dm: Implement OF_PLATDATA_INST in driver model (part E)
Subsystems (4)
cover dm: Implement OF_PLATDATA_INST in driver model (part E)
This series builds on the recent dtoc implementation of build-time device instantiation.
It adds the required driver model support, which is basically a few more data structures.
With this, sandbox_spl and chromebook_coral both use the new feature.
For coral TPL there is a 1.5KB code-size reduction and a 1.75KB data-size increase:
text data bss dec hex 18836 3080 12 21928 55a8 original 17229 4896 12 22137 5679 with OF_PLATDATA_INST 17277 4896 12 22185 56a9 with OF_PLATDATA_RT
The extra data size is due to the build-time devices which are now included in the image instead of being built at runtime. Also the private data for each device is allocated in the data region at present, even through much of it is just zeroes.
The reduction in code size is due to not needing to bind devices at runtime, as a well as a simplified probe() function. Coral requires that data be copied out to RAM before being updated, so that adds a small amount to size (shown in the third line).
Quite a lot of future work is possible, including reducing the size of data structures. See [1] for more ideas. But this series implements the basic feature.
To try this out on your board, define CONFIG_SPL_OF_PLATDATA_INST and see what you get.
Note: SPL tests do not yet pass with this series. The driver_rt struct is not set up so device_get_by_driver_info_idx() does not work. This means that looking up phandles will fail. This will be addressed in a v2 series, along with documentation updates and a little more information on code-size impact.
This series is available at u-boot-dm/tin-working
[1] https://lists.denx.de/pipermail/u-boot/2020-July/418433.html
Simon Glass (30):
sandbox: Drop debug message in os_spl_to_uboot()
linker_lists: Allow use in data structures
dm: core: Add macros to access the new linker lists
dm: core: Allow dropping run-time binding of devices
dm: core: Adjust uclass setup with of-platdata
dm: core: Set up driver model for OF_PLATDATA_INST
dm: core: Skip adding uclasses with OF_PLATDATA_INST
dm: Add the new dtoc-generated files to the build
dm: core: Include dt-decl.h automatically
dm: test: Avoid destroying uclasses with of-platdata-inst
clk: sandbox: Move priv/plat data to a header file
clk: fixed-rate: Export driver parts for OF_PLATDATA_INST
clk: sandbox: Create a special fixed-rate driver
sandbox: Create a new sandbox_noinst build
test: Run sandbox_spl tests on sandbox_noinst azure/gitlab: Add tests for sandbox_noinst
dm: core: Add an option to support SPL in read-only memory
dm: core: Create a struct for device runtime info
dm: core: Move flags to device-runtime info
dm: core: Allow storing priv/plat data separately
sandbox: Define a region for device priv/plat data
dm: core: Use separate priv/plat data region
x86: Define a region for device priv/plat data
x86: apl: Fix the header order in pmc
x86: apl: Tell of-platdata about a required header file
x86: itss: Tidy up bind() for of-platdata-inst
x86: Support a fake PCI device with of-platdata-inst
x86: Don't include reset driver in SPL
x86: coral: Drop ACPI properties from of-platdata
x86: apl: Use read-only SPL and new of-platdata
.azure-pipelines.yml | 3 + .gitlab-ci.yml | 10 +- arch/sandbox/cpu/os.c | 1 - arch/sandbox/cpu/u-boot-spl.lds | 8 + arch/sandbox/dts/sandbox.dtsi | 2 +- arch/sandbox/include/asm/clk.h | 24 +++ arch/x86/cpu/apollolake/Kconfig | 2 + arch/x86/cpu/apollolake/pmc.c | 2 +- arch/x86/cpu/apollolake/punit.c | 1 + arch/x86/cpu/intel_common/itss.c | 5 +- arch/x86/cpu/u-boot-spl.lds | 8 + arch/x86/dts/reset.dtsi | 2 +- arch/x86/lib/tpl.c | 1 + board/sandbox/MAINTAINERS | 7 + common/spl/Kconfig | 24 +++ configs/chromebook_coral_defconfig | 1 + configs/sandbox_noinst_defconfig | 229 +++++++++++++++++++++++++++++ configs/sandbox_spl_defconfig | 1 + drivers/clk/clk_fixed_rate.c | 14 +- drivers/clk/clk_sandbox.c | 40 ++++- drivers/clk/clk_sandbox_test.c | 6 - drivers/core/device.c | 83 ++++++++--- drivers/core/root.c | 83 +++++++++-- drivers/core/uclass.c | 5 +- dts/Kconfig | 38 +++++ include/asm-generic/global_data.h | 22 +++ include/asm-generic/sections.h | 3 + include/dm/device-internal.h | 26 ++++ include/dm/device.h | 38 ++++- include/dm/root.h | 3 + include/dm/uclass-internal.h | 23 +++ include/dm/uclass.h | 20 +++ include/dm/util.h | 9 ++ include/dt-structs.h | 2 + include/linker_lists.h | 28 ++++ include/linux/clk-provider.h | 5 + scripts/Makefile.spl | 3 +- test/dm/of_platdata.c | 2 +- test/dm/test-main.c | 3 +- test/run | 4 + 40 files changed, 724 insertions(+), 67 deletions(-) create mode 100644 configs/sandbox_noinst_defconfig
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 1/30 sandbox: Drop debug message in os_spl_to_uboot()
This is not needed in normal operation. Drop it.
---
arch/sandbox/cpu/os.c | 1 - 1 file changed, 1 deletion(-)
arch/sandbox/cpu/os.c +0 -1
@@ -792,7 +792,6 @@ int os_spl_to_uboot(const char *fname) {struct sandbox_state *state =state_get_current (); -printf ("%s\n" , __func__);/* U-Boot will delete ram buffer after read: "--rm_memory"*/ state->ram_buf_rm = true;return os_jump_to_file (fname);
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 2/30 linker_lists: Allow use in data structures
In reply to Simon GlassAt present linker lists are designed for use in code. They make use of statements within expressions ({...}), for example.
It is possible to generate a reference to a linker_list entry that can be used in data structures, where such features are not permitted. It requires that the reference first be declared as extern. In other words the existing macro needs to be split into two parts.
Add new macros to support this.
---
include/linker_lists.h | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+)
include/linker_lists.h +28 -0
@@ -211,6 +211,34 @@ _ll_result; \ }) +/** +* ll_entry_decl() - Declare a linker-generated array entry reference as extern +* +* This declares a reference to a list entry so that it can be used later, +* without needing the code in ll_entry_get(). +* +* To use this, put ll_entry_decl() somewhere in your file, then use +* ll_entry_ref() later on, to reference the entry. +* +* @_type: Data type of the entry +* @_name: Name of the entry +* @_list: Name of the list in which this entry is placed +*/ +#define ll_entry_decl (_type, _name, _list) \ + extern _type _u_boot_list_2_##_list##_2_##_name + +/** +* ll_entry_ref() - Get a reference to a linker-generated array entry +* +* Once ll_entry_decl() has been used to declare the reference, this macro +* allows the entry to be accessed. +* +* This is like ll_entry_get(), but without the extra code, so it is suitable +* for putting into data structures. +*/ +#define ll_entry_ref (_type, _name, _list) \ + ((_type *)&_u_boot_list_2_##_list##_2_##_name) +/** * ll_start() - Point to first entry of first linker-generated array * @_type: Data type of the entry
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 3/30 dm: core: Add macros to access the new linker lists
In reply to Simon GlassAdd macros which work with instantiated devices and uclasses, as created at build time by dtoc. Include variants that can be used in data structures.
---
include/dm/device-internal.h | 26 ++++++++++++++++++++++++++ include/dm/device.h | 11 +++++++++++ include/dm/uclass-internal.h | 23 +++++++++++++++++++++++ include/dm/uclass.h | 20 ++++++++++++++++++++ 4 files changed, 80 insertions(+)
include/dm/device-internal.h +26 -0
@@ -10,11 +10,37 @@#ifndef _DM_DEVICE_INTERNAL_H#define _DM_DEVICE_INTERNAL_H +#include <linker_lists.h> #include <dm/ofnode.h> struct device_node ;struct udevice ; +/* +* These are only allowed these in code generated by dtoc, because the ordering +* is important and if other instances creep in then they may mess up the +* ordering expected by dtoc. +*/ + +/* Declare a bound device ready for run-time use */ +#define DM_DEVICE_INST (__name) \ + ll_entry_declare(struct udevice, __name, udevice) + +/* Declare a bound device as an extern, so it can be referenced at build time */ +#define DM_DEVICE_DECL (__name) \ + ll_entry_decl(struct udevice, __name, udevice) + +/* +* Get a pointer to a given device, for use in data structures. This requires +* that the symbol be declared with DM_DRIVER_DECL() first +*/ +#define DM_DEVICE_REF (__name) \ + ll_entry_ref(struct udevice, __name, udevice) + +/* Get a pointer to a given device */ +#define DM_DEVICE_GET (__name) \ + ll_entry_get(struct udevice, __name, udevice) +/** * device_bind() - Create a device and bind it to a driver *
include/dm/device.h +11 -0
@@ -338,6 +338,17 @@ struct driver {#define DM_DRIVER_GET (__name) \ ll_entry_get(struct driver, __name, driver) +/* Declare a driver as an extern, so it can be referenced at build time */ +#define DM_DRIVER_DECL (__name) \ + ll_entry_decl(struct driver, __name, driver) + +/* +* Get a pointer to a given driver, for use in data structures. This requires +* that the symbol be declared with DM_DRIVER_DECL() first +*/ +#define DM_DRIVER_REF (__name) \ + ll_entry_ref(struct driver, __name, driver) +/** * Declare a macro to state a alias for a driver name. This macro will * produce no code but its information will be parsed by tools like
include/dm/uclass-internal.h +23 -0
@@ -11,6 +11,29 @@#include <dm/ofnode.h> +/* +* These are only allowed these in code generated by dtoc, because the ordering +* is important and if other instances creep in then they may mess up the +* ordering expected by dtoc. +*/ + +/* +* Declare a uclass ready for run-time use. This adds an actual struct uclass +* to a list which is found by driver model on start-up. +*/ +#define DM_UCLASS_INST (__name) \ + ll_entry_declare(struct uclass, __name, uclass) + +#define DM_UCLASS_DECL (__name) \ + ll_entry_decl(struct uclass, __name, uclass) + +/* +* Declare a uclass as an extern, so it can be referenced at build time. This +* is an extern for DM_UCLASS_INST(). +*/ +#define DM_UCLASS_REF (__name) \ + ll_entry_ref(struct uclass, __name, uclass) +/** * uclass_set_priv() - Set the private data for a uclass *
include/dm/uclass.h +20 -0
@@ -114,6 +114,26 @@ struct uclass_driver {#define UCLASS_DRIVER (__name) \ ll_entry_declare(struct uclass_driver, __name, uclass_driver) +/* +* These two macros are related to of-platdata, and normally only used in +* code generated by dtoc +*/ + +/* +* Declare a uclass driver as an extern, so it can be referenced at build time +* This is the extern equivalent of UCLASS_DRIVER(). You need to place this at +* the top level before you use DM_UCLASS_DRIVER_REF() in a file. +*/ +#define DM_UCLASS_DRIVER_DECL (__name) \ + ll_entry_decl(struct uclass_driver, __name, uclass_driver) + +/* +* Get a pointer to a given uclass driver, for use in data structures. This +* requires that the symbol be declared with DM_UCLASS_DRIVER_DECL() first +*/ +#define DM_UCLASS_DRIVER_REF (__name) \ + ll_entry_ref(struct uclass_driver, __name, uclass_driver) +/** * uclass_get_priv() - Get the private data for a uclass *
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 4/30 dm: core: Allow dropping run-time binding of devices
In reply to Simon GlassWith OF_PLATDATA_INST devices are bound at build time. we should not allow binding of devices at runtime. Add an option to control this.
Update the driver model core so that it does not bind devices. Update device_bind() to return an error if called.
---
drivers/core/device.c | 38 +++++++++++++++++++++----------------- dts/Kconfig | 16 ++++++++++++++++ 2 files changed, 37 insertions(+), 17 deletions(-)
drivers/core/device.c +21 -17
@@ -44,6 +44,9 @@ static int device_bind_common(struct udevice *parent, const struct driver *drv,bool auto_seq = true;void *ptr; +if (CONFIG_IS_ENABLED (OF_PLATDATA_NO_BIND )) +return -ENOSYS ; +if (devp) *devp =NULL ;if (!name)
@@ -382,26 +385,27 @@ int device_of_to_plat(struct udevice *dev)if (dev_get_flags (dev) &DM_FLAG_PLATDATA_VALID )return 0 ; -/* Ensure all parents have ofdata */ -if (dev->parent) { - ret =device_of_to_plat (dev->parent); +if (!CONFIG_IS_ENABLED (OF_PLATDATA_NO_BIND )) { +/* Ensure all parents have ofdata */ +if (dev->parent) { + ret =device_of_to_plat (dev->parent); +if (ret) + goto fail; + +/* +* The device might have already been probed during +* the call to device_probe() on its parent device +* (e.g. PCI bridge devices). Test the flags again +* so that we don't mess up the device. +*/ +if (dev_get_flags (dev) &DM_FLAG_PLATDATA_VALID ) +return 0 ; + } + + ret =device_alloc_priv (dev);if (ret) goto fail; - -/* -* The device might have already been probed during -* the call to device_probe() on its parent device -* (e.g. PCI bridge devices). Test the flags again -* so that we don't mess up the device. -*/ -if (dev_get_flags (dev) &DM_FLAG_PLATDATA_VALID ) -return 0 ; } - - ret =device_alloc_priv (dev); -if (ret) - goto fail; - drv = dev->driver;assert (drv);
dts/Kconfig +16 -0
@@ -371,6 +371,14 @@ config SPL_OF_PLATDATA_INST Declare devices as udevice instances so that they do not need to be bound when U-Boot starts. This can save time and code space. +config SPL_OF_PLATDATA_NO_BIND + bool "Don't allow run-time binding of devices" + depends on SPL_OF_PLATDATA_INST + default y + help + This removes the ability to bind devices at run time, thus saving + some code space in U-Boot. + endif config TPL_OF_PLATDATA
@@ -411,6 +419,14 @@ config TPL_OF_PLATDATA_INST Declare devices as udevice instances so that they do not need to be bound when U-Boot starts. This can save time and code space. +config TPL_OF_PLATDATA_NO_BIND + bool "Don't allow run-time binding of devices" + depends on TPL_OF_PLATDATA_INST + default y + help + This removes the ability to bind devices at run time, thus saving + some code space in U-Boot. + endif endmenu
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 5/30 dm: core: Adjust uclass setup with of-platdata
In reply to Simon GlassWhen OF_PLATDATA_INST is enabled we don't need to create the uclass list. Instead we just need to point to the existing list. Update the code accordingly.
---
drivers/core/root.c | 8 ++++++-- include/dm/root.h | 3 +++ 2 files changed, 9 insertions(+), 2 deletions(-)
drivers/core/root.c +6 -2
@@ -136,8 +136,12 @@ int dm_init(bool of_live)dm_warn ("Virtual root driver already exists!\n" );return -EINVAL ; } - gd->uclass_root = &DM_UCLASS_ROOT_S_NON_CONST ; -INIT_LIST_HEAD (DM_UCLASS_ROOT_NON_CONST ); +if (CONFIG_IS_ENABLED (OF_PLATDATA_INST )) { + gd->uclass_root = &uclass_head; + }else { + gd->uclass_root = &DM_UCLASS_ROOT_S_NON_CONST ; +INIT_LIST_HEAD (DM_UCLASS_ROOT_NON_CONST ); + }if (IS_ENABLED (CONFIG_NEEDS_MANUAL_RELOC )) {fix_drivers ();
include/dm/root.h +3 -0
@@ -11,6 +11,9 @@struct udevice ; +/* Head of the uclass list if CONFIG_OF_PLATDATA_INST is enabled */ +extern struct list_head uclass_head; +/** * dm_root() - Return pointer to the top of the driver tree *
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 6/30 dm: core: Set up driver model for OF_PLATDATA_INST
In reply to Simon GlassWith this we don't need to scan and bind drivers, not even the root device. We just need to locate the root device that was set up at build time, then set our root in global_data to point to it.
Update the code to handle this case.
---
drivers/core/root.c | 42 ++++++++++++++++++++++++++++++------------ 1 file changed, 30 insertions(+), 12 deletions(-)
drivers/core/root.c +30 -12
@@ -128,6 +128,13 @@ void fix_devices(void) } } +static int dm_setup_inst (void ) +{ +DM_ROOT_NON_CONST =DM_DEVICE_GET (root); + +return 0 ; +} +int dm_init (bool of_live) {int ret;
@@ -149,14 +156,23 @@ int dm_init(bool of_live)fix_devices (); } - ret =device_bind_by_name (NULL , false, &root_info, &DM_ROOT_NON_CONST ); -if (ret) -return ret; -if (CONFIG_IS_ENABLED (OF_CONTROL )) -dev_set_ofnode (DM_ROOT_NON_CONST ,ofnode_root ()); - ret =device_probe (DM_ROOT_NON_CONST ); -if (ret) -return ret; +if (CONFIG_IS_ENABLED (OF_PLATDATA_INST )) { + ret =dm_setup_inst (); +if (ret) { +log_debug ("dm_setup_inst() failed: %d\n" , ret); +return ret; + } + }else { + ret =device_bind_by_name (NULL , false, &root_info, + &DM_ROOT_NON_CONST ); +if (ret) +return ret; +if (CONFIG_IS_ENABLED (OF_CONTROL )) +dev_set_ofnode (DM_ROOT_NON_CONST ,ofnode_root ()); + ret =device_probe (DM_ROOT_NON_CONST ); +if (ret) +return ret; + }return 0 ; }
@@ -345,10 +361,12 @@ int dm_init_and_scan(bool pre_reloc_only)debug ("dm_init() failed: %d\n" , ret);return ret; } - ret =dm_scan (pre_reloc_only); -if (ret) { -log_debug ("dm_scan() failed: %d\n" , ret); -return ret; +if (!CONFIG_IS_ENABLED (OF_PLATDATA_INST )) { + ret =dm_scan (pre_reloc_only); +if (ret) { +log_debug ("dm_scan() failed: %d\n" , ret); +return ret; + } }return 0 ;
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 7/30 dm: core: Skip adding uclasses with OF_PLATDATA_INST
In reply to Simon GlassThere is no need to ever add new uclasses since these are set up at build time. Update the code to return an error if this is attempted.
---
drivers/core/uclass.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/core/uclass.c +4 -1
@@ -147,8 +147,11 @@ int uclass_get(enum uclass_id id, struct uclass **ucp) *ucp =NULL ; uc =uclass_find (id); -if (!uc) +if (!uc) { +if (CONFIG_IS_ENABLED (OF_PLATDATA_INST )) +return -ENOENT ;return uclass_add (id, ucp); + } *ucp = uc;return 0 ;
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 8/30 dm: Add the new dtoc-generated files to the build
In reply to Simon GlassNow that dtoc generates some new C files, add these to the build so that the instantiated devices and uclasses can be used.
---
scripts/Makefile.spl | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
scripts/Makefile.spl +2 -1
@@ -120,7 +120,8 @@ endif u-boot-spl-init := $(head-y) u-boot-spl-main := $(libs-y) ifdef CONFIG_$(SPL_TPL_)OF_PLATDATA -u-boot-spl-platdata := $(obj)/dts/dt-plat.o +u-boot-spl-platdata := $(obj)/dts/dt-plat.o $(obj)/dts/dt-uclass.o \ + $(obj)/dts/dt-device.o u-boot-spl-platdata_c := $(patsubst %.o,%.c,$(u-boot-spl-platdata)) endif
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 9/30 dm: core: Include dt-decl.h automatically
In reply to Simon GlassWhen dt-structs.h is used, include the dt-decl.h header as well, so that these declarations are available.
---
include/dt-structs.h | 2 ++ 1 file changed, 2 insertions(+)
include/dt-structs.h +2 -0
@@ -24,7 +24,9 @@ struct phandle_2_arg {uint idx;int arg[2 ]; }; +#include <generated/dt-structs-gen.h> +#include <generated/dt-decl.h> #endif #endif
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 10/30 dm: test: Avoid destroying uclasses with of-platdata-inst
In reply to Simon GlassIt is not possible to destroy the uclasses when they are created at build time. Skip this step so that SPL test can complete successfully.
---
test/dm/test-main.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
test/dm/test-main.c +2 -1
@@ -107,7 +107,8 @@ static int dm_do_test(struct unit_test_state *uts, struct unit_test *test, gd->flags &= ~(GD_FLG_SILENT |GD_FLG_RECORD );state_set_skip_delays (false); -ut_assertok (dm_test_destroy (uts)); +if (!CONFIG_IS_ENABLED (OF_PLATDATA_INST )) +ut_assertok (dm_test_destroy (uts));return 0 ; }
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 11/30 clk: sandbox: Move priv/plat data to a header file
In reply to Simon GlassAt present the structs used by this driver are not accessible outside it, so cannot be used with OF_PLATDATA_INST. Move them to a header file to fix this.
---
arch/sandbox/include/asm/clk.h | 16 ++++++++++++++++ drivers/clk/clk_sandbox.c | 7 ------- drivers/clk/clk_sandbox_test.c | 6 ------ 3 files changed, 16 insertions(+), 13 deletions(-)
arch/sandbox/include/asm/clk.h +16 -0
@@ -7,6 +7,9 @@#define __SANDBOX_CLK_H#include <common.h> +#include <clk.h> +#include <dt-structs.h> +#include <linux/clk-provider.h> struct udevice ;
@@ -45,6 +48,19 @@ enum sandbox_clk_test_id {#define SANDBOX_CLK_TEST_NON_DEVM_COUNT SANDBOX_CLK_TEST_ID_DEVM1 +struct sandbox_clk_priv { +bool probed; +ulong rate[SANDBOX_CLK_ID_COUNT ]; +bool enabled[SANDBOX_CLK_ID_COUNT ]; +bool requested[SANDBOX_CLK_ID_COUNT ]; +}; + +struct sandbox_clk_test { +struct clk clks[SANDBOX_CLK_TEST_NON_DEVM_COUNT ]; +struct clk *clkps[SANDBOX_CLK_TEST_ID_COUNT ]; +struct clk_bulk bulk; +}; +/** * sandbox_clk_query_rate - Query the current rate of a sandbox clock. *
drivers/clk/clk_sandbox.c +0 -7
@@ -10,13 +10,6 @@#include <malloc.h> #include <asm/clk.h> -struct sandbox_clk_priv { -bool probed; -ulong rate[SANDBOX_CLK_ID_COUNT ]; -bool enabled[SANDBOX_CLK_ID_COUNT ]; -bool requested[SANDBOX_CLK_ID_COUNT ]; -}; -static ulong sandbox_clk_get_rate (struct clk *clk) {struct sandbox_clk_priv *priv =dev_get_priv (clk->dev);
drivers/clk/clk_sandbox_test.c +0 -6
@@ -11,12 +11,6 @@#include <dm/device_compat.h> #include <linux/err.h> -struct sandbox_clk_test { -struct clk clks[SANDBOX_CLK_TEST_NON_DEVM_COUNT ]; -struct clk *clkps[SANDBOX_CLK_TEST_ID_COUNT ]; -struct clk_bulk bulk; -}; -static const char *const sandbox_clk_test_names[] = { [SANDBOX_CLK_TEST_ID_FIXED ] ="fixed" , [SANDBOX_CLK_TEST_ID_SPI ] ="spi" ,
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 12/30 clk: fixed-rate: Export driver parts for OF_PLATDATA_INST
In reply to Simon GlassWe need to allow SoCs to create their own drivers for this so that they can use their own of-platdata structs. To minimise code duplication, export the driver operations and the ofdata_to_plat() setup function.
---
drivers/clk/clk_fixed_rate.c | 14 ++++++++++---- include/linux/clk-provider.h | 5 +++++ 2 files changed, 15 insertions(+), 4 deletions(-)
drivers/clk/clk_fixed_rate.c +10 -4
@@ -25,18 +25,24 @@ const struct clk_ops clk_fixed_rate_ops = { .enable = dummy_enable, }; -static int clk_fixed_rate_of_to_plat (struct udevice *dev) +void clk_fixed_rate_ofdata_to_plat_ (struct udevice *dev, +struct clk_fixed_rate *plat) { -struct clk *clk = &to_clk_fixed_rate (dev)->clk; +struct clk *clk = &plat->clk;#if !CONFIG_IS_ENABLED (OF_PLATDATA ) -to_clk_fixed_rate (dev)->fixed_rate = -dev_read_u32_default (dev,"clock-frequency" ,0 ); + plat->fixed_rate =dev_read_u32_default (dev,"clock-frequency" ,0 );#endif /* Make fixed rate clock accessible from higher level struct clk */ /* FIXME: This is not allowed */ dev_set_uclass_priv (dev, clk); + clk->dev = dev; clk->enable_count =0 ; +} + +static int clk_fixed_rate_of_to_plat (struct udevice *dev) +{ +clk_fixed_rate_ofdata_to_plat_ (dev,to_clk_fixed_rate (dev));return 0 ; }
include/linux/clk-provider.h +5 -0
@@ -136,6 +136,8 @@ struct clk_fixed_factor {unsigned int div; }; +extern const struct clk_ops clk_fixed_rate_ops; +#define to_clk_fixed_factor (_clk) container_of(_clk, struct clk_fixed_factor,\ clk)
@@ -146,6 +148,9 @@ struct clk_fixed_rate {#define to_clk_fixed_rate (dev) ((struct clk_fixed_rate *)dev_get_plat(dev)) +void clk_fixed_rate_ofdata_to_plat_ (struct udevice *dev, +struct clk_fixed_rate *plat); +struct clk_composite {struct clk clk;struct clk_ops ops;
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 13/30 clk: sandbox: Create a special fixed-rate driver
In reply to Simon GlassCreate a version of this driver for sandbox so that it can use the of-platdata struct.
---
arch/sandbox/dts/sandbox.dtsi | 2 +- arch/sandbox/include/asm/clk.h | 8 ++++++++ drivers/clk/clk_sandbox.c | 33 +++++++++++++++++++++++++++++++++ test/dm/of_platdata.c | 2 +- 4 files changed, 43 insertions(+), 2 deletions(-)
arch/sandbox/dts/sandbox.dtsi +1 -1
@@ -31,7 +31,7 @@ clk_fixed: clk-fixed { u-boot,dm-pre-reloc; - compatible = "fixed-clock"; + compatible = "sandbox,fixed-clock"; #clock-cells = <0>; clock-frequency = <1234>; };
arch/sandbox/include/asm/clk.h +8 -0
@@ -61,6 +61,14 @@ struct sandbox_clk_test {struct clk_bulk bulk; }; +/* Platform data for the sandbox fixed-rate clock driver */ +struct sandbox_clk_fixed_rate_plat { +#if CONFIG_IS_ENABLED (OF_PLATDATA ) +struct dtd_sandbox_fixed_clock dtplat; +#endif +struct clk_fixed_rate fixed; +}; +/** * sandbox_clk_query_rate - Query the current rate of a sandbox clock. *
drivers/clk/clk_sandbox.c +33 -0
@@ -9,6 +9,7 @@#include <errno.h> #include <malloc.h> #include <asm/clk.h> +#include <linux/clk-provider.h> static ulong sandbox_clk_get_rate (struct clk *clk) {
@@ -154,3 +155,35 @@ int sandbox_clk_query_requested(struct udevice *dev, int id)return -EINVAL ;return priv->requested[id]; } + +int clk_fixed_rate_of_to_plat (struct udevice *dev) +{ +struct clk_fixed_rate *cplat; + +#if CONFIG_IS_ENABLED (OF_PLATDATA ) +struct sandbox_clk_fixed_rate_plat *plat =dev_get_plat (dev); + + cplat = &plat->fixed; + cplat->fixed_rate = plat->dtplat.clock_frequency; +#else + cplat =to_clk_fixed_rate (dev); +#endif +clk_fixed_rate_ofdata_to_plat_ (dev, cplat); + +return 0 ; +} + +static const struct udevice_id sandbox_clk_fixed_rate_match[] = { + { .compatible ="sandbox,fixed-clock" }, + {/* sentinel */ } +}; + +U_BOOT_DRIVER (sandbox_fixed_clock) = { + .name ="sandbox_fixed_clock" , + .id =UCLASS_CLK , + .of_match = sandbox_clk_fixed_rate_match, + .of_to_plat = clk_fixed_rate_of_to_plat, + .plat_auto =sizeof (struct sandbox_clk_fixed_rate_plat ), + .ops = &clk_fixed_rate_ops, + .flags =DM_FLAG_PRE_RELOC , +};
test/dm/of_platdata.c +1 -1
@@ -184,7 +184,7 @@ static int dm_test_of_plat_phandle(struct unit_test_state *uts) plat =dev_get_plat (dev);ut_assertok (device_get_by_driver_info_idx (plat->clocks[0 ].idx, &clk)); -ut_asserteq_str ("fixed_clock" , clk->name); +ut_asserteq_str ("sandbox_fixed_clock" , clk->name);ut_assertok (device_get_by_driver_info_idx (plat->clocks[1 ].idx, &clk));ut_asserteq_str ("sandbox_clk" , clk->name);
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 14/30 sandbox: Create a new sandbox_noinst build
In reply to Simon GlassMove sandbox_spl over to use OF_PLATDATA_INST. Create a new board to test the case when this is not enabled, since we will be keeping that code around for several months and want to avoid regressions.
---
board/sandbox/MAINTAINERS | 7 + configs/sandbox_noinst_defconfig | 229 +++++++++++++++++++++++++++++++ configs/sandbox_spl_defconfig | 1 + 3 files changed, 237 insertions(+) create mode 100644 configs/sandbox_noinst_defconfig
board/sandbox/MAINTAINERS +7 -0
@@ -20,6 +20,13 @@ F: board/sandbox/ F: include/configs/sandbox_spl.h F: configs/sandbox_spl_defconfig +SANDBOX NOINST BOARD +M: Simon Glass <[email protected]> +S: Maintained +F: board/sandbox/ +F: include/configs/sandbox_spl.h +F: configs/sandbox_noinst_defconfig + SANDBOX FLAT TREE BOARD M: Simon Glass <[email protected]> S: Maintained
configs/sandbox_noinst_defconfig new file +229 -0
229 lines not on this page · raw message
configs/sandbox_spl_defconfig +1 -0
@@ -87,6 +87,7 @@ CONFIG_OF_CONTROL=y CONFIG_SPL_OF_CONTROL=y CONFIG_OF_HOSTFILE=y CONFIG_SPL_OF_PLATDATA=y +CONFIG_SPL_OF_PLATDATA_INST=y CONFIG_ENV_IS_NOWHERE=y CONFIG_ENV_IS_IN_EXT4=y CONFIG_ENV_EXT4_INTERFACE="host"
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 15/30 test: Run sandbox_spl tests on sandbox_noinst
In reply to Simon GlassRun the tests on this build too, to prevent regressions.
---
test/run | 4 ++++ 1 file changed, 4 insertions(+)
test/run +4 -0
@@ -30,6 +30,10 @@ fi run_test "sandbox_spl" ./test/py/test.py --bd sandbox_spl --build \ -k 'test_ofplatdata or test_handoff or test_spl' +# Run the sane tests with sandbox_noinst (i.e. without OF_PLATDATA_INST) +run_test "sandbox_spl" ./test/py/test.py --bd sandbox_noinst --build \ + -k 'test_ofplatdata or test_handoff or test_spl' + if [ -z "$tools_only" ]; then # Run tests for the flat-device-tree version of sandbox. This is a special # build which does not enable CONFIG_OF_LIVE for the live device tree, so we can
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 16/30 azure/gitlab: Add tests for sandbox_noinst
In reply to Simon GlassAdd this new board to the test plans. Travis-CI is left out, since it is being removed soon due to lack of capacity.
---
.azure-pipelines.yml | 3 +++ .gitlab-ci.yml | 10 +++++++++- 2 files changed, 12 insertions(+), 1 deletion(-)
.azure-pipelines.yml +3 -0
@@ -183,6 +183,9 @@ jobs: sandbox_spl: TEST_PY_BD: "sandbox_spl" TEST_PY_TEST_SPEC: "test_ofplatdata or test_handoff or test_spl" + sandbox_noinst: + TEST_PY_BD: "sandbox_noinst" + TEST_PY_TEST_SPEC: "test_ofplatdata or test_handoff or test_spl" sandbox_flattree: TEST_PY_BD: "sandbox_flattree" evb_ast2500:
.gitlab-ci.yml +9 -1
@@ -165,7 +165,8 @@ Run binman, buildman, dtoc, Kconfig and patman testsuites: export UBOOT_TRAVIS_BUILD_DIR=/tmp/sandbox_spl; export PYTHONPATH="${UBOOT_TRAVIS_BUILD_DIR}/scripts/dtc/pylibfdt"; export PATH="${UBOOT_TRAVIS_BUILD_DIR}/scripts/dtc:${PATH}"; - ./tools/buildman/buildman -o ${UBOOT_TRAVIS_BUILD_DIR} -w sandbox_spl; + ./tools/buildman/buildman -o ${UBOOT_TRAVIS_BUILD_DIR} -w + --board sandbox_spl; ./tools/binman/binman --toolpath ${UBOOT_TRAVIS_BUILD_DIR}/tools test; ./tools/buildman/buildman -t; ./tools/dtoc/dtoc -t;
@@ -201,6 +202,13 @@ sandbox_spl test.py: TEST_PY_TEST_SPEC: "test_ofplatdata or test_handoff or test_spl" <<: *buildman_and_testpy_dfn +sandbox_noinst_test.py: + tags: [ 'all' ] + variables: + TEST_PY_BD: "sandbox_noinst" + TEST_PY_TEST_SPEC: "test_ofplatdata or test_handoff or test_spl" + <<: *buildman_and_testpy_dfn + evb-ast2500 test.py: tags: [ 'all' ] variables:
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 17/30 dm: core: Add an option to support SPL in read-only memory
In reply to Simon GlassSome systems (e.g. x86 APL) run SPL from read-only memory. The device instances created by dtoc are therefore not writeable. To make things work we would need to copy the devices to read/write memory.
To avoid this, add an option to use a separate runtime struct for devices, just as is done for drivers. This can be used to hold information that changes at runtime, avoiding the need for a copy.
Also add a Kconfig option for read-only SPL, which selects this feature.
---
common/spl/Kconfig | 24 ++++++++++++++++++++++++ dts/Kconfig | 22 ++++++++++++++++++++++ 2 files changed, 46 insertions(+)
common/spl/Kconfig +24 -0
@@ -276,6 +276,19 @@ config SPL_SEPARATE_BSS location is used. Normally we put the device tree at the end of BSS but with this option enabled, it goes at _image_binary_end. +config SPL_READ_ONLY + bool + depends on SPL_OF_PLATDATA + # Bind cannot be supported because the udevice structs are in read-only + # memory so we cannot update the linked lists. + select SPL_OF_PLATDATA_NO_BIND + select SPL_OF_PLATDATA_RT + help + Some platforms (e.g. x86 Apollo Lake) load SPL into a read-only + section of memory. This means that of-platdata must make a copy (in + writeable memory) of anything it wants to modify, such as + device-private data. + config SPL_BANNER_PRINT bool "Enable output of the SPL banner 'U-Boot SPL ...'" default y
@@ -1432,6 +1445,17 @@ config TPL_STACK The address of the initial stack-pointer for the TPL stage. Usually this will be the (aligned) top-of-stack. +config TPL_READ_ONLY + bool + depends on TPL_OF_PLATDATA + select TPL_OF_PLATDATA_NO_BIND + select TPL_OF_PLATDATA_RT + help + Some platforms (e.g. x86 Apollo Lake) load SPL into a read-only + section of memory. This means that of-platdata must make a copy (in + writeable memory) of anything it wants to modify, such as + device-private data. + config TPL_BOOTROM_SUPPORT bool "Support returning to the BOOTROM (from TPL)" help
dts/Kconfig +22 -0
@@ -379,6 +379,17 @@ config SPL_OF_PLATDATA_NO_BIND This removes the ability to bind devices at run time, thus saving some code space in U-Boot. +config SPL_OF_PLATDATA_RT + bool "Use a separate struct for device runtime data" + depends on SPL_OF_PLATDATA_INST + default y + help + For systems running SPL from read-only memory it is convenient to + separate out the runtime information, so that the devices don't need + to be copied before being used. This moves the read-write parts of + struct udevice (at present just the flags) into a separate struct, + which is allocated at runtime. + endif config TPL_OF_PLATDATA
@@ -427,6 +438,17 @@ config TPL_OF_PLATDATA_NO_BIND This removes the ability to bind devices at run time, thus saving some code space in U-Boot. +config TPL_OF_PLATDATA_RT + bool "Use a separate struct for device runtime data" + depends on TPL_OF_PLATDATA_INST + default y + help + For systems running TPL from read-only memory it is convenient to + separate out the runtime information, so that the devices don't need + to be copied before being used. This moves the read-write parts of + struct udevice (at present just the flags) into a separate struct, + which is allocated at runtime. + endif endmenu
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 18/30 dm: core: Create a struct for device runtime info
In reply to Simon GlassAt present when driver model needs to change a device it simply updates the struct udevice structure. But with of-platdata-inst most of the fields are not modified at runtime. In fact, typically only the flags need to change.
For systems running SPL from read-only memory it is convenient to separate out the runtime information, so that the devices don't need to be copied before being used.
Create a new udevice_rt table, similar to the existing driver_rt. For now it just holds the flags, although they are not used in this patch.
---
drivers/core/root.c | 12 ++++++++++++ include/asm-generic/global_data.h | 12 ++++++++++++ include/dm/device.h | 17 ++++++++++++++++- 3 files changed, 40 insertions(+), 1 deletion(-)
drivers/core/root.c +12 -0
@@ -132,6 +132,18 @@ static int dm_setup_inst(void) {DM_ROOT_NON_CONST =DM_DEVICE_GET (root); +if (CONFIG_IS_ENABLED (OF_PLATDATA_RT )) { +struct udevice_rt *urt; +int n_ents; + +/* Allocate the udevice_rt table */ + n_ents = ll_entry_count(struct udevice, udevice); + urt =calloc (n_ents,sizeof (struct udevice_rt )); +if (!urt) +return log_msg_ret ("urt" , -ENOMEM ); +gd_set_dm_udevice_rt (urt); + } +return 0 ; }
include/asm-generic/global_data.h +12 -0
@@ -205,6 +205,10 @@ struct global_data {/** Dynamic info about the driver */ struct driver_rt *dm_driver_rt;# endif +#if CONFIG_IS_ENABLED (OF_PLATDATA_RT ) +/** @dm_udevice_rt: Dynamic info about the udevice */ +struct udevice_rt *dm_udevice_rt; +# endif #endif #ifdef CONFIG_TIMER /**
@@ -465,6 +469,14 @@ struct global_data {#define gd_dm_driver_rt () NULL#endif +#if CONFIG_IS_ENABLED (OF_PLATDATA_RT ) +#define gd_set_dm_udevice_rt (dyn) gd->dm_udevice_rt = dyn +#define gd_dm_udevice_rt () gd->dm_udevice_rt +#else +#define gd_set_dm_udevice_rt (dyn) +#define gd_dm_udevice_rt () NULL +#endif +#ifdef CONFIG_GENERATE_ACPI_TABLE #define gd_acpi_ctx () gd->acpi_ctx#else
include/dm/device.h +16 -1
@@ -166,7 +166,7 @@ struct udevice {struct list_head sibling_node;u32 flags_;int seq_; -#if !CONFIG_IS_ENABLED (OF_PLATDATA ) +#if !CONFIG_IS_ENABLED (OF_PLATDATA_RT )ofnode node_;#endif #ifdef CONFIG_DEVRES
@@ -174,6 +174,21 @@ struct udevice {#endif }; +/** +* udevice_rt - runtime information set up by U-Boot +* +* This is only used with OF_PLATDATA_RT +* +* There is one of these for every udevice in the linker list, indexed by +* the udevice_info idx value. +* +* @flags_: Flags for this device DM_FLAG_... (do not access outside driver +* model) +*/ +struct udevice_rt { +u32 flags_; +}; +/* Maximum sequence number supported */ #define DM_MAX_SEQ 999
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 19/30 dm: core: Move flags to device-runtime info
In reply to Simon GlassWhen of-platdata-inst is active, use the flags in the new udevice_rt table, dropping them from the main struct udevice. This ensures that the latter is not updated at runtime.
---
drivers/core/device.c | 33 +++++++++++++++++++++++++++++++++ include/dm/device.h | 10 +++++++++- 2 files changed, 42 insertions(+), 1 deletion(-)
drivers/core/device.c +33 -0
@@ -1077,3 +1077,36 @@ int dev_enable_by_path(const char *path)return lists_bind_fdt (parent, node,NULL , false); }#endif + +#if CONFIG_IS_ENABLED (OF_PLATDATA_RT ) +static struct udevice_rt *dev_get_rt (const struct udevice *dev) +{ + struct udevice *base = ll_entry_start(struct udevice, udevice); +int idx = dev - base; + +struct udevice_rt *urt =gd_dm_udevice_rt () + idx; + +return urt; +} + +u32 dev_get_flags (const struct udevice *dev) +{ +const struct udevice_rt *urt =dev_get_rt (dev); + +return urt->flags_; +} + +void dev_or_flags (const struct udevice *dev,u32 or) +{ +struct udevice_rt *urt =dev_get_rt (dev); + + urt->flags_ |= or; +} + +void dev_bic_flags (const struct udevice *dev,u32 bic) +{ +struct udevice_rt *urt =dev_get_rt (dev); + + urt->flags_ &= ~bic; +} +#endif /* OF_PLATDATA_RT */
include/dm/device.h +9 -1
@@ -164,11 +164,13 @@ struct udevice {struct list_head uclass_node;struct list_head child_head;struct list_head sibling_node; -u32 flags_;int seq_;#if !CONFIG_IS_ENABLED (OF_PLATDATA_RT )ofnode node_;#endif +#if !CONFIG_IS_ENABLED (OF_PLATDATA_RT ) +u32 flags_; +#endif #ifdef CONFIG_DEVRES struct list_head devres_head;#endif
@@ -195,6 +197,11 @@ struct udevice_rt {/* Returns the operations for a device */ #define device_get_ops (dev) (dev->driver->ops) +#if CONFIG_IS_ENABLED (OF_PLATDATA_RT ) +u32 dev_get_flags (const struct udevice *dev); +void dev_or_flags (const struct udevice *dev,u32 or); +void dev_bic_flags (const struct udevice *dev,u32 bic); +#else static inline u32 dev_get_flags (const struct udevice *dev) {return dev->flags_;
@@ -209,6 +216,7 @@ static inline void dev_bic_flags(struct udevice *dev, u32 bic) { dev->flags_ &= ~bic; } +#endif /* OF_PLATDATA_RT */ /** * dev_ofnode() - get the DT node reference associated with a udevice
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 20/30 dm: core: Allow storing priv/plat data separately
In reply to Simon GlassAt present the device priv/data data allocated by dtoc is stored in the data section along with other variables. On some platforms it is better to allocate space for it separately, e.g. if SPL is running from read-only memory.
Create a new space with the same size as that allocated by dtoc, ready for use.
---
drivers/core/root.c | 12 ++++++++++++ include/asm-generic/global_data.h | 10 ++++++++++ include/asm-generic/sections.h | 3 +++ 3 files changed, 25 insertions(+)
drivers/core/root.c +12 -0
@@ -11,6 +11,7 @@#include <fdtdec.h> #include <log.h> #include <malloc.h> +#include <asm-generic/sections.h> #include <linux/libfdt.h> #include <dm/acpi.h> #include <dm/device.h>
@@ -134,7 +135,9 @@ static int dm_setup_inst(void)if (CONFIG_IS_ENABLED (OF_PLATDATA_RT )) {struct udevice_rt *urt; +void *base;int n_ents; +uint size;/* Allocate the udevice_rt table */ n_ents = ll_entry_count(struct udevice, udevice);
@@ -142,6 +145,15 @@ static int dm_setup_inst(void)if (!urt)return log_msg_ret ("urt" , -ENOMEM );gd_set_dm_udevice_rt (urt); + +/* Now allocate space for the priv/plat data, and copy it in */ + size = __priv_data_end - __priv_data_start; + + base =calloc (1 , size); +if (!base) +return log_msg_ret ("priv" , -ENOMEM ); +memcpy (base, __priv_data_start, size); +gd_set_dm_priv_base (base); }return 0 ;
include/asm-generic/global_data.h +10 -0
@@ -208,6 +208,12 @@ struct global_data {#if CONFIG_IS_ENABLED (OF_PLATDATA_RT )/** @dm_udevice_rt: Dynamic info about the udevice */ struct udevice_rt *dm_udevice_rt; +/** +* @dm_priv_base: Base address of the priv/plat region used when +* udevices and uclasses are in read-only memory. This is NULL if not +* used +*/ +void *dm_priv_base;# endif #endif #ifdef CONFIG_TIMER
@@ -472,9 +478,13 @@ struct global_data {#if CONFIG_IS_ENABLED (OF_PLATDATA_RT )#define gd_set_dm_udevice_rt (dyn) gd->dm_udevice_rt = dyn#define gd_dm_udevice_rt () gd->dm_udevice_rt +#define gd_set_dm_priv_base (dyn) gd->dm_priv_base = dyn +#define gd_dm_priv_base () gd->dm_priv_base#else #define gd_set_dm_udevice_rt (dyn)#define gd_dm_udevice_rt () NULL +#define gd_set_dm_priv_base (dyn) +#define gd_dm_priv_base () NULL#endif #ifdef CONFIG_GENERATE_ACPI_TABLE
include/asm-generic/sections.h +3 -0
@@ -28,6 +28,9 @@ extern char __efi_helloworld_end[];extern char __efi_var_file_begin[];extern char __efi_var_file_end[]; +/* Private data used by of-platdata devices/uclasses */ +extern char __priv_data_start[], __priv_data_end[]; +/* Start and end of .ctors section - used for constructor calls. */ extern char __ctors_start[], __ctors_end[];
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 21/30 sandbox: Define a region for device priv/plat data
In reply to Simon GlassCollect this together in one place, so driver model can access set it up in a new place if needed.
---
arch/sandbox/cpu/u-boot-spl.lds | 8 ++++++++ 1 file changed, 8 insertions(+)
arch/sandbox/cpu/u-boot-spl.lds +8 -0
@@ -13,6 +13,14 @@ SECTIONS KEEP(*(SORT(.u_boot_list*))); } + /* Private data for devices with OF_PLATDATA_RT */ + . = ALIGN(4); + .priv_data : { + __priv_data_start = .; + *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.priv_data*))) + __priv_data_end = .; + } + __u_boot_sandbox_option_start = .; _u_boot_sandbox_getopt : { KEEP(*(.u_boot_sandbox_getopt)) } __u_boot_sandbox_option_end = .;
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 22/30 dm: core: Use separate priv/plat data region
In reply to Simon GlassMake use of the new priv/plat data region if enabled. This is implemented as a simple offset from the position set up by dtoc to the new position.
So long as all access goes through dm_priv_to_rw() this is safe.
---
drivers/core/device.c | 12 ++++++------ drivers/core/root.c | 9 +++++++++ include/dm/util.h | 9 +++++++++ 3 files changed, 24 insertions(+), 6 deletions(-)
drivers/core/device.c +6 -6
@@ -533,7 +533,7 @@ void *dev_get_plat(const struct udevice *dev)return NULL ; } -return dev->plat_; +return dm_priv_to_rw (dev->plat_); }void *dev_get_parent_plat (const struct udevice *dev)
@@ -543,7 +543,7 @@ void *dev_get_parent_plat(const struct udevice *dev)return NULL ; } -return dev->parent_plat_; +return dm_priv_to_rw (dev->parent_plat_); }void *dev_get_uclass_plat (const struct udevice *dev)
@@ -553,7 +553,7 @@ void *dev_get_uclass_plat(const struct udevice *dev)return NULL ; } -return dev->uclass_plat_; +return dm_priv_to_rw (dev->uclass_plat_); }void *dev_get_priv (const struct udevice *dev)
@@ -563,7 +563,7 @@ void *dev_get_priv(const struct udevice *dev)return NULL ; } -return dev->priv_; +return dm_priv_to_rw (dev->priv_); }void *dev_get_uclass_priv (const struct udevice *dev)
@@ -573,7 +573,7 @@ void *dev_get_uclass_priv(const struct udevice *dev)return NULL ; } -return dev->uclass_priv_; +return dm_priv_to_rw (dev->uclass_priv_); }void *dev_get_parent_priv (const struct udevice *dev)
@@ -583,7 +583,7 @@ void *dev_get_parent_priv(const struct udevice *dev)return NULL ; } -return dev->parent_priv_; +return dm_priv_to_rw (dev->parent_priv_); } static int device_get_device_tail(struct udevice *dev, int ret,
drivers/core/root.c +9 -0
@@ -341,6 +341,15 @@ __weak int dm_scan_other(bool pre_reloc_only)return 0 ; } +#if CONFIG_IS_ENABLED (OF_PLATDATA_INST ) &&CONFIG_IS_ENABLED (READ_ONLY ) +void *dm_priv_to_rw (void *priv) +{ +long offset = priv - (void *)__priv_data_start; + +return gd_dm_priv_base () + offset; +} +#endif +/** * dm_scan() - Scan tables to bind devices *
include/dm/util.h +9 -0
@@ -49,3 +49,12 @@ void dm_dump_driver_compat(void);void dm_dump_static_driver_info (void );#endif + +#if CONFIG_IS_ENABLED (OF_PLATDATA_INST ) &&CONFIG_IS_ENABLED (READ_ONLY ) +void *dm_priv_to_rw (void *priv); +#else +static inline void *dm_priv_to_rw (void *priv) +{ +return priv; +} +#endif
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 23/30 x86: Define a region for device priv/plat data
In reply to Simon GlassCollect this together in one place, so driver model can access set it up in a new place if needed.
---
arch/x86/cpu/u-boot-spl.lds | 8 ++++++++ 1 file changed, 8 insertions(+)
arch/x86/cpu/u-boot-spl.lds +8 -0
@@ -32,6 +32,14 @@ SECTIONS . = ALIGN(4); .rodata : { *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*))) } + . = ALIGN(4); + + .priv_data : { + __priv_data_start = .; + *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.priv_data*))) + __priv_data_end = .; + } + . = ALIGN(4); .data : { *(.data*) }
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 24/30 x86: apl: Fix the header order in pmc
In reply to Simon GlassThe dm.h header should come first. In fact it needs to, since otherwise the driver model definitions are not available to dt-structs.h
Fix this, since it causes problems with OF_PLATDATA_INST.
---
arch/x86/cpu/apollolake/pmc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/cpu/apollolake/pmc.c +1 -1
@@ -9,8 +9,8 @@#define LOG_CATEGORY UCLASS_ACPI_PMC#include <common.h> -#include <dt-structs.h> #include <dm.h> +#include <dt-structs.h> #include <log.h> #include <spl.h> #include <acpi/acpi_s3.h>
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 25/30 x86: apl: Tell of-platdata about a required header file
In reply to Simon GlassThis enum is needed to generate build-time devices. Tell dtoc where to find the header, to avoid compile errors in the generated code.
---
arch/x86/cpu/apollolake/punit.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/cpu/apollolake/punit.c +1 -0
@@ -93,4 +93,5 @@ U_BOOT_DRIVER(intel_apl_punit) = { .id = UCLASS_SYSCON, .of_match = apl_syscon_ids, .probe = apl_punit_probe, + DM_HEADER(<asm/cpu.h>) /* for X86_SYSCON_PUNIT */ };
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 26/30 x86: itss: Tidy up bind() for of-platdata-inst
In reply to Simon GlassWith the standard of-platdata we must fix up driver_data manually. With of-platadata-inst this is not necessary, since it is added to the device by dtoc.
Update the code to handle this.
---
arch/x86/cpu/intel_common/itss.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
arch/x86/cpu/intel_common/itss.c +3 -2
@@ -152,8 +152,9 @@ static int route_pmc_gpio_gpe(struct udevice *dev, uint pmc_gpe_num)static int itss_bind (struct udevice *dev) { -/* This is not set with of-platdata, so set it manually */ -if (CONFIG_IS_ENABLED (OF_PLATDATA )) +/* This is not set with basic of-platdata, so set it manually */ +if (CONFIG_IS_ENABLED (OF_PLATDATA ) && + !CONFIG_IS_ENABLED (OF_PLATDATA_INST )) dev->driver_data =X86_IRQT_ITSS ;return 0 ;
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 27/30 x86: Support a fake PCI device with of-platdata-inst
In reply to Simon GlassWith TPL we don't need full PCI support and it adds to code size. Instead, a simple_bus driver is good enough to be able to read and write the PCI config and do a little basic setup.
So at present there are two drivers in U-Boot called pci_x86. One is in UCLASS_PCI, used in SPL and U-Boot proper. The other is in UCLASS_SIMPLE_BUS and used only in TPL.
Add a tag to tell dtoc about this, so it knows which one to use when generating the devices and uclasses.
---
arch/x86/lib/tpl.c | 1 + 1 file changed, 1 insertion(+)
arch/x86/lib/tpl.c +1 -0
@@ -144,5 +144,6 @@ U_BOOT_DRIVER(pci_x86) = { .name = "pci_x86", .id = UCLASS_SIMPLE_BUS, .of_match =of_match_ptr (tpl_fake_pci_ids), +DM_PHASE (tpl) };#endif
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 28/30 x86: Don't include reset driver in SPL
In reply to Simon GlassWe don't normally need this driver in TPL/SPL, so drop it for now. It can be enabled by individual boards if needed.
---
arch/x86/dts/reset.dtsi | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/dts/reset.dtsi +1 -1
@@ -1,6 +1,6 @@ / { reset: reset { compatible = "x86,reset"; - u-boot,dm-pre-reloc; + u-boot,dm-pre-proper; }; };
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 29/30 x86: coral: Drop ACPI properties from of-platdata
In reply to Simon GlassWe don't use these in TPL or SPL, so drop them. Also drop 'ranges' since we don't have a full PCI driver.
---
configs/chromebook_coral_defconfig | 1 + 1 file changed, 1 insertion(+)
configs/chromebook_coral_defconfig +1 -0
@@ -71,6 +71,7 @@ CONFIG_MAC_PARTITION=y CONFIG_ISO_PARTITION=y CONFIG_EFI_PARTITION=y # CONFIG_SPL_EFI_PARTITION is not set +CONFIG_OF_SPL_REMOVE_PROPS="clocks clock-names interrupt-parent interrupts linux-name acpi,name acpi,path u-boot,acpi-dsdt-order u-boot,acpi-ssdt-order ranges" CONFIG_ENV_OVERWRITE=y CONFIG_REGMAP=y CONFIG_SYSCON=y
Signature · 1 line
2.29.2.729.g45daf8777d-goog
PATCH 30/30 x86: apl: Use read-only SPL and new of-platdata
In reply to Simon GlassWith Apollo Lake, SPL is placed in read-only memory. Set this new option so that OF_PLATDATA_INST can be used.
---
arch/x86/cpu/apollolake/Kconfig | 2 ++ 1 file changed, 2 insertions(+)
arch/x86/cpu/apollolake/Kconfig +2 -0
@@ -20,6 +20,8 @@ config INTEL_APOLLOLAKE select INTEL_GMA_SWSMISCI select ACPI_GNVS_EXTERNAL select TPL_OF_PLATDATA_PARENT + select TPL_OF_PLATDATA_INST + select TPL_READ_ONLY imply ENABLE_MRC_CACHE imply AHCI_PCI imply SCSI
Signature · 1 line
2.29.2.729.g45daf8777d-goog
cover dm: Implement OF_PLATDATA_INST in driver model (part E)
In reply to Simon GlassHi Simon,
Quoted 39 lines · Simon Glass
This series builds on the recent dtoc implementation of build-time device instantiation.
It adds the required driver model support, which is basically a few more data structures.
With this,
sandbox_splandchromebook_coralboth use the new feature.For coral TPL there is a 1.5KB code-size reduction and a 1.75KB data-size increase:
text data bss dec hex 18836 3080 12 21928 55a8 original 17229 4896 12 22137 5679 with OF_PLATDATA_INST 17277 4896 12 22185 56a9 with OF_PLATDATA_RTThe extra data size is due to the build-time devices which are now included in the image instead of being built at runtime. Also the private data for each device is allocated in the data region at present, even through much of it is just zeroes.
The reduction in code size is due to not needing to bind devices at runtime, as a well as a simplified
probe()function. Coral requires that data be copied out to RAM before being updated, so that adds a small amount to size (shown in the third line).Quite a lot of future work is possible, including reducing the size of data structures. See [1] for more ideas. But this series implements the basic feature.
To try this out on your board, define
CONFIG_SPL_OF_PLATDATA_INSTand see what you get.Note: SPL tests do not yet pass with this series. The
driver_rtstruct is not set up sodevice_get_by_driver_info_idx()does not work. This means that looking up phandles will fail. This will be addressed in a v2 series, along with documentation updates and a little more information on code-size impact.
Quoted 3 lines
This series is available at u-boot-dm/tin-working
[1] https://lists.denx.de/pipermail/u-boot/2020-July/418433.html
Thanks for this series. I have done a quick review, but I plan to do a deeper one and some testing during the following days.
Regards,
Walter
Quoted 76 lines
Simon Glass (30):
sandbox: Drop debug message inos_spl_to_uboot()
linker_lists: Allow use in data structures
dm: core: Add macros to access the new linker lists
dm: core: Allow dropping run-time binding of devices
dm: core: Adjust uclass setup with of-platdata
dm: core: Set up driver model forOF_PLATDATA_INST
dm: core: Skip adding uclasses withOF_PLATDATA_INST
dm: Add the new dtoc-generated files to the build
dm: core: Includedt-decl.hautomatically
dm: test: Avoid destroying uclasses with of-platdata-inst
clk: sandbox: Move priv/plat data to a header file
clk: fixed-rate: Export driver parts forOF_PLATDATA_INST
clk: sandbox: Create a special fixed-rate driver
sandbox: Create a newsandbox_noinstbuild
test: Runsandbox_spltests onsandbox_noinstazure/gitlab: Add tests forsandbox_noinst
dm: core: Add an option to support SPL in read-only memory
dm: core: Create a struct for device runtime info
dm: core: Move flags to device-runtime info
dm: core: Allow storing priv/plat data separately
sandbox: Define a region for device priv/plat data
dm: core: Use separate priv/plat data region
x86: Define a region for device priv/plat data
x86: apl: Fix the header order in pmc
x86: apl: Tell of-platdata about a required header file
x86: itss: Tidy upbind()for of-platdata-inst
x86: Support a fake PCI device with of-platdata-inst
x86: Don't include reset driver in SPL
x86: coral: Drop ACPI properties from of-platdata
x86: apl: Use read-only SPL and new of-platdata.azure-pipelines.yml | 3 + .gitlab-ci.yml | 10 +- arch/sandbox/cpu/os.c | 1 - arch/sandbox/cpu/u-boot-spl.lds | 8 + arch/sandbox/dts/sandbox.dtsi | 2 +- arch/sandbox/include/asm/clk.h | 24 +++ arch/x86/cpu/apollolake/Kconfig | 2 + arch/x86/cpu/apollolake/pmc.c | 2 +- arch/x86/cpu/apollolake/punit.c | 1 + arch/x86/cpu/intel_common/itss.c | 5 +- arch/x86/cpu/u-boot-spl.lds | 8 + arch/x86/dts/reset.dtsi | 2 +- arch/x86/lib/tpl.c | 1 + board/sandbox/MAINTAINERS | 7 + common/spl/Kconfig | 24 +++ configs/chromebook_coral_defconfig | 1 + configs/sandbox_noinst_defconfig | 229 +++++++++++++++++++++++++++++ configs/sandbox_spl_defconfig | 1 + drivers/clk/clk_fixed_rate.c | 14 +- drivers/clk/clk_sandbox.c | 40 ++++- drivers/clk/clk_sandbox_test.c | 6 - drivers/core/device.c | 83 ++++++++--- drivers/core/root.c | 83 +++++++++-- drivers/core/uclass.c | 5 +- dts/Kconfig | 38 +++++ include/asm-generic/global_data.h | 22 +++ include/asm-generic/sections.h | 3 + include/dm/device-internal.h | 26 ++++ include/dm/device.h | 38 ++++- include/dm/root.h | 3 + include/dm/uclass-internal.h | 23 +++ include/dm/uclass.h | 20 +++ include/dm/util.h | 9 ++ include/dt-structs.h | 2 + include/linker_lists.h | 28 ++++ include/linux/clk-provider.h | 5 + scripts/Makefile.spl | 3 +- test/dm/of_platdata.c | 2 +- test/dm/test-main.c | 3 +- test/run | 4 + 40 files changed, 724 insertions(+), 67 deletions(-)create mode 100644 configs/sandbox_noinst_defconfig