Skip to contentSkip to conversation

kho: skip KHO for crash kernel

PATCH Applied Evangelos Petrongonas 3 participants 6 messages Apr 10 – Apr 28

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 kho: skip KHO for crash kernel

kho_fill_kimage() unconditionally populates the kimage with KHO metadata for every kexec image type. When the image is a crash kernel, this can be problematic as the crash kernel can run in a small reserved region and the KHO scratch areas can sit outside it.
The crash kernel then faults during kho_memory_init() when it tries phys_to_virt() on the KHO FDT address:

  Unable to handle kernel paging request at virtual address xxxxxxxx
  ...
    fdt_offset_ptr+...
    fdt_check_node_offset_+...
    fdt_first_property_offset+...
    fdt_get_property_namelen_+...
    fdt_getprop+...
    kho_memory_init+...
    mm_core_init+...
    start_kernel+...

kho_locate_mem_hole() already skips KHO logic for KEXEC_TYPE_CRASH images, but kho_fill_kimage() was missing the same guard. As kho_fill_kimage() is the single point that populates image->kho.fdt and image->kho.scratch, fixing it here is sufficient for both arm64 and x86 as the FDT and boot_params path are bailing out when these fields are unset.

Fixes d7255959b69a ("kho: allow kexec load before KHO finalization")

---

v2: Per Mike's review [1], move the guard into kho_fill_kimage() instead
of patching the arch-level producers and consumers. This fixes both arm64 and x86 in one place and avoids redundant checks. Tested again.

Note regarding backporting
The offending commit was deployed with 6.19. The only other supported kernel version with 6.18, unless I miss someting uses

```
if (!kho_out.finalized)
```

which in the case of crash kernel it shouldn't be finalised.

[1] https://lore.kernel.org/all/[email protected]/

 kernel/liveupdate/kexec_handover.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/liveupdate/kexec_handover.c +1 -1
index cc68a3692905..1029fe8778f2 100644
@@ -1551,7 +1551,7 @@ int kho_fill_kimage(struct kimage *image)
 	int err = 0;
 	struct kexec_buf scratch;
 
-	if (!kho_enable)
+	if (!kho_enable || image->type == KEXEC_TYPE_CRASH)
 		return 0;
 
 	image->kho.fdt = virt_to_phys(kho_out.fdt);
Signature · 12 lines
2.47.3




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597
maintainer Reviewed-by on v2 1/1
#1 lore raw
Quoted 25 lines · Evangelos Petrongonas

kho_fill_kimage() unconditionally populates the kimage with KHO metadata for every kexec image type. When the image is a crash kernel, this can be problematic as the crash kernel can run in a small reserved region and the KHO scratch areas can sit outside it.
The crash kernel then faults during kho_memory_init() when it tries phys_to_virt() on the KHO FDT address:

  Unable to handle kernel paging request at virtual address xxxxxxxx
  ...
    fdt_offset_ptr+...
    fdt_check_node_offset_+...
    fdt_first_property_offset+...
    fdt_get_property_namelen_+...
    fdt_getprop+...
    kho_memory_init+...
    mm_core_init+...
    start_kernel+...

kho_locate_mem_hole() already skips KHO logic for KEXEC_TYPE_CRASH images, but kho_fill_kimage() was missing the same guard. As kho_fill_kimage() is the single point that populates image->kho.fdt and image->kho.scratch, fixing it here is sufficient for both arm64 and x86 as the FDT and boot_params path are bailing out when these fields are unset.

Quoted 2 lines
Fixes d7255959b69a ("kho: allow kexec load before KHO finalization")
Quoted 6 lines

---

v2: Per Mike's review [1], move the guard into kho_fill_kimage() instead
of patching the arch-level producers and consumers. This fixes both arm64 and x86 in one place and avoids redundant checks. Tested again.

Quoted 7 lines

Note regarding backporting
The offending commit was deployed with 6.19. The only other supported kernel version with 6.18, unless I miss someting uses

```
if (!kho_out.finalized)
```

which in the case of crash kernel it shouldn't be finalised.

Yes, this seems about right :)
The only released kernel that has this issue is v6.19 and it will be EOL in less than a week.

Quoted 32 lines

[1] https://lore.kernel.org/all/[email protected]/

 kernel/liveupdate/kexec_handover.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/liveupdate/kexec_handover.c +1 -1
index cc68a3692905..1029fe8778f2 100644
@@ -1551,7 +1551,7 @@ int kho_fill_kimage(struct kimage *image)
 	int err = 0;
 	struct kexec_buf scratch;
 
-	if (!kho_enable)
+	if (!kho_enable || image->type == KEXEC_TYPE_CRASH)
 		return 0;
 
 	image->kho.fdt = virt_to_phys(kho_out.fdt);
Signature · 12 lines
2.47.3




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597
Signature · 2 lines
Sincerely yours,
Mike.
maintainer
#2 lore raw
In reply to Evangelos Petrongonas

Hi Evangelos,

Quoted 33 lines · Evangelos Petrongonas

kho_fill_kimage() unconditionally populates the kimage with KHO metadata for every kexec image type. When the image is a crash kernel, this can be problematic as the crash kernel can run in a small reserved region and the KHO scratch areas can sit outside it.
The crash kernel then faults during kho_memory_init() when it tries phys_to_virt() on the KHO FDT address:

  Unable to handle kernel paging request at virtual address xxxxxxxx
  ...
    fdt_offset_ptr+...
    fdt_check_node_offset_+...
    fdt_first_property_offset+...
    fdt_get_property_namelen_+...
    fdt_getprop+...
    kho_memory_init+...
    mm_core_init+...
    start_kernel+...

kho_locate_mem_hole() already skips KHO logic for KEXEC_TYPE_CRASH images, but kho_fill_kimage() was missing the same guard. As kho_fill_kimage() is the single point that populates image->kho.fdt and image->kho.scratch, fixing it here is sufficient for both arm64 and x86 as the FDT and boot_params path are bailing out when these fields are unset.

Fixes d7255959b69a ("kho: allow kexec load before KHO finalization")

---

v2: Per Mike's review [1], move the guard into kho_fill_kimage() instead
of patching the arch-level producers and consumers. This fixes both arm64 and x86 in one place and avoids redundant checks. Tested again.

Quoted 7 lines

Note regarding backporting
The offending commit was deployed with 6.19. The only other supported kernel version with 6.18, unless I miss someting uses

```
if (!kho_out.finalized)
```

which in the case of crash kernel it shouldn't be finalised.

While normally you should load the crash kernel early in boot and at that point KHO should not be finalized, I don't see anything that prevents crash kernel from being loaded after finalize. In which case, you can trigger this bug before d7255959b69a ("kho: allow kexec load before KHO finalization") as well. Also, before f322a97aeb2a ("kho: only fill kimage if KHO is finalized") (landed in v6.18) kho_fill_kimage() was also guarded by if (!kho_enable). So you'd hit this bug in all kernels before that point in the very same way as today.

So should we update Fixes to 3bdecc3c93f9 ("kexec: add KHO support to kexec file loads") and Cc stable?

Quoted 19 lines

[1] https://lore.kernel.org/all/[email protected]/

 kernel/liveupdate/kexec_handover.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/liveupdate/kexec_handover.c +1 -1
index cc68a3692905..1029fe8778f2 100644
@@ -1551,7 +1551,7 @@ int kho_fill_kimage(struct kimage *image)
 	int err = 0;
 	struct kexec_buf scratch;
 
-	if (!kho_enable)
+	if (!kho_enable || image->type == KEXEC_TYPE_CRASH)
 		return 0;
 
 	image->kho.fdt = virt_to_phys(kho_out.fdt);
Signature · 2 lines
Regards,
Pratyush Yadav
#3 lore raw
Quoted 54 lines · On Mon, Apr 13, 2026 at 01:52:40PM +0000 Pratyush Yadav wrote:

Hi Evangelos,

Quoted 33 lines · Evangelos Petrongonas

kho_fill_kimage() unconditionally populates the kimage with KHO metadata for every kexec image type. When the image is a crash kernel, this can be problematic as the crash kernel can run in a small reserved region and the KHO scratch areas can sit outside it.
The crash kernel then faults during kho_memory_init() when it tries phys_to_virt() on the KHO FDT address:

  Unable to handle kernel paging request at virtual address xxxxxxxx
  ...
    fdt_offset_ptr+...
    fdt_check_node_offset_+...
    fdt_first_property_offset+...
    fdt_get_property_namelen_+...
    fdt_getprop+...
    kho_memory_init+...
    mm_core_init+...
    start_kernel+...

kho_locate_mem_hole() already skips KHO logic for KEXEC_TYPE_CRASH images, but kho_fill_kimage() was missing the same guard. As kho_fill_kimage() is the single point that populates image->kho.fdt and image->kho.scratch, fixing it here is sufficient for both arm64 and x86 as the FDT and boot_params path are bailing out when these fields are unset.

Fixes d7255959b69a ("kho: allow kexec load before KHO finalization")

---

v2: Per Mike's review [1], move the guard into kho_fill_kimage() instead
of patching the arch-level producers and consumers. This fixes both arm64 and x86 in one place and avoids redundant checks. Tested again.

Quoted 7 lines

Note regarding backporting
The offending commit was deployed with 6.19. The only other supported kernel version with 6.18, unless I miss someting uses

```
if (!kho_out.finalized)
```

which in the case of crash kernel it shouldn't be finalised.

While normally you should load the crash kernel early in boot and at that point KHO should not be finalized, I don't see anything that prevents crash kernel from being loaded after finalize. In which case, you can trigger this bug before d7255959b69a ("kho: allow kexec load before KHO finalization") as well. Also, before f322a97aeb2a ("kho: only fill kimage if KHO is finalized") (landed in v6.18) kho_fill_kimage() was also guarded by if (!kho_enable). So you'd hit this bug in all kernels before that point in the very same way as today.

Quoted 3 lines

So should we update Fixes to 3bdecc3c93f9 ("kexec: add KHO support to kexec file loads") and Cc stable?

But in this case it seems a userspace misuse if it finalizes kho for crash kernel. Whereas with the current state we hit the bug with a sane userspace. Yeap we would hit that in earlier kernel than 6.18, but none with KHO is supported is it?

I don't have strong objection for backporting it to 6.18, but it feels unnecessary.

Quoted 14 lines
Quoted 14 lines

[1] https://lore.kernel.org/all/[email protected]/

 kernel/liveupdate/kexec_handover.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/liveupdate/kexec_handover.c +0 -0
index cc68a3692905..1029fe8778f2 100644
@@ -1551,7 +1551,7 @@ int kho_fill_kimage(struct kimage *image)
 	int err = 0;
 	struct kexec_buf scratch;
 
Quoted 9 lines
Quoted 5 lines
kernel/liveupdate/kexec_handover.c +1 -1
-	if (!kho_enable)
+	if (!kho_enable || image->type == KEXEC_TYPE_CRASH)
 		return 0;
 
 	image->kho.fdt = virt_to_phys(kho_out.fdt);
Signature · 2 lines
Regards,
Pratyush Yadav

Kind Regards,
Evangelos

Amazon Web Services Development Center Germany GmbH Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

maintainer
#4 lore raw
Quoted 4 lines · Evangelos Petrongonas
Quoted 3 lines · On Mon, Apr 13, 2026 at 01:52:40PM +0000 Pratyush Yadav wrote:

Hi Evangelos,

On Fri, Apr 10 2026, Evangelos Petrongonas wrote:

[...]

Quoted 20 lines
Quoted 17 lines
Quoted 7 lines

Note regarding backporting
The offending commit was deployed with 6.19. The only other supported kernel version with 6.18, unless I miss someting uses

```
if (!kho_out.finalized)
```

which in the case of crash kernel it shouldn't be finalised.

While normally you should load the crash kernel early in boot and at that point KHO should not be finalized, I don't see anything that prevents crash kernel from being loaded after finalize. In which case, you can trigger this bug before d7255959b69a ("kho: allow kexec load before KHO finalization") as well. Also, before f322a97aeb2a ("kho: only fill kimage if KHO is finalized") (landed in v6.18) kho_fill_kimage() was also guarded by if (!kho_enable). So you'd hit this bug in all kernels before that point in the very same way as today.

Quoted 3 lines

So should we update Fixes to 3bdecc3c93f9 ("kexec: add KHO support to kexec file loads") and Cc stable?

Quoted 7 lines

But in this case it seems a userspace misuse if it finalizes kho for crash kernel. Whereas with the current state we hit the bug with a sane userspace. Yeap we would hit that in earlier kernel than 6.18, but none with KHO is supported is it?

I don't have strong objection for backporting it to 6.18, but it feels unnecessary.

Fair enough. Let's leave it as it is then.

[...]

Signature · 2 lines
Regards,
Pratyush Yadav
maintainer Applied
#5 lore raw
In reply to Evangelos Petrongonas

From: Mike Rapoport (Microsoft) <[email protected]>

Quoted 8 lines · Evangelos Petrongonas

kho_fill_kimage() unconditionally populates the kimage with KHO metadata for every kexec image type. When the image is a crash kernel, this can be problematic as the crash kernel can run in a small reserved region and the KHO scratch areas can sit outside it.
The crash kernel then faults during kho_memory_init() when it tries phys_to_virt() on the KHO FDT address:

[...]

Applied to fixes branch of liveupdate/linux.git tree, thanks!

[1/1] kho: skip KHO for crash kernel

      commit: a6715d7ec472a476db17787697a4abda62962284

tree: https://git.kernel.org/pub/scm/linux/kernel/git/liveupdate/linux
branch: fixes

--
Sincerely yours,
Mike.