Re: [PATCH master 3/4] ddr: imx8m: workaround old spreadsheets not initializing ADDRMAP7

[Teresa Remmet <T.Remmet@phytec.de>]

Hello Ahmad,

Am Donnerstag, dem 23.06.2022 um 13:26 +0200 schrieb Teresa Remmet:
> Am Donnerstag, dem 23.06.2022 um 13:14 +0200 schrieb Ahmad Fatoum:
> > Hi Teresa,
> > 
> > On 23.06.22 12:59, Teresa Remmet wrote:
> > > Hello Ahmad,
> > > 
> > > Am Donnerstag, dem 23.06.2022 um 12:30 +0200 schrieb Ahmad
> > > Fatoum:
> > > > Older NXP DDR spreadsheets don't initialize ADDRMAP7, leaving
> > > > it
> > > > at
> > > > its
> > > > POR default of zero. Now that barebox looks at ADDRMAP7 to be
> > > > able to
> > > > correctly detect bigger memory sizes, barebox proper on out-of-
> > > > tree
> > > > boards with older spreadsheets may read back 4x times as much
> > > > RAM
> > > > as actually fitted.
> > > > 
> > > > Work around this by writing a trailing 0xf0f (the neutral
> > > > ignore-
> > > > me
> > > > value for the register) if the register wasn't written through
> > > > dram_timing_info::ddrc_cfg. We consider this safe to do,
> > > > because
> > > > the DDRC is held in reset while these values are programmed.
> > > 
> > > have you tried this patch without actually updating the RAM
> > > Timings
> > > (
> > > so patch 2/4)?
> > 
> > Yes. I've an out-of-tree i.MX8MM board with 1G of LPDDR4 and old
> > RAM
> > setup.
> > 
> > With upstream/next: 4G
> > With Workaround: 1G + Warning
> > With fixed ADDRMAP7: 1G, no warning
> > 
> > Reading back 0x3d40021c gives me 0xf0f as expected.
> > 
> > > I have just played around also setting ADDRMAP7 right before 
> > > ddr_cfg_umctl2() without fixing the timings and the register was
> > > 0
> > > again after startup and so the RAM size wrong.
> > > So it seems to me that the value is overwritten/resetted at some
> > > other
> > > point again ... or maybe I just did something wrong.
> > 
> > Strange. What SoC/board is this on?
> 
> phyCORE-i.MX8MQ but with a not upstreamed 2GB RAM configuration.
> 
> I applied this patch now with the same result. Wrong RAM size. 

I now have also tried some out of tree phyCORE-i.MX8MM implementation
without and with this patch and the patch works for our MM. too. RAM
size is corrected.
So maybe the MQ is somehow different here?

Teresa




> Regards,
> Teresa
> 
> > Cheers,
> > Ahmad
> > 
> > > Regards,
> > > Teresa
> > > 
> > > > Fixes: dad2b5636bd8 ("ARM: imx: Add imx8 support for 18 bit
> > > > SDRAM
> > > > row
> > > > size handle")
> > > > Fixes: 6cf197fa61f9 ("arm: imx: mmdc_size: Increase row_max for
> > > > imx8m")
> > > > Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
> > > > ---
> > > >  drivers/ddr/imx8m/ddr_init.c | 18 ++++++++++++++++++
> > > >  drivers/ddr/imx8m/helper.c   |  6 ++++++
> > > >  include/soc/imx8m/ddr.h      |  1 +
> > > >  3 files changed, 25 insertions(+)
> > > > 
> > > > diff --git a/drivers/ddr/imx8m/ddr_init.c
> > > > b/drivers/ddr/imx8m/ddr_init.c
> > > > index ae05b136229c..9a4b4e2ca88a 100644
> > > > --- a/drivers/ddr/imx8m/ddr_init.c
> > > > +++ b/drivers/ddr/imx8m/ddr_init.c
> > > > @@ -13,14 +13,32 @@
> > > >  #include <mach/imx8m-regs.h>
> > > >  #include <mach/imx8m-ccm-regs.h>
> > > >  
> > > > +bool imx8m_ddr_old_spreadsheet = true;
> > > > +
> > > >  static void ddr_cfg_umctl2(struct dram_cfg_param *ddrc_cfg,
> > > > int
> > > > num)
> > > >  {
> > > >  	int i = 0;
> > > >  
> > > >  	for (i = 0; i < num; i++) {
> > > > +		if (ddrc_cfg->reg == DDRC_ADDRMAP7(0))
> > > > +		    imx8m_ddr_old_spreadsheet = false;
> > > >  		reg32_write((unsigned long)ddrc_cfg->reg,
> > > > ddrc_cfg-
> > > > > val);
> > > >  		ddrc_cfg++;
> > > >  	}
> > > > +
> > > > +	/*
> > > > +	 * Older NXP DDR configuration spreadsheets don't
> > > > initialize
> > > > ADDRMAP7,
> > > > +	 * which falsifies the memory size read back from the
> > > > controller
> > > > +	 * in barebox proper.
> > > > +	 */
> > > > +	if (imx8m_ddr_old_spreadsheet) {
> > > > +		pr_warn("Working around old spreadsheet. Please
> > > > regenerate\n");
> > > > +		/*
> > > > +		 * Alternatively, stick { DDRC_ADDRMAP7(0),
> > > > 0xf0f }
> > > > into
> > > > +		 * struct dram_timing_info::ddrc_cfg of your
> > > > old timing
> > > > file
> > > > +		 */
> > > > +		reg32_write(DDRC_ADDRMAP7(0), 0xf0f);
> > > > +	}
> > > >  }
> > > >  
> > > >  /*
> > > > diff --git a/drivers/ddr/imx8m/helper.c
> > > > b/drivers/ddr/imx8m/helper.c
> > > > index 94bbb811576d..98e40849584b 100644
> > > > --- a/drivers/ddr/imx8m/helper.c
> > > > +++ b/drivers/ddr/imx8m/helper.c
> > > > @@ -62,6 +62,12 @@ void dram_config_save(struct
> > > > dram_timing_info
> > > > *timing_info,
> > > >  		cfg++;
> > > >  	}
> > > >  
> > > > +	if (imx8m_ddr_old_spreadsheet) {
> > > > +		cfg->reg = DDRC_ADDRMAP7(0);
> > > > +		cfg->val = 0xf0f;
> > > > +		cfg++;
> > > > +	}
> > > > +
> > > >  	/* save ddrphy config */
> > > >  	saved_timing->ddrphy_cfg = cfg;
> > > >  	for (i = 0; i < timing_info->ddrphy_cfg_num; i++) {
> > > > diff --git a/include/soc/imx8m/ddr.h b/include/soc/imx8m/ddr.h
> > > > index 9ae7cb877686..147a7d499aaf 100644
> > > > --- a/include/soc/imx8m/ddr.h
> > > > +++ b/include/soc/imx8m/ddr.h
> > > > @@ -407,6 +407,7 @@ static inline void reg32setbit(unsigned
> > > > long
> > > > addr, u32 bit)
> > > >  #define dwc_ddrphy_apb_rd(addr) \
> > > >  	reg32_read(IOMEM(IP2APB_DDRPHY_IPS_BASE_ADDR(0)) + 4 *
> > > > (addr))
> > > >  
> > > > +extern bool imx8m_ddr_old_spreadsheet;
> > > >  extern struct dram_cfg_param ddrphy_trained_csr[];
> > > >  extern uint32_t ddrphy_trained_csr_num;
> > > >  
-- 
PHYTEC Messtechnik GmbH | Robert-Koch-Str. 39 | 55129 Mainz, Germany

Geschäftsführer: Dipl.-Ing. Michael Mitezki, Dipl.-Ing. Bodo Huber |
Handelsregister Mainz HRB 4656 | Finanzamt Mainz | St.Nr. 266500608, DE
149059855