[U-Boot] [PATCH] FSL PCI: Configure PCIe reference ratio

[Joakim Tjernlund <Joakim.Tjernlund@infinera.com>]

On Tue, 2017-11-21 at 18:04 +0000, York Sun wrote:
> 
> 
> On 11/21/2017 09:52 AM, Joakim Tjernlund wrote:
> > On Tue, 2017-11-21 at 17:45 +0000, York Sun wrote:
> > > 
> > > On 11/21/2017 09:41 AM, Joakim Tjernlund wrote:
> > > > On Tue, 2017-11-21 at 17:32 +0000, York Sun wrote:
> > > > > CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
> > > > > 
> > > > > 
> > > > > On 11/21/2017 09:29 AM, Joakim Tjernlund wrote:
> > > > > > On Tue, 2017-11-21 at 17:23 +0000, York Sun wrote:
> > > > > > > CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
> > > > > > > 
> > > > > > > 
> > > > > > > On 11/21/2017 09:18 AM, Joakim Tjernlund wrote:
> > > > > > > > On Tue, 2017-09-12 at 19:56 +0200, Joakim Tjernlund wrote:
> > > > > > > > > Most FSL PCIe controllers expects 333 MHz PCI reference clock.
> > > > > > > > > This clock is derived from the CCB but in many cases the ref.
> > > > > > > > > clock is not 333 MHz and a divisor needs to be configured.
> > > > > > > > > 
> > > > > > > > > This adds PEX_CCB_DIV #define which can be defined for each
> > > > > > > > > type of CPU/platform.
> > > > > > > > > 
> > > > > > > > > Signed-off-by: Joakim Tjernlund <joakim.tjernlund@infinera.com>
> > > > > > > > > ---
> > > > > > > > >  drivers/pci/fsl_pci_init.c | 6 ++++++
> > > > > > > > >  1 file changed, 6 insertions(+)
> > > > > > > > > 
> > > > > > > > > diff --git a/drivers/pci/fsl_pci_init.c b/drivers/pci/fsl_pci_init.c
> > > > > > > > > index 52792dcd59..4d00b3f26c 100644
> > > > > > > > > --- a/drivers/pci/fsl_pci_init.c
> > > > > > > > > +++ b/drivers/pci/fsl_pci_init.c
> > > > > > > > > @@ -322,6 +322,12 @@ void fsl_pci_init(struct pci_controller *hose, struct fsl_pci_info *pci_info)
> > > > > > > > > 
> > > > > > > > >      pci_setup_indirect(hose, cfg_addr, cfg_data);
> > > > > > > > > 
> > > > > > > > > +#ifdef PEX_CCB_DIV
> > > > > > > > > +    /* Configure the PCIE controller core clock ratio */
> > > > > > > > > +    pci_hose_write_config_dword(hose, dev, 0x440,
> > > > > > > > > +                                ((gd->bus_clk / 1000000) *
> > > > > > > > > +                                 (16 / PEX_CCB_DIV)) / 333);
> > > > > > > > > +#endif
> 
> 
> 
> I took another look at this patch. Would it be appropriate to alway
> write to this register with correct clock?

I sure hope so, the docs I have only mentions this reg having a default value which
needs to be changed in most cases I guess.

> 
> 
> 
> > > > > > > > >      block_rev = in_be32(&pci->block_rev1);
> > > > > > > > >      if (PEX_IP_BLK_REV_2_2 <= block_rev) {
> > > > > > > > >              pi = &pci->pit[2];      /* 0xDC0 */
> > > > > > > > 
> > > > > > > > Ping?
> > > > > > > > 
> > > > > > > >  Jocke
> > > > > > > > 
> > > > > > > 
> > > > > > > I believe Mingkai's last comment was "to add the PCIe clock in gd".
> > > > > > 
> > > > > > Right, I seem to have forgotten to comment on that ...
> > > > > > Why add that complexity/bloat? This is a constant defined by the CPU/SOC so
> > > > > > it makes perfect sense to just #define it.
> > > > > > 
> > > > > 
> > > > > I am leaning to agree with you. The clock is either CCB clock, or CCB/2.
> > > > > Would it be better if you use a Kconfig option and select the divisor by
> > > > > SoC?
> > > > 
> > > > No, this is not something that needs Kconfig, just add the PEX_CCB_DIV #define
> > > > in relevant PPC CPU, like in config_mpc85xx.h
> > > > 
> > > 
> > > So what should be in Kconfig, and what shouldn't? This is per SoC macro.
> > > Sounds like CONFIG_SYS_ in old days.
> > 
> > I must confess, I am not up to date with Kconfig stuff. I based my suggestion on
> > what already exists in config_mpc85xx.h:
> > #elif defined(CONFIG_PPC_T1040) || defined(CONFIG_PPC_T1042) ||\
> > defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
> > #define CONFIG_E5500
> > #define CONFIG_FSL_CORENET            /* Freescale CoreNet platform */
> > #define CONFIG_SYS_FSL_QORIQ_CHASSIS2 /* Freescale Chassis generation 2 */
> > #define CONFIG_SYS_FSL_CORES_PER_CLUSTER 1
> > #define CONFIG_SYS_FSL_QMAN_V3                /* QMAN version 3 */
> > #ifdef CONFIG_SYS_FSL_DDR4
> > #define CONFIG_SYS_FSL_DDRC_GEN4
> > #endif
> > #if defined(CONFIG_PPC_T1040) || defined(CONFIG_PPC_T1042)
> > #define CONFIG_MAX_CPUS                       4
> > #elif defined(CONFIG_PPC_T1020) || defined(CONFIG_PPC_T1022)
> > #define CONFIG_MAX_CPUS                       2
> > #endif
> > #define CONFIG_SYS_FSL_NUM_CC_PLLS    2
> > #define CONFIG_SYS_FSL_CLUSTER_CLOCKS   { 1, 1, 1, 1 }
> > #define CONFIG_SYS_FSL_NUM_LAWS               16
> > #define CONFIG_SYS_FSL_SRDS_1
> > #define CONFIG_SYS_FSL_SEC_COMPAT     5
> > #define CONFIG_SYS_NUM_FMAN           1
> > #define CONFIG_SYS_NUM_FM1_DTSEC      5
> > ....
> > etc.
> > 
> > Seems like a good place to put another CPU constant.
> > 
> 
> Yeah. We have been moving things out of header files and into Kconfig. I
> would avoid adding new macro to header files.

OK, I am happy if don't have to manually select the constant.

> 
> > Anyhow, that patch stands of its own I think. Where to put all constants
> > is another patch for NXP.
> 
> Yes, we can add another patch to define this option for SoCs.

 :)

 Jocke