| Message ID | c7063c79-0e34-f1be-35ec-56d1cf993a2f@nokia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | arm: Support initrd with address in boot alias region | expand |
+Rob, Russell, On 9/29/2020 12:54 AM, Matija Glavinic Pecotic wrote: > If bootloader passes address of initrd in boot alias region, initrd > will fail on memblock_is_region_memory as memblock with such address > doesn't exist. > > Issue was observed with kexec which passed initrd address in boot alias > region, while bootloader will typically pass physical address. Commit > fe7db7570379 ("of/fdt: Populate phys_initrd_start/phys_initrd_size from FDT") > uncovered problem by removing virt_to_phys on the initrd physical > address. __virt_to_phys on our platform coincidentally fixed address. > > Fix by trying to correct address in case of invalid physical address. > > Signed-off-by: Matija Glavinic Pecotic <matija.glavinic-pecotic.ext@nokia.com> This should have a Fixes tag: fe7db7570379 ("of/fdt: Populate phys_initrd_start/phys_initrd_size from FDT") I am not familiar enough with how the identity map works, but it seems to do what you want it to.
On Tue, Sep 29, 2020 at 10:21:35AM -0700, Florian Fainelli wrote: > +Rob, Russell, > > On 9/29/2020 12:54 AM, Matija Glavinic Pecotic wrote: > > If bootloader passes address of initrd in boot alias region, initrd > > will fail on memblock_is_region_memory as memblock with such address > > doesn't exist. > > > > Issue was observed with kexec which passed initrd address in boot alias > > region, while bootloader will typically pass physical address. Commit > > fe7db7570379 ("of/fdt: Populate phys_initrd_start/phys_initrd_size from FDT") > > uncovered problem by removing virt_to_phys on the initrd physical > > address. __virt_to_phys on our platform coincidentally fixed address. > > > > Fix by trying to correct address in case of invalid physical address. > > > > Signed-off-by: Matija Glavinic Pecotic <matija.glavinic-pecotic.ext@nokia.com> > > This should have a Fixes tag: > > fe7db7570379 ("of/fdt: Populate phys_initrd_start/phys_initrd_size from > FDT") > > I am not familiar enough with how the identity map works, but it seems to do > what you want it to. It seems strange to pass the initrd in using an address that is not listed as part of the system memory map, especially as the initrd is freed during kernel initialisation. Which boot loader is this, and is there a reason it operates this way?
On 09/30/2020 12:35 PM, Russell King - ARM Linux admin wrote: > It seems strange to pass the initrd in using an address that is not > listed as part of the system memory map, especially as the initrd is > freed during kernel initialisation. > > Which boot loader is this, and is there a reason it operates this way? It is kexec which populates this address. Kexec is allowed to do that as this region is listed as boot alias: $ cat /proc/iomem [cut] c0000000-d4da1fff : System RAM (boot alias) 840000000-854da1fff : System RAM 840008000-840bfffff : Kernel code 840e00000-840ee56fb : Kernel data [cut] 880000000-8ffffffff : System RAM it is arch/arm/kernel/setup.c which populates this entry on line 870: /* * Some systems have a special memory alias which is only * used for booting. We need to advertise this region to * kexec-tools so they know where bootable RAM is located. */ boot_alias_start = phys_to_idmap(start); if (arm_has_idmap_alias() && boot_alias_start != IDMAP_INVALID_ADDR) { res = memblock_alloc(sizeof(*res), SMP_CACHE_BYTES); if (!res) panic("%s: Failed to allocate %zu bytes\n", __func__, sizeof(*res)); res->name = "System RAM (boot alias)"; res->start = boot_alias_start; res->end = phys_to_idmap(end); res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; request_resource(&iomem_resource, res); } This is likely reason why problem has gone unnoticed for so long. I dont think there is large number of platforms which have boot alias regions + they use kexec. As noted earlier, bootloader passes correct address, it is kexec which passes invalid one resulting with: [ 0.000000] INITRD: 0xc49bb000+0x01a1a000 is not a memory region - disabling initrd and later with failure to mount it. Regards, Matija
diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c index 000c1b48e973..6cbf64383cfc 100644 --- a/arch/arm/mm/init.c +++ b/arch/arm/mm/init.c @@ -164,6 +164,17 @@ static void __init arm_initrd_init(void) if (!phys_initrd_size) return; + /* + * In case initrd is not valid physical address, try with boot + * memory region alias as bootloader may have passed address + * within. + */ + if (!pfn_valid(__phys_to_pfn(phys_initrd_start))) { + pr_info("initrd at invalid address, trying with boot alias\n"); + if (arm_has_idmap_alias()) + phys_initrd_start = idmap_to_phys(phys_initrd_start); + } + /* * Round the memory region to page boundaries as per free_initrd_mem() * This allows us to detect whether the pages overlapping the initrd
If bootloader passes address of initrd in boot alias region, initrd will fail on memblock_is_region_memory as memblock with such address doesn't exist. Issue was observed with kexec which passed initrd address in boot alias region, while bootloader will typically pass physical address. Commit fe7db7570379 ("of/fdt: Populate phys_initrd_start/phys_initrd_size from FDT") uncovered problem by removing virt_to_phys on the initrd physical address. __virt_to_phys on our platform coincidentally fixed address. Fix by trying to correct address in case of invalid physical address. Signed-off-by: Matija Glavinic Pecotic <matija.glavinic-pecotic.ext@nokia.com> --- arch/arm/mm/init.c | 11 +++++++++++ 1 file changed, 11 insertions(+)