On Tue, May 07, 2019 at 11:39:12AM +0200, Niel Fourie wrote:
Hi Tom,
On 5/6/19 7:24 PM, Tom Rini wrote:
On Mon, May 06, 2019 at 06:44:48PM +0200, Niel Fourie wrote:
Hi Tom,
On 5/6/19 4:18 PM, Tom Rini wrote:
On Mon, May 06, 2019 at 04:02:53PM +0200, Niel Fourie wrote:
Support for Phytech phyCORE AM335x R2 SOM (PCL060) on the Phytec phyBOARD-Wega AM335x.
CPU : AM335X-GP rev 2.1 Model: Phytec AM335x phyBOARD-WEGA DRAM: 256 MiB NAND: 256 MiB MMC: OMAP SD/MMC: 0 eth0: ethernet@4a100000
Working:
- Eth0
- i2C
- MMC/SD
- NAND
- UART
- USB (host)
Device trees were taken from Linux mainline: commit 37624b58542f ("Linux 5.1-rc7")
Signed-off-by: Niel Fourie lusus@denx.de
[snip]
void sdram_init(void) { /* Configure memory to maximum supported size for detection */ int ram_type_index = PHYCORE_R2_MT41K512M16HA125IT_1024MB; config_ddr(DDR_CLK_MHZ, &ioregs, &physom_timings[ram_type_index].ddr3_data, &ddr3_cmd_ctrl_data, &physom_timings[ram_type_index].ddr3_emif_reg_data, 0);
/* Detect memory physically present */ gd->ram_size = get_ram_size((void *)CONFIG_SYS_SDRAM_BASE, CONFIG_MAX_RAM_BANK_SIZE);
/* Reconfigure memory for actual detected size */ switch (gd->ram_size) { case SZ_1G: ram_type_index = PHYCORE_R2_MT41K512M16HA125IT_1024MB; break; case SZ_512M: ram_type_index = PHYCORE_R2_MT41K256M16TW107IT_512MB; break; case SZ_256M: default: ram_type_index = PHYCORE_R2_MT41K128M16JT_256MB; break; } config_ddr(DDR_CLK_MHZ, &ioregs, &physom_timings[ram_type_index].ddr3_data, &ddr3_cmd_ctrl_data, &physom_timings[ram_type_index].ddr3_emif_reg_data, 0); }
The ugliest part of this is, as you pointed out, that directly after this is called, get_ram_size() will be called again from sdram_init(). But it at least noninvasive, and no longer requires the device tree.
I don't think it's safe to call config_ddr twice, especially with the possibly wrong parameters. What's barebox doing in this case, being told the presumably correct DDR size in the device tree?
Good point. Barebox uses the above mechanism to detect the memory size, and I could find no equivalent memory size specified in its internal device tree.
Configure for 1GB and then see how much we can actually talk to?
Marek originally proposed using the memory size specified in the device tree as an improvement over specifying the size in the defconfig (as in v1 of the patch).
But then you aren't populating 3 device trees nor making it clear / easy to say which module you're on, and then still need to change the config for which DT you're picking up. These SOMs really don't provide any run-time method to see which one you're on? There's no GPIOs to poke?