Hi Miquel
On Tue, Nov 28, 2023 at 11:56 AM Miquel Raynal miquel.raynal@bootlin.com wrote:
This is a debug command to monitor the retention state of the data on the array. The command needs a duplication of the mtd_read_oob() function to actually return the maximum number of bitflips encountered while reading the page. We could write a specific implementation for the Sunxi driver but this is probably enough.
nand watch <off> <size> - check an area for bitflips nand watch.part <part> - check a partition for bitflips nand watch.chip - check the whole device for bitflipsThe output may be a bit verbose and could look like:
I have rebased and I will take a look on this patch tomorrow. I think having it is ok.
Michael
=> nand watch.chip device 0 whole chip size adjusted to 0xff60000 (5 bad blocks) NAND watch for bitflips in area 0x0-0xff60000: Page 0 (0x00000000) -> error -74 Page 1 (0x00000800) -> error -74 Page 2 (0x00001000) -> error -74 Page 3 (0x00001800) -> error -74 Page 4 (0x00002000) -> error -74 Page 5 (0x00002800) -> error -74 Page 6 (0x00003000) -> error -74 Page 7 (0x00003800) -> error -74 Page 8 (0x00004000) -> error -74 Page 9 (0x00004800) -> error -74 Page 10 (0x00005000) -> error -74 Page 11 (0x00005800) -> error -74 Page 12 (0x00006000) -> error -74 Page 13 (0x00006800) -> error -74 Page 14 (0x00007000) -> error -74 Page 15 (0x00007800) -> error -74 Page 16 (0x00008000) -> error -74 Page 17 (0x00008800) -> error -74 Page 18 (0x00009000) -> error -74 Page 19 (0x00009800) -> error -74 Page 20 (0x0000a000) -> error -74 Page 21 (0x0000a800) -> error -74 Page 22 (0x0000b000) -> error -74 Page 23 (0x0000b800) -> error -74 Page 1110 (0x0022b000) -> up to 1 bf/chunk Page 1122 (0x00231000) -> up to 1 bf/chunk Page 1132 (0x00236000) -> up to 1 bf/chunk Page 1362 (0x002a9000) -> up to 1 bf/chunk Page 4990 (0x009bf000) -> up to 1 bf/chunk Page 5728 (0x00b30000) -> up to 1 bf/chunk Page 7116 (0x00de6000) -> up to 1 bf/chunk Page 7160 (0x00dfc000) -> up to 1 bf/chunk Page 7494 (0x00ea3000) -> up to 1 bf/chunk Page 10842 (0x0152d000) -> up to 1 bf/chunk Page 11614 (0x016af000) -> up to 1 bf/chunk Page 11970 (0x01761000) -> up to 1 bf/chunk Page 12536 (0x0187c000) -> up to 1 bf/chunk Page 12687 (0x018c7800) -> up to 1 bf/chunk Page 14298 (0x01bed000) -> up to 1 bf/chunk Page 18268 (0x023ae000) -> up to 1 bf/chunk Page 18760 (0x024a4000) -> up to 1 bf/chunk Page 21440 (0x029e0000) -> up to 1 bf/chunk Page 22336 (0x02ba0000) -> up to 1 bf/chunk Page 22592 (0x02c20000) -> up to 1 bf/chunk Page 23872 (0x02ea0000) -> up to 1 bf/chunk Page 27584 (0x035e0000) -> up to 1 bf/chunk Page 35008 (0x04460000) -> up to 1 bf/chunk Page 37184 (0x048a0000) -> up to 1 bf/chunk Page 41728 (0x05180000) -> up to 1 bf/chunk Page 42176 (0x05260000) -> up to 1 bf/chunk Page 43200 (0x05460000) -> up to 1 bf/chunk Page 43328 (0x054a0000) -> up to 1 bf/chunk Page 45376 (0x058a0000) -> up to 1 bf/chunk Page 47040 (0x05be0000) -> up to 1 bf/chunk Page 47552 (0x05ce0000) -> up to 1 bf/chunk Page 49344 (0x06060000) -> up to 1 bf/chunk Page 49856 (0x06160000) -> up to 1 bf/chunk Page 62784 (0x07aa0000) -> up to 1 bf/chunk Page 65153 (0x07f40800) -> up to 1 bf/chunk Page 65228 (0x07f66000) -> up to 1 bf/chunk Page 65382 (0x07fb3000) -> up to 1 bf/chunk Page 98624 (0x0c0a0000) -> up to 1 bf/chunk Page 101952 (0x0c720000) -> up to 1 bf/chunk Page 107584 (0x0d220000) -> up to 1 bf/chunk Page 118208 (0x0e6e0000) -> up to 1 bf/chunk Page 126656 (0x0f760000) -> up to 1 bf/chunk Page 127680 (0x0f960000) -> up to 1 bf/chunk Page 129920 (0x0fdc0000) -> up to 1 bf/chunk Maximum number of bitflips: 1 Pages with bitflips: 44/130752It is also possible to reduce the output with the .quiet suffix in order to just show the summary.
=> nand watch.chip device 0 whole chip size adjusted to 0xff60000 (5 bad blocks) NAND watch for bitflips in area 0x0-0xff60000: Maximum number of bitflips: 1 Pages with bitflips: 44/130752Signed-off-by: Miquel Raynal miquel.raynal@bootlin.com
Hello, I recently came across a batch of NANDs with a lot of "natural" bitflips so in order to easily and objectively characterize how unstable these parts were, I wrote this little tool which was pretty handy to have in U-Boot. I believe it can be useful for others as well, so here is the patch. Cheers, Miquèl
cmd/Kconfig | 5 ++ cmd/nand.c | 103 ++++++++++++++++++++++++++++++++++++++++ drivers/mtd/mtdcore.c | 22 +++++++++ include/linux/mtd/mtd.h | 1 + 4 files changed, 131 insertions(+)
diff --git a/cmd/Kconfig b/cmd/Kconfig index 451baa3ecac..0524328d373 100644 --- a/cmd/Kconfig +++ b/cmd/Kconfig @@ -1384,6 +1384,11 @@ config CMD_NAND_TORTURE help NAND torture support.
+config CMD_NAND_WATCH
bool "nand watch"helpNAND watch bitflip support.endif # CMD_NAND
config CMD_NVME diff --git a/cmd/nand.c b/cmd/nand.c index 71b8f964429..3bf67f5b65e 100644 --- a/cmd/nand.c +++ b/cmd/nand.c @@ -231,6 +231,54 @@ free_dat: return ret; }
+#ifdef CONFIG_CMD_NAND_WATCH +static int nand_watch_bf(struct mtd_info *mtd, ulong off, ulong size, bool quiet) +{
unsigned int max_bf = 0, pages_wbf = 0;unsigned int first_page, pages, i;struct mtd_oob_ops ops = {};u_char *buf;int ret;buf = memalign(ARCH_DMA_MINALIGN, mtd->writesize);if (!buf) {puts("No memory for page buffer\n");return 1;}first_page = off / mtd->writesize;pages = size / mtd->writesize;ops.datbuf = buf;ops.len = mtd->writesize;for (i = first_page; i < first_page + pages; i++) {ulong addr = mtd->writesize * i;ret = mtd_read_oob_bf(mtd, addr, &ops);if (ret < 0) {if (quiet)continue;printf("Page %7d (0x%08lx) -> error %d\n",i, addr, ret);} else if (ret) {max_bf = max(max_bf, (unsigned int)ret);pages_wbf++;if (quiet)continue;printf("Page %7d (0x%08lx) -> up to %2d bf/chunk\n",i, addr, ret);}}printf("Maximum number of bitflips: %u\n", max_bf);printf("Pages with bitflips: %u/%u\n", pages_wbf, pages);free(buf);return 0;+} +#endif
/* ------------------------------------------------------------------------- */
static int set_dev(int dev) @@ -778,6 +826,55 @@ static int do_nand(struct cmd_tbl *cmdtp, int flag, int argc, return ret == 0 ? 0 : 1; }
+#ifdef CONFIG_CMD_NAND_WATCH
if (strncmp(cmd, "watch", 5) == 0) {int args = 2;if (cmd[5]) {if (!strncmp(&cmd[5], ".part", 5)) {args = 1;} else if (!strncmp(&cmd[5], ".chip", 5)) {args = 0;} else {goto usage;}}if (cmd[10])if (!strncmp(&cmd[10], ".quiet", 6))quiet = true;if (argc != 2 + args)goto usage;ret = mtd_arg_off_size(argc - 2, argv + 2, &dev, &off, &size,&maxsize, MTD_DEV_TYPE_NAND, mtd->size);if (ret)return ret;/* size is unspecified */if (argc < 4)adjust_size_for_badblocks(&size, off, dev);if ((off & (mtd->writesize - 1)) ||(size & (mtd->writesize - 1))) {printf("Attempt to read non page-aligned data\n");return -EINVAL;}ret = set_dev(dev);if (ret)return ret;mtd = get_nand_dev_by_index(dev);printf("\nNAND watch for bitflips in area 0x%llx-0x%llx:\n",off, off + size);return nand_watch_bf(mtd, off, size, quiet);}+#endif
#ifdef CONFIG_CMD_NAND_TORTURE if (strcmp(cmd, "torture") == 0) { loff_t endoff; @@ -943,6 +1040,12 @@ U_BOOT_LONGHELP(nand, "nand erase.chip [clean] - erase entire chip'\n" "nand bad - show bad blocks\n" "nand dump[.oob] off - dump page\n" +#ifdef CONFIG_CMD_NAND_WATCH
"nand watch <off> <size> - check an area for bitflips\n""nand watch.part <part> - check a partition for bitflips\n""nand watch.chip - check the whole device for bitflips\n""\t\t.quiet - Query only the summary, not the details\n"+#endif #ifdef CONFIG_CMD_NAND_TORTURE "nand torture off - torture one block at offset\n" "nand torture off [size] - torture blocks from off to off+size\n" diff --git a/drivers/mtd/mtdcore.c b/drivers/mtd/mtdcore.c index aa78d41a55e..2baf92a9056 100644 --- a/drivers/mtd/mtdcore.c +++ b/drivers/mtd/mtdcore.c @@ -1126,6 +1126,28 @@ int mtd_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops) } EXPORT_SYMBOL_GPL(mtd_read_oob);
+/* This is a bare copy of mtd_read_oob returning the actual number of bitflips */ +int mtd_read_oob_bf(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops) +{
int ret_code;ops->retlen = ops->oobretlen = 0;if (!mtd->_read_oob)return -EOPNOTSUPP;/** In cases where ops->datbuf != NULL, mtd->_read_oob() has semantics* similar to mtd->_read(), returning a non-negative integer* representing max bitflips. In other cases, mtd->_read_oob() may* return -EUCLEAN. In all cases, perform similar logic to mtd_read().*/ret_code = mtd->_read_oob(mtd, from, ops);if (unlikely(ret_code < 0))return ret_code;if (mtd->ecc_strength == 0)return 0; /* device lacks ecc */return ret_code;+} +EXPORT_SYMBOL_GPL(mtd_read_oob_bf);
int mtd_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops) { diff --git a/include/linux/mtd/mtd.h b/include/linux/mtd/mtd.h index 09f52698877..28afbb86ea9 100644 --- a/include/linux/mtd/mtd.h +++ b/include/linux/mtd/mtd.h @@ -413,6 +413,7 @@ int mtd_panic_write(struct mtd_info *mtd, loff_t to, size_t len, size_t *retlen, const u_char *buf);
int mtd_read_oob(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops); +int mtd_read_oob_bf(struct mtd_info *mtd, loff_t from, struct mtd_oob_ops *ops); int mtd_write_oob(struct mtd_info *mtd, loff_t to, struct mtd_oob_ops *ops);
int mtd_get_fact_prot_info(struct mtd_info *mtd, size_t len, size_t *retlen,
2.34.1