Hello,
I am trying to replace the xway_nand driver (which is still using the
legacy NAND API) with the intel-nand-controller driver. The Intel LGM
IP (for which intel-nand-controller was implemented) uses a newer
version of the EBU NAND and HSNAND IP found in Lantiq XWAY SoCs. The
most notable change is the addition of HSNAND Intel LGM SoCs (it's not
clear to me if/which Lantiq SoCs also have this DMA engine).
While testing my changes on a Lantiq xRX200 SoC I came across some
issues with the intel-nand-controller driver. The problems I found are:
1) Mismatch between dt-bindings and driver implementation (compatible
string, patch #1 and patch #4) and hardware capabilities (number of
CS lines, patch #1).
2) The driver reads the CS (chip select) line from the NAND controller's
reg property. In the dt-bindings example this is 0xe0f00000. Instead
it must be read from the NAND chip (child node).
3) A few smaller code cleanups to make the driver easier to understand
(patches #5 to #8)
4) I tried to understand the timing parameter calculation code but found
that it probably doesn't work on the Intel LGM SoCs either. The
dt-bindings example use clock ID 125 which is LGM_GCLK_EBU. So far
this is fine because EBU is the actual IP block for the NAND
interface. However, drivers/clk/x86/clk-lgm.c defines this clock as
a gate without a parent, so it's rate (as read by Linux) is always 0.
The intel-nand-controller driver then tries to calculate:
rate = clk_get_rate(ctrl->clk) / HZ_PER_MHZ
(rate will be 0 because clk_get_rate() returns 0) and then:
DIV_ROUND_UP(USEC_PER_SEC, rate)
(this then tries to divide by zero)
For me to move forward with the transition from xway_nand to the
intel-nand-controller driver I to understand a few more details:
- Who from Maxlinear (who took over Intel's AnyWAN division, which
previously worked on the drivers for the Intel LGM SoCs) can send a
patch to correct the LGM_GCLK_EBU clock rate in
drivers/clk/x86/clk-lgm.c? Or is LGM dead and the various drivers
should be removed instead?
- Who from Maxlinear can provide insights into which clock is connected
to the EBU NAND controller on Lantiq XWAY (Danube, xRX100, xRX200,
xRX300) SoCs as well as newer GRX350/GRX550 SoCs so that I can make
the intel-nand-controller work without hardcoded timing settings on
the XWAY SoCs?
Due to the severity of issues 2) and 4) above I am targeting linux-next
with this series. In my opinion there's no point in backporting these
fixes to a driver which has been broken since it was upstreamed.
Changes since v1 from [0]:
- Thanks to Miguel for confirming that the reg property of the NAND chip
is the chip select number of the NAND controller. I removed a question
about this from the cover-letter.
- Fixed accidental $id change in patch #1 which fixes a binding error
reported by Rob's bot
- Dropped RFC status
Changes since v2 from [1]:
- Renamed the binding file (in patch #1) to match the new compatible
string as suggested by Rob
- Added Rob's Acked-by to patch #2 (thank you!)
Best regards,
Martin
[0] https://lore.kernel.org/linux-mtd/20220628163850.17c56935@xps-13/T/#m4b2b6e1c970adf074a17ab9568637aff90e6ca36
[1] https://lore.kernel.org/linux-mtd/20220629213508.1989600-1-martin.blumenstingl@xxxxxxxxxxxxxx/T/#u
Martin Blumenstingl (8):
dt-bindings: mtd: intel: lgm-nand: Fix compatible string
dt-bindings: mtd: intel: lgm-nand: Fix maximum chip select value
mtd: rawnand: intel: Read the chip-select line from the correct OF
node
mtd: rawnand: intel: Remove undocumented compatible string
mtd: rawnand: intel: Don't re-define NAND_DATA_IFACE_CHECK_ONLY
mtd: rawnand: intel: Remove unused nand_pa member from ebu_nand_cs
mtd: rawnand: intel: Remove unused clk_rate member from struct
ebu_nand
mtd: rawnand: intel: Use devm_platform_ioremap_resource_byname()
...l,lgm-nand.yaml => intel,lgm-ebunand.yaml} | 8 +++---
drivers/mtd/nand/raw/intel-nand-controller.c | 28 +++++++++----------
2 files changed, 17 insertions(+), 19 deletions(-)
rename Documentation/devicetree/bindings/mtd/{intel,lgm-nand.yaml => intel,lgm-ebunand.yaml} (91%)
--
2.37.0
