Re: Why we echo a invalid start_address_of_new_memory succeeded ?

[David Rientjes <rientjes@google.com>]

On Fri, 20 Jun 2014, Zhang Zhen wrote:

> Hi,
> 
> I am testing mem-hotplug on a qemu virtual machine. I executed the following command
> to notify memory hot-add event by hand.
> 
> % echo start_address_of_new_memory > /sys/devices/system/memory/probe
> 
> To a different start_address_of_new_memory I got different results.
> The results are as follows:
> 
> MBSC-x86_64 /sys/devices/system/memory # ls
> block_size_bytes  memory2           memory5           power
> memory0           memory3           memory6           probe
> memory1           memory4           memory7           uevent
> MBSC-x86_64 /sys/devices/system/memory # echo 0x70000000 > probe

Since block_size_bytes is 0x8000000 == 128MB, this is 0x70000000 / 
0x8000000 = section number 14.  Successfully hot added.  Presumably you're 
reporting that there is no physical memory there, so this would default to 
the online node of the first memory block, probably node 0.

> MBSC-x86_64 /sys/devices/system/memory # echo 0x78000000 > probe
> -sh: echo: write error: File exists

EEXIST gets returned when the resource already exists, mostly likely 
system RAM or reserved memory as reported by your BIOS.  You report this 
is a 2GB machine, no reason to believe memory at 1920MB isn't already 
online (including reserved).

> MBSC-x86_64 /sys/devices/system/memory # echo 0x80000000 > probe
> -sh: echo: write error: File exists
> MBSC-x86_64 /sys/devices/system/memory # echo 0x88000000 > probe
> -sh: echo: write error: File exists

Same.

> MBSC-x86_64 /sys/devices/system/memory # echo 0x8f000000 > probe
> -sh: echo: write error: Invalid argument

Returns EINVAL because it's not a multiple of block_size_bytes, it's not 
aligned properly.

> MBSC-x86_64 /sys/devices/system/memory # echo 0x90000000 > probe
> -sh: echo: write error: File exists

See above, the resoure already exists.  Check your e820 your dmesg, which 
is missing from this report, to determine what already exists and may be 
already online or reserved.

> MBSC-x86_64 /sys/devices/system/memory # echo 0xff0000000 > probe

0xff0000000 / 0x8000000 is section 510, successfully onlined.

> MBSC-x86_64 /sys/devices/system/memory # ls
> block_size_bytes  memory2           memory510         probe
> memory0           memory3           memory6           uevent
> memory1           memory4           memory7
> memory14          memory5           power

Looks good, you onlined sections 14 and 510 above.

> MBSC-x86_64 /sys/devices/system/memory # echo 0xfff0000000 > probe

Same for section 8190.

> MBSC-x86_64 /sys/devices/system/memory # ls
> block_size_bytes  memory2           memory510         power
> memory0           memory3           memory6           probe
> memory1           memory4           memory7           uevent
> memory14          memory5           memory8190
> 

Confirmed it's onlined.

> The qemu virtual machine's physical memory size is 2048M, and the boot memory is 1024M.
> 
> MBSC-x86_64 / # cat /proc/meminfo
> MemTotal:        1018356 kB
> MBSC-x86_64 / # cat /sys/devices/system/memory/block_size_bytes
> 8000000
> 

That's irrelevant, you've explicitly onlined memory that doesn't exist.  
Not sure why you're using the probe interface unless you need it for x86, 
is ACPI not registering it correctly?

> Three questions:
> 1. The machine's physical memory size is 2048M, why echo 0x78000000 as the start_address_of_new_memory failed ?
> 

Copy your e820 map from your dmesg, it's probably reserved or already 
online, this is lower than 2048M.

> 2. Why echo 0x8f000000 as the start_address_of_new_memory, the error message is different ?
> 

Not properly aligned to block_size_bytes.  It's a nuance, but 
block_size_bytes is exported in hex, not decimal.

> 3. Why echo 0xfff0000000 as the start_address_of_new_memory succeeded ? 0xfff0000000 has exceeded the machine's physical memory size.
> 

You're telling the kernel differently.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>