Loading kernel from nand boot time optimization

RAJ.M

Part Number: DM385

Tool/software: Linux

Hello All,

We are using custom DM385 with ipnc 3.8 rdk.

We are doing boot time optimization below log shows the loading kernel and uncompressing(From nand boot) takes mush time.

Anybody have idea to reduce time ?

[0.000001 0.000001]
[0.000093 0.000092] U-Boot 2010.06 (Dec 12 2016 - 11:13:13) DM385_IPNC_3.50.00
[0.003725 0.003632]
[0.004920 0.001195] NAND device: Manufacturer ID: 0x01, Chip ID: 0xf1 (Spansion/AMD NAND 128MiB 3,3V 8-bit)
[0.012803 0.007883]
[0.014935 0.002132]
[0.015004 0.000069]
[0.015055 0.000051] NAND read: device 0 offset 0x20000, size 0x40000
[0.436970 0.421915] 262144 bytes read: OK
[0.453942 0.016972]
[0.453984 0.000042] U-Boot 2010.06 (Dec 12 2016 - 11:45:55) DM385_IPNC_3.50.00
[0.465980 0.011996]
[0.466035 0.000055] NAND device: Manufacturer ID: 0x01, Chip ID: 0xf1 (Spansion/AMD NAND 128MiB 3,3V 8-bit)
[0.473533 0.007498]
[0.947992 0.474459]
[0.948064 0.000072]
[0.948100 0.000036] Loading from nand0, offset 0x280000
[0.951921 0.003821]    Image Name:   Linux-2.6.37_DM385_CARDVR_3.50.0
[0.955458 0.003537]    Image Type:   ARM Linux Kernel Image (uncompressed)
[0.960461 0.005003]    Data Size:    1852212 Bytes = 1.8 MiB
[0.963663 0.003202]    Load Address: 80008000
[0.966474 0.002811]    Entry Point: 80008000
[2.634972 1.668498] ## Booting kernel from Legacy Image at 81000000 ...
[2.637300 0.002328]    Image Name:   Linux-2.6.37_DM385_CARDVR_3.50.0
[2.642113 0.004813]    Image Type:   ARM Linux Kernel Image (uncompressed)
[2.647925 0.005812]    Data Size:    1852212 Bytes = 1.8 MiB
[2.651202 0.003277]    Load Address: 80008000
[2.654003 0.002801]    Entry Point: 80008000
[2.657921 0.003918]    Loading Kernel Image ... OK
[3.325104 0.667183] OK
[3.325172 0.000068]
[3.325191 0.000019] Starting kernel ...
[3.326921 0.001730]
[3.329926 0.003005] Uncompressing Linux... done, booting the kernel.
[3.997961 0.668035] Trying to install type control for IRQ375

Regards,

RAJ M

over 7 years ago

0 Margarita Gashova over 7 years ago

TI__Guru* 80190 points

Hello,

I will notify the IPNC team for help.

BR
Margarita

0 Pavel Botev over 7 years ago

TI__Guru**** 170625 points

RAJ,

Meanwhile you might look in the below wiki pages:

processors.wiki.ti.com/.../Sitara_Linux_Training:_Boot_Time_Reduction_Update
processors.wiki.ti.com/.../Sitara_Linux_Training:_Boot_Time_Reduction
processors.wiki.ti.com/.../Optimize_Linux_Boot_Time
processors.wiki.ti.com/.../All_This_For_1_Second_Boot

onedrive.live.com/view.aspx

Regards,
Pavel

0 RAJ.M over 7 years ago in reply to Pavel Botev

Expert 2440 points

Hi Pavel,

how to avoid double copy of the kernel image

[2.634972 1.668498] ## Booting kernel from Legacy Image at 81000000 ...
[2.637300 0.002328] Image Name: Linux-2.6.37_DM385_CARDVR_3.50.0
[2.642113 0.004813] Image Type: ARM Linux Kernel Image (uncompressed)
[2.647925 0.005812] Data Size: 1852212 Bytes = 1.8 MiB
[2.651202 0.003277] Load Address: 80008000
[2.654003 0.002801] Entry Point: 80008000
[2.657921 0.003918] Loading Kernel Image ... OK
[3.325104 0.667183] OK

In above one of the link
changing the environment variables to use kloadaddr for the load address and bootm addresses rather than loadaddr

How to do that in DM385?

Regards
RAJ M

0 Pavel Botev over 7 years ago in reply to RAJ.M

TI__Guru**** 170625 points

RAJ,

kloadaddr is used to that the kernel will not have to be relocated to the boot address by u-boot during the boot process. If you use loadaddr you will spend time copying the kernel from the loadaddr location to the kloadaddr location during boot.

You can try the below sequence:

setenv kloadaddr <addr>
nand read ${kloadaddr} <offset> <len>
bootm ${kloadaddr}

See the below e2e thread for more info:
e2e.ti.com/.../209047

Regards,
Pavel

0 RAJ.M over 7 years ago in reply to Pavel Botev

Expert 2440 points

Hi Pavel,

Thank you for the reply. its helpful.

How to optimize the NAND driver to increase the throughput for Copying the kernel,uboot to the RAM location?

NAND read: device 0 offset 0x20000, size 0x40000
[0.436970 0.421915] 262144 bytes read: OK

[2.634972 1.668498] ## Booting kernel from Legacy Image at 81000000 ...

Regards,
RAJ M

0 Pavel Botev over 7 years ago in reply to RAJ.M

TI__Guru**** 170625 points

RAJ,

Refer to the below e2e thread (and the links to other e2e threads from within):

e2e.ti.com/.../549002

Regards,
Pavel

0 RAJ.M over 7 years ago in reply to Pavel Botev

Expert 2440 points

Hi Pavel,

For nand driver optimization (prefetch read), I applied below patch (attached).

In omap_nand_read_buf function()

Its hang when *buf++ = *(volatile uint8_t *)(nand_fifo_add); instruction execting.

Is there any memory remap issue because when i try to access that memory only hangs?

Regards,

RAJ M

6237.prefetch_nand_read.patch.tar.gz

0 Pavel Botev over 7 years ago in reply to RAJ.M

TI__Guru**** 170625 points

Raj,

Can you print at which address exactly your code hang?

Note that NAND prefetch is enabled by default in linux kernel, so you might reuse the approach there:

linux-kernel/drivers/mtd/nand/omap2.c, see function omap_read_buf_pref()
processors.wiki.ti.com/.../TI81XX_PSP_NAND_Driver_User_Guide
processors.wiki.ti.com/.../TI81XX_PSP_04.04.00.02_Feature_Performance_Guide

Note also that NAND prefetch is enabled in the latest u-boot, which comes with AM335x, DRA7x, you can also see how it is done there:

u-boot/drivers/mtd/nand/omap_gpmc.c
processors.wiki.ti.com/.../Linux_Core_NAND_User's_Guide

Regards,
Pavel

0 RAJ.M over 7 years ago in reply to Pavel Botev

Expert 2440 points

Hi Pavel,

Note that NAND prefetch is enabled by default in linux kernel, so you might reuse the approach there:

linux-kernel/drivers/mtd/nand/omap2.c, see function omap_read_buf_pref()

I have done same thing as u suggest check my patch file as attached above reply.

changing the macro "PISMO1_NAND_SIZE GPMC_SIZE_256M" to "PISMO1_NAND_SIZE GPMC_SIZE_16M"

Now I can read but still the time is not improved both prefetch and default read took same time for reading image from nand.

Below is my log for reference

[0.000001 0.000001]
[0.000111 0.000110] U-Boot 2010.06 (Jan 18 2017 - 10:23:06) DM385_IPNC_3.50.00
[0.003783 0.003672]
[0.005940 0.002157] NAND device: Manufacturer ID: 0x01, Chip ID: 0xf1 (Spansion/AMD NAND 128MiB 3,3V 8-bit)
[0.013199 0.007259]
[0.014948 0.001749]
[0.015019 0.000071]
[0.015069 0.000050] NAND read: device 0 offset 0x20000, size 0x1e000
[0.307967 0.292898] 122880 bytes read: OK
[0.313943 0.005976]
[0.314000 0.000057] U-Boot 2010.06 (Jan 09 2017 - 20:04:24) DM385_IPNC_3.50.00
[0.325943 0.011943]
[0.325998 0.000055] NAND device: Manufacturer ID: 0x01, Chip ID: 0xf1 (Spansion/AMD NAND 128MiB 3,3V 8-bit)
[0.333900 0.007902]
[0.838992 0.505092]
[0.840940 0.001948]
[0.840983 0.000043] Loading from NAND 128MiB 3,3V 8-bit, offset 0x280000
[0.846919 0.005936]    Image Name:   Linux-2.6.37_DM385_CARDVR_3.50.0
[0.850596 0.003677]    Image Type:   ARM Linux Kernel Image (uncompressed)
[0.856333 0.005737]    Data Size:    1763436 Bytes = 1.7 MiB
[0.859676 0.003343]    Load Address: 80008000
[0.864342 0.004666]    Entry Point: 80008000
[2.459966 1.595624] ## Booting kernel from Legacy Image at 80007fc0 ...

Also this is gpmc config is it ok?

#define M_NAND_GPMC_CONFIG1     0x00001800
#define M_NAND_GPMC_CONFIG2     0x00060600
#define M_NAND_GPMC_CONFIG3     0x00060601
#define M_NAND_GPMC_CONFIG4     0x04010401
#define M_NAND_GPMC_CONFIG5     0x00040506
#define M_NAND_GPMC_CONFIG6     0x04000580
#define M_NAND_GPMC_CONFIG7     0x00000008

GPMC clock frequency

#define L3_M (opp_val_dm385(800, 800))

Regards,

RAJ M

0 Pavel Botev over 7 years ago in reply to RAJ.M

TI__Guru**** 170625 points

RAJ,

See the below e2e thread:

e2e.ti.com/.../2028337

Regards,
Pavel

Processors

Processors forum

Loading kernel from nand boot time optimization