[PATCH v3] spl: Add support for Device Firmware Upgrade (DFU) over PCIe

[PATCH v3] spl: Add support for Device Firmware Upgrade (DFU) over PCIe

[Hrushikesh Salunke]

Introduces support for Device Firmware Upgrade (DFU) over PCIe in
U-Boot. Traditionally, the DFU protocol is used over USB, where a
device enters DFU mode and allows a host to upload firmware or binary
images directly via the USB interface. This is a widely adopted and
convenient method for updating firmware.

In the context of Texas Instruments (TI) SoCs, PCIe can be used as a
boot interface in a manner that differs from the conventional
"PCIe Boot" process, which typically refers to booting an OS or
firmware image from an NVMe SSD or other PCIe-attached storage devices.
Instead, TI SoCs can be configured as a PCIe Endpoint, allowing a
connected PCIe Root Complex (host) to transfer images directly into the
device’s memory over the PCIe bus for boot purposes. This mechanism is
analogous to DFU over USB, but leverages the high-speed PCIe link and
does not depend on traditional storage devices.

By extending the DFU framework in U-Boot to support PCIe, it will be
possible to flash images over PCIe. While this implementation is
motivated by TI SoC use cases, the framework is generic and can be
adopted by everyone for platforms that support PCIe Endpoint mode.
Platforms with hardware support for PCIe-based memory loading can use
this to implement PCIe as a boot mode, as well as to enable flashing
and recovery scenarios similar to DFU over USB.

In summary, enable support for:
- DFU-style flashing of firmware/images over PCIe, analogous to existing
USB DFU workflows
- PCIe as a boot mode where a host can load images directly into device
memory using DFU over PCIe

Signed-off-by: Hrushikesh Salunke <h-salunke@ti.com>
---
This patch is based on commit
9de873b4c30 Merge patch series "linux/bitfield.h: sync <linux/bitfield.h> from Linux 6.15 + winbond"

Changes since v2
Put the usage of macro BOOT_DEVICE_PCIE under #ifdef CONFIG_SPL_PCI_DFU
to avoid potential compilation errors. This is required as 
BOOT_DEVICE_PCIE macro is defined for only those platforms which support
PCIE as a valid boot device.

v2 : https://patchwork.ozlabs.org/project/uboot/patch/20250616164444.631575-1-h-salunke@ti.com/

 common/spl/Kconfig   | 62 +++++++++++++++++++++++++++++
 common/spl/spl_dfu.c | 93 ++++++++++++++++++++++++++++++++++++++++++++
 common/spl/spl_ram.c | 18 +++++++++
 drivers/Makefile     |  1 +
 4 files changed, 174 insertions(+)

diff --git a/common/spl/Kconfig b/common/spl/Kconfig
index 880192043c4..212b4fc4e9e 100644
--- a/common/spl/Kconfig
+++ b/common/spl/Kconfig
@@ -1268,6 +1268,14 @@ config SPL_PCI
 	  necessary driver support. This enables the drivers in drivers/pci
 	  as part of an SPL build.
 
+config SPL_PCI_ENDPOINT
+	bool "Support for PCI endpoint drivers"
+	help
+	   Enable this configuration option to support configurable PCI
+	   endpoints at SPL. This should be enabled if the platform has
+	   a PCI controllers that can operate in endpoint mode (as a device
+	   connected to PCI host or bridge).
+
 config SPL_PCH
 	bool "Support PCH drivers"
 	help
@@ -1328,6 +1336,60 @@ config SPL_RAM_DEVICE
 	  be already in memory when SPL takes over, e.g. loaded by the boot
 	  ROM.
 
+config SPL_PCI_DFU
+	bool "PCIe boot support"
+	depends on SPL_PCI_ENDPOINT
+	help
+	  This config enables support to download bootloaders over PCIe
+	  when device is acting as an PCI endpoint.
+
+if SPL_PCI_DFU
+
+config PCI_DFU_SPL_LOAD_FIT_ADDRESS
+	hex "Address to load FIT image when booting via DFU over PCIe"
+	help
+	  Specify the load address of the fit image that will be loaded
+	  by SPL via DFU over PCIe.
+
+config PCI_DFU_BAR_SIZE
+	hex "BAR size to advertise for PCIe DFU"
+	default 0x800000
+	help
+	  This config sets the size of BAR to be advertised to the Root
+	  Complex. The size should be large enough to fit the FIT image
+	  being downloaded via DFU over PCIe.
+
+config PCI_DFU_MAGIC_WORD
+	hex "Completion magic word for PCIe DFU boot"
+	default 0xdeadbeef
+	help
+	  Specify the magic word which will be written to a specific
+	  address to signal the completion of transfer of FIT image
+	  when using DFU over PCIe to download the image. Size of magic
+	  word should be 32-bit.
+
+config PCI_DFU_VENDOR_ID
+	hex "PCI Vendor ID for PCI endpoint"
+	help
+	  PCI Vendor ID for endpoint device for DFU over PCIe. This should
+	  be set to your assigned 16-bit PCI Vendor ID.
+
+config PCI_DFU_DEVICE_ID
+	hex "PCI Vendor ID for PCI endpoint"
+	help
+	  A 16-bit PCI Vendor ID for endpoint device for DFU over PCIe.
+
+config PCI_DFU_BOOT_PHASE
+	string "Current boot phase for PCI DFU boot"
+	help
+	  Specify the current boot phase when booting via DFU over PCIe.
+	  This value can be read by the root complex to determine the
+	  current boot phase. Value of this config is written to memory
+	  location (BAR_start + PCI_DFU_BAR_SIZE - 70). Max size of this
+	  config is 63 bytes.
+
+endif
+
 config SPL_REMOTEPROC
 	bool "Support REMOTEPROCS"
 	default y if (CPU_V7R && ARCH_K3)
diff --git a/common/spl/spl_dfu.c b/common/spl/spl_dfu.c
index e9f381c392c..1fb5fb2bf6b 100644
--- a/common/spl/spl_dfu.c
+++ b/common/spl/spl_dfu.c
@@ -15,6 +15,17 @@
 #include <usb.h>
 #include <dfu.h>
 #include <linux/printk.h>
+#include <pci_ep.h>
+#include <dm/uclass.h>
+#include <cpu_func.h>
+#include <linux/io.h>
+
+/*
+ * Macros define size of magic word and boot phase string
+ * in bytes.
+ */
+#define MAGIC_WORD_SIZE 4
+#define BOOT_PHASE_STRING_SIZE 64
 
 static int run_dfu(int usb_index, char *interface, char *devstring)
 {
@@ -32,11 +43,93 @@ exit:
 	return ret;
 }
 
+#ifdef CONFIG_SPL_PCI_DFU
+static int dfu_over_pcie(void)
+{
+	u32 offset, magic_word;
+	void *addr;
+	struct udevice *dev;
+	struct pci_bar bar;
+	struct pci_ep_header hdr;
+	uint fn = 0;
+	int ret;
+	char *bootphase;
+
+	uclass_get_device_by_seq(UCLASS_PCI_EP, 0, &dev);
+	if (!dev) {
+		pr_err("Failed to get pci ep device\n");
+		return -ENODEV;
+	}
+
+	printf("%s: Configuring PCIe endpoint\n", __func__);
+	hdr.deviceid = CONFIG_PCI_DFU_DEVICE_ID;
+	hdr.vendorid = CONFIG_PCI_DFU_VENDOR_ID;
+	hdr.baseclass_code = PCI_BASE_CLASS_MEMORY;
+
+	ret = pci_ep_write_header(dev, fn, &hdr);
+	if (ret) {
+		pr_err("Failed to write header: %d\n", ret);
+		return ret;
+	}
+
+	bar.barno = BAR_0;
+	bar.phys_addr = (dma_addr_t)CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
+	bar.flags = PCI_BASE_ADDRESS_SPACE_MEMORY |
+			  PCI_BASE_ADDRESS_MEM_TYPE_32 |
+			  PCI_BASE_ADDRESS_MEM_PREFETCH;
+
+	bar.size = CONFIG_PCI_DFU_BAR_SIZE;
+
+	ret = pci_ep_set_bar(dev, fn, &bar);
+	if (ret) {
+		pr_err("Failed to set bar: %d\n", ret);
+		return ret;
+	}
+
+	ret = pci_ep_start(dev);
+	if (ret) {
+		pr_err("Failed to start ep: %d\n", ret);
+		return ret;
+	}
+
+	addr = (void *)CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
+	offset = CONFIG_PCI_DFU_BAR_SIZE - MAGIC_WORD_SIZE;
+
+	if (sizeof(CONFIG_PCI_DFU_BOOT_PHASE) > BOOT_PHASE_STRING_SIZE) {
+		pr_err("Boot phase string too long\n");
+	} else {
+		bootphase = (char *)(addr + CONFIG_PCI_DFU_BAR_SIZE -
+				  (BOOT_PHASE_STRING_SIZE + MAGIC_WORD_SIZE));
+		strncpy(bootphase, CONFIG_PCI_DFU_BOOT_PHASE,
+			sizeof(CONFIG_PCI_DFU_BOOT_PHASE));
+		bootphase[sizeof(CONFIG_PCI_DFU_BOOT_PHASE)] = '\0';
+	}
+
+	addr = addr + offset;
+	magic_word = CONFIG_PCI_DFU_MAGIC_WORD;
+	(*(int *)addr) = 0;
+	printf("%s: Waiting for an image .....\n", __func__);
+	for (;;) {
+		if (*(int *)addr == magic_word)
+			break;
+		invalidate_dcache_all();
+	}
+	printf("Image loaded\n");
+
+	return 0;
+}
+#endif
+
 int spl_dfu_cmd(int usbctrl, char *dfu_alt_info, char *interface, char *devstr)
 {
 	char *str_env;
 	int ret;
 
+#ifdef CONFIG_SPL_PCI_DFU
+	if (spl_boot_device() == BOOT_DEVICE_PCIE)
+		return dfu_over_pcie();
+#endif
+
 	/* set default environment */
 	env_set_default(NULL, 0);
 	str_env = env_get(dfu_alt_info);
diff --git a/common/spl/spl_ram.c b/common/spl/spl_ram.c
index 71b7a8374bb..c8a62a438bb 100644
--- a/common/spl/spl_ram.c
+++ b/common/spl/spl_ram.c
@@ -27,6 +27,11 @@ static ulong spl_ram_load_read(struct spl_load_info *load, ulong sector,
 	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT)) {
 		addr = IF_ENABLED_INT(CONFIG_SPL_LOAD_FIT,
 				      CONFIG_SPL_LOAD_FIT_ADDRESS);
+
+#ifdef	CONFIG_SPL_PCI_DFU
+		if (spl_boot_device() == BOOT_DEVICE_PCIE)
+			addr = CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
+#endif
 	}
 	addr += sector;
 	if (CONFIG_IS_ENABLED(IMAGE_PRE_LOAD))
@@ -47,6 +52,11 @@ static int spl_ram_load_image(struct spl_image_info *spl_image,
 	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT)) {
 		addr = IF_ENABLED_INT(CONFIG_SPL_LOAD_FIT,
 				      CONFIG_SPL_LOAD_FIT_ADDRESS);
+
+#ifdef	CONFIG_SPL_PCI_DFU
+		if (spl_boot_device() == BOOT_DEVICE_PCIE)
+			addr = CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
+#endif
 	}
 
 	if (CONFIG_IS_ENABLED(IMAGE_PRE_LOAD)) {
@@ -64,6 +74,11 @@ static int spl_ram_load_image(struct spl_image_info *spl_image,
 		spl_dfu_cmd(0, "dfu_alt_info_ram", "ram", "0");
 #endif
 
+#if CONFIG_IS_ENABLED(PCI_DFU)
+	if (bootdev->boot_device == BOOT_DEVICE_PCIE)
+		spl_dfu_cmd(0, "dfu_alt_info_ram", "ram", "0");
+#endif
+
 	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT) &&
 	    image_get_magic(header) == FDT_MAGIC) {
 		struct spl_load_info load;
@@ -102,3 +117,6 @@ SPL_LOAD_IMAGE_METHOD("RAM", 0, BOOT_DEVICE_RAM, spl_ram_load_image);
 #if CONFIG_IS_ENABLED(DFU)
 SPL_LOAD_IMAGE_METHOD("DFU", 0, BOOT_DEVICE_DFU, spl_ram_load_image);
 #endif
+#if CONFIG_IS_ENABLED(PCI_DFU)
+SPL_LOAD_IMAGE_METHOD("PCIE", 0, BOOT_DEVICE_PCIE, spl_ram_load_image);
+#endif
diff --git a/drivers/Makefile b/drivers/Makefile
index 3c0ad17138f..bd15e89adb0 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -25,6 +25,7 @@ obj-$(CONFIG_MULTIPLEXER) += mux/
 obj-$(CONFIG_$(PHASE_)ETH) += net/
 obj-$(CONFIG_$(PHASE_)PCH) += pch/
 obj-$(CONFIG_$(PHASE_)PCI) += pci/
+obj-$(CONFIG_$(PHASE_)PCI_ENDPOINT) += pci_endpoint/
 obj-$(CONFIG_$(PHASE_)PHY) += phy/
 obj-$(CONFIG_$(PHASE_)PINCTRL) += pinctrl/
 obj-$(CONFIG_$(PHASE_)POWER) += power/
-- 
2.34.1

Re: [PATCH v3] spl: Add support for Device Firmware Upgrade (DFU) over PCIe

[Mattijs Korpershoek]

Hi Hrushikesh,

Thank you for the patch.

On Thu, Jun 19, 2025 at 12:48, Hrushikesh Salunke <h-salunke@ti.com> wrote:

> Introduces support for Device Firmware Upgrade (DFU) over PCIe in
> U-Boot. Traditionally, the DFU protocol is used over USB, where a
> device enters DFU mode and allows a host to upload firmware or binary
> images directly via the USB interface. This is a widely adopted and
> convenient method for updating firmware.
>
> In the context of Texas Instruments (TI) SoCs, PCIe can be used as a
> boot interface in a manner that differs from the conventional
> "PCIe Boot" process, which typically refers to booting an OS or
> firmware image from an NVMe SSD or other PCIe-attached storage devices.
> Instead, TI SoCs can be configured as a PCIe Endpoint, allowing a
> connected PCIe Root Complex (host) to transfer images directly into the
> device’s memory over the PCIe bus for boot purposes. This mechanism is
> analogous to DFU over USB, but leverages the high-speed PCIe link and
> does not depend on traditional storage devices.
>
> By extending the DFU framework in U-Boot to support PCIe, it will be
> possible to flash images over PCIe. While this implementation is
> motivated by TI SoC use cases, the framework is generic and can be
> adopted by everyone for platforms that support PCIe Endpoint mode.
> Platforms with hardware support for PCIe-based memory loading can use
> this to implement PCIe as a boot mode, as well as to enable flashing
> and recovery scenarios similar to DFU over USB.
>
> In summary, enable support for:
> - DFU-style flashing of firmware/images over PCIe, analogous to existing
> USB DFU workflows
> - PCIe as a boot mode where a host can load images directly into device
> memory using DFU over PCIe
>
> Signed-off-by: Hrushikesh Salunke <h-salunke@ti.com>
> ---
> This patch is based on commit
> 9de873b4c30 Merge patch series "linux/bitfield.h: sync <linux/bitfield.h> from Linux 6.15 + winbond"
>
> Changes since v2
> Put the usage of macro BOOT_DEVICE_PCIE under #ifdef CONFIG_SPL_PCI_DFU
> to avoid potential compilation errors. This is required as 
> BOOT_DEVICE_PCIE macro is defined for only those platforms which support
> PCIE as a valid boot device.
>
> v2 : https://patchwork.ozlabs.org/project/uboot/patch/20250616164444.631575-1-h-salunke@ti.com/
>
>  common/spl/Kconfig   | 62 +++++++++++++++++++++++++++++
>  common/spl/spl_dfu.c | 93 ++++++++++++++++++++++++++++++++++++++++++++
>  common/spl/spl_ram.c | 18 +++++++++
>  drivers/Makefile     |  1 +
>  4 files changed, 174 insertions(+)
>
> diff --git a/common/spl/Kconfig b/common/spl/Kconfig
> index 880192043c4..212b4fc4e9e 100644
> --- a/common/spl/Kconfig
> +++ b/common/spl/Kconfig
> @@ -1268,6 +1268,14 @@ config SPL_PCI
>  	  necessary driver support. This enables the drivers in drivers/pci
>  	  as part of an SPL build.
>  
> +config SPL_PCI_ENDPOINT
> +	bool "Support for PCI endpoint drivers"
> +	help
> +	   Enable this configuration option to support configurable PCI
> +	   endpoints at SPL. This should be enabled if the platform has
> +	   a PCI controllers that can operate in endpoint mode (as a device
> +	   connected to PCI host or bridge).
> +
>  config SPL_PCH
>  	bool "Support PCH drivers"
>  	help
> @@ -1328,6 +1336,60 @@ config SPL_RAM_DEVICE
>  	  be already in memory when SPL takes over, e.g. loaded by the boot
>  	  ROM.
>  
> +config SPL_PCI_DFU
> +	bool "PCIe boot support"
> +	depends on SPL_PCI_ENDPOINT
> +	help
> +	  This config enables support to download bootloaders over PCIe
> +	  when device is acting as an PCI endpoint.
> +
> +if SPL_PCI_DFU
> +
> +config PCI_DFU_SPL_LOAD_FIT_ADDRESS

Can we please prefix this symbol and all others with SPL_ as well?

SPL_PCI_DFU_LOAD_FIT_ADDRESS

Rationale: if we ever get DFU over PCIe in U-Boot proper, there will be
a naming clash.

> +	hex "Address to load FIT image when booting via DFU over PCIe"
> +	help
> +	  Specify the load address of the fit image that will be loaded
> +	  by SPL via DFU over PCIe.
> +
> +config PCI_DFU_BAR_SIZE
> +	hex "BAR size to advertise for PCIe DFU"
> +	default 0x800000
> +	help
> +	  This config sets the size of BAR to be advertised to the Root
> +	  Complex. The size should be large enough to fit the FIT image
> +	  being downloaded via DFU over PCIe.
> +
> +config PCI_DFU_MAGIC_WORD
> +	hex "Completion magic word for PCIe DFU boot"
> +	default 0xdeadbeef
> +	help
> +	  Specify the magic word which will be written to a specific
> +	  address to signal the completion of transfer of FIT image
> +	  when using DFU over PCIe to download the image. Size of magic
> +	  word should be 32-bit.
> +
> +config PCI_DFU_VENDOR_ID
> +	hex "PCI Vendor ID for PCI endpoint"
> +	help
> +	  PCI Vendor ID for endpoint device for DFU over PCIe. This should
> +	  be set to your assigned 16-bit PCI Vendor ID.
> +
> +config PCI_DFU_DEVICE_ID
> +	hex "PCI Vendor ID for PCI endpoint"
> +	help
> +	  A 16-bit PCI Vendor ID for endpoint device for DFU over PCIe.
> +
> +config PCI_DFU_BOOT_PHASE
> +	string "Current boot phase for PCI DFU boot"
> +	help
> +	  Specify the current boot phase when booting via DFU over PCIe.
> +	  This value can be read by the root complex to determine the
> +	  current boot phase. Value of this config is written to memory
> +	  location (BAR_start + PCI_DFU_BAR_SIZE - 70). Max size of this
> +	  config is 63 bytes.
> +
> +endif
> +
>  config SPL_REMOTEPROC
>  	bool "Support REMOTEPROCS"
>  	default y if (CPU_V7R && ARCH_K3)
> diff --git a/common/spl/spl_dfu.c b/common/spl/spl_dfu.c
> index e9f381c392c..1fb5fb2bf6b 100644
> --- a/common/spl/spl_dfu.c
> +++ b/common/spl/spl_dfu.c
> @@ -15,6 +15,17 @@
>  #include <usb.h>
>  #include <dfu.h>
>  #include <linux/printk.h>
> +#include <pci_ep.h>
> +#include <dm/uclass.h>
> +#include <cpu_func.h>
> +#include <linux/io.h>
> +
> +/*
> + * Macros define size of magic word and boot phase string
> + * in bytes.
> + */
> +#define MAGIC_WORD_SIZE 4
> +#define BOOT_PHASE_STRING_SIZE 64
>  
>  static int run_dfu(int usb_index, char *interface, char *devstring)
>  {
> @@ -32,11 +43,93 @@ exit:
>  	return ret;
>  }
>  
> +#ifdef CONFIG_SPL_PCI_DFU
> +static int dfu_over_pcie(void)
> +{
> +	u32 offset, magic_word;
> +	void *addr;
> +	struct udevice *dev;
> +	struct pci_bar bar;
> +	struct pci_ep_header hdr;
> +	uint fn = 0;
> +	int ret;
> +	char *bootphase;
> +
> +	uclass_get_device_by_seq(UCLASS_PCI_EP, 0, &dev);
> +	if (!dev) {
> +		pr_err("Failed to get pci ep device\n");
> +		return -ENODEV;
> +	}
> +
> +	printf("%s: Configuring PCIe endpoint\n", __func__);

These seem like debug leftovers. Are we sure we need to be this verbose
in a normal operation mode?

> +	hdr.deviceid = CONFIG_PCI_DFU_DEVICE_ID;
> +	hdr.vendorid = CONFIG_PCI_DFU_VENDOR_ID;
> +	hdr.baseclass_code = PCI_BASE_CLASS_MEMORY;
> +
> +	ret = pci_ep_write_header(dev, fn, &hdr);
> +	if (ret) {
> +		pr_err("Failed to write header: %d\n", ret);
> +		return ret;
> +	}
> +
> +	bar.barno = BAR_0;
> +	bar.phys_addr = (dma_addr_t)CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
> +	bar.flags = PCI_BASE_ADDRESS_SPACE_MEMORY |
> +			  PCI_BASE_ADDRESS_MEM_TYPE_32 |
> +			  PCI_BASE_ADDRESS_MEM_PREFETCH;
> +
> +	bar.size = CONFIG_PCI_DFU_BAR_SIZE;
> +
> +	ret = pci_ep_set_bar(dev, fn, &bar);
> +	if (ret) {
> +		pr_err("Failed to set bar: %d\n", ret);
> +		return ret;
> +	}
> +
> +	ret = pci_ep_start(dev);
> +	if (ret) {
> +		pr_err("Failed to start ep: %d\n", ret);
> +		return ret;
> +	}
> +
> +	addr = (void *)CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
> +	offset = CONFIG_PCI_DFU_BAR_SIZE - MAGIC_WORD_SIZE;
> +
> +	if (sizeof(CONFIG_PCI_DFU_BOOT_PHASE) > BOOT_PHASE_STRING_SIZE) {
> +		pr_err("Boot phase string too long\n");

Maybe be more specific: "Not copying boot phase. String too long\n"

This way, we inform the user about both the problem and the action the
code has taken (to not copy)

> +	} else {
> +		bootphase = (char *)(addr + CONFIG_PCI_DFU_BAR_SIZE -
> +				  (BOOT_PHASE_STRING_SIZE + MAGIC_WORD_SIZE));
> +		strncpy(bootphase, CONFIG_PCI_DFU_BOOT_PHASE,

Can't we use strlcpy here instead of strlcpy ?

> +			sizeof(CONFIG_PCI_DFU_BOOT_PHASE));
> +		bootphase[sizeof(CONFIG_PCI_DFU_BOOT_PHASE)] = '\0';

What happens when CONFIG_PCI_DFU_BOOT_PHASE is exactly 64 characters?
Won't we truncate the last character here in order to put '\0' ?

If so, maybe we should test for 63 characters max (like mentioned in the
KConfig entry description) ?

> +	}
> +
> +	addr = addr + offset;
> +	magic_word = CONFIG_PCI_DFU_MAGIC_WORD;
> +	(*(int *)addr) = 0;
> +	printf("%s: Waiting for an image .....\n", __func__);

Is this also a debug leftover?

> +	for (;;) {
> +		if (*(int *)addr == magic_word)
> +			break;
> +		invalidate_dcache_all();
> +	}
> +	printf("Image loaded\n");
> +
> +	return 0;
> +}
> +#endif
> +
>  int spl_dfu_cmd(int usbctrl, char *dfu_alt_info, char *interface, char *devstr)
>  {
>  	char *str_env;
>  	int ret;
>  
> +#ifdef CONFIG_SPL_PCI_DFU

Can't we use if (IS_ENABLED(CONFIG...)) instead of #ifdef?

> +	if (spl_boot_device() == BOOT_DEVICE_PCIE)
> +		return dfu_over_pcie();
> +#endif
> +
>  	/* set default environment */
>  	env_set_default(NULL, 0);
>  	str_env = env_get(dfu_alt_info);
> diff --git a/common/spl/spl_ram.c b/common/spl/spl_ram.c
> index 71b7a8374bb..c8a62a438bb 100644
> --- a/common/spl/spl_ram.c
> +++ b/common/spl/spl_ram.c
> @@ -27,6 +27,11 @@ static ulong spl_ram_load_read(struct spl_load_info *load, ulong sector,
>  	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT)) {
>  		addr = IF_ENABLED_INT(CONFIG_SPL_LOAD_FIT,
>  				      CONFIG_SPL_LOAD_FIT_ADDRESS);
> +
> +#ifdef	CONFIG_SPL_PCI_DFU

Can't we use if (IS_ENABLED(CONFIG...)) instead of #ifdef?

Same for all occurences below

> +		if (spl_boot_device() == BOOT_DEVICE_PCIE)
> +			addr = CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
> +#endif
>  	}
>  	addr += sector;
>  	if (CONFIG_IS_ENABLED(IMAGE_PRE_LOAD))
> @@ -47,6 +52,11 @@ static int spl_ram_load_image(struct spl_image_info *spl_image,
>  	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT)) {
>  		addr = IF_ENABLED_INT(CONFIG_SPL_LOAD_FIT,
>  				      CONFIG_SPL_LOAD_FIT_ADDRESS);
> +
> +#ifdef	CONFIG_SPL_PCI_DFU
> +		if (spl_boot_device() == BOOT_DEVICE_PCIE)
> +			addr = CONFIG_PCI_DFU_SPL_LOAD_FIT_ADDRESS;
> +#endif
>  	}
>  
>  	if (CONFIG_IS_ENABLED(IMAGE_PRE_LOAD)) {
> @@ -64,6 +74,11 @@ static int spl_ram_load_image(struct spl_image_info *spl_image,
>  		spl_dfu_cmd(0, "dfu_alt_info_ram", "ram", "0");
>  #endif
>  
> +#if CONFIG_IS_ENABLED(PCI_DFU)
> +	if (bootdev->boot_device == BOOT_DEVICE_PCIE)
> +		spl_dfu_cmd(0, "dfu_alt_info_ram", "ram", "0");
> +#endif
> +
>  	if (IS_ENABLED(CONFIG_SPL_LOAD_FIT) &&
>  	    image_get_magic(header) == FDT_MAGIC) {
>  		struct spl_load_info load;
> @@ -102,3 +117,6 @@ SPL_LOAD_IMAGE_METHOD("RAM", 0, BOOT_DEVICE_RAM, spl_ram_load_image);
>  #if CONFIG_IS_ENABLED(DFU)
>  SPL_LOAD_IMAGE_METHOD("DFU", 0, BOOT_DEVICE_DFU, spl_ram_load_image);
>  #endif
> +#if CONFIG_IS_ENABLED(PCI_DFU)
> +SPL_LOAD_IMAGE_METHOD("PCIE", 0, BOOT_DEVICE_PCIE, spl_ram_load_image);
> +#endif
> diff --git a/drivers/Makefile b/drivers/Makefile
> index 3c0ad17138f..bd15e89adb0 100644
> --- a/drivers/Makefile
> +++ b/drivers/Makefile
> @@ -25,6 +25,7 @@ obj-$(CONFIG_MULTIPLEXER) += mux/
>  obj-$(CONFIG_$(PHASE_)ETH) += net/
>  obj-$(CONFIG_$(PHASE_)PCH) += pch/
>  obj-$(CONFIG_$(PHASE_)PCI) += pci/
> +obj-$(CONFIG_$(PHASE_)PCI_ENDPOINT) += pci_endpoint/
>  obj-$(CONFIG_$(PHASE_)PHY) += phy/
>  obj-$(CONFIG_$(PHASE_)PINCTRL) += pinctrl/
>  obj-$(CONFIG_$(PHASE_)POWER) += power/
> -- 
> 2.34.1