Re: [v10,2/2] PCI: Rockchip: Add Rockchip PCIe controller support

From: Bjorn Helgaas
Date: Thu Sep 01 2016 - 17:26:31 EST


On Thu, Sep 01, 2016 at 10:14:01AM -0700, Brian Norris wrote:
> On Thu, Sep 01, 2016 at 11:34:55AM -0500, Bjorn Helgaas wrote:
> > In the interest of making progress, I made most of the changes Guenter
> > suggested (thanks very much, Guenter!) and made some more of my own on
> > top of those.
>
> I'm not sure, but was this change your idea Bjorn?

Yep.

> commit 88cb3f59d73e261a3cd1957ff392f98c9916a5d1
> Author: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>
> Date: Thu Sep 1 10:27:44 2016 -0500
>
> Simplify the confusing HIWORD_UPDATE scheme.
>
> [Which I'll summarize here; you're replacing the macro:
>
> -/*
> - * The higher 16-bit of this register is used for write protection
> - * only if BIT(x + 16) set to 1 the BIT(x) can be written.
> - */
> -#define HIWORD_UPDATE(val, mask, shift) \
> - ((val) << (shift) | (mask) << ((shift) + 16))
>
> with its expansions:
>
> +
> +/*
> + * The higher 16-bit of this register is used for write protection
> + * only if BIT(x + 16) set to 1 the BIT(x) can be written.
> + */
> +#define PCIE_CLIENT_CONF_ENABLE (0x00010000 | 0x0001)
> +#define PCIE_CLIENT_LINK_TRAIN_ENABLE (0x00020000 | 0x0002)
> +#define PCIE_CLIENT_ARI_ENABLE (0x00080000 | 0x0008)
> +#define PCIE_CLIENT_CONF_LANE_NUM(x) (0x00300000 | (((x >> 1) & 3) << 4)
> +#define PCIE_CLIENT_MODE_RC (0x00400000 | 0x0040)
> +#define PCIE_CLIENT_GEN_SEL(x) (0x00800000 | ((x & 1) << 7)
> +#define PCIE_CLIENT_GEN_SEL_0 0
> +#define PCIE_CLIENT_GEN_SEL_2 1
>
> Full change can be had by:
> git fetch git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci.git pci/host-rockchip-wip
> ]
>
>
> The use of HIWORD_UPDATE can indeed be a bit confusing, IMO, but this is
> really a common Rockchip-ism that, once you read several of their
> drivers, can make a little more sense. If you grep around, it's in at
> least their clock, ethernet, SDHCI, PHY, and DP/DRM drivers. I might
> defer to Heiko (upstream maintainer of Rockchip code) for a decision.
> Maybe there's some intermediate ground where we keep the HIWORK_UPDATE()
> logic (it does make sure we get the 16-bit shift right, I think) while
> still refactoring a few other bits (like PCIE_CLIENT_CONF_LANE_NUM() and
> PCIE_CLIENT_GEN_SEL() for wrapping HIWORK_UPDATE()?).
>
> Anyway, just a thought. If it really is most understandable to write it
> this way, then maybe it's fine to be different than the other 16
> rockchip files that have this pattern.

Hmm. I did look at the other files, so I know this isn't a PCIe
special. HIWORD_UPDATE is ugly as sin, especially in cases like this
where we're always writing a constant value and it takes three
#defines plus HIWORD_UPDATE itself, e.g.,

- HIWORD_UPDATE(PCIE_CLIENT_CONF_ENABLE,
- PCIE_CLIENT_CONF_ENABLE_MASK,
- PCIE_CLIENT_CONF_ENABLE_SHIFT) |

vs this:

+ PCIE_CLIENT_CONF_ENABLE |

For these constant cases, a HIWORD_UPDATE that generated the
associated macro names would help readability a bit, though we'd still
need the ugly #defines:

#define HIWORD_UPDATE_CONSTANT(x) (HIWORD_UPDATE(x, x ## _MASK, x ## _SHIFT)

#define PCIE_CLIENT_CONF_ENABLE 0x1
#define PCIE_CLIENT_CONF_ENABLE_SHIFT 0
#define PCIE_CLIENT_CONF_ENABLE_MASK 0x1

HIWORD_UPDATE_CONSTANT(PCIE_CLIENT_CONF_ENABLE) |