Skip to contentSkip to conversation

fsl_elbc_nand bug

U-Boot Andrei Yakimov 2 participants 4 messages May 15, 2015 – May 18, 2015

lore
Share & tools
lore
#0 lore raw

Hi,
I just found, if NAND ONFI parameter page first copy is
bad, u-boot will never read extra copes:

fsl_elbc_nand.c:
342 out_be32(&lbc->fbcr, 256);
343 ctrl->read_bytes = 256;

this code cause "read beyond buffer" error message:

nand_base.c:
chip->cmdfunc(mtd, NAND_CMD_PARAM, 0, -1);
3243 for (i = 0; i < 3; i++) {
3244 for (j = 0; j < sizeof(*p); j++)
3245 ((uint8_t *)p)[j] = chip->read_byte(mtd);
3246 if (onfi_crc16(ONFI_CRC_BASE, (uint8_t *)p, 254) ==
3247 le16_to_cpu(p->crc)) {
3248 break;
3249 }
3250 }

Unfortunately I do not working with latest u-boot now,
so can not provide a patch, read size should be increased to at least 768 ( 3 copy mandatory for ONFI spec). Linux kernel should be updated too.

I am also not expert for other NAND controllers drivers.

But first copy of ONFI parameter page CAN GO BAD. I seen it my self.

Andrei

#1 lore raw
Quoted 21 lines · Andrei Yakimov

Hi,
I just found, if NAND ONFI parameter page first copy is
bad, u-boot will never read extra copes:

fsl_elbc_nand.c:
342 out_be32(&lbc->fbcr, 256);
343 ctrl->read_bytes = 256;

this code cause "read beyond buffer" error message:

nand_base.c:

   chip->cmdfunc(mtd, NAND_CMD_PARAM, 0, -1);

3243 for (i = 0; i < 3; i++) {
3244 for (j = 0; j < sizeof(*p); j++)
3245 ((uint8_t *)p)[j] = chip->read_byte(mtd);
3246 if (onfi_crc16(ONFI_CRC_BASE, (uint8_t *)p, 254) ==
3247 le16_to_cpu(p->crc)) {
3248 break;
3249 }
3250 }

Quoted 4 lines

Unfortunately I do not working with latest u-boot now,
so can not provide a patch, read size should be increased to at least 768 ( 3 copy mandatory for ONFI spec). Linux kernel should be updated too.

Could you send a patch (both here and Linux)? Especially since you have a test case. :-)

Though a better fix than just increasing the byte count would be to add a read_id() callback so that the NAND subsystem can actually tell the driver how much data it's expecting, rather than forcing it to go through cmdfunc that makes high-level drivers (poorly) emulate low-level hardware.

Quoted 1 line

I am also not expert for other NAND controllers drivers.

At least IFC will have the same issue.

-Scott

#2 lore raw

I will try, It will take a while due to I am not working with latest code. I do have a chip with first 2 parameter blocks corrupted. So I can test it. I can not send it to Linux - I am not subscribed to any Linux mailing lists.
As of patch structure:

1) separate READID and PARAM command.
2) READID will read 5 bytes as per ONFI spec.
3) PARAM command will read 768 bytes as per ONFI spec.

I will check is path to send read size are valid for this command - but it will affect every NAND driver, It is dangers changing. First approach is save, will not eliminate other divers problems if they exist.

I will update only fsl_elbc_nand driver.

This is least invasive patch.

Should I get just master for generation of this patch?

Andrei.

Quoted 43 lines · Scott Wood
Quoted 21 lines · Andrei Yakimov

Hi,
I just found, if NAND ONFI parameter page first copy is
bad, u-boot will never read extra copes:

fsl_elbc_nand.c:
342 out_be32(&lbc->fbcr, 256);
343 ctrl->read_bytes = 256;

this code cause "read beyond buffer" error message:

nand_base.c:

   chip->cmdfunc(mtd, NAND_CMD_PARAM, 0, -1);

3243 for (i = 0; i < 3; i++) {
3244 for (j = 0; j < sizeof(*p); j++)
3245 ((uint8_t *)p)[j] = chip->read_byte(mtd);
3246 if (onfi_crc16(ONFI_CRC_BASE, (uint8_t *)p, 254) ==
3247 le16_to_cpu(p->crc)) {
3248 break;
3249 }
3250 }

Quoted 4 lines

Unfortunately I do not working with latest u-boot now,
so can not provide a patch, read size should be increased to at least 768 ( 3 copy mandatory for ONFI spec). Linux kernel should be updated too.

Could you send a patch (both here and Linux)? Especially since you have a test case. :-)

Though a better fix than just increasing the byte count would be to add a read_id() callback so that the NAND subsystem can actually tell the driver how much data it's expecting, rather than forcing it to go through cmdfunc that makes high-level drivers (poorly) emulate low-level hardware.

Quoted 1 line

I am also not expert for other NAND controllers drivers.

At least IFC will have the same issue.

-Scott

#3 lore raw
Quoted 4 lines · Andrei Yakimov

I will try, It will take a while due to I am not working with latest code. I do have a chip with first 2 parameter blocks corrupted. So I can test it. I can not send it to Linux - I am not subscribed to any Linux mailing lists.

You don't need to be subscribed to a list to send a patch (but it does need to be based on the latest upstream tree).

Quoted 8 lines

As of patch structure:

1) separate READID and PARAM command.
2) READID will read 5 bytes as per ONFI spec.
3) PARAM command will read 768 bytes as per ONFI spec.

I will check is path to send read size are valid for this command - but it will affect every NAND driver,

It wouldn't affect every NAND driver. There should be a default implementation that calls cmdfunc, so that things work the same way as now unless a driver overrides it.

-Scott