OFW FDT getprop itself. Then, if the fdtmemreserv property exists, excise_fdt_reserved() is called,= and it does 2 main things: 1- Excludes reserved memory from available memory regions 2- Excludes FDT memory region itself from available memory In my case, it's the second item that causes a problem later, at moea64_early_bootstrap(). There is one part of this function that checks for overlaps between availab= le memory regions and the region where the kernel was loaded. When this overla= p is found, the available memory list is supposed to be adjusted to exclude kern= el memory. But this code doesn't work correctly for some available memory regions, like the one that is produced by excise_fdt_reserved() when there is an overlap between FDT and kernel end. This is the relevant part of the available regions in my case: 0x00004000 - 0x01800074 (~24MB) 0x0180b04c - 0x02c00000 (~20MB) 0x0344d3f0 - 0x7dbe0000 (~2GB) kernelstart =3D 0x0100100 kernelend =3D 0x1b60000 fdt =3D 0x1b5b000 - 0x1b5f000 (upper bound rounded to page size) When FDT is not used, moea64_early_bootstrap() will adjust the regions like this: 0x00004000 - 0x01000000 0x01b60000 - 0x02c00000 0x0344d3f0 - 0x7dbe0000 And everything works fine. However, when FDT is used, excise_fdt_reserved() will change the available regions to this: 0x00004000 - 0x01800074 0x0180b04c - 0x01b5b000 0x01b5f000 - 0x02c00000 0x0344d3f0 - 0x7dbe0000 And then moea64_early_bootstrap() will adjust the regions and produce this: 0x00004000 - 0x01000000 0x0180b04c - 0x01b5b000 0x01b61000 - 0x02c00000 0x0344d3f0 - 0x7dbe0000 Which turns a large part of the kernel memory (0x0180b04c - 0x01b5b000) into available memory, causing the issue when this region starts being used as s= uch. I'm not sure about the best way to fix this. Maybe one (or more) of the following: 1- Change moea64_early_bootstrap() to make it able to detect and adjust reg= ions such as (0x0180b04c - 0x01b5b000), in which case the whole region should be removed from available list. 2- Change the boot loader to place the FDT after kernelend. And one last thing that I noticed and seems odd to me is that region (0x01800074 - 0x0180b04c) is not reported as available, but even so the ker= nel is loaded at (0x0100100 - 0x1b60000), which includes the not available regi= on. --=20 You are receiving this mail because: You are the assignee for the bug.=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-226974-21-ewvYjNFRjm>