[PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[Brian Mak]

Hi all,

Here is v2 with the comments from v1 addressed. I also added a patch to
the series to add KEXEC_FILE_NO_CMA as a legal flag. I noticed that this
was missing when I went to add my KEXEC_FILE_FORCE_DTB flag.

Thanks,
Brian

Changes in v2:
- Added a patch to add KEXEC_FILE_NO_CMA as a legal flag
- Added a KEXEC_FILE_FORCE_DTB flag to enable carrying over the current
  boot's DTB on x86.
- Modified the commit message to include more reasoning for the change.
- Changed a pr_info print to a pr_debug print.

Brian Mak (2):
  kexec: Add KEXEC_FILE_NO_CMA as a legal flag
  x86/kexec: Carry forward the boot DTB on kexec

 arch/x86/kernel/kexec-bzimage64.c | 47 +++++++++++++++++++++++++++++--
 include/linux/kexec.h             |  6 +++-
 include/uapi/linux/kexec.h        |  4 +++
 kernel/kexec_file.c               |  1 +
 4 files changed, 54 insertions(+), 4 deletions(-)


base-commit: 35a813e010b99894bb4706c56c16a580bf7959c2
-- 
2.25.1

[PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Brian Mak]

Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
introduces logic to use CMA-based allocation in kexec by default. As
part of the changes, it introduces a kexec_file_load flag to disable the
use of CMA allocations from userspace. However, this flag is broken
since it is missing from the list of legal flags for kexec_file_load.
kexec_file_load returns EINVAL when attempting to use the flag.

Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
for kexec_file_load.

Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")
Signed-off-by: Brian Mak <makb@juniper.net>
---
 include/linux/kexec.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/linux/kexec.h b/include/linux/kexec.h
index 1b10a5d84b68..39fe3e6cd282 100644
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -460,7 +460,8 @@ bool kexec_load_permitted(int kexec_image_type);
 
 /* List of defined/legal kexec file flags */
 #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
-				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG)
+				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
+				 KEXEC_FILE_NO_CMA)
 
 /* flag to track if kexec reboot is in progress */
 extern bool kexec_in_progress;
-- 
2.25.1

[PATCH v2 2/2] x86/kexec: Carry forward the boot DTB on kexec

[Brian Mak]

Currently, the kexec_file_load syscall on x86 does not support passing a
device tree blob to the new kernel. Some embedded x86 systems use device
trees. On these systems, failing to pass a device tree to the new kernel
causes a boot failure.

To add support for this, we copy the behavior of ARM64 and PowerPC and
copy the current boot's device tree blob for use in the new kernel. We
do this on x86 by passing the device tree blob as a setup_data entry in
accordance with the x86 boot protocol.

This behavior is gated behind the KEXEC_FILE_FORCE_DTB flag.

Signed-off-by: Brian Mak <makb@juniper.net>
---
 arch/x86/kernel/kexec-bzimage64.c | 47 +++++++++++++++++++++++++++++--
 include/linux/kexec.h             |  5 +++-
 include/uapi/linux/kexec.h        |  4 +++
 kernel/kexec_file.c               |  1 +
 4 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/arch/x86/kernel/kexec-bzimage64.c b/arch/x86/kernel/kexec-bzimage64.c
index 24a41f0e0cf1..c3244ac680d1 100644
--- a/arch/x86/kernel/kexec-bzimage64.c
+++ b/arch/x86/kernel/kexec-bzimage64.c
@@ -16,6 +16,8 @@
 #include <linux/kexec.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/libfdt.h>
+#include <linux/of_fdt.h>
 #include <linux/efi.h>
 #include <linux/random.h>
 
@@ -212,6 +214,28 @@ setup_efi_state(struct boot_params *params, unsigned long params_load_addr,
 }
 #endif /* CONFIG_EFI */
 
+#ifdef CONFIG_OF_FLATTREE
+static void setup_dtb(struct boot_params *params,
+		      unsigned long params_load_addr,
+		      unsigned int dtb_setup_data_offset)
+{
+	struct setup_data *sd = (void *)params + dtb_setup_data_offset;
+	unsigned long setup_data_phys, dtb_len;
+
+	dtb_len = fdt_totalsize(initial_boot_params);
+	sd->type = SETUP_DTB;
+	sd->len = dtb_len;
+
+	/* Carry over current boot DTB with setup_data */
+	memcpy(sd->data, initial_boot_params, dtb_len);
+
+	/* Add setup data */
+	setup_data_phys = params_load_addr + dtb_setup_data_offset;
+	sd->next = params->hdr.setup_data;
+	params->hdr.setup_data = setup_data_phys;
+}
+#endif /* CONFIG_OF_FLATTREE */
+
 static void
 setup_ima_state(const struct kimage *image, struct boot_params *params,
 		unsigned long params_load_addr,
@@ -336,6 +360,17 @@ setup_boot_parameters(struct kimage *image, struct boot_params *params,
 			sizeof(struct efi_setup_data);
 #endif
 
+#ifdef CONFIG_OF_FLATTREE
+	if (image->force_dtb && initial_boot_params) {
+		setup_dtb(params, params_load_addr, setup_data_offset);
+		setup_data_offset += sizeof(struct setup_data) +
+				     fdt_totalsize(initial_boot_params);
+	} else {
+		pr_debug("Not carrying over DTB, force_dtb = %d\n",
+			 image->force_dtb);
+	}
+#endif
+
 	if (IS_ENABLED(CONFIG_IMA_KEXEC)) {
 		/* Setup IMA log buffer state */
 		setup_ima_state(image, params, params_load_addr,
@@ -529,6 +564,12 @@ static void *bzImage64_load(struct kimage *image, char *kernel,
 				sizeof(struct setup_data) +
 				RNG_SEED_LENGTH;
 
+#ifdef CONFIG_OF_FLATTREE
+	if (image->force_dtb && initial_boot_params)
+		kbuf.bufsz += sizeof(struct setup_data) +
+			      fdt_totalsize(initial_boot_params);
+#endif
+
 	if (IS_ENABLED(CONFIG_IMA_KEXEC))
 		kbuf.bufsz += sizeof(struct setup_data) +
 			      sizeof(struct ima_setup_data);
@@ -537,7 +578,7 @@ static void *bzImage64_load(struct kimage *image, char *kernel,
 		kbuf.bufsz += sizeof(struct setup_data) +
 			      sizeof(struct kho_data);
 
-	params = kzalloc(kbuf.bufsz, GFP_KERNEL);
+	params = kvzalloc(kbuf.bufsz, GFP_KERNEL);
 	if (!params)
 		return ERR_PTR(-ENOMEM);
 	efi_map_offset = params_cmdline_sz;
@@ -647,7 +688,7 @@ static void *bzImage64_load(struct kimage *image, char *kernel,
 	return ldata;
 
 out_free_params:
-	kfree(params);
+	kvfree(params);
 	return ERR_PTR(ret);
 }
 
@@ -659,7 +700,7 @@ static int bzImage64_cleanup(void *loader_data)
 	if (!ldata)
 		return 0;
 
-	kfree(ldata->bootparams_buf);
+	kvfree(ldata->bootparams_buf);
 	ldata->bootparams_buf = NULL;
 
 	return 0;
diff --git a/include/linux/kexec.h b/include/linux/kexec.h
index 39fe3e6cd282..ff7e231b0485 100644
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -395,6 +395,9 @@ struct kimage {
 
 	/* Information for loading purgatory */
 	struct purgatory_info purgatory_info;
+
+	/* Force carrying over the DTB from the current boot */
+	bool force_dtb;
 #endif
 
 #ifdef CONFIG_CRASH_HOTPLUG
@@ -461,7 +464,7 @@ bool kexec_load_permitted(int kexec_image_type);
 /* List of defined/legal kexec file flags */
 #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
 				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
-				 KEXEC_FILE_NO_CMA)
+				 KEXEC_FILE_NO_CMA | KEXEC_FILE_FORCE_DTB)
 
 /* flag to track if kexec reboot is in progress */
 extern bool kexec_in_progress;
diff --git a/include/uapi/linux/kexec.h b/include/uapi/linux/kexec.h
index 8958ebfcff94..55749cb0b81d 100644
--- a/include/uapi/linux/kexec.h
+++ b/include/uapi/linux/kexec.h
@@ -22,12 +22,16 @@
  * KEXEC_FILE_ON_CRASH : Load/unload operation belongs to kdump image.
  * KEXEC_FILE_NO_INITRAMFS : No initramfs is being loaded. Ignore the initrd
  *                           fd field.
+ * KEXEC_FILE_FORCE_DTB : Force carrying over the current boot's DTB to the new
+ *                        kernel on x86. This is already the default behavior on
+ *                        some other architectures, like ARM64 and PowerPC.
  */
 #define KEXEC_FILE_UNLOAD	0x00000001
 #define KEXEC_FILE_ON_CRASH	0x00000002
 #define KEXEC_FILE_NO_INITRAMFS	0x00000004
 #define KEXEC_FILE_DEBUG	0x00000008
 #define KEXEC_FILE_NO_CMA	0x00000010
+#define KEXEC_FILE_FORCE_DTB	0x00000020
 
 /* These values match the ELF architecture values.
  * Unless there is a good reason that should continue to be the case.
diff --git a/kernel/kexec_file.c b/kernel/kexec_file.c
index 91d46502a817..eb62a9794242 100644
--- a/kernel/kexec_file.c
+++ b/kernel/kexec_file.c
@@ -255,6 +255,7 @@ kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
 	}
 
 	image->no_cma = !!(flags & KEXEC_FILE_NO_CMA);
+	image->force_dtb = flags & KEXEC_FILE_FORCE_DTB;
 
 	if (cmdline_len) {
 		image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
-- 
2.25.1

Re: [PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[Brian Mak]

On Aug 5, 2025, at 2:15 PM, Brian Mak <makb@juniper.net> wrote:

> Hi all,
> 
> Here is v2 with the comments from v1 addressed. I also added a patch to
> the series to add KEXEC_FILE_NO_CMA as a legal flag. I noticed that this
> was missing when I went to add my KEXEC_FILE_FORCE_DTB flag.
> 
> Thanks,
> Brian
> 
> Changes in v2:
> - Added a patch to add KEXEC_FILE_NO_CMA as a legal flag
> - Added a KEXEC_FILE_FORCE_DTB flag to enable carrying over the current
>  boot's DTB on x86.
> - Modified the commit message to include more reasoning for the change.
> - Changed a pr_info print to a pr_debug print.
> 
> Brian Mak (2):
>  kexec: Add KEXEC_FILE_NO_CMA as a legal flag
>  x86/kexec: Carry forward the boot DTB on kexec
> 
> arch/x86/kernel/kexec-bzimage64.c | 47 +++++++++++++++++++++++++++++--
> include/linux/kexec.h             |  6 +++-
> include/uapi/linux/kexec.h        |  4 +++
> kernel/kexec_file.c               |  1 +
> 4 files changed, 54 insertions(+), 4 deletions(-)
> 
> 
> base-commit: 35a813e010b99894bb4706c56c16a580bf7959c2
> -- 
> 2.25.1

I see Andrew has sent these patches to linux-next. Thanks for that,
Andrew!

Dave and Baoquan, have you two had a chance to peek at the updated patch
set yet?

Thanks,
Brian

Re: [PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[Dave Young]

On Wed, 13 Aug 2025 at 02:01, Brian Mak <makb@juniper.net> wrote:
>
> On Aug 5, 2025, at 2:15 PM, Brian Mak <makb@juniper.net> wrote:
>
> > Hi all,
> >
> > Here is v2 with the comments from v1 addressed. I also added a patch to
> > the series to add KEXEC_FILE_NO_CMA as a legal flag. I noticed that this
> > was missing when I went to add my KEXEC_FILE_FORCE_DTB flag.
> >
> > Thanks,
> > Brian
> >
> > Changes in v2:
> > - Added a patch to add KEXEC_FILE_NO_CMA as a legal flag
> > - Added a KEXEC_FILE_FORCE_DTB flag to enable carrying over the current
> >  boot's DTB on x86.
> > - Modified the commit message to include more reasoning for the change.
> > - Changed a pr_info print to a pr_debug print.
> >
> > Brian Mak (2):
> >  kexec: Add KEXEC_FILE_NO_CMA as a legal flag
> >  x86/kexec: Carry forward the boot DTB on kexec
> >
> > arch/x86/kernel/kexec-bzimage64.c | 47 +++++++++++++++++++++++++++++--
> > include/linux/kexec.h             |  6 +++-
> > include/uapi/linux/kexec.h        |  4 +++
> > kernel/kexec_file.c               |  1 +
> > 4 files changed, 54 insertions(+), 4 deletions(-)
> >
> >
> > base-commit: 35a813e010b99894bb4706c56c16a580bf7959c2
> > --
> > 2.25.1
>
> I see Andrew has sent these patches to linux-next. Thanks for that,
> Andrew!
>
> Dave and Baoquan, have you two had a chance to peek at the updated patch
> set yet?

Hi, The #ifdef CONFIG_* which can be replaced by #if defined(), but I
do not have a strong opinion.  other than that do you have kexec-tools
patch ready? It would be better to provide a link so that people can
try and test it.

Thanks
Dave

Re: [PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[Brian Mak]

On Aug 12, 2025, at 8:54 PM, Dave Young <dyoung@redhat.com> wrote:

> Hi, The #ifdef CONFIG_* which can be replaced by #if defined(), but I
> do not have a strong opinion.  other than that do you have kexec-tools
> patch ready? It would be better to provide a link so that people can
> try and test it.

Hi Dave,

Thanks for looking at this! The #ifdef CONFIG_* is used, as it matches
the style of other ifdefs in this file. I'd like to keep it as-is for
consistency.

I do have a prototype kexec-tools patch ready for testing. The changes
can be found at this repo:

https://github.com/makb-juniper/kexec-tools.git

I've also placed the diff below.

Thanks,
Brian

diff --git a/kexec/kexec-syscall.h b/kexec/kexec-syscall.h
index 9b17578..8419b23 100644
--- a/kexec/kexec-syscall.h
+++ b/kexec/kexec-syscall.h
@@ -124,6 +124,8 @@ static inline long kexec_file_load(int kernel_fd, int initrd_fd,
 #define KEXEC_FILE_ON_CRASH	0x00000002
 #define KEXEC_FILE_NO_INITRAMFS	0x00000004
 #define KEXEC_FILE_DEBUG	0x00000008
+#define KEXEC_FILE_NO_CMA	0x00000010
+#define KEXEC_FILE_FORCE_DTB	0x00000020
 
 /* These values match the ELF architecture values. 
  * Unless there is a good reason that should continue to be the case.
diff --git a/kexec/kexec.c b/kexec/kexec.c
index 6bf12d7..0e13b6b 100644
--- a/kexec/kexec.c
+++ b/kexec/kexec.c
@@ -1102,6 +1102,7 @@ void usage(void)
 	       " --hotplug            Do in-kernel update of kexec segments on CPU/Memory\n"
 	       "                      hot add/remove events, avoiding the need to reload\n"
 	       "                      kdump kernel on online/offline events.\n"
+	       " --force-dtb          Carry over the current boot's device tree blob (x86 only).\n"
 	       " -d, --debug          Enable debugging to help spot a failure.\n"
 	       " -S, --status         Return 1 if the type (by default crash) is loaded,\n"
 	       "                      0 if not.\n"
@@ -1640,6 +1641,9 @@ int main(int argc, char *argv[])
 		case OPT_HOTPLUG:
 			do_hotplug = 1;
 			break;
+		case OPT_FORCE_DTB:
+			kexec_file_flags |= KEXEC_FILE_FORCE_DTB;
+			break;
 		default:
 			break;
 		}
diff --git a/kexec/kexec.h b/kexec/kexec.h
index a2e19c4..2235aa2 100644
--- a/kexec/kexec.h
+++ b/kexec/kexec.h
@@ -235,7 +235,8 @@ extern int file_types;
 #define OPT_LOAD_LIVE_UPDATE	263
 #define OPT_EXEC_LIVE_UPDATE	264
 #define OPT_HOTPLUG		        265
-#define OPT_MAX		266
+#define OPT_FORCE_DTB		266
+#define OPT_MAX		267
 #define KEXEC_OPTIONS \
 	{ "help",		0, 0, OPT_HELP }, \
 	{ "version",		0, 0, OPT_VERSION }, \
@@ -263,6 +264,7 @@ extern int file_types;
 	{ "status",		0, 0, OPT_STATUS }, \
 	{ "print-ckr-size",     0, 0, OPT_PRINT_CKR_SIZE }, \
 	{ "hotplug",		    0, 0, OPT_HOTPLUG }, \
+	{ "force-dtb",          0, 0, OPT_FORCE_DTB }, \
 
 #define KEXEC_OPT_STR "h?vdfixyluet:pscaS"
 

base-commit: daa29443819d3045338792b5ba950ed90e79d7a5
-- 
2.25.1

Re: [PATCH v2 0/2] x86/kexec: Carry forward the boot DTB on kexec

[Baoquan He]

On 08/13/25 at 07:24pm, Brian Mak wrote:
> On Aug 12, 2025, at 8:54 PM, Dave Young <dyoung@redhat.com> wrote:
> 
> > Hi, The #ifdef CONFIG_* which can be replaced by #if defined(), but I
> > do not have a strong opinion.  other than that do you have kexec-tools
> > patch ready? It would be better to provide a link so that people can
> > try and test it.
> 
> Hi Dave,
> 
> Thanks for looking at this! The #ifdef CONFIG_* is used, as it matches
> the style of other ifdefs in this file. I'd like to keep it as-is for
> consistency.
> 
> I do have a prototype kexec-tools patch ready for testing. The changes
> can be found at this repo:
> 
> https://github.com/makb-juniper/kexec-tools.git
> 
> I've also placed the diff below.

Thanks, I will apply your patch and take a test.

Usually, Andrew will merge kexec/kdump patch if the patch is OK to him.
I will add comment if I have concern. Otherwise, I don't want Andrew to
bother to add my ACK.

> 
> diff --git a/kexec/kexec-syscall.h b/kexec/kexec-syscall.h
> index 9b17578..8419b23 100644
> --- a/kexec/kexec-syscall.h
> +++ b/kexec/kexec-syscall.h
> @@ -124,6 +124,8 @@ static inline long kexec_file_load(int kernel_fd, int initrd_fd,
>  #define KEXEC_FILE_ON_CRASH	0x00000002
>  #define KEXEC_FILE_NO_INITRAMFS	0x00000004
>  #define KEXEC_FILE_DEBUG	0x00000008
> +#define KEXEC_FILE_NO_CMA	0x00000010
> +#define KEXEC_FILE_FORCE_DTB	0x00000020
>  
>  /* These values match the ELF architecture values. 
>   * Unless there is a good reason that should continue to be the case.
> diff --git a/kexec/kexec.c b/kexec/kexec.c
> index 6bf12d7..0e13b6b 100644
> --- a/kexec/kexec.c
> +++ b/kexec/kexec.c
> @@ -1102,6 +1102,7 @@ void usage(void)
>  	       " --hotplug            Do in-kernel update of kexec segments on CPU/Memory\n"
>  	       "                      hot add/remove events, avoiding the need to reload\n"
>  	       "                      kdump kernel on online/offline events.\n"
> +	       " --force-dtb          Carry over the current boot's device tree blob (x86 only).\n"
>  	       " -d, --debug          Enable debugging to help spot a failure.\n"
>  	       " -S, --status         Return 1 if the type (by default crash) is loaded,\n"
>  	       "                      0 if not.\n"
> @@ -1640,6 +1641,9 @@ int main(int argc, char *argv[])
>  		case OPT_HOTPLUG:
>  			do_hotplug = 1;
>  			break;
> +		case OPT_FORCE_DTB:
> +			kexec_file_flags |= KEXEC_FILE_FORCE_DTB;
> +			break;
>  		default:
>  			break;
>  		}
> diff --git a/kexec/kexec.h b/kexec/kexec.h
> index a2e19c4..2235aa2 100644
> --- a/kexec/kexec.h
> +++ b/kexec/kexec.h
> @@ -235,7 +235,8 @@ extern int file_types;
>  #define OPT_LOAD_LIVE_UPDATE	263
>  #define OPT_EXEC_LIVE_UPDATE	264
>  #define OPT_HOTPLUG		        265
> -#define OPT_MAX		266
> +#define OPT_FORCE_DTB		266
> +#define OPT_MAX		267
>  #define KEXEC_OPTIONS \
>  	{ "help",		0, 0, OPT_HELP }, \
>  	{ "version",		0, 0, OPT_VERSION }, \
> @@ -263,6 +264,7 @@ extern int file_types;
>  	{ "status",		0, 0, OPT_STATUS }, \
>  	{ "print-ckr-size",     0, 0, OPT_PRINT_CKR_SIZE }, \
>  	{ "hotplug",		    0, 0, OPT_HOTPLUG }, \
> +	{ "force-dtb",          0, 0, OPT_FORCE_DTB }, \
>  
>  #define KEXEC_OPT_STR "h?vdfixyluet:pscaS"
>  
> 
> base-commit: daa29443819d3045338792b5ba950ed90e79d7a5
> -- 
> 2.25.1
> 

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Andrew Morton]

On Tue, 5 Aug 2025 14:15:26 -0700 Brian Mak <makb@juniper.net> wrote:

> Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
> introduces logic to use CMA-based allocation in kexec by default. As
> part of the changes, it introduces a kexec_file_load flag to disable the
> use of CMA allocations from userspace. However, this flag is broken
> since it is missing from the list of legal flags for kexec_file_load.
> kexec_file_load returns EINVAL when attempting to use the flag.
> 
> Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
> for kexec_file_load.
> 
> Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")

A description of the userspace-visible runtime effects of this bug
would be very helpful, please.  A lot more than "is broken"!

Also, could we please have some reviewer input on this change?

Thanks.

> --- a/include/linux/kexec.h
> +++ b/include/linux/kexec.h
> @@ -460,7 +460,8 @@ bool kexec_load_permitted(int kexec_image_type);
>  
>  /* List of defined/legal kexec file flags */
>  #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
> -				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG)
> +				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
> +				 KEXEC_FILE_NO_CMA)
>  
>  /* flag to track if kexec reboot is in progress */
>  extern bool kexec_in_progress;
> -- 
> 2.25.1
> 

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Baoquan He]

On 08/20/25 at 09:47pm, Andrew Morton wrote:
> On Tue, 5 Aug 2025 14:15:26 -0700 Brian Mak <makb@juniper.net> wrote:
> 
> > Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
> > introduces logic to use CMA-based allocation in kexec by default. As
> > part of the changes, it introduces a kexec_file_load flag to disable the
> > use of CMA allocations from userspace. However, this flag is broken
> > since it is missing from the list of legal flags for kexec_file_load.
> > kexec_file_load returns EINVAL when attempting to use the flag.
> > 
> > Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
> > for kexec_file_load.
> > 
> > Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")
> 
> A description of the userspace-visible runtime effects of this bug
> would be very helpful, please.  A lot more than "is broken"!
> 
> Also, could we please have some reviewer input on this change?

I didn't receive this patchset, and kexec mailing list is not in CC.
I don't know what happened.

> 
> 
> > --- a/include/linux/kexec.h
> > +++ b/include/linux/kexec.h
> > @@ -460,7 +460,8 @@ bool kexec_load_permitted(int kexec_image_type);
> >  
> >  /* List of defined/legal kexec file flags */
> >  #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
> > -				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG)
> > +				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
> > +				 KEXEC_FILE_NO_CMA)
> >  
> >  /* flag to track if kexec reboot is in progress */
> >  extern bool kexec_in_progress;
> > -- 
> > 2.25.1
> > 
> 

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Andrew Morton]

On Thu, 21 Aug 2025 16:33:26 +0800 Baoquan He <bhe@redhat.com> wrote:

> On 08/20/25 at 09:47pm, Andrew Morton wrote:
> > On Tue, 5 Aug 2025 14:15:26 -0700 Brian Mak <makb@juniper.net> wrote:
> > 
> > > Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
> > > introduces logic to use CMA-based allocation in kexec by default. As
> > > part of the changes, it introduces a kexec_file_load flag to disable the
> > > use of CMA allocations from userspace. However, this flag is broken
> > > since it is missing from the list of legal flags for kexec_file_load.
> > > kexec_file_load returns EINVAL when attempting to use the flag.
> > > 
> > > Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
> > > for kexec_file_load.
> > > 
> > > Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")
> > 
> > A description of the userspace-visible runtime effects of this bug
> > would be very helpful, please.  A lot more than "is broken"!
> > 
> > Also, could we please have some reviewer input on this change?
> 
> I didn't receive this patchset, and kexec mailing list is not in CC.
> I don't know what happened.

Well that's annoying.  kexec@ and linux-kernel were on the cc line.

Here's what's in mm.git's mm-hotfixes branch:

From: Brian Mak <makb@juniper.net>
Subject: kexec: add KEXEC_FILE_NO_CMA as a legal flag
Date: Tue, 5 Aug 2025 14:15:26 -0700

Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
introduces logic to use CMA-based allocation in kexec by default.  As part
of the changes, it introduces a kexec_file_load flag to disable the use of
CMA allocations from userspace.  However, this flag is broken since it is
missing from the list of legal flags for kexec_file_load.  kexec_file_load
returns EINVAL when attempting to use the flag.

Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
for kexec_file_load.

Link: https://lkml.kernel.org/r/20250805211527.122367-2-makb@juniper.net
Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")
Signed-off-by: Brian Mak <makb@juniper.net>
Cc: Alexander Graf <graf@amazon.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Young <dyoung@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Rob Herring <robh@kernel.org>
Cc: Saravana Kannan <saravanak@google.com>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 include/linux/kexec.h |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

--- a/include/linux/kexec.h~kexec-add-kexec_file_no_cma-as-a-legal-flag
+++ a/include/linux/kexec.h
@@ -460,7 +460,8 @@ bool kexec_load_permitted(int kexec_imag
 
 /* List of defined/legal kexec file flags */
 #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
-				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG)
+				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
+				 KEXEC_FILE_NO_CMA)
 
 /* flag to track if kexec reboot is in progress */
 extern bool kexec_in_progress;
_


and the second patch I placed in mm-unstable:

From: Brian Mak <makb@juniper.net>
Subject: x86/kexec: carry forward the boot DTB on kexec
Date: Tue, 5 Aug 2025 14:15:27 -0700

Currently, the kexec_file_load syscall on x86 does not support passing a
device tree blob to the new kernel.  Some embedded x86 systems use device
trees.  On these systems, failing to pass a device tree to the new kernel
causes a boot failure.

To add support for this, we copy the behavior of ARM64 and PowerPC and
copy the current boot's device tree blob for use in the new kernel.  We do
this on x86 by passing the device tree blob as a setup_data entry in
accordance with the x86 boot protocol.

This behavior is gated behind the KEXEC_FILE_FORCE_DTB flag.

Link: https://lkml.kernel.org/r/20250805211527.122367-3-makb@juniper.net
Signed-off-by: Brian Mak <makb@juniper.net>
Cc: Alexander Graf <graf@amazon.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Young <dyoung@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Rob Herring <robh@kernel.org>
Cc: Saravana Kannan <saravanak@google.com>
Cc: Thomas Gleinxer <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 arch/x86/kernel/kexec-bzimage64.c |   47 ++++++++++++++++++++++++++--
 include/linux/kexec.h             |    5 ++
 include/uapi/linux/kexec.h        |    4 ++
 kernel/kexec_file.c               |    1 
 4 files changed, 53 insertions(+), 4 deletions(-)

--- a/arch/x86/kernel/kexec-bzimage64.c~x86-kexec-carry-forward-the-boot-dtb-on-kexec
+++ a/arch/x86/kernel/kexec-bzimage64.c
@@ -16,6 +16,8 @@
 #include <linux/kexec.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/libfdt.h>
+#include <linux/of_fdt.h>
 #include <linux/efi.h>
 #include <linux/random.h>
 
@@ -212,6 +214,28 @@ setup_efi_state(struct boot_params *para
 }
 #endif /* CONFIG_EFI */
 
+#ifdef CONFIG_OF_FLATTREE
+static void setup_dtb(struct boot_params *params,
+		      unsigned long params_load_addr,
+		      unsigned int dtb_setup_data_offset)
+{
+	struct setup_data *sd = (void *)params + dtb_setup_data_offset;
+	unsigned long setup_data_phys, dtb_len;
+
+	dtb_len = fdt_totalsize(initial_boot_params);
+	sd->type = SETUP_DTB;
+	sd->len = dtb_len;
+
+	/* Carry over current boot DTB with setup_data */
+	memcpy(sd->data, initial_boot_params, dtb_len);
+
+	/* Add setup data */
+	setup_data_phys = params_load_addr + dtb_setup_data_offset;
+	sd->next = params->hdr.setup_data;
+	params->hdr.setup_data = setup_data_phys;
+}
+#endif /* CONFIG_OF_FLATTREE */
+
 static void
 setup_ima_state(const struct kimage *image, struct boot_params *params,
 		unsigned long params_load_addr,
@@ -336,6 +360,17 @@ setup_boot_parameters(struct kimage *ima
 			sizeof(struct efi_setup_data);
 #endif
 
+#ifdef CONFIG_OF_FLATTREE
+	if (image->force_dtb && initial_boot_params) {
+		setup_dtb(params, params_load_addr, setup_data_offset);
+		setup_data_offset += sizeof(struct setup_data) +
+				     fdt_totalsize(initial_boot_params);
+	} else {
+		pr_debug("Not carrying over DTB, force_dtb = %d\n",
+			 image->force_dtb);
+	}
+#endif
+
 	if (IS_ENABLED(CONFIG_IMA_KEXEC)) {
 		/* Setup IMA log buffer state */
 		setup_ima_state(image, params, params_load_addr,
@@ -529,6 +564,12 @@ static void *bzImage64_load(struct kimag
 				sizeof(struct setup_data) +
 				RNG_SEED_LENGTH;
 
+#ifdef CONFIG_OF_FLATTREE
+	if (image->force_dtb && initial_boot_params)
+		kbuf.bufsz += sizeof(struct setup_data) +
+			      fdt_totalsize(initial_boot_params);
+#endif
+
 	if (IS_ENABLED(CONFIG_IMA_KEXEC))
 		kbuf.bufsz += sizeof(struct setup_data) +
 			      sizeof(struct ima_setup_data);
@@ -537,7 +578,7 @@ static void *bzImage64_load(struct kimag
 		kbuf.bufsz += sizeof(struct setup_data) +
 			      sizeof(struct kho_data);
 
-	params = kzalloc(kbuf.bufsz, GFP_KERNEL);
+	params = kvzalloc(kbuf.bufsz, GFP_KERNEL);
 	if (!params)
 		return ERR_PTR(-ENOMEM);
 	efi_map_offset = params_cmdline_sz;
@@ -647,7 +688,7 @@ static void *bzImage64_load(struct kimag
 	return ldata;
 
 out_free_params:
-	kfree(params);
+	kvfree(params);
 	return ERR_PTR(ret);
 }
 
@@ -659,7 +700,7 @@ static int bzImage64_cleanup(void *loade
 	if (!ldata)
 		return 0;
 
-	kfree(ldata->bootparams_buf);
+	kvfree(ldata->bootparams_buf);
 	ldata->bootparams_buf = NULL;
 
 	return 0;
--- a/include/linux/kexec.h~x86-kexec-carry-forward-the-boot-dtb-on-kexec
+++ a/include/linux/kexec.h
@@ -395,6 +395,9 @@ struct kimage {
 
 	/* Information for loading purgatory */
 	struct purgatory_info purgatory_info;
+
+	/* Force carrying over the DTB from the current boot */
+	bool force_dtb;
 #endif
 
 #ifdef CONFIG_CRASH_HOTPLUG
@@ -461,7 +464,7 @@ bool kexec_load_permitted(int kexec_imag
 /* List of defined/legal kexec file flags */
 #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
 				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
-				 KEXEC_FILE_NO_CMA)
+				 KEXEC_FILE_NO_CMA | KEXEC_FILE_FORCE_DTB)
 
 /* flag to track if kexec reboot is in progress */
 extern bool kexec_in_progress;
--- a/include/uapi/linux/kexec.h~x86-kexec-carry-forward-the-boot-dtb-on-kexec
+++ a/include/uapi/linux/kexec.h
@@ -22,12 +22,16 @@
  * KEXEC_FILE_ON_CRASH : Load/unload operation belongs to kdump image.
  * KEXEC_FILE_NO_INITRAMFS : No initramfs is being loaded. Ignore the initrd
  *                           fd field.
+ * KEXEC_FILE_FORCE_DTB : Force carrying over the current boot's DTB to the new
+ *                        kernel on x86. This is already the default behavior on
+ *                        some other architectures, like ARM64 and PowerPC.
  */
 #define KEXEC_FILE_UNLOAD	0x00000001
 #define KEXEC_FILE_ON_CRASH	0x00000002
 #define KEXEC_FILE_NO_INITRAMFS	0x00000004
 #define KEXEC_FILE_DEBUG	0x00000008
 #define KEXEC_FILE_NO_CMA	0x00000010
+#define KEXEC_FILE_FORCE_DTB	0x00000020
 
 /* These values match the ELF architecture values.
  * Unless there is a good reason that should continue to be the case.
--- a/kernel/kexec_file.c~x86-kexec-carry-forward-the-boot-dtb-on-kexec
+++ a/kernel/kexec_file.c
@@ -255,6 +255,7 @@ kimage_file_prepare_segments(struct kima
 	}
 
 	image->no_cma = !!(flags & KEXEC_FILE_NO_CMA);
+	image->force_dtb = flags & KEXEC_FILE_FORCE_DTB;
 
 	if (cmdline_len) {
 		image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
_

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Brian Mak]

On Aug 20, 2025, at 9:47 PM, Andrew Morton <akpm@linux-foundation.org> wrote:

> On Tue, 5 Aug 2025 14:15:26 -0700 Brian Mak <makb@juniper.net> wrote:
> 
>> Commit 07d24902977e ("kexec: enable CMA based contiguous allocation")
>> introduces logic to use CMA-based allocation in kexec by default. As
>> part of the changes, it introduces a kexec_file_load flag to disable the
>> use of CMA allocations from userspace. However, this flag is broken
>> since it is missing from the list of legal flags for kexec_file_load.
>> kexec_file_load returns EINVAL when attempting to use the flag.
>> 
>> Fix this by adding the KEXEC_FILE_NO_CMA flag to the list of legal flags
>> for kexec_file_load.
>> 
>> Fixes: 07d24902977e ("kexec: enable CMA based contiguous allocation")
> 
> A description of the userspace-visible runtime effects of this bug
> would be very helpful, please.  A lot more than "is broken"!

Hi Andrew,

Thanks for your comment. The userspace-visible runtime effect of this
bug is that kexec_file_load returns EINVAL when attempting to use the
KEXEC_FILE_NO_CMA flag. This is stated in the sentence following the
"this flag is broken" statement.

Is there something else that you wanted me to add here?

Thanks,
Brian

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Baoquan He]

On 08/21/25 at 04:53am, Andrew Morton wrote:
> On Thu, 21 Aug 2025 16:33:26 +0800 Baoquan He <bhe@redhat.com> wrote:
> 
> > On 08/20/25 at 09:47pm, Andrew Morton wrote:
> > > On Tue, 5 Aug 2025 14:15:26 -0700 Brian Mak <makb@juniper.net> wrote:
......snip.....
> ---
> 
>  include/linux/kexec.h |    3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> --- a/include/linux/kexec.h~kexec-add-kexec_file_no_cma-as-a-legal-flag
> +++ a/include/linux/kexec.h
> @@ -460,7 +460,8 @@ bool kexec_load_permitted(int kexec_imag
>  
>  /* List of defined/legal kexec file flags */
>  #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
> -				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG)
> +				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
> +				 KEXEC_FILE_NO_CMA)
>  
>  /* flag to track if kexec reboot is in progress */
>  extern bool kexec_in_progress;

Yeah, this is a good catch and great fix. Without this fix,
kexec_file_load syscall will failed and return '-EINVAL' when
KEXEC_FILE_NO_CMA is specified just as below code shows. So, for this
patch, 

Acked-by: Baoquan He <bhe@redhat.com>


And, by the way, has the user space kexec-tools got the change merged
to allow KEXEC_FILE_NO_CMA specified?

And, Alexander, I am wondering why this is not captured when you test
specifying KEXEC_FILE_NO_CMA case. Or you just skip the no_cma case
testing?

===================================================================
SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, 
                unsigned long, cmdline_len, const char __user *, cmdline_ptr,
                unsigned long, flags)
{               
        int image_type = (flags & KEXEC_FILE_ON_CRASH) ?
                         KEXEC_TYPE_CRASH : KEXEC_TYPE_DEFAULT;
        struct kimage **dest_image, *image;
        int ret = 0, i;

        /* We only trust the superuser with rebooting the system. */
        if (!kexec_load_permitted(image_type))
                return -EPERM;

        /* Make sure we have a legal set of flags */
        if (flags != (flags & KEXEC_FILE_FLAGS))
                return -EINVAL;
         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
	......
}
=====================================================

> _
> 
> 
> and the second patch I placed in mm-unstable:
> 
> From: Brian Mak <makb@juniper.net>
> Subject: x86/kexec: carry forward the boot DTB on kexec
> Date: Tue, 5 Aug 2025 14:15:27 -0700
> 
> Currently, the kexec_file_load syscall on x86 does not support passing a
> device tree blob to the new kernel.  Some embedded x86 systems use device
> trees.  On these systems, failing to pass a device tree to the new kernel
> causes a boot failure.
> 
> To add support for this, we copy the behavior of ARM64 and PowerPC and
> copy the current boot's device tree blob for use in the new kernel.  We do
> this on x86 by passing the device tree blob as a setup_data entry in
> accordance with the x86 boot protocol.
> 
> This behavior is gated behind the KEXEC_FILE_FORCE_DTB flag.
> 
> Link: https://lkml.kernel.org/r/20250805211527.122367-3-makb@juniper.net
> Signed-off-by: Brian Mak <makb@juniper.net>
> Cc: Alexander Graf <graf@amazon.com>
> Cc: Baoquan He <bhe@redhat.com>
> Cc: Borislav Betkov <bp@alien8.de>
> Cc: Dave Young <dyoung@redhat.com>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Rob Herring <robh@kernel.org>
> Cc: Saravana Kannan <saravanak@google.com>
> Cc: Thomas Gleinxer <tglx@linutronix.de>
> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> ---
> 
>  arch/x86/kernel/kexec-bzimage64.c |   47 ++++++++++++++++++++++++++--
>  include/linux/kexec.h             |    5 ++
>  include/uapi/linux/kexec.h        |    4 ++
>  kernel/kexec_file.c               |    1 
>  4 files changed, 53 insertions(+), 4 deletions(-)
> 
> --- a/arch/x86/kernel/kexec-bzimage64.c~x86-kexec-carry-forward-the-boot-dtb-on-kexec
> +++ a/arch/x86/kernel/kexec-bzimage64.c
> @@ -16,6 +16,8 @@
>  #include <linux/kexec.h>
>  #include <linux/kernel.h>
>  #include <linux/mm.h>
> +#include <linux/libfdt.h>
> +#include <linux/of_fdt.h>
>  #include <linux/efi.h>
>  #include <linux/random.h>
>  
> @@ -212,6 +214,28 @@ setup_efi_state(struct boot_params *para
>  }
>  #endif /* CONFIG_EFI */
>  
> +#ifdef CONFIG_OF_FLATTREE
> +static void setup_dtb(struct boot_params *params,
> +		      unsigned long params_load_addr,
> +		      unsigned int dtb_setup_data_offset)
> +{
> +	struct setup_data *sd = (void *)params + dtb_setup_data_offset;
> +	unsigned long setup_data_phys, dtb_len;
> +
> +	dtb_len = fdt_totalsize(initial_boot_params);
> +	sd->type = SETUP_DTB;
> +	sd->len = dtb_len;
> +
> +	/* Carry over current boot DTB with setup_data */
> +	memcpy(sd->data, initial_boot_params, dtb_len);
> +
> +	/* Add setup data */
> +	setup_data_phys = params_load_addr + dtb_setup_data_offset;
> +	sd->next = params->hdr.setup_data;
> +	params->hdr.setup_data = setup_data_phys;
> +}
> +#endif /* CONFIG_OF_FLATTREE */
> +
>  static void
>  setup_ima_state(const struct kimage *image, struct boot_params *params,
>  		unsigned long params_load_addr,
> @@ -336,6 +360,17 @@ setup_boot_parameters(struct kimage *ima
>  			sizeof(struct efi_setup_data);
>  #endif
>  
> +#ifdef CONFIG_OF_FLATTREE
> +	if (image->force_dtb && initial_boot_params) {
> +		setup_dtb(params, params_load_addr, setup_data_offset);
> +		setup_data_offset += sizeof(struct setup_data) +
> +				     fdt_totalsize(initial_boot_params);
> +	} else {
> +		pr_debug("Not carrying over DTB, force_dtb = %d\n",
> +			 image->force_dtb);
> +	}
> +#endif
> +
>  	if (IS_ENABLED(CONFIG_IMA_KEXEC)) {
>  		/* Setup IMA log buffer state */
>  		setup_ima_state(image, params, params_load_addr,
> @@ -529,6 +564,12 @@ static void *bzImage64_load(struct kimag
>  				sizeof(struct setup_data) +
>  				RNG_SEED_LENGTH;
>  
> +#ifdef CONFIG_OF_FLATTREE
> +	if (image->force_dtb && initial_boot_params)
> +		kbuf.bufsz += sizeof(struct setup_data) +
> +			      fdt_totalsize(initial_boot_params);
> +#endif
> +
>  	if (IS_ENABLED(CONFIG_IMA_KEXEC))
>  		kbuf.bufsz += sizeof(struct setup_data) +
>  			      sizeof(struct ima_setup_data);
> @@ -537,7 +578,7 @@ static void *bzImage64_load(struct kimag
>  		kbuf.bufsz += sizeof(struct setup_data) +
>  			      sizeof(struct kho_data);
>  
> -	params = kzalloc(kbuf.bufsz, GFP_KERNEL);
> +	params = kvzalloc(kbuf.bufsz, GFP_KERNEL);

Wondering how big the dtb blob is, can you explain a little bit about
the kvzalloc usage here?

Except of this, I have no other concern about this patch.

And what's your plan about the user space kexec-tool change?

>  	if (!params)
>  		return ERR_PTR(-ENOMEM);
>  	efi_map_offset = params_cmdline_sz;
> @@ -647,7 +688,7 @@ static void *bzImage64_load(struct kimag
>  	return ldata;
>  
>  out_free_params:
> -	kfree(params);
> +	kvfree(params);
>  	return ERR_PTR(ret);
>  }
>  
> @@ -659,7 +700,7 @@ static int bzImage64_cleanup(void *loade
>  	if (!ldata)
>  		return 0;
>  
> -	kfree(ldata->bootparams_buf);
> +	kvfree(ldata->bootparams_buf);
>  	ldata->bootparams_buf = NULL;
>  
>  	return 0;
> --- a/include/linux/kexec.h~x86-kexec-carry-forward-the-boot-dtb-on-kexec
> +++ a/include/linux/kexec.h
> @@ -395,6 +395,9 @@ struct kimage {
>  
>  	/* Information for loading purgatory */
>  	struct purgatory_info purgatory_info;
> +
> +	/* Force carrying over the DTB from the current boot */
> +	bool force_dtb;
>  #endif
>  
>  #ifdef CONFIG_CRASH_HOTPLUG
> @@ -461,7 +464,7 @@ bool kexec_load_permitted(int kexec_imag
>  /* List of defined/legal kexec file flags */
>  #define KEXEC_FILE_FLAGS	(KEXEC_FILE_UNLOAD | KEXEC_FILE_ON_CRASH | \
>  				 KEXEC_FILE_NO_INITRAMFS | KEXEC_FILE_DEBUG | \
> -				 KEXEC_FILE_NO_CMA)
> +				 KEXEC_FILE_NO_CMA | KEXEC_FILE_FORCE_DTB)
>  
>  /* flag to track if kexec reboot is in progress */
>  extern bool kexec_in_progress;
> --- a/include/uapi/linux/kexec.h~x86-kexec-carry-forward-the-boot-dtb-on-kexec
> +++ a/include/uapi/linux/kexec.h
> @@ -22,12 +22,16 @@
>   * KEXEC_FILE_ON_CRASH : Load/unload operation belongs to kdump image.
>   * KEXEC_FILE_NO_INITRAMFS : No initramfs is being loaded. Ignore the initrd
>   *                           fd field.
> + * KEXEC_FILE_FORCE_DTB : Force carrying over the current boot's DTB to the new
> + *                        kernel on x86. This is already the default behavior on
> + *                        some other architectures, like ARM64 and PowerPC.
>   */
>  #define KEXEC_FILE_UNLOAD	0x00000001
>  #define KEXEC_FILE_ON_CRASH	0x00000002
>  #define KEXEC_FILE_NO_INITRAMFS	0x00000004
>  #define KEXEC_FILE_DEBUG	0x00000008
>  #define KEXEC_FILE_NO_CMA	0x00000010
> +#define KEXEC_FILE_FORCE_DTB	0x00000020
>  
>  /* These values match the ELF architecture values.
>   * Unless there is a good reason that should continue to be the case.
> --- a/kernel/kexec_file.c~x86-kexec-carry-forward-the-boot-dtb-on-kexec
> +++ a/kernel/kexec_file.c
> @@ -255,6 +255,7 @@ kimage_file_prepare_segments(struct kima
>  	}
>  
>  	image->no_cma = !!(flags & KEXEC_FILE_NO_CMA);
> +	image->force_dtb = flags & KEXEC_FILE_FORCE_DTB;
>  
>  	if (cmdline_len) {
>  		image->cmdline_buf = memdup_user(cmdline_ptr, cmdline_len);
> _
> 

Re: [PATCH v2 1/2] kexec: Add KEXEC_FILE_NO_CMA as a legal flag

[Brian Mak]

On Aug 21, 2025, at 8:33 PM, Baoquan He <bhe@redhat.com> wrote:

> Yeah, this is a good catch and great fix. Without this fix,
> kexec_file_load syscall will failed and return '-EINVAL' when
> KEXEC_FILE_NO_CMA is specified just as below code shows. So, for this
> patch,
> 
> Acked-by: Baoquan He <bhe@redhat.com>

Hi Baoquan,

Thanks for the ACK!

> And, by the way, has the user space kexec-tools got the change merged
> to allow KEXEC_FILE_NO_CMA specified?

I don't see any recent commits to kexec-tools to support
KEXEC_FILE_NO_CMA.

>> From: Brian Mak <makb@juniper.net>
>> Subject: x86/kexec: carry forward the boot DTB on kexec
>> Date: Tue, 5 Aug 2025 14:15:27 -0700
>> 
>> Currently, the kexec_file_load syscall on x86 does not support passing a
>> device tree blob to the new kernel.  Some embedded x86 systems use device
>> trees.  On these systems, failing to pass a device tree to the new kernel
>> causes a boot failure.
>> 
>> To add support for this, we copy the behavior of ARM64 and PowerPC and
>> copy the current boot's device tree blob for use in the new kernel.  We do
>> this on x86 by passing the device tree blob as a setup_data entry in
>> accordance with the x86 boot protocol.
>> 
>> This behavior is gated behind the KEXEC_FILE_FORCE_DTB flag.
>> 
>> Link: https://urldefense.com/v3/__https://lkml.kernel.org/r/20250805211527.122367-3-makb@juniper.net__;!!NEt6yMaO-gk!EbJyF8xO2E51MyYdN3_zqCBBMj0JKoiKoPuG_8vEctQMk9uCyjX0LdSEH_FGkPDV8egxzc7w$
>> Signed-off-by: Brian Mak <makb@juniper.net>
>> Cc: Alexander Graf <graf@amazon.com>
>> Cc: Baoquan He <bhe@redhat.com>
>> Cc: Borislav Betkov <bp@alien8.de>
>> Cc: Dave Young <dyoung@redhat.com>
>> Cc: "H. Peter Anvin" <hpa@zytor.com>
>> Cc: Ingo Molnar <mingo@redhat.com>
>> Cc: Rob Herring <robh@kernel.org>
>> Cc: Saravana Kannan <saravanak@google.com>
>> Cc: Thomas Gleinxer <tglx@linutronix.de>
>> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
>> ---
>> 
>> arch/x86/kernel/kexec-bzimage64.c |   47 ++++++++++++++++++++++++++--
>> include/linux/kexec.h             |    5 ++
>> include/uapi/linux/kexec.h        |    4 ++
>> kernel/kexec_file.c               |    1
>> 4 files changed, 53 insertions(+), 4 deletions(-)
>> 
>> --- a/arch/x86/kernel/kexec-bzimage64.c~x86-kexec-carry-forward-the-boot-dtb-on-kexec
>> +++ a/arch/x86/kernel/kexec-bzimage64.c
>> @@ -16,6 +16,8 @@
>> #include <linux/kexec.h>
>> #include <linux/kernel.h>
>> #include <linux/mm.h>
>> +#include <linux/libfdt.h>
>> +#include <linux/of_fdt.h>
>> #include <linux/efi.h>
>> #include <linux/random.h>
>> 
>> @@ -212,6 +214,28 @@ setup_efi_state(struct boot_params *para
>> }
>> #endif /* CONFIG_EFI */
>> 
>> +#ifdef CONFIG_OF_FLATTREE
>> +static void setup_dtb(struct boot_params *params,
>> +                   unsigned long params_load_addr,
>> +                   unsigned int dtb_setup_data_offset)
>> +{
>> +     struct setup_data *sd = (void *)params + dtb_setup_data_offset;
>> +     unsigned long setup_data_phys, dtb_len;
>> +
>> +     dtb_len = fdt_totalsize(initial_boot_params);
>> +     sd->type = SETUP_DTB;
>> +     sd->len = dtb_len;
>> +
>> +     /* Carry over current boot DTB with setup_data */
>> +     memcpy(sd->data, initial_boot_params, dtb_len);
>> +
>> +     /* Add setup data */
>> +     setup_data_phys = params_load_addr + dtb_setup_data_offset;
>> +     sd->next = params->hdr.setup_data;
>> +     params->hdr.setup_data = setup_data_phys;
>> +}
>> +#endif /* CONFIG_OF_FLATTREE */
>> +
>> static void
>> setup_ima_state(const struct kimage *image, struct boot_params *params,
>>              unsigned long params_load_addr,
>> @@ -336,6 +360,17 @@ setup_boot_parameters(struct kimage *ima
>>                      sizeof(struct efi_setup_data);
>> #endif
>> 
>> +#ifdef CONFIG_OF_FLATTREE
>> +     if (image->force_dtb && initial_boot_params) {
>> +             setup_dtb(params, params_load_addr, setup_data_offset);
>> +             setup_data_offset += sizeof(struct setup_data) +
>> +                                  fdt_totalsize(initial_boot_params);
>> +     } else {
>> +             pr_debug("Not carrying over DTB, force_dtb = %d\n",
>> +                      image->force_dtb);
>> +     }
>> +#endif
>> +
>>      if (IS_ENABLED(CONFIG_IMA_KEXEC)) {
>>              /* Setup IMA log buffer state */
>>              setup_ima_state(image, params, params_load_addr,
>> @@ -529,6 +564,12 @@ static void *bzImage64_load(struct kimag
>>                              sizeof(struct setup_data) +
>>                              RNG_SEED_LENGTH;
>> 
>> +#ifdef CONFIG_OF_FLATTREE
>> +     if (image->force_dtb && initial_boot_params)
>> +             kbuf.bufsz += sizeof(struct setup_data) +
>> +                           fdt_totalsize(initial_boot_params);
>> +#endif
>> +
>>      if (IS_ENABLED(CONFIG_IMA_KEXEC))
>>              kbuf.bufsz += sizeof(struct setup_data) +
>>                            sizeof(struct ima_setup_data);
>> @@ -537,7 +578,7 @@ static void *bzImage64_load(struct kimag
>>              kbuf.bufsz += sizeof(struct setup_data) +
>>                            sizeof(struct kho_data);
>> 
>> -     params = kzalloc(kbuf.bufsz, GFP_KERNEL);
>> +     params = kvzalloc(kbuf.bufsz, GFP_KERNEL);
> 
> Wondering how big the dtb blob is, can you explain a little bit about
> the kvzalloc usage here?
> 
> Except of this, I have no other concern about this patch.
> 
> And what's your plan about the user space kexec-tool change?

When I tested this earlier on x86, the DTB was allowed to be up to just
under 64 pages large before the DTB failed to load. This is because it
has to fit into an early_memremap() mapping (relevant code snippet at
the bottom). Since the allocation can be many pages, I changed the
kzalloc to a kvzalloc.

For the kexec-tools change, I have a draft change that I've already
shared on this thread for testing purposes. I believe you said you were
going to test it, but I haven't heard anything back from that yet. I'll
raise that change for review properly once this kernel commit is in
mainline.

---------

void __init x86_flattree_get_config(void)
{
#ifdef CONFIG_OF_EARLY_FLATTREE
	u32 size, map_len;
	void *dt;

	if (initial_dtb) {
		map_len = max(PAGE_SIZE - (initial_dtb & ~PAGE_MASK), (u64)128);

		dt = early_memremap(initial_dtb, map_len);
		size = fdt_totalsize(dt);
		if (map_len < size) {
			early_memunmap(dt, map_len);
			dt = early_memremap(initial_dtb, size);
			map_len = size;
		}

		early_init_dt_verify(dt, __pa(dt));
	}

	unflatten_and_copy_device_tree();

	if (initial_dtb)
		early_memunmap(dt, map_len);
#endif
	if (acpi_disabled && of_have_populated_dt())
		x86_init.mpparse.parse_smp_cfg = x86_dtb_parse_smp_config;
}

---------

Thanks,
Brian