Re: [PATCH] mtd: spinand: confirm ready state on SkyHigh S35ML chips
From: Miquel Raynal
Date: Tue Nov 17 2026 - 03:36:55 EST
Hi Vitaly,
On 16/11/2026 at 21:27:14 +02, Vitaliy Sochnev <sochnev.v.74@xxxxxxxxx> wrote:
> Hi Miquèl,
>
> I instrumented the SkyHigh SPI-NAND path on a Nokia XG-040G-MD with
> ktime_get_ns() and bounded kernel log messages, without tracing. With
> INFO messages suppressed on the serial console, 512 page-load waits, 510
> program waits and 8 erase waits returned READY on both the polled result
> and the subsequent status read. The pairs were 0x00/0x00, except for
> nine 0x10/0x10 read pairs (1–2 corrected bit-flips according to the
> SkyHigh driver; STATUS_BUSY clear). I did not capture a READY-to-BUSY
> transition in these attempts.
>
> For 128 read waits, I also recorded the individual GET_FEATURE
> transactions: each had BUSY (0x01 or 0x11), then READY (0x00 or 0x10),
> followed by a READY confirmation. The poll transactions took 10.80–17.08
> us. The interval from completion of the READY poll to the start of
> confirmation was 6.84–11.52 us (median 7.22 us); confirmation took
> 11.12–15.24 us. One sequence of ktime_get_ns() start/end timestamps in
> ns was BUSY 0x01 at 38810096120/38810107840, READY 0x00 at
> 38810158640/38810170520, and confirmation 0x00 at
> 38810178320/38810189840. Logging in the poll path adds latency, so these
> intervals do not represent uninstrumented timings or exclude a shorter
> transient.
>
> For SPINAND_RESET_INITIAL_DELAY_US, I measured three boots each at 5, 9
> and 20 us. The reset wait returned READY 19.68–19.80, 23.56–24.12 and
> 105.76–358.68 us after the reset command, respectively; an additional
> status read returned 0x00 each time. At boot, reset runs before the
> chip's quirk flag is assigned. spi_mem_poll_status() uses udelay() below
> 10 us and usleep_range() at 10 us and above, accounting for the larger
> and more variable 20 us result. Increasing the RESET delay changes this
> wait, but does not add a second status sample to the page-read, program
> or erase waits. Whether it addresses the reported events remains
> unverified.
Thanks a lot for these numbers. Based on those results, and based on the
fact that you never experienced a RDY -> BUSY -> RDY situation as
claimed in the original commit log, how do you explain the proposed
diff? I also did not fully understand if all the commands could
potentially report faulty statuses or only one?
Do you have several dies on this chips? Could it be a problem with the
status bit synchronization?
Miquèl