On 3/6/23 19:20, Simon Glass wrote:
Hi Johan,
On Thu, 2 Mar 2023 at 17:15, Johan Jonker jbx6244@gmail.com wrote:
The fdt_addr_t and phys_addr_t size have been decoupled. A 32bit CPU can expect 64-bit data from the device tree parser, so use
Why is that? It seems quite inefficient.
1: === Because the device tree does describes more then just only the internal io/mem range. When a NAND chip is connected it must be able to describe partitions far beyond that 32bit range. This change only changes the PARSE capacity defined by fdt_addr_t and fdt_size_t and not the phys_addr_t and phys_size_t. Most drivers make a little mess when taking a DT reg value and then trying to make it fit in a structure of multiple registers with various offsets. Fixing all of that is beyond my capacity/this serie and more a MAINTAINER task.
https://elixir.bootlin.com/u-boot/latest/source/drivers/mtd/mtdpart.c#L933
struct mtd_partition { const char *name; /* identifier string */ uint64_t size; /* partition size */ uint64_t offset; /* offset within the master MTD space */ uint32_t mask_flags; /* master MTD flags to mask out for this partition */ struct nand_ecclayout *ecclayout; /* out of band layout for this partition (NAND only) */ };
int add_mtd_partitions_of(struct mtd_info *master) struct mtd_partition part = { 0 }; fdt_addr_t offset; fdt_size_t size;
[..] part.offset = offset; part.size = size;
While the mtd_partition structure is ready with uint64_t size all the parse functions that export this reg value were typedef to a 32bit value.
=== Given mk808 rk3066a with NAND:
[ 38.750789] nand: Hynix H27UCG8T2ATR-BC 64G 3.3V 8-bit [ 38.756650] nand: 8192 MiB, MLC, erase size: 2048 KiB, page size: 8192, OOB size: 640
=== BEFORE:
List of MTD devices: * nand0 - type: MLC NAND flash - block size: 0x200000 bytes - min I/O: 0x2000 bytes - OOB size: 640 bytes - OOB available: 4294967290 bytes - ECC strength: 40 bits - ECC step size: 1024 bytes - bitflip threshold: 30 bits - 0x000000000000-0x000200000000 : "nand0" - 0x000000400000-0x000000600000 : "boot-blk-0" - 0x000000600000-0x000000800000 : "boot-blk-1" - 0x000000800000-0x000000a00000 : "boot-blk-2" - 0x000000a00000-0x000000c00000 : "boot-blk-3" - 0x000000c00000-0x000000e00000 : "boot-blk-4" - 0x000001000000-0x0000fe000000 : "rootfs" - 0x0000fe000000-0x000100000000 : "bbt"
# This output is corrupted.
=== AFTER:
List of MTD devices: * nand0 - type: MLC NAND flash - block size: 0x200000 bytes - min I/O: 0x2000 bytes - OOB size: 640 bytes - OOB available: 4294967290 bytes - ECC strength: 40 bits - ECC step size: 1024 bytes - bitflip threshold: 30 bits - 0x000000000000-0x000200000000 : "nand0" - 0x000000400000-0x000000600000 : "boot-blk-0" - 0x000000600000-0x000000800000 : "boot-blk-1" - 0x000000800000-0x000000a00000 : "boot-blk-2" - 0x000000a00000-0x000000c00000 : "boot-blk-3" - 0x000000c00000-0x000000e00000 : "boot-blk-4" - 0x000001000000-0x0001fe000000 : "rootfs" - 0x0001fe000000-0x000200000000 : "bbt" === Example: partitions { compatible = "fixed-partitions"; #address-cells = <2>; #size-cells = <2>;
partition@400000 { reg = <0x0 0x00400000 0x0 0x00200000>; label = "boot-blk-0"; };
partition@600000 { reg = <0x0 0x00600000 0x0 0x00200000>; label = "boot-blk-1"; };
partition@800000 { reg = <0x0 0x00800000 0x0 0x00200000>; label = "boot-blk-2"; };
partition@a00000 { reg = <0x0 0x00a00000 0x0 0x00200000>; label = "boot-blk-3"; };
partition@c00000 { reg = <0x0 0x00c00000 0x0 0x00200000>; label = "boot-blk-4"; };
partition@1000000 { reg = <0x0 0x01000000 0x1 0xfd000000>; label = "rootfs"; };
partition@1fe000000 { reg = <0x1 0xfe000000 0x0 0x02000000>; label = "bbt"; }; }; ===
2: As a side effect Rockchip rk3288 with 32bit reg should be able to parse 64bit size device tree synced from Linux. (No need to maintain simply copy and paste) How ever popmetal-rk3288 and phycore-rk3288 must be changed to use TPL/SPL to fit in size again. (Could someone finally make some progress in making these boards future proof? And could Simon Glass make/help with rk3288 video/mipi/lvds/hdmi Uboot drivers for work with Linux nodes?) (This seems to work only on Google boards that nobody has, but it blocks all progress for the rest)
dev_read_addr_index_ptr instead of the dev_read_addr_index function in the various files in the drivers directory that cast to a pointer.
Signed-off-by: Johan Jonker jbx6244@gmail.com Reviewed-by: Michael Trimarchi michael@amarulasolutions.com
Changed V6: use -EINVAL on return drop cast
drivers/mtd/nand/raw/cortina_nand.c | 4 ++-- drivers/net/dm9000x.c | 2 +- drivers/net/dwmac_meson8b.c | 4 ++-- drivers/pci/pcie_dw_meson.c | 8 ++++---- drivers/pci/pcie_dw_rockchip.c | 8 ++++---- drivers/watchdog/sbsa_gwdt.c | 12 ++++++------ 6 files changed, 19 insertions(+), 19 deletions(-)
[..]
Regards, SImon