Hi Quentin, On 2024-04-24 11:11, Quentin Schulz wrote: > Hi Jonas, > > On 4/24/24 00:40, Jonas Karlman wrote: >> Hi Quentin, >> >> On 2024-04-15 16:16, Quentin Schulz wrote: >>> From: Quentin Schulz <[email protected]> > [...] > >>> + if (!(tmp_mem_map->attrs & PTE_BLOCK_NON_SHARE)) { >> >> This check does not seem to work because PTE_BLOCK_NON_SHARE evaluates >> to 0. Because of this the logic to split the 0-8 GiB bank reported on >> rk3568 is never split in two. >> > > Oof, that's a bit oversight, thanks for the catch. > > Can you test the following please? > > """ > diff --git a/arch/arm/mach-rockchip/sdram.c b/arch/arm/mach-rockchip/sdram.c > index 5b1ff1e5495..0492f9b9f41 100644 > --- a/arch/arm/mach-rockchip/sdram.c > +++ b/arch/arm/mach-rockchip/sdram.c > @@ -196,7 +196,23 @@ static int rockchip_dram_init_banksize(void) > const phys_size_t rsrv_size = tmp_mem_map->size; > const phys_addr_t rsrv_end = rsrv_start + rsrv_size; > > - if (!(tmp_mem_map->attrs & PTE_BLOCK_NON_SHARE)) { > + /* > + * DRAM memories are expected by Arm to be marked as > + * Normal Write-back cacheable, Inner shareable[1], so > + * let's filter on that to put holes in non-DRAM areas. > + * > + * [1] > https://developer.arm.com/documentation/102376/0200/Cacheability-and-shareability-attributes > + */ > + const u64 dram_attrs = PTE_BLOCK_MEMTYPE(MT_NORMAL) | > + PTE_BLOCK_INNER_SHARE; > + /* > + * (AttrIndx | SH) in Lower Attributes of Block > + * Descriptor[2]. > + * [2] > https://developer.arm.com/documentation/102376/0200/Describing-memory-in-AArch64 > + */ > + const u64 attrs_mask = PMD_ATTRINDX_MASK | GENMASK(9, > 8); > + > + if ((tmp_mem_map->attrs & attrs_mask) == dram_attrs) { > tmp_mem_map++; > continue; > } > """ > > The DRAM mem_map entry for Rockchip devices seems to have those > attributes and no other non-DRAM entry seems to have > PTE_BLOCK_MEMTYPE(MT_NORMAL). > > We may have an issue in the future if we also want to mark SRAM, ROM or > flash > (https://developer.arm.com/documentation/102376/0200/Normal-memory > because it's likely those would also match the same attributes but we > would need to put holes for those so that they aren't thought to be > DRAM, but I guess we can tackle this the day this happens :) > > Thanks again for the catch, let me know if this helps and makes sense > and I'll send a v4 for it.
I think this makes sense and seem to work on at least two of my RK356x 8GB boards using the generic-rk3568 target. U-Boot: => bdinfo boot_params = 0x0000000000000000 DRAM bank = 0x0000000000000000 -> start = 0x0000000000200000 -> size = 0x00000000efe00000 DRAM bank = 0x0000000000000001 -> start = 0x0000000100000000 -> size = 0x0000000100000000 flashstart = 0x0000000000000000 flashsize = 0x0000000000000000 flashoffset = 0x0000000000000000 baudrate = 1500000 bps relocaddr = 0x00000000eff3a000 reloc off = 0x00000000ef53a000 Build = 64-bit fdt_blob = 0x00000000edf0b390 new_fdt = 0x00000000edf0b390 fdt_size = 0x000000000000f980 lmb_dump_all: memory.cnt = 0x2 / max = 0x10 memory[0] [0x200000-0xefffffff], 0xefe00000 bytes flags: 0 memory[1] [0x100000000-0x1ffffffff], 0x100000000 bytes flags: 0 reserved.cnt = 0x2 / max = 0x10 reserved[0] [0xecf07000-0xefffffff], 0x030f9000 bytes flags: 0 reserved[1] [0x100000000-0x1ffffffff], 0x100000000 bytes flags: 0 devicetree = separate serial addr = 0x00000000fe660000 width = 0x0000000000000004 shift = 0x0000000000000002 offset = 0x0000000000000000 clock = 0x00000000016e3600 arch_number = 0x0000000000000000 TLB addr = 0x00000000efff0000 irq_sp = 0x00000000edf0b0c0 sp start = 0x00000000edf0b0c0 Early malloc usage: 2338 / 10000 Linux: [ 0.000000] Machine model: Generic RK3566/RK3568 [ 0.000000] efi: UEFI not found. [ 0.000000] NUMA: No NUMA configuration found [ 0.000000] NUMA: Faking a node at [mem 0x0000000000200000-0x00000001ffffffff] [ 0.000000] NUMA: NODE_DATA [mem 0x1ff0389c0-0x1ff03afff] [ 0.000000] Zone ranges: [ 0.000000] DMA [mem 0x0000000000200000-0x00000000ffffffff] [ 0.000000] DMA32 empty [ 0.000000] Normal [mem 0x0000000100000000-0x00000001ffffffff] [ 0.000000] Movable zone start for each node [ 0.000000] Early memory node ranges [ 0.000000] node 0: [mem 0x0000000000200000-0x00000000efffffff] [ 0.000000] node 0: [mem 0x0000000100000000-0x00000001ffffffff] [ 0.000000] Initmem setup node 0 [mem 0x0000000000200000-0x00000001ffffffff] [ 0.000000] On node 0, zone DMA: 512 pages in unavailable ranges [ 0.000000] cma: Reserved 32 MiB at 0x00000000ee000000 on node -1 Regards, Jonas > > Cheers, > Quentin
