Hi Stephen,
On 4 August 2016 at 12:52, Stephen Warren swarren@wwwdotorg.org wrote:
On 08/03/2016 07:16 PM, Simon Glass wrote:
Hi Stephen,
On 1 August 2016 at 09:22, Stephen Warren swarren@wwwdotorg.org wrote:
On 07/31/2016 07:02 PM, Simon Glass wrote:
Hi Stephen,
On 27 July 2016 at 15:48, Stephen Warren swarren@wwwdotorg.org wrote:
From: Stephen Warren swarren@nvidia.com
Tegra186 supports the new standard clock, reset, and power domain APIs. Older Tegra SoCs still use custom APIs. Enhance the Tegra PCIe driver so that it can operate with either set of APIs.
On Tegra186, the BPMP handles all aspects of PCIe PHY (UPHY) programming. Consequently, this logic is disabled too.
Signed-off-by: Stephen Warren swarren@nvidia.com
This whole series builds on the other Tegra186 series that I just sent.
drivers/pci/Kconfig | 1 + drivers/pci/pci_tegra.c | 154 ++++++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 150 insertions(+), 5 deletions(-)
diff --git a/drivers/pci/Kconfig b/drivers/pci/Kconfig index 26aa2b0930a0..669e37bb5dc5 100644 --- a/drivers/pci/Kconfig +++ b/drivers/pci/Kconfig @@ -31,6 +31,7 @@ config PCI_SANDBOX config PCI_TEGRA bool "Tegra PCI support" depends on TEGRA
depends on (TEGRA186 && POWER_DOMAIN) || (!TEGRA186) help Enable support for the PCIe controller found on somegenerations of Tegra. Tegra20 has 2 root ports with a total of 4 lanes, Tegra30 has diff --git a/drivers/pci/pci_tegra.c b/drivers/pci/pci_tegra.c index 352cdef56ab4..a6785ad0bbee 100644 --- a/drivers/pci/pci_tegra.c +++ b/drivers/pci/pci_tegra.c @@ -13,22 +13,26 @@ #define pr_fmt(fmt) "tegra-pcie: " fmt
#include <common.h> +#include <clk.h> #include <dm.h> #include <errno.h> #include <fdtdec.h> #include <malloc.h> #include <pci.h> +#include <power-domain.h> +#include <reset.h>
#include <asm/io.h> #include <asm/gpio.h>
+#include <linux/list.h>
+#ifndef CONFIG_TEGRA186 #include <asm/arch/clock.h> #include <asm/arch/powergate.h> #include <asm/arch-tegra/xusb-padctl.h>
-#include <linux/list.h>
#include <dt-bindings/pinctrl/pinctrl-tegra-xusb.h> +#endif
DECLARE_GLOBAL_DATA_PTR;
@@ -103,6 +107,9 @@ DECLARE_GLOBAL_DATA_PTR; #define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_222 (0x1 << 20) #define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_X4_X1 (0x1 << 20) #define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_411 (0x2 << 20) +#define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_T186_401 (0x0 << 20) +#define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_T186_211 (0x1 << 20) +#define AFI_PCIE_CONFIG_SM2TMS0_XBAR_CONFIG_T186_111 (0x2 << 20)
#define AFI_FUSE 0x104 #define AFI_FUSE_PCIE_T0_GEN2_DIS (1 << 2) @@ -110,6 +117,7 @@ DECLARE_GLOBAL_DATA_PTR; #define AFI_PEX0_CTRL 0x110 #define AFI_PEX1_CTRL 0x118 #define AFI_PEX2_CTRL 0x128 +#define AFI_PEX2_CTRL_T186 0x19c #define AFI_PEX_CTRL_RST (1 << 0) #define AFI_PEX_CTRL_CLKREQ_EN (1 << 1) #define AFI_PEX_CTRL_REFCLK_EN (1 << 3) @@ -173,6 +181,7 @@ enum tegra_pci_id { TEGRA30_PCIE, TEGRA124_PCIE, TEGRA210_PCIE,
TEGRA186_PCIE,};
struct tegra_pcie_port { @@ -189,6 +198,7 @@ struct tegra_pcie_soc { unsigned int num_ports; unsigned long pads_pll_ctl; unsigned long tx_ref_sel;
unsigned long afi_pex2_ctrl; u32 pads_refclk_cfg0; u32 pads_refclk_cfg1; bool has_pex_clkreq_en;@@ -209,7 +219,17 @@ struct tegra_pcie { unsigned long xbar;
const struct tegra_pcie_soc *soc;
+#ifdef CONFIG_TEGRA186
struct clk clk_afi;struct clk clk_pex;struct reset_ctl reset_afi;struct reset_ctl reset_pex;struct reset_ctl reset_pcie_x;struct power_domain pwrdom;+#else
Eek. This is a compile-time driver configuration. Can you use the device tree compatible string to detect which to use?
No. The legacy APIs are simply not available (not compiled in) on the newer SoCs, so there must be a compile-time condition.
It is theoretically possible to convert all the old SoCs to the new APIs so that this conditional goes away in the future, but we don't have that work scheduled at present.
What did you do with Linux?
We lucked out and life was a bit simpler:
When we started adding Tegra support, there was already a standard clock API header, albeit with an entirely separate implementation for each SoC. When the common clock framework/core appeared, it maintained this same API so drivers weren't affected, except for a change in a #include filename.
As far as converting the clock driver from a standalone/custom implementation to being based on the common clock framework/core, life was much easier. At that time, we only supported 2 Tegra SoC generations and many fewer boards and drivers, and so significantly less testing was required for the converted implementation. Many SoC generations, board, and drivers were only written once the new clock implementation was in place, so never needed conversion.
Tegra started out with a completely custom reset API and implementation. When Tegra converted from its custom standalone clock implementation to the common clock framework/core, that same API was re-implemented using the new clock code (as you know, clocks and resets on Tegra are very closely tied), and so there was no effect on drivers.
Later, we converted to the new reset API/framework/core. At this point, we did support 4 SoC generations and many more boards. However, IIRC during this time, there was a huge focus on converting to standard APIs, and a lull in new SoC and board support, so there was time to spend on large-scale conversions; they didn't take time away from other work required to support new SoCs/boards. This conversion implemented the new standard reset for all SoCs first, converted all drivers to use the new API roughly at once, and then removed the old API.
So we got away without ifdefs.
In U-Boot, we now support many more SoCs, boards, and perhaps even drivers, so a "convert everything at once" approach is quite a lot more work than it was in the kernel, especially on the clock/reset driver implementation side. Equally, to be honest, my primary focus is currently on adding Tegra186 support at the moment. Given we needed a completely new clock/reset/... implementation anyway due to the existence of the BPMP, it was a no-brainer to create and/or use the new standard clock/reset/... APIs for Tegra186, rather than basing the new implementation on the old custom APIs. However, converting all the other SoCs/boards first is just going to delay any forward progress on Tegra186 and lead to a mega series of code conversions that I'd rather avoid doing all at once. Adding these ifdefs, allows incremental forward progress to be made pretty easily. They should be removed when older SoC generations are converted to the new APIs, so everything is unified again. Granted, I have no definite timescale for that at present.
OK, well I'd like to see some TODOs in the code around these #ifdefs. They are definitely not what we want, and should be marked as interim. Please see if you can find someone to convert things over. I can perhaps help too.
Regards, Simon