Skip to contentSkip to conversation

x86: Improve support for chain-loading U-Boot

PATCH New Simon Glass 3 participants 23 messages Dec 21, 2019 – Feb 5, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover x86: Improve support for chain-loading U-Boot

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Simon Glass (5):
x86: fsp: Allow skipping init code when chain loading
x86: apl: Skip init code when chain loading
x86: cpu: Skip init code when chain loading
dm: Avoid initing built-in devices when chain loading
pci: Avoid auto-config when chain loading

 arch/x86/cpu/apollolake/fsp_s.c |  2 ++
 arch/x86/cpu/cpu.c              |  4 +++-
 arch/x86/cpu/i386/interrupt.c   |  3 +++
 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 arch/x86/lib/init_helpers.c     |  3 +++
 common/board_r.c                |  3 +++
 drivers/pci/pci-uclass.c        |  4 ++--
 10 files changed, 38 insertions(+), 4 deletions(-)
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#1 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +8 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
 	phys_addr_t low_end;
 	uint bank;
 
+	if (!ll_boot_init()) {
+		gd->bd->bi_dram[0].start = 0;
+		gd->bd->bi_dram[0].size = gd->ram_size;
+
+		mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+		return 0;
+	}
+
 	low_end = 0;
 	for (bank = 1, hdr = gd->arch.hob_list;
 	     bank < CONFIG_NR_DRAM_BANKS && !end_of_hob(hdr);
arch/x86/lib/fsp/fsp_graphics.c +3 -0
index 226c7e66b3..98b762209f 100644
@@ -78,6 +78,9 @@ static int fsp_video_probe(struct udevice *dev)
 	struct vesa_mode_info *vesa = &mode_info.vesa;
 	int ret;
 
+	if (!ll_boot_init())
+		return 0;
+
 	printf("Video: ");
 
 	/* Initialize vesa_mode_info structure */
arch/x86/lib/fsp2/fsp_dram.c +10 -0
index 90a238a224..74835eebce 100644
@@ -12,11 +12,18 @@
 #include <asm/fsp/fsp_support.h>
 #include <asm/fsp2/fsp_api.h>
 #include <asm/fsp2/fsp_internal.h>
+#include <linux/sizes.h>
 
 int dram_init(void)
 {
 	int ret;
 
+	if (!ll_boot_init()) {
+		/* Use a small and safe amount of 1GB */
+		gd->ram_size = SZ_1G;
+
+		return 0;
+	}
 	if (spl_phase() == PHASE_SPL) {
 #ifdef CONFIG_HAVE_ACPI_RESUME
 		bool s3wake = gd->arch.prev_sleep_state == ACPI_S3;
@@ -68,6 +75,9 @@ int dram_init(void)
 
 ulong board_get_usable_ram_top(ulong total_size)
 {
+	if (!ll_boot_init())
+		return gd->ram_size;
+
 #if CONFIG_IS_ENABLED(HANDOFF)
 	struct spl_handoff *ho = gd->spl_handoff;
 
arch/x86/lib/fsp2/fsp_init.c +1 -1
index da9bd6b45c..c7dc2ea257 100644
@@ -23,7 +23,7 @@ int arch_cpu_init_dm(void)
 	int ret;
 
 	/* Make sure pads are set up early in U-Boot */
-	if (spl_phase() != PHASE_BOARD_F)
+	if (!ll_boot_init() || spl_phase() != PHASE_BOARD_F)
 		return 0;
 
 	/* Probe all pinctrl devices to set up the pads */
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#2 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Simon,

Quoted 23 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 10 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
        uint bank;
 
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +0 -0
        low_end = 0;
        for (bank = 1, hdr = gd->arch.hob_list;
             bank < CONFIG_NR_DRAM_BANKS && !end_of_hob(hdr);
arch/x86/lib/fsp/fsp_graphics.c +0 -0
index 226c7e66b3..98b762209f 100644

arch/x86/lib/fsp/fsp_graphics.c static int fsp_video_probe(struct udevice *dev) · line 83

Quoted 6 lines
arch/x86/lib/fsp/fsp_graphics.c +3 -0
@@ -78,6 +78,9 @@ static int fsp_video_probe(struct udevice *dev)
        struct vesa_mode_info *vesa = &mode_info.vesa;
        int ret;
 
+       if (!ll_boot_init())
+               return 0;
+

the same question as above

Quoted 11 lines
arch/x86/lib/fsp/fsp_graphics.c +0 -0
        printf("Video: ");
 
        /* Initialize vesa_mode_info structure */
arch/x86/lib/fsp2/fsp_dram.c +0 -0
index 90a238a224..74835eebce 100644
@@ -12,11 +12,18 @@
 #include <asm/fsp/fsp_support.h>
 #include <asm/fsp2/fsp_api.h>
 #include <asm/fsp2/fsp_internal.h>

arch/x86/lib/fsp2/fsp_dram.c line 24

Quoted 10 lines
arch/x86/lib/fsp2/fsp_dram.c +5 -0
+#include <linux/sizes.h>
 
 int dram_init(void)
 {
        int ret;
 
+       if (!ll_boot_init()) {
+               /* Use a small and safe amount of 1GB */
+               gd->ram_size = SZ_1G;
+

ditto. Maybe we should explain on what configuration ll_boot_init() is set to true / false?

Quoted 20 lines
arch/x86/lib/fsp2/fsp_dram.c +5 -0
+               return 0;
+       }
        if (spl_phase() == PHASE_SPL) {
 #ifdef CONFIG_HAVE_ACPI_RESUME
                bool s3wake = gd->arch.prev_sleep_state == ACPI_S3;
@@ -68,6 +75,9 @@ int dram_init(void)
 
 ulong board_get_usable_ram_top(ulong total_size)
 {
+       if (!ll_boot_init())
+               return gd->ram_size;
+
 #if CONFIG_IS_ENABLED(HANDOFF)
        struct spl_handoff *ho = gd->spl_handoff;
 
arch/x86/lib/fsp2/fsp_init.c +0 -0
index da9bd6b45c..c7dc2ea257 100644

arch/x86/lib/fsp2/fsp_init.c int arch_cpu_init_dm(void)

Quoted 9 lines
arch/x86/lib/fsp2/fsp_init.c +1 -2
@@ -23,7 +23,7 @@ int arch_cpu_init_dm(void)
        int ret;
 
        /* Make sure pads are set up early in U-Boot */
-       if (spl_phase() != PHASE_BOARD_F)
+       if (!ll_boot_init() || spl_phase() != PHASE_BOARD_F)
                return 0;
 
        /* Probe all pinctrl devices to set up the pads */
--

Regards,
Bin

#3 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Bin,

Quoted 29 lines · Bin Meng

Hi Simon,

Quoted 25 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;
        uint bank;
 
Quoted 11 lines

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

How about this:

It is useful to be able to boot the same x86 image on a device with or without a first-stage bootloader. For example, with chromebook_coral, it is helpful for testing to be able to boot the same U-Boot (complete with FSP) on bare metal and from coreboot. It allows checking of things like CPU speed, comparing registers, ACPI tables and the like.

The idea is to change ll_boot_init() to false, and rebuild without changing anything else.

Regards,
Simon

#4 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Simon,

Quoted 47 lines · Simon Glass

Hi Bin,

Quoted 29 lines · Bin Meng

Hi Simon,

Quoted 25 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;
        uint bank;
 
Quoted 11 lines

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

How about this:

Quoted 8 lines

It is useful to be able to boot the same x86 image on a device with or without a first-stage bootloader. For example, with chromebook_coral, it is helpful for testing to be able to boot the same U-Boot (complete with FSP) on bare metal and from coreboot. It allows checking of things like CPU speed, comparing registers, ACPI tables and the like.

The idea is to change ll_boot_init() to false, and rebuild without changing anything else.

This sounds like for a debugging purpose, instead of a supported configuration. So when we boot from previous bootloader, we really should use its configuration, for the coreboot case, coreboot-x86_defconfig should be used, and we can compare the registers, ACPI tables in that configuration, instead of "hacking" current U-Boot codes like in this series.

Regards,
Bin

#5 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Bin,

Quoted 60 lines · Bin Meng

Hi Simon,

Quoted 47 lines · Simon Glass

Hi Bin,

Quoted 29 lines · Bin Meng

Hi Simon,

Quoted 25 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;
        uint bank;
 
Quoted 11 lines

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

How about this:

Quoted 8 lines

It is useful to be able to boot the same x86 image on a device with or without a first-stage bootloader. For example, with chromebook_coral, it is helpful for testing to be able to boot the same U-Boot (complete with FSP) on bare metal and from coreboot. It allows checking of things like CPU speed, comparing registers, ACPI tables and the like.

The idea is to change ll_boot_init() to false, and rebuild without changing anything else.

Quoted 6 lines

This sounds like for a debugging purpose, instead of a supported configuration. So when we boot from previous bootloader, we really should use its configuration, for the coreboot case, coreboot-x86_defconfig should be used, and we can compare the registers, ACPI tables in that configuration, instead of "hacking" current U-Boot codes like in this series.

Well the problem is then that we cannot boot the same image with or without coreboot. Also there are various of device-tree things in coral that we need for the laptop to work properly. My idea is to detect coreboot and set ll_boot_init() dynamically if needed.

Regards,
Simon

#6 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Simon,

Quoted 71 lines · Simon Glass

Hi Bin,

Quoted 60 lines · Bin Meng

Hi Simon,

Quoted 47 lines · Simon Glass

Hi Bin,

Quoted 29 lines · Bin Meng

Hi Simon,

Quoted 25 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;
        uint bank;
 
Quoted 11 lines

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

How about this:

Quoted 8 lines

It is useful to be able to boot the same x86 image on a device with or without a first-stage bootloader. For example, with chromebook_coral, it is helpful for testing to be able to boot the same U-Boot (complete with FSP) on bare metal and from coreboot. It allows checking of things like CPU speed, comparing registers, ACPI tables and the like.

The idea is to change ll_boot_init() to false, and rebuild without changing anything else.

Quoted 6 lines

This sounds like for a debugging purpose, instead of a supported configuration. So when we boot from previous bootloader, we really should use its configuration, for the coreboot case, coreboot-x86_defconfig should be used, and we can compare the registers, ACPI tables in that configuration, instead of "hacking" current U-Boot codes like in this series.

Quoted 4 lines

Well the problem is then that we cannot boot the same image with or without coreboot. Also there are various of device-tree things in coral that we need for the laptop to work properly. My idea is to detect coreboot and set ll_boot_init() dynamically if needed.

If so, why should we keep coreboot_defconfig for U-Boot as a supported target?

I am more interested in what you proposed here: "detect coreboot and set ll_boot_init() dynamically if needed." The name ll_boot_init() does not sound great. It's written like a function call but actually it is a macro for true/false. Can we do something in the gd->flags to indicate we are booting from previous stage bootloaders instead of relying on ll_boot_init()?

Regards,
Bin

#7 lore raw

PATCH 1/5 x86: fsp: Allow skipping init code when chain loading

Hi Bin,

Quoted 80 lines · Bin Meng

Hi Simon,

Quoted 71 lines · Simon Glass

Hi Bin,

Quoted 60 lines · Bin Meng

Hi Simon,

Quoted 47 lines · Simon Glass

Hi Bin,

Quoted 29 lines · Bin Meng

Hi Simon,

Quoted 25 lines · Simon Glass

When U-Boot is no the first-stage bootloader much of this code is not needed and can break booting. Add checks for this to the FSP code.

Rather than checking for the amount of available SDRAM, just use 1GB in this situation, which should be safe. Using 2GB may run into a memory hole on some SoCs.

---

 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 4 files changed, 22 insertions(+), 1 deletion(-)
arch/x86/lib/fsp/fsp_dram.c +0 -0
index 9ce0ddf0d3..15e82de2fe 100644
@@ -44,6 +44,14 @@ int dram_init_banksize(void)
        phys_addr_t low_end;
        uint bank;
 
Quoted 11 lines

arch/x86/lib/fsp/fsp_dram.c int dram_init_banksize(void) · line 54

Quoted 8 lines
arch/x86/lib/fsp/fsp_dram.c +8 -0
+       if (!ll_boot_init()) {
+               gd->bd->bi_dram[0].start = 0;
+               gd->bd->bi_dram[0].size = gd->ram_size;
+
+               mtrr_add_request(MTRR_TYPE_WRBACK, 0, gd->ram_size);
+               return 0;
+       }
+

I wonder why this change is needed. When booting from other bootloader, this dram_init_banksize() will not be called, no?

How about this:

Quoted 8 lines

It is useful to be able to boot the same x86 image on a device with or without a first-stage bootloader. For example, with chromebook_coral, it is helpful for testing to be able to boot the same U-Boot (complete with FSP) on bare metal and from coreboot. It allows checking of things like CPU speed, comparing registers, ACPI tables and the like.

The idea is to change ll_boot_init() to false, and rebuild without changing anything else.

Quoted 6 lines

This sounds like for a debugging purpose, instead of a supported configuration. So when we boot from previous bootloader, we really should use its configuration, for the coreboot case, coreboot-x86_defconfig should be used, and we can compare the registers, ACPI tables in that configuration, instead of "hacking" current U-Boot codes like in this series.

Quoted 4 lines

Well the problem is then that we cannot boot the same image with or without coreboot. Also there are various of device-tree things in coral that we need for the laptop to work properly. My idea is to detect coreboot and set ll_boot_init() dynamically if needed.

Quoted 8 lines

If so, why should we keep coreboot_defconfig for U-Boot as a supported target?

I am more interested in what you proposed here: "detect coreboot and set ll_boot_init() dynamically if needed." The name ll_boot_init() does not sound great. It's written like a function call but actually it is a macro for true/false. Can we do something in the gd->flags to indicate we are booting from previous stage bootloaders instead of relying on ll_boot_init()?

That sounds like a good idea, although at present it is statically configured. But we don't really care about code size here I think.

Regards,
Simon

#8 lore raw

PATCH 2/5 x86: apl: Skip init code when chain loading

In reply to Simon Glass

When U-Boot is not the first-stage bootloader the FSP-S init must be skipped. Update it to add a check.

---

 arch/x86/cpu/apollolake/fsp_s.c | 2 ++
 1 file changed, 2 insertions(+)
arch/x86/cpu/apollolake/fsp_s.c +2 -0
index 9804227f80..a3d147ec14 100644
@@ -623,6 +623,8 @@ int arch_fsp_init_r(void)
 	struct udevice *dev, *itss;
 	int ret;
 
+	if (!ll_boot_init())
+		return 0;
 	/*
 	 * This must be called before any devices are probed. Put any probing
 	 * into arch_fsps_preinit() above.
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#9 lore raw

PATCH 3/5 x86: cpu: Skip init code when chain loading

In reply to Simon Glass

When U-Boot is not the first-stage bootloader the interrupt and cache init must be skipped, as well as init for various peripherals. Update the code to add checks for this.

---

 arch/x86/cpu/cpu.c            | 4 +++-
 arch/x86/cpu/i386/interrupt.c | 3 +++
 arch/x86/lib/init_helpers.c   | 3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)
arch/x86/cpu/cpu.c +3 -1
index d626e38fd1..d82305303a 100644
@@ -238,8 +238,10 @@ int cpu_init_r(void)
 	struct udevice *dev;
 	int ret;
 
-	if (!ll_boot_init())
+	if (!ll_boot_init()) {
+		uclass_first_device(UCLASS_PCI, &dev);
 		return 0;
+	}
 
 	ret = x86_init_cpus();
 	if (ret)
arch/x86/cpu/i386/interrupt.c +3 -0
index 78aa51a3ea..c33ca165d8 100644
@@ -261,6 +261,9 @@ int interrupt_init(void)
 	struct udevice *dev;
 	int ret;
 
+	if (!ll_boot_init())
+		return 0;
+
 	/* Try to set up the interrupt router, but don't require one */
 	ret = uclass_first_device_err(UCLASS_IRQ, &dev);
 	if (ret && ret != -ENODEV)
arch/x86/lib/init_helpers.c +3 -0
index 5bb55e256f..d906b528b3 100644
@@ -30,6 +30,9 @@ int init_cache_f_r(void)
 			return ret;
 	}
 
+	if (!ll_boot_init())
+		return 0;
+
 	/* Initialise the CPU cache(s) */
 	return init_cache();
 }
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#10 lore raw

PATCH 3/5 x86: cpu: Skip init code when chain loading

Hi Simon,

Quoted 34 lines · Simon Glass

When U-Boot is not the first-stage bootloader the interrupt and cache init must be skipped, as well as init for various peripherals. Update the code to add checks for this.

---

 arch/x86/cpu/cpu.c            | 4 +++-
 arch/x86/cpu/i386/interrupt.c | 3 +++
 arch/x86/lib/init_helpers.c   | 3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)
arch/x86/cpu/cpu.c +3 -1
index d626e38fd1..d82305303a 100644
@@ -238,8 +238,10 @@ int cpu_init_r(void)
        struct udevice *dev;
        int ret;
 
-       if (!ll_boot_init())
+       if (!ll_boot_init()) {
+               uclass_first_device(UCLASS_PCI, &dev);
                return 0;
+       }
 
        ret = x86_init_cpus();
        if (ret)
arch/x86/cpu/i386/interrupt.c +0 -0
index 78aa51a3ea..c33ca165d8 100644

arch/x86/cpu/i386/interrupt.c int interrupt_init(void) · line 266

Quoted 6 lines
arch/x86/cpu/i386/interrupt.c +3 -0
@@ -261,6 +261,9 @@ int interrupt_init(void)
        struct udevice *dev;
        int ret;
 
+       if (!ll_boot_init())
+               return 0;
+

If you scroll down a little bit in this function, you see there is already a ll_boot_init() check. We need some consolidation here.

arch/x86/lib/init_helpers.c int init_cache_f_r(void) · line 34

Quoted 13 lines
arch/x86/cpu/i386/interrupt.c +0 -0
        /* Try to set up the interrupt router, but don't require one */
        ret = uclass_first_device_err(UCLASS_IRQ, &dev);
        if (ret && ret != -ENODEV)
arch/x86/lib/init_helpers.c +2 -0
index 5bb55e256f..d906b528b3 100644
@@ -30,6 +30,9 @@ int init_cache_f_r(void)
                        return ret;
        }
 
+       if (!ll_boot_init())
+               return 0;

I am not convinced that we should add the ll_boot_init() check here. This function already has a do_mtrr variable that is based on the config options.

arch/x86/lib/init_helpers.c int init_cache_f_r(void) · line 38

Quoted 4 lines
arch/x86/lib/init_helpers.c +1 -0
+
        /* Initialise the CPU cache(s) */
        return init_cache();
 }

Regards,
Bin

#11 lore raw

PATCH 3/5 x86: cpu: Skip init code when chain loading

Hi Bin,

Quoted 59 lines · Bin Meng

Hi Simon,

Quoted 34 lines · Simon Glass

When U-Boot is not the first-stage bootloader the interrupt and cache init must be skipped, as well as init for various peripherals. Update the code to add checks for this.

---

 arch/x86/cpu/cpu.c            | 4 +++-
 arch/x86/cpu/i386/interrupt.c | 3 +++
 arch/x86/lib/init_helpers.c   | 3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)
arch/x86/cpu/cpu.c +3 -1
index d626e38fd1..d82305303a 100644
@@ -238,8 +238,10 @@ int cpu_init_r(void)
        struct udevice *dev;
        int ret;
 
-       if (!ll_boot_init())
+       if (!ll_boot_init()) {
+               uclass_first_device(UCLASS_PCI, &dev);
                return 0;
+       }
 
        ret = x86_init_cpus();
        if (ret)
arch/x86/cpu/i386/interrupt.c +0 -0
index 78aa51a3ea..c33ca165d8 100644

arch/x86/cpu/i386/interrupt.c int interrupt_init(void) · line 266

Quoted 6 lines
arch/x86/cpu/i386/interrupt.c +3 -0
@@ -261,6 +261,9 @@ int interrupt_init(void)
        struct udevice *dev;
        int ret;
 
+       if (!ll_boot_init())
+               return 0;
+

If you scroll down a little bit in this function, you see there is already a ll_boot_init() check. We need some consolidation here.

Quoted 11 lines
arch/x86/cpu/i386/interrupt.c +0 -0
        /* Try to set up the interrupt router, but don't require one */
        ret = uclass_first_device_err(UCLASS_IRQ, &dev);
        if (ret && ret != -ENODEV)
arch/x86/lib/init_helpers.c +0 -0
index 5bb55e256f..d906b528b3 100644
@@ -30,6 +30,9 @@ int init_cache_f_r(void)
                        return ret;
        }
 
Quoted 6 lines

arch/x86/lib/init_helpers.c int init_cache_f_r(void) · line 34

Quoted 2 lines
arch/x86/lib/init_helpers.c +2 -0
+       if (!ll_boot_init())
+               return 0;

I am not convinced that we should add the ll_boot_init() check here. This function already has a do_mtrr variable that is based on the config options.

Yes but as per my other explanation, we are not using those config options. The idea is to boot an image as close as possible to the bare-metal one.

Regards,
Simon

#12 lore raw

PATCH 4/5 dm: Avoid initing built-in devices when chain loading

In reply to Simon Glass

When U-Boot is not the first-stage bootloader we don't want to init devices early during boot. Add a check to avoid this.

---

 common/board_r.c | 3 +++
 1 file changed, 3 insertions(+)
common/board_r.c +3 -0
index e711de64b5..4e0dfac4fc 100644
@@ -320,6 +320,9 @@ static int initr_dm_devices(void)
 {
 	int ret;
 
+	if (!ll_boot_init())
+		return 0;
+
 	if (IS_ENABLED(CONFIG_TIMER_EARLY)) {
 		ret = dm_timer_init();
 		if (ret)
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#13 lore raw

PATCH 4/5 dm: Avoid initing built-in devices when chain loading

Hi Simon,

Quoted 15 lines · Simon Glass

When U-Boot is not the first-stage bootloader we don't want to init devices early during boot. Add a check to avoid this.

---

 common/board_r.c | 3 +++
 1 file changed, 3 insertions(+)
common/board_r.c +0 -0
index e711de64b5..4e0dfac4fc 100644

common/board_r.c static int initr_dm_devices(void) · line 325

Quoted 6 lines
common/board_r.c +3 -0
@@ -320,6 +320,9 @@ static int initr_dm_devices(void)
 {
        int ret;
 
+       if (!ll_boot_init())
+               return 0;
+

I can't think of a reason why dm_timer_init() cannot be called in this case.

common/board_r.c static int initr_dm_devices(void)

Quoted 4 lines
common/board_r.c +0 -1
        if (IS_ENABLED(CONFIG_TIMER_EARLY)) {
                ret = dm_timer_init();
                if (ret)
--

Regards,
Bin

#14 lore raw

PATCH 4/5 dm: Avoid initing built-in devices when chain loading

Hi Bin,

Quoted 22 lines · Bin Meng

Hi Simon,

Quoted 18 lines · Simon Glass

When U-Boot is not the first-stage bootloader we don't want to init devices early during boot. Add a check to avoid this.

---

 common/board_r.c | 3 +++
 1 file changed, 3 insertions(+)
common/board_r.c +0 -0
index e711de64b5..4e0dfac4fc 100644
@@ -320,6 +320,9 @@ static int initr_dm_devices(void)
 {
        int ret;
 
Quoted 5 lines

common/board_r.c static int initr_dm_devices(void) · line 325

Quoted 3 lines
common/board_r.c +3 -0
+       if (!ll_boot_init())
+               return 0;
+

I can't think of a reason why dm_timer_init() cannot be called in this case.

Perhaps we can drop this one, since we don't reset the timer base anymore.

Quoted 5 lines

common/board_r.c static int initr_dm_devices(void)

Quoted 5 lines
common/board_r.c +0 -1
        if (IS_ENABLED(CONFIG_TIMER_EARLY)) {
                ret = dm_timer_init();
                if (ret)
--

Regards,
SImon

#15 lore raw

PATCH 5/5 pci: Avoid auto-config when chain loading

In reply to Simon Glass

When U-Boot is not the first-stage bootloader we don't want to re-configure the PCI devices, since this has already been done. Add a check to avoid this.

---

 drivers/pci/pci-uclass.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/pci/pci-uclass.c +2 -2
index 7308f612b6..6af97cd797 100644
@@ -991,7 +991,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
 	if (ret)
 		return ret;
 
-	if (CONFIG_IS_ENABLED(PCI_PNP) &&
+	if (CONFIG_IS_ENABLED(PCI_PNP) && ll_boot_init() &&
 	    (!hose->skip_auto_config_until_reloc ||
 	     (gd->flags & GD_FLG_RELOC))) {
 		ret = pci_auto_config_devices(bus);
@@ -1013,7 +1013,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
 	 * Note we only call this 1) after U-Boot is relocated, and 2)
 	 * root bus has finished probing.
 	 */
-	if ((gd->flags & GD_FLG_RELOC) && (bus->seq == 0)) {
+	if ((gd->flags & GD_FLG_RELOC) && bus->seq == 0 && ll_boot_init()) {
 		ret = fsp_init_phase_pci();
 		if (ret)
 			return ret;
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#16 lore raw

PATCH 5/5 pci: Avoid auto-config when chain loading

Hi Simon,

Quoted 16 lines · Simon Glass

When U-Boot is not the first-stage bootloader we don't want to re-configure the PCI devices, since this has already been done. Add a check to avoid this.

---

 drivers/pci/pci-uclass.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/pci/pci-uclass.c +0 -0
index 7308f612b6..6af97cd797 100644

drivers/pci/pci-uclass.c static int pci_uclass_post_probe(struct udevice *bus) · line 994

Quoted 5 lines
drivers/pci/pci-uclass.c +1 -1
@@ -991,7 +991,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
        if (ret)
                return ret;
 
-       if (CONFIG_IS_ENABLED(PCI_PNP) &&
+       if (CONFIG_IS_ENABLED(PCI_PNP) && ll_boot_init() &&

Doesn't the PCI_PNP option already cover such situation?

drivers/pci/pci-uclass.c static int pci_uclass_post_probe(struct udevice *bus) · line 1016

Quoted 9 lines
drivers/pci/pci-uclass.c +1 -1
            (!hose->skip_auto_config_until_reloc ||
             (gd->flags & GD_FLG_RELOC))) {
                ret = pci_auto_config_devices(bus);
@@ -1013,7 +1013,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
         * Note we only call this 1) after U-Boot is relocated, and 2)
         * root bus has finished probing.
         */
-       if ((gd->flags & GD_FLG_RELOC) && (bus->seq == 0)) {
+       if ((gd->flags & GD_FLG_RELOC) && bus->seq == 0 && ll_boot_init()) {

So CONFIG_HAVE_FSP already covers this. In such configuration, U-Boot should not turn on CONFIG_HAVE_FSP hence there is no need to check ll_boot_init().

drivers/pci/pci-uclass.c static int pci_uclass_post_probe(struct udevice *bus)

Quoted 4 lines
drivers/pci/pci-uclass.c +0 -1
                ret = fsp_init_phase_pci();
                if (ret)
                        return ret;
--

Regards,
Bin

#17 lore raw

PATCH 5/5 pci: Avoid auto-config when chain loading

Hi Bin,

Quoted 28 lines · Bin Meng

Hi Simon,

Quoted 16 lines · Simon Glass

When U-Boot is not the first-stage bootloader we don't want to re-configure the PCI devices, since this has already been done. Add a check to avoid this.

---

 drivers/pci/pci-uclass.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/pci/pci-uclass.c +0 -0
index 7308f612b6..6af97cd797 100644

drivers/pci/pci-uclass.c static int pci_uclass_post_probe(struct udevice *bus) · line 994

Quoted 5 lines
drivers/pci/pci-uclass.c +1 -1
@@ -991,7 +991,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
        if (ret)
                return ret;
 
-       if (CONFIG_IS_ENABLED(PCI_PNP) &&
+       if (CONFIG_IS_ENABLED(PCI_PNP) && ll_boot_init() &&

Doesn't the PCI_PNP option already cover such situation?

Quoted 13 lines

drivers/pci/pci-uclass.c static int pci_uclass_post_probe(struct udevice *bus) · line 1016

Quoted 9 lines
drivers/pci/pci-uclass.c +1 -1
            (!hose->skip_auto_config_until_reloc ||
             (gd->flags & GD_FLG_RELOC))) {
                ret = pci_auto_config_devices(bus);
@@ -1013,7 +1013,7 @@ static int pci_uclass_post_probe(struct udevice *bus)
         * Note we only call this 1) after U-Boot is relocated, and 2)
         * root bus has finished probing.
         */
-       if ((gd->flags & GD_FLG_RELOC) && (bus->seq == 0)) {
+       if ((gd->flags & GD_FLG_RELOC) && bus->seq == 0 && ll_boot_init()) {

So CONFIG_HAVE_FSP already covers this. In such configuration, U-Boot should not turn on CONFIG_HAVE_FSP hence there is no need to check ll_boot_init().

Same comment as with patch 1. See if that explains things enough, and if so I can send a v2.

Regards,
Simon

#18 lore raw

cover x86: Improve support for chain-loading U-Boot

In reply to Simon Glass
Quoted 10 lines · Simon Glass

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Thank you for sharing. May I ask what devices and configurations you have tested this on?

P.S. I'll be able to test my case after xmas vacations (mid January).

Quoted 22 lines

Simon Glass (5):
x86: fsp: Allow skipping init code when chain loading
x86: apl: Skip init code when chain loading
x86: cpu: Skip init code when chain loading
dm: Avoid initing built-in devices when chain loading
pci: Avoid auto-config when chain loading

 arch/x86/cpu/apollolake/fsp_s.c |  2 ++
 arch/x86/cpu/cpu.c              |  4 +++-
 arch/x86/cpu/i386/interrupt.c   |  3 +++
 arch/x86/lib/fsp/fsp_dram.c     |  8 ++++++++
 arch/x86/lib/fsp/fsp_graphics.c |  3 +++
 arch/x86/lib/fsp2/fsp_dram.c    | 10 ++++++++++
 arch/x86/lib/fsp2/fsp_init.c    |  2 +-
 arch/x86/lib/init_helpers.c     |  3 +++
 common/board_r.c                |  3 +++
 drivers/pci/pci-uclass.c        |  4 ++--
 10 files changed, 38 insertions(+), 4 deletions(-)

--
2.24.1.735.g03f4e72817-goog

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#19 lore raw

cover x86: Improve support for chain-loading U-Boot

Hi Andy,

Quoted 13 lines · Andy Shevchenko
Quoted 10 lines · Simon Glass

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Quoted 4 lines

Thank you for sharing. May I ask what devices and configurations you have tested this on?

P.S. I'll be able to test my case after xmas vacations (mid January).

This is only on chromebook_coral. But I think it could work in a similar way with any of the boards.

Merry Christmas!
Simon

#20 lore raw

cover x86: Improve support for chain-loading U-Boot

In reply to Andy Shevchenko

Hi Andy,

Quoted 13 lines · Andy Shevchenko
Quoted 10 lines · Simon Glass

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Quoted 5 lines

Thank you for sharing. May I ask what devices and configurations you have tested this on?

P.S. I'll be able to test my case after xmas vacations (mid January).

Do you have any review comments to share?

Regards,
Bin

#21 lore raw

cover x86: Improve support for chain-loading U-Boot

Quoted 15 lines · Bin Meng
Quoted 13 lines · Andy Shevchenko
Quoted 10 lines · Simon Glass

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Quoted 7 lines
Quoted 5 lines

Thank you for sharing. May I ask what devices and configurations you have tested this on?

P.S. I'll be able to test my case after xmas vacations (mid January).

Do you have any review comments to share?

Not for time being. Sorry, I'm busy with something else, but if there is a branch with these patches available thru x86 U-Boot tree, it will be easier to me to have a chance to test.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#22 lore raw

cover x86: Improve support for chain-loading U-Boot

Hi Andy,

Quoted 20 lines · Andy Shevchenko
Quoted 18 lines · Bin Meng
Quoted 16 lines · Andy Shevchenko
Quoted 10 lines · Simon Glass

This little series adds a few checks into the code to allow better operation when booting a build from a previous-state loader such as coreboot.

At present we have a 'coreboot' target, but sometimes it is useful to boot a bare-metal target, such as coral, from coreboot. That allows comparison of operation between the bare metal version doing all the init itself and relying on coreboot for some init.

Thank you for sharing. May I ask what devices and configurations you have tested this on?

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

P.S. I'll be able to test my case after xmas vacations (mid January).

Do you have any review comments to share?

Not for time being. Sorry, I'm busy with something else, but if there is a branch with these patches available thru x86 U-Boot tree, it will be easier to me to have a chance to test.

This one is u-boot-dm/cor3chain-working

I always push to the -dm tree before sending patches out, but don't always mention it in the cover letter!

The ACPI patches are at dm/coral-working

Regards,
Simon