Syslinux - Jul 2008 - [PATCH 1/3] Fix loading of *.lkrn images from gPXE

Attempts to load *.lkrn images were failing with "Not enough memory
to load specified image".  These images have the su_ramdisk_max
header field set to zero, and the code in core/runkernel.inc was
limiting MyHighMemSize too early (before the load_high call used to
load the main part of the kernel).

Signed-off-by: Sergey Vlasov <vsu at altlinux.ru>
---
 core/runkernel.inc |   13 ++++++-------
 1 files changed, 6 insertions(+), 7 deletions(-)

diff --git a/core/runkernel.inc b/core/runkernel.inc
index 4d62778..c7af6cc 100644
--- a/core/runkernel.inc
+++ b/core/runkernel.inc
@@ -216,13 +216,6 @@ new_kernel:
                 mov al,[es:su_loadflags]
 		mov [LoadFlags],al
 
-		; Cap the ramdisk memory range if appropriate
-		mov eax,[RamdiskMax]
-		cmp eax,[MyHighMemSize]
-		ja .ok
-		mov [MyHighMemSize],eax
-.ok:
-
 any_kernel:
 
 ;
@@ -291,6 +284,12 @@ high_load_done:
 ; if we tried to load initrd using an old kernel
 ;
 load_initrd:
+		; Cap the ramdisk memory range if appropriate
+		mov eax,[RamdiskMax]
+		cmp eax,[MyHighMemSize]
+		ja .ok
+		mov [MyHighMemSize],eax
+.ok:
 		xor eax,eax
                 cmp [InitRDPtr],ax
                 jz .noinitrd
-- 
1.5.6.2.305.g2938b

The address in EDI returned by the load_high call used to load the
kernel needs to be passed to parse_load_initrd, but the code which
clears memory after setup sectors for 1.2.x kernels was corrupting
low 16 bits of EDI.  In most cases this corruption was not noticed,
because with usual setup sizes DI was set to 0xf800, therefore the
chance of getting the kernel size such that initrd would actually
overwrite the kernel was about 3%.

Signed-off-by: Sergey Vlasov <vsu at altlinux.ru>
---
 core/runkernel.inc |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/core/runkernel.inc b/core/runkernel.inc
index c7af6cc..5748b0b 100644
--- a/core/runkernel.inc
+++ b/core/runkernel.inc
@@ -270,6 +270,7 @@ high_load_done:
 ; if they see protected-mode kernel data after the setup sectors, so
 ; clear that memory.
 ;
+		push di
 		mov di,[SetupSecs]
 		shl di,9
 		xor eax,eax
@@ -277,6 +278,7 @@ high_load_done:
 		sub cx,di
 		shr cx,2
 		rep stosd
+		pop di
 
 ;
 ; Now see if we have an initial RAMdisk; if so, do requisite computation
-- 
1.5.6.2.305.g2938b

Sergey Vlasov wrote:
> Attempts to load *.lkrn images were failing with "Not enough memory
> to load specified image".  These images have the su_ramdisk_max
> header field set to zero, and the code in core/runkernel.inc was
> limiting MyHighMemSize too early (before the load_high call used to
> load the main part of the kernel).
Hm.  I don't know what *.lkrn images you're referring to (the newer gPXE
images?) but that would be a bug in those images.  If they don't have 
any intention of using an initrd it should be set to -1.

Doesn't mean it wouldn't be worthwhile to work around this issue, but 
it's still wrong.

	-hpa