[PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[He Zhe]

OE uses qemumips to simulate a Malta board by default.

As upstream qemu introduced:
https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b

The Malta board can support up to 2GiB of RAM which should
be able to boot a Linux kernel built with CONFIG_HIGHMEM
enabled and passing "-m 2048" to QEMU and appending the
following kernel parameters:
...
mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
...

But the following commit in kernel broke above mem=X@Y setting
which added the memory as reserved memory area.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
...
commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
Date:   Wed Nov 23 14:43:49 2016 +0100

    MIPS: fix mem=X@Y commandline processing
...

So remove `mem=*' to disable user-defined physical RAM map
which let kernel itself caculates memory ranges.

Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: He Zhe <zhe.he@windriver.com>
---
 scripts/runqemu | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/scripts/runqemu b/scripts/runqemu
index de42d0f323..597e7e9a79 100755
--- a/scripts/runqemu
+++ b/scripts/runqemu
@@ -668,7 +668,10 @@ class BaseConfig(object):
             logger.info('QB_MEM is not set, use 512M by default')
             self.set('QB_MEM', '-m 512')
 
-        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
+        mach = self.get('MACHINE')
+        if mach != 'qemumips':
+            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
+
         self.qemu_opt_script += ' %s' % self.get('QB_MEM')
 
     def check_tcpserial(self):
-- 
2.11.0

[PATCH 2/2] scripts/runqemu: fix qemumips64 with 512M RAM caused kernel panic

[He Zhe]

$ runqemu qemumips64 core-image-minimal nographic qemuparams="-m 512"
...
[    0.000000] Call Trace:
[    0.000000] [<ffffffff801268c0>] clear_page+0x0/0x128
[    0.000000] [<ffffffff80238158>] get_page_from_freelist+0xab8/0xc00
[    0.000000] [<ffffffff80238964>] __alloc_pages_nodemask+0xdc/0xf68
[    0.000000] [<ffffffff80239808>] __get_free_pages+0x18/0x70
[    0.000000] [<ffffffff80122a4c>] setup_zero_pages+0x1c/0xb8
[    0.000000] [<ffffffff80c7c998>] mem_init+0x54/0xa0
[    0.000000] [<ffffffff80c74904>] start_kernel+0x204/0x4d8
[    0.000000] [<ffffffff8091dfb0>] kernel_entry+0x0/0x40
[    0.000000] Code: 02002025  1000f8d9  8e634d7c <34860f80> cc9e0000
cc9e0020  cc9e0040  cc9e0060  cc9e0080
[    0.000000]
[    0.000000] ---[ end trace 0000000000000000 ]---
[    0.000000] Kernel panic - not syncing: Attempted to kill the idle task!
[    0.000000] ---[ end Kernel panic - not syncing: Attempted to kill the idle task!
...

Remove `mem=*' to disable user-defined physical RAM map which
let kernel itself caculates memory ranges.

Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: He Zhe <zhe.he@windriver.com>
---
 scripts/runqemu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/runqemu b/scripts/runqemu
index 597e7e9a79..78f5c8b778 100755
--- a/scripts/runqemu
+++ b/scripts/runqemu
@@ -669,7 +669,7 @@ class BaseConfig(object):
             self.set('QB_MEM', '-m 512')
 
         mach = self.get('MACHINE')
-        if mach != 'qemumips':
+        if mach != 'qemumips' and mach != 'qemumips64':
             self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
 
         self.qemu_opt_script += ' %s' % self.get('QB_MEM')
-- 
2.11.0

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Khem Raj]

i think you can squash both patches into one. Additionally we should
be able to cover mipsel case as well
On Tue, Jul 3, 2018 at 11:57 PM He Zhe <zhe.he@windriver.com> wrote:
>
> OE uses qemumips to simulate a Malta board by default.
>
> As upstream qemu introduced:
> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b
>
> The Malta board can support up to 2GiB of RAM which should
> be able to boot a Linux kernel built with CONFIG_HIGHMEM
> enabled and passing "-m 2048" to QEMU and appending the
> following kernel parameters:
> ...
> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
> ...
>
> But the following commit in kernel broke above mem=X@Y setting
> which added the memory as reserved memory area.
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> ...
> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
> Date:   Wed Nov 23 14:43:49 2016 +0100
>
>     MIPS: fix mem=X@Y commandline processing
> ...
>
> So remove `mem=*' to disable user-defined physical RAM map
> which let kernel itself caculates memory ranges.
>
> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
> Signed-off-by: He Zhe <zhe.he@windriver.com>
> ---
>  scripts/runqemu | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/runqemu b/scripts/runqemu
> index de42d0f323..597e7e9a79 100755
> --- a/scripts/runqemu
> +++ b/scripts/runqemu
> @@ -668,7 +668,10 @@ class BaseConfig(object):
>              logger.info('QB_MEM is not set, use 512M by default')
>              self.set('QB_MEM', '-m 512')
>
> -        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +        mach = self.get('MACHINE')
> +        if mach != 'qemumips':
> +            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +
>          self.qemu_opt_script += ' %s' % self.get('QB_MEM')
>
>      def check_tcpserial(self):
> --
> 2.11.0
>
> --
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[He Zhe]

On 2018年07月04日 21:29, Khem Raj wrote:
> i think you can squash both patches into one. Additionally we should
> be able to cover mipsel case as well

I started a build with latest poky and only set DEFAULTTUE=mipsel. Is this the right way to use little endian?

Then I got the following error and found kernel and kernel modules output are all big endian.

ERROR: linux-yocto-4.14.48+gitAUTOINC+94457657b8_6c2433d7c5-r0 do_package_qa: QA Issue: Endiannes did not match (1 to 0) on /work/qemumips-poky-linux/linux-yocto/4.14.48+gitAUTOINC+94457657b8_6c2433d7c5-r0/packages-split/kernel-module-i2c-dev-4.14.48-yocto-standard/lib/modules/4.14.48-yocto-standard/kernel/drivers/i2c/i2c-dev.ko [arch]

Then I tried to build a empty file.
tmp/work/x86_64-linux/gcc-cross-mipsel/8.1.0-r0/recipe-sysroot-native/usr/bin/mipsel-poky-linux/mipsel-poky-linux-gcc -c a.c
as: unrecognized option '-EL'

Does it mean there something wrong with little endian toolchain?

Zhe

> On Tue, Jul 3, 2018 at 11:57 PM He Zhe <zhe.he@windriver.com> wrote:
>> OE uses qemumips to simulate a Malta board by default.
>>
>> As upstream qemu introduced:
>> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b
>>
>> The Malta board can support up to 2GiB of RAM which should
>> be able to boot a Linux kernel built with CONFIG_HIGHMEM
>> enabled and passing "-m 2048" to QEMU and appending the
>> following kernel parameters:
>> ...
>> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
>> ...
>>
>> But the following commit in kernel broke above mem=X@Y setting
>> which added the memory as reserved memory area.
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
>> ...
>> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
>> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
>> Date:   Wed Nov 23 14:43:49 2016 +0100
>>
>>     MIPS: fix mem=X@Y commandline processing
>> ...
>>
>> So remove `mem=*' to disable user-defined physical RAM map
>> which let kernel itself caculates memory ranges.
>>
>> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
>> Signed-off-by: He Zhe <zhe.he@windriver.com>
>> ---
>>  scripts/runqemu | 5 ++++-
>>  1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/scripts/runqemu b/scripts/runqemu
>> index de42d0f323..597e7e9a79 100755
>> --- a/scripts/runqemu
>> +++ b/scripts/runqemu
>> @@ -668,7 +668,10 @@ class BaseConfig(object):
>>              logger.info('QB_MEM is not set, use 512M by default')
>>              self.set('QB_MEM', '-m 512')
>>
>> -        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
>> +        mach = self.get('MACHINE')
>> +        if mach != 'qemumips':
>> +            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
>> +
>>          self.qemu_opt_script += ' %s' % self.get('QB_MEM')
>>
>>      def check_tcpserial(self):
>> --
>> 2.11.0
>>
>> --
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core@lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-core

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Khem Raj]

we need to define a new machine config for qemumipsel.conf something like below

#@TYPE: Machine
#@NAME: mti_malta32_le MIPS
#@DESCRIPTION: mti_malta32_le

require conf/machine/include/qemu.inc
require conf/machine/include/tune-mips32r2.inc
require conf/machine/include/qemuboot-mips.inc

DEFAULTTUNE = "mips32r2el"

KERNEL_IMAGETYPE = "vmlinux"
KERNEL_ALT_IMAGETYPE = "vmlinux.bin"

SERIAL_CONSOLES ?= "115200;ttyS0 115200;ttyS1"

MACHINE_EXTRA_RRECOMMENDS = " kernel-modules"
On Thu, Jul 5, 2018 at 2:37 AM He Zhe <zhe.he@windriver.com> wrote:
>
>
>
> On 2018年07月04日 21:29, Khem Raj wrote:
> > i think you can squash both patches into one. Additionally we should
> > be able to cover mipsel case as well
>
> I started a build with latest poky and only set DEFAULTTUE=mipsel. Is this the right way to use little endian?
>
> Then I got the following error and found kernel and kernel modules output are all big endian.
>
> ERROR: linux-yocto-4.14.48+gitAUTOINC+94457657b8_6c2433d7c5-r0 do_package_qa: QA Issue: Endiannes did not match (1 to 0) on /work/qemumips-poky-linux/linux-yocto/4.14.48+gitAUTOINC+94457657b8_6c2433d7c5-r0/packages-split/kernel-module-i2c-dev-4.14.48-yocto-standard/lib/modules/4.14.48-yocto-standard/kernel/drivers/i2c/i2c-dev.ko [arch]
>
> Then I tried to build a empty file.
> tmp/work/x86_64-linux/gcc-cross-mipsel/8.1.0-r0/recipe-sysroot-native/usr/bin/mipsel-poky-linux/mipsel-poky-linux-gcc -c a.c
> as: unrecognized option '-EL'
>
> Does it mean there something wrong with little endian toolchain?
>
> Zhe
>
> > On Tue, Jul 3, 2018 at 11:57 PM He Zhe <zhe.he@windriver.com> wrote:
> >> OE uses qemumips to simulate a Malta board by default.
> >>
> >> As upstream qemu introduced:
> >> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b
> >>
> >> The Malta board can support up to 2GiB of RAM which should
> >> be able to boot a Linux kernel built with CONFIG_HIGHMEM
> >> enabled and passing "-m 2048" to QEMU and appending the
> >> following kernel parameters:
> >> ...
> >> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
> >> ...
> >>
> >> But the following commit in kernel broke above mem=X@Y setting
> >> which added the memory as reserved memory area.
> >> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> >> ...
> >> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> >> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
> >> Date:   Wed Nov 23 14:43:49 2016 +0100
> >>
> >>     MIPS: fix mem=X@Y commandline processing
> >> ...
> >>
> >> So remove `mem=*' to disable user-defined physical RAM map
> >> which let kernel itself caculates memory ranges.
> >>
> >> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
> >> Signed-off-by: He Zhe <zhe.he@windriver.com>
> >> ---
> >>  scripts/runqemu | 5 ++++-
> >>  1 file changed, 4 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/scripts/runqemu b/scripts/runqemu
> >> index de42d0f323..597e7e9a79 100755
> >> --- a/scripts/runqemu
> >> +++ b/scripts/runqemu
> >> @@ -668,7 +668,10 @@ class BaseConfig(object):
> >>              logger.info('QB_MEM is not set, use 512M by default')
> >>              self.set('QB_MEM', '-m 512')
> >>
> >> -        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> >> +        mach = self.get('MACHINE')
> >> +        if mach != 'qemumips':
> >> +            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> >> +
> >>          self.qemu_opt_script += ' %s' % self.get('QB_MEM')
> >>
> >>      def check_tcpserial(self):
> >> --
> >> 2.11.0
> >>
> >> --
> >> _______________________________________________
> >> Openembedded-core mailing list
> >> Openembedded-core@lists.openembedded.org
> >> http://lists.openembedded.org/mailman/listinfo/openembedded-core
>

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Hongxu Jia]

On 2017年12月20日 09:53, Hongxu Jia wrote:
> Ping Ross
>

Ping

//Hongxu

> //Hongxu
>
> On 2017年12月06日 10:46, Hongxu Jia wrote:
>> OE uses qemumips to simulate a Malta board by default.
>>
>> As upstream qemu introduced:
>> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b 
>>
>>
>> The Malta board can support up to 2GiB of RAM which should
>> be able to boot a Linux kernel built with CONFIG_HIGHMEM
>> enabled and passing "-m 2048" to QEMU and appending the
>> following kernel parameters:
>> ...
>> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
>> ...
>>
>> But the following commit in kernel broke above mem=X@Y setting
>> which added the memory as reserved memory area.
>> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411 
>>
>> ...
>> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
>> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
>> Date:   Wed Nov 23 14:43:49 2016 +0100
>>
>>      MIPS: fix mem=X@Y commandline processing
>> ...
>>
>> So remove `mem=*' to disable user-defined physical RAM map
>> which let kernel itself caculates memory ranges.
>>
>> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
>> ---
>>   scripts/runqemu | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/scripts/runqemu b/scripts/runqemu
>> index f2b4b3c..0ae2f85 100755
>> --- a/scripts/runqemu
>> +++ b/scripts/runqemu
>> @@ -658,7 +658,10 @@ class BaseConfig(object):
>>               logger.info('QB_MEM is not set, use 512M by default')
>>               self.set('QB_MEM', '-m 512')
>>   -        self.kernel_cmdline_script += ' mem=%s' % 
>> self.get('QB_MEM').replace('-m','').strip() + 'M'
>> +        mach = self.get('MACHINE')
>> +        if mach != 'qemumips':
>> +            self.kernel_cmdline_script += ' mem=%s' % 
>> self.get('QB_MEM').replace('-m','').strip() + 'M'
>> +
>>           self.qemu_opt_script += ' %s' % self.get('QB_MEM')
>>         def check_tcpserial(self):
>
>

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Hongxu Jia]

Ping Ross

//Hongxu

On 2017年12月06日 10:46, Hongxu Jia wrote:
> OE uses qemumips to simulate a Malta board by default.
>
> As upstream qemu introduced:
> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b
>
> The Malta board can support up to 2GiB of RAM which should
> be able to boot a Linux kernel built with CONFIG_HIGHMEM
> enabled and passing "-m 2048" to QEMU and appending the
> following kernel parameters:
> ...
> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
> ...
>
> But the following commit in kernel broke above mem=X@Y setting
> which added the memory as reserved memory area.
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> ...
> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
> Date:   Wed Nov 23 14:43:49 2016 +0100
>
>      MIPS: fix mem=X@Y commandline processing
> ...
>
> So remove `mem=*' to disable user-defined physical RAM map
> which let kernel itself caculates memory ranges.
>
> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
> ---
>   scripts/runqemu | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/runqemu b/scripts/runqemu
> index f2b4b3c..0ae2f85 100755
> --- a/scripts/runqemu
> +++ b/scripts/runqemu
> @@ -658,7 +658,10 @@ class BaseConfig(object):
>               logger.info('QB_MEM is not set, use 512M by default')
>               self.set('QB_MEM', '-m 512')
>   
> -        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +        mach = self.get('MACHINE')
> +        if mach != 'qemumips':
> +            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +
>           self.qemu_opt_script += ' %s' % self.get('QB_MEM')
>   
>       def check_tcpserial(self):

Re: [PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Hongxu Jia]

[-- Attachment #1: Type: text/plain, Size: 2074 bytes --]

On 2017年12月06日 10:46, Hongxu Jia wrote:
> OE uses qemumips to simulate a Malta board by default.
>
> As upstream qemu introduced:
> https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b
>
> The Malta board can support up to 2GiB of RAM which should
> be able to boot a Linux kernel built with CONFIG_HIGHMEM
> enabled and passing "-m 2048" to QEMU and appending the
> following kernel parameters:

Hi Ross,

I have sent another fix to linux-yocto to enable CONFIG_HIGHMEM
for qemumips (mti-malta32).

//Hongxu

> ...
> mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
> ...
>
> But the following commit in kernel broke above mem=X@Y setting
> which added the memory as reserved memory area.
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> ...
> commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
> Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
> Date:   Wed Nov 23 14:43:49 2016 +0100
>
>      MIPS: fix mem=X@Y commandline processing
> ...
>
> So remove `mem=*' to disable user-defined physical RAM map
> which let kernel itself caculates memory ranges.
>
> Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
> ---
>   scripts/runqemu | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/runqemu b/scripts/runqemu
> index f2b4b3c..0ae2f85 100755
> --- a/scripts/runqemu
> +++ b/scripts/runqemu
> @@ -658,7 +658,10 @@ class BaseConfig(object):
>               logger.info('QB_MEM is not set, use 512M by default')
>               self.set('QB_MEM', '-m 512')
>   
> -        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +        mach = self.get('MACHINE')
> +        if mach != 'qemumips':
> +            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
> +
>           self.qemu_opt_script += ' %s' % self.get('QB_MEM')
>   
>       def check_tcpserial(self):



[-- Attachment #2: Type: text/html, Size: 3610 bytes --]

[PATCH 1/2] scripts/runqemu: qemumips support up to 2GiB RAM

[Hongxu Jia]

OE uses qemumips to simulate a Malta board by default.

As upstream qemu introduced:
https://git.qemu.org/?p=qemu.git;a=commit;h=94c2b6aff43cdfcfdfb552773a6b6b973a72ef0b

The Malta board can support up to 2GiB of RAM which should
be able to boot a Linux kernel built with CONFIG_HIGHMEM
enabled and passing "-m 2048" to QEMU and appending the
following kernel parameters:
...
mem=256M@0x0 mem=256M@0x90000000 mem=1536M@0x20000000
...

But the following commit in kernel broke above mem=X@Y setting
which added the memory as reserved memory area.
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
...
commit 73fbc1eba7ffa3bf0ad12486232a8a1edb4e4411
Author: Marcin Nowakowski <marcin.nowakowski@imgtec.com>
Date:   Wed Nov 23 14:43:49 2016 +0100

    MIPS: fix mem=X@Y commandline processing
...

So remove `mem=*' to disable user-defined physical RAM map
which let kernel itself caculates memory ranges.

Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
---
 scripts/runqemu | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/scripts/runqemu b/scripts/runqemu
index f2b4b3c..0ae2f85 100755
--- a/scripts/runqemu
+++ b/scripts/runqemu
@@ -658,7 +658,10 @@ class BaseConfig(object):
             logger.info('QB_MEM is not set, use 512M by default')
             self.set('QB_MEM', '-m 512')
 
-        self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
+        mach = self.get('MACHINE')
+        if mach != 'qemumips':
+            self.kernel_cmdline_script += ' mem=%s' % self.get('QB_MEM').replace('-m','').strip() + 'M'
+
         self.qemu_opt_script += ' %s' % self.get('QB_MEM')
 
     def check_tcpserial(self):
-- 
2.8.1