[lkml]   [2022]   [Mar]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v4 1/4] MIPS: Refactor early_parse_mem() to fix mem= parameter
On Fri, Mar 04, 2022 at 05:11:44PM +0000, Maciej W. Rozycki wrote:
> On Fri, 4 Mar 2022, Thomas Bogendoerfer wrote:
> > > > With this patch, when add "mem=3G" to the command-line, the
> > > > kernel boots successfully, we can see the following messages:
> > >
> > > unfortunately this patch would break platforms without memory detection,
> > > which simply use mem=32M for memory configuration. Not sure how many
> > > rely on this mechanism. If we can make sure nobody uses it, I'm fine
> > > with your patch.
> >
> > maybe we could add a CONFIG option, which will be selected by
> > platforms, which don't need/want this usermem thing.
> FWIW I don't understand what the issue is here beyond that we have a bug
> that causes a system to hang when "mem=3G" is passed on the kernel command
> line. That is assuming that system does have contiguous RAM available for
> the kernel to use from address 0 up to 3GiB; otherwise it's a user error
> to tell the kernel it has that memory available (I did get bitten by that
> myself too): garbage in, garbage out.

I did a quick test with an IP30:

>> bootp(): ip=dhcp root=/dev/nfs console=ttyS0 mem=384M
Setting $netaddr to (from server )
Obtaining from server
9012640+181664 entry: 0xa800000020664a60
Linux version 5.17.0-rc3+ (tbogendoerfer@adalid) (mips64-linux-gnu-gcc (GCC) 6.1.1 20160621 (Red Hat Cross 6.1.1-2), GNU ld version 2.27-3.fc24) #155 SMP Mon Mar 7 13:12:01 CET 2022
PROMLIB: ARC firmware Version 64 Revision 0
printk: bootconsole [early0] enabled
CPU0 revision is: 00000934 (R10000)
FPU revision is: 00000900
Detected 512MB of physical memory.
User-defined physical RAM map overwrite
Kernel sections are not in the memory maps
IP30: Slot: 0, PrID: 00000934, PhyID: 0, VirtID: 0
IP30: Slot: 1, PrID: 00000934, PhyID: 1, VirtID: 1
IP30: Detected 2 CPU(s) present.
Primary instruction cache 32kB, VIPT, 2-way, linesize 64 bytes.
Primary data cache 32kB, 2-way, VIPT, no aliases, linesize 32 bytes
Unified secondary cache 1024kB 2-way, linesize 128 bytes.
Zone ranges:
DMA32 [mem 0x0000000000000000-0x00000000ffffffff]
Normal empty
Movable zone start for each node
Early memory node ranges
node 0: [mem 0x0000000000000000-0x0000000017ffffff]
node 0: [mem 0x0000000020004000-0x00000000208c7fff]
Initmem setup node 0 [mem 0x0000000000000000-0x00000000208c7fff]

after that it's dead (it doesn't have memory starting at 0x0).
Most SGI systems will act broken with mem= in one way or another.
And I already had the need to limit the amount of memory.

> I think having a CONFIG option automatically selected to disable the
> ability to give a memory map override would handicap people in debugging
> their systems or working around firmware bugs, so I would rather be
> against it.

I'm thinking about a CONFIG option, which isn't user selectable, but
selected via Kconfig only. But that would give to differents semantics
for mem=

So can I just limit amount of memory without interfering with normal
memory detection ?


Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea. [ RFC1925, 2.3 ]

 \ /
  Last update: 2022-03-07 17:29    [W:1.874 / U:0.032 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site