8148 2GB DDR3

mike A

Genius 3490 points

Hello,

We have a custom6148 board with 2GB DDR3, What are the LISA_MAP_X register values to access this complete 2gb ?

Thanks,

Mike

over 9 years ago

0 Pavel Botev over 9 years ago

TI__Guru**** 170625 points

Hi Mike,

Do you mean custom DM8148 board (not 6148) ?

BR
Pavel

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

Hi Pavel,

My bad sorry, I meant custom DM8148.

Thanks,

Mike

0 Pavel Botev over 9 years ago in reply to mike A

TI__Guru**** 170625 points

Mike,

One possible configurations is:

DMM_LISA_MAP__0 = 0
DMM_LISA_MAP__1 = 0
DMM_LISA_MAP__2 = 0x80640300
DMM_LISA_MAP__3 = 0xC0640320

See the below wiki for more info regarding this configuration:
processors.wiki.ti.com/.../EZSDK_Memory_Map

Other possible configuration is:

DMM_LISA_MAP__0 = 0
DMM_LISA_MAP__1 = 0
DMM_LISA_MAP__2 = 0
DMM_LISA_MAP__3 = 0x80740300

See DM814x TRM, chapter 6 DMM/TILER
6.2.1.6 Section Mapping
6.3.1 DMM Basic Register Setup
6.3.4 Address Management Using LISA Sections

See also DM814x silicon errata, advisory 3.0.31

BR
Pavel

0 Marat Shchuchinsky over 9 years ago in reply to Pavel Botev

Expert 2595 points

Hi Pavel,

Please let me know, is enlisted above LISA setting (define 1 or 2 mapped regions) may influent on DDR read/write performance, specially on EDMA transfers?

0 Pavel Botev over 9 years ago in reply to Marat Shchuchinsky

TI__Guru**** 170625 points

For optimal system performance, it is recommended to enable interleaving between the 2 EMIF banks and thus have same sized memory on both the EMIF banks. Example : 512MB of DDR3 on EMIF bank 0 and 512MB of DDR3 on EMIF bank 1, for a total system DDR3 memory of 1GB.

For optimal system performance, symmetric configuration is HIGHLY recommended. Unless dictated by system cost and other constraints, Asymmetrical distribution of memory should not be considered.

BR
Pavel

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

Hi Pavel,

Does that mean:
DMM_LISA_MAP__0 = 0
DMM_LISA_MAP__1 = 0
DMM_LISA_MAP__2 = 0x80640300
DMM_LISA_MAP__3 = 0xC0640320

This option is optimal.

Thanks,
Mike

0 Pavel Botev over 9 years ago in reply to mike A

TI__Guru**** 170625 points

Yes, I think so.

BR
Pavel

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

Hi Pavel,

Thanks for your response.

I had one more question, the kernel runs out of memory, how do I configure low memory

cat /proc/meminfo
MemTotal: 1561076 kB
MemFree: 1483556 kB
Buffers: 0 kB
Cached: 41464 kB
SwapCached: 0 kB
Active: 18396 kB
Inactive: 36520 kB
Active(anon): 13456 kB
Inactive(anon): 564 kB
Active(file): 4940 kB
Inactive(file): 35956 kB
Unevictable: 0 kB
Mlocked: 0 kB
HighTotal: 1207296 kB
HighFree: 1144800 kB
LowTotal: 353780 kB
LowFree: 338756 kB
SwapTotal: 0 kB
SwapFree: 0 kB
Dirty: 0 kB
Writeback: 0 kB
AnonPages: 13476 kB
Mapped: 17120 kB
Shmem: 568 kB
Slab: 7644 kB
SReclaimable: 2884 kB
SUnreclaim: 4760 kB
KernelStack: 656 kB
PageTables: 588 kB
NFS_Unstable: 0 kB
Bounce: 0 kB
WritebackTmp: 0 kB
CommitLimit: 780536 kB
Committed_AS: 90180 kB
VmallocTotal: 540672 kB
VmallocUsed: 375696 kB
VmallocChunk: 133052 kB

Thanks,
Mike

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

# dmesg |grep Memory:
[ 0.000000] Memory: 364MB 270MB 908MB 1MB = 1543MB total
[ 0.000000] Memory: 1560900k/1560900k available, 71356k reserved, 1207296K highmem

I want to Increase the low mem as the kernel is failing to allocate memory.

My bootargs are:
console=ttyO0,115200n8 rootwait ubi.mtd=14,2048 rootfstype=ubifs root=ubi0:rootfs rw mem=364M@0x80000000 mem=270M@0x9FC00000 mem=960M@0xC0000000 vmalloc=500M notifyk.vpssm3_sva=0xBF900000

Is there a config to do it ?

Thanks,
Mike

0 Pavel Botev over 9 years ago in reply to mike A

TI__Guru**** 170625 points

Already discussed in the below e2e thread:

e2e.ti.com/.../459590

BR
Pavel

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

Hi Pavel,

I am still unclear, as why there is so much high mem as despite telling in in bootargs with mem.

Kernel command line: console=ttyO0,115200n8 rootwait ubi.mtd=14,2048 rootfstype=ubifs root=ubi0:rootfs rw mem=364M@0x80000000 mem=320M@0x9FC00000 mem=960M@0xC0000000 vmalloc=500M notifyk.vpssm3_sva=0xBF900000

[ 0.000000] PID hash table entries: 2048 (order: 1, 8192 bytes)
[ 0.000000] Dentry cache hash table entries: 65536 (order: 6, 262144 bytes)
[ 0.000000] Inode-cache hash table entries: 32768 (order: 5, 131072 bytes)
[ 0.000000] Memory: 364MB 320MB 908MB 1MB = 1593MB total
[ 0.000000] Memory: 1560900k/1560900k available, 71356k reserved, 1207296K highmem
[ 0.000000] Virtual kernel memory layout:
[ 0.000000] vector : 0xffff0000 - 0xffff1000 ( 4 kB)
[ 0.000000] fixmap : 0xfff00000 - 0xfffe0000 ( 896 kB)
[ 0.000000] DMA : 0xffc00000 - 0xffe00000 ( 2 MB)
[ 0.000000] vmalloc : 0xd7000000 - 0xf8000000 ( 528 MB)
[ 0.000000] lowmem : 0xc0000000 - 0xd6c00000 ( 364 MB)
[ 0.000000] pkmap : 0xbfe00000 - 0xc0000000 ( 2 MB)
[ 0.000000] modules : 0xbf000000 - 0xbfe00000 ( 14 MB)
[ 0.000000] .init : 0xc0008000 - 0xc0034000 ( 176 kB)
[ 0.000000] .text : 0xc0034000 - 0xc0532000 (5112 kB)
[ 0.000000] .data : 0xc0532000 - 0xc0589f80 ( 352 kB)

Thanks,
Mike

0 Pavel Botev over 9 years ago in reply to mike A

TI__Guru**** 170625 points

Mike,

Around 50MB of RAM is reserved for FB driver by default, thus when in example you set "mem=200M" you will get "Memory: 150MB = 150MB total"

You can use HIGHMEM support to allocate more memory to Linux. This can be enabled/disable by menuconfig->Kernel Features->HIGH Memory Support.

Refer to the below wiki for more info regarding HIGHMEM:

processors.wiki.ti.com/.../TI81XX_PSP_04.04.00.02_Feature_Performance_Guide

BR
Pavel

0 mike A over 9 years ago in reply to Pavel Botev

Genius 3490 points

Hi Pavel,

If I am not wrong, highmem is user space and lowmem is kernel space ? So I need to increase kernel space ? Is there a option to do it ?

Thanks,
Mike

0 Pavel Botev over 9 years ago in reply to mike A

TI__Guru**** 170625 points

Default kernel build is set up with 3G/1G split for User/Kernel space. In addition, "High Memory" support in kernel (CONFIG_HIGHMEM) is enabled by default to accommodate larger physical memory/address space.

It should be possible to allow larger direct mapped memory into kernel space by changing User/Kernel split to 2/2 or 1/3. Please note that these are NOT TESTED and may lead to unpredictable behavior - particularly some applications may fail.

It is possible to indicate the kernel that the usable RAM is spanned across holes in between. This is achieved from passing multiple "mem=<size>@<start-address>" arguments to kernel.

Even when passing memory with holes, the kernel reserves contiguous space incorporating the whole memory passed through all 'mem' arguments. This means, the actual lowmem mapped will be more than the total size of all memory arguments combined together.

Without HIGHMEM support, the vmalloc and lowmem sizes are dependent - on RAM available to kernel (as specified by 'mem=<size-in-MB>M' boot argument) and vmalloc size required - restricted by the amound of space that can be directly mapped into kernel. As you provide more memory for kernel to map, the vmalloc space will be lowered. Vice versa is true when vmalloc region is changed by passing 'vmalloc=<size-in-MB>M' argument.

Though HIGHMEM configuration is enabled by default, the mapping for highmem will only be created on need basis and in case where total space comprising of specified RAM, vmalloc and memory holes (see the section on Memory Holes Configuration) exceeds directly mappable space (that is, 888MB).

BR
Pavel

Processors

Processors forum

8148 2GB DDR3