On Mon Jun 1, 2026 at 2:56 AM CEST, Cheng Ming Lin wrote: > Hi Michael, > > Michael Walle 於 2026年5月28日週四 下午3:52寫道： >> >> Hi, >> >> On Thu May 28, 2026 at 9:42 AM CEST, Miquel Raynal wrote: >> > Hi Cheng Ming, >> > >> >>> It looks like you're getting bitten by the ID reuse. You can't just >> >>> unconditionally add the quad PP because as far as I can see the >> >>> MX25L12805D [1] is just a standard single bit i/o flash and doesn't >> >>> support the 4PP. >> >> >> >> You are absolutely right. Thanks for catching this. >> >> >> >> The MX25L12805D is indeed a much older product (released around 2009). >> >> Since the initial JESD216 SFDP standard wasn't published until 2011, >> >> I double-checked with our internal PM and confirmed that the MX25L12805D >> >> does not support SFDP at all. >> >> >> >> Since the newer flashes (MX25L12833F and MX25L12845G) do support SFDP, >> >> we could use this as a differentiator to distinguish them from the legacy >> >> MX25L12805D. >> >> >> >> What if we try to read the SFDP signature (RDSFDP) in the fixup hook? >> >> If a valid SFDP signature is detected, we can safely identify it as the >> >> newer flash and apply the SNOR_HWCAPS_PP_1_4_4 capability. If there is >> >> no SFDP signature, we leave it as is for the legacy MX25L12805D. >> >> >> >> Do you think this approach is feasible and acceptable? If so, I will >> >> implement this logic and submit a v2 patch. >> > >> > The ->post_sfdp() fixup hook is documented as "not called for SPI NORs >> > that do not support SFDP". Alternatively, I believe an earlier hook, >> > like ->post_bfpt() could also work since it does not seem to run on >> > non SFDP compatible flashes. >> >> Moreover: >> >> * spi_nor_post_sfdp_fixups() is called after the SFDP tables are parsed. >> * It is used to tweak various flash parameters when information provided >> * by the SFDP tables are wrong. >> >> And it looks like that is the case for these flashes. Also, please >> provide a comment on the fixup describing that the flash misses that >> property. Basically, what you've written in the commit message. > > I am currently working on the flash part 0xc22018 and I am planning to > add its size back to the flash info table. However, I encountered an > issue regarding the fixup hooks. > > Since this specific flash does not support dual, quad, or octal reads, > the core logic in spi_nor_init_params_deprecated() determines that SFDP > parsing is not needed. Because the SFDP parsing is skipped, the post_bfpt > or post_sfdp hooks are never triggered, which prevents us from applying > the necessary fixups as we expected. > > I initially considered asking about the possibility of removing this > specific check in spi_nor_init_params_deprecated(), which would allow > all SPI-NOR flashes to go through SFDP parsing by default (unless > SPI_NOR_SKIP_SFDP is explicitly set). Oops, I've forgot to add you on CC. Sorry. I've just send a series to get rid of that legacy handling. Please give it a try on your flashes. [1] https://lore.kernel.org/linux-mtd/20260601125438.3481722-1-mwalle@kernel.org/ -michael > However, I am aware that forcing SFDP parsing globally on all legacy > 1-1-1 flashes might be risky. Flashes predating the JESD216 standard > might return garbage data when the SFDP opcode (0x5A) is issued, > potentially causing regressions if the garbage data accidentally matches > the SFDP signature. > > If dropping this global check is indeed considered too risky for older > hardware, would it be acceptable to use the late_init hook to handle > this flash individually? The idea would be to manually check and parse > the SFDP table exclusively for 0xc22018 within its own fixup, ensuring > that no other legacy 1-1-1 flashes are affected by this change. > > Any thoughts or guidance on whether this late_init approach is the right > way to go would be greatly appreciated.