Skip to contentSkip to conversation

IA64: better calculate PT_LOAD segment size

PATCH In review Jay Lan 2 participants 3 messages Sep 20, 2008 – Sep 27, 2008

Subsystems (2)
lore
Share & tools
lore

Jay Lan

author
#0 lore raw

PATCH 1/1 IA64: better calculate PT_LOAD segment size

This patch combines consecutive PL_LOAD segments into one. The end address of the last PL_LOAD segment, calculated by adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE, will be the end address of this loaded_segments[] entry.

This patch fixes the kdump kernel MCA problem caused by under- allocation of memory and a "kdump broken on ALtix 350" problem reported by Bernhard Walle.

Simon, this patch replaces my previous patch I submitted on the underallocation issue.

---
 kexec/arch/ia64/crashdump-ia64.c |   15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
kexec/arch/ia64/crashdump-ia64.c +8 -7
Index: kexec-tools/kexec/arch/ia64/crashdump-ia64.c
===================================================================
@@ -86,19 +86,20 @@ static void add_loaded_segments_info(str
                 loaded_segments[loaded_segments_num].end =
 			loaded_segments[loaded_segments_num].start;
 
+		/* Consolidate consecutive PL_LOAD segments into one.
+		 * The end addr of the last PL_LOAD segment, calculated by
+		 * adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE,
+		 * will be the end address of this loaded_segments entry.
+		 */
 		while (i < ehdr->e_phnum) {
 			phdr = &ehdr->e_phdr[i];
 	                if (phdr->p_type != PT_LOAD)
 	                        break;
-			if (loaded_segments[loaded_segments_num].end !=
-				phdr->p_paddr & ~(ELF_PAGE_SIZE-1))
-				break;
-			loaded_segments[loaded_segments_num].end +=
-				(phdr->p_memsz + ELF_PAGE_SIZE - 1) &
-				~(ELF_PAGE_SIZE - 1);
+			loaded_segments[loaded_segments_num].end =
+				(phdr->p_paddr + phdr->p_memsz +
+				ELF_PAGE_SIZE - 1) & ~(ELF_PAGE_SIZE - 1);
 			i++;
 		}
-
 		loaded_segments_num++;
 	}
 }

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#1 lore raw
Quoted 5 lines · Jay Lan

This patch combines consecutive PL_LOAD segments into one. The end address of the last PL_LOAD segment, calculated by adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE, will be the end address of this loaded_segments[] entry.

Quoted 9 lines

This patch fixes the kdump kernel MCA problem caused by under- allocation of memory and a "kdump broken on ALtix 350" problem reported by Bernhard Walle.

Simon, this patch replaces my previous patch I submitted on the underallocation issue.

The logic-bug portion of this fix I am fine with. As you pointed out previously & has higher precedence than += which makes the following bogus:

                        loaded_segments[loaded_segments_num].end +=
                                (phdr->p_memsz + ELF_PAGE_SIZE - 1) &
                                ~(ELF_PAGE_SIZE - 1)

The segment merging portion I am less clear about. Why are there gaps in the first place? If these multiple PT_LOAD segments are really one segment, why aren't they being fed in as one segment?

Quoted 36 lines
---
 kexec/arch/ia64/crashdump-ia64.c |   15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

Index: kexec-tools/kexec/arch/ia64/crashdump-ia64.c
===================================================================

kexec/arch/ia64/crashdump-ia64.c +8 -7
@@ -86,19 +86,20 @@ static void add_loaded_segments_info(str
                 loaded_segments[loaded_segments_num].end =
 			loaded_segments[loaded_segments_num].start;
 
+		/* Consolidate consecutive PL_LOAD segments into one.
+		 * The end addr of the last PL_LOAD segment, calculated by
+		 * adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE,
+		 * will be the end address of this loaded_segments entry.
+		 */
 		while (i < ehdr->e_phnum) {
 			phdr = &ehdr->e_phdr[i];
 	                if (phdr->p_type != PT_LOAD)
 	                        break;
-			if (loaded_segments[loaded_segments_num].end !=
-				phdr->p_paddr & ~(ELF_PAGE_SIZE-1))
-				break;
-			loaded_segments[loaded_segments_num].end +=
-				(phdr->p_memsz + ELF_PAGE_SIZE - 1) &
-				~(ELF_PAGE_SIZE - 1);
+			loaded_segments[loaded_segments_num].end =
+				(phdr->p_paddr + phdr->p_memsz +
+				ELF_PAGE_SIZE - 1) & ~(ELF_PAGE_SIZE - 1);
 			i++;
 		}
-
 		loaded_segments_num++;
 	}
 }
Signature · 9 lines
Simon Horman
  VA Linux Systems Japan K.K., Sydney, Australia Satellite Office
  H: www.vergenet.net/~horms/             W: www.valinux.co.jp/en


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
author
#2 lore raw
Quoted 20 lines · Simon Horman
Quoted 5 lines · Jay Lan

This patch combines consecutive PL_LOAD segments into one. The end address of the last PL_LOAD segment, calculated by adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE, will be the end address of this loaded_segments[] entry.

Quoted 9 lines

This patch fixes the kdump kernel MCA problem caused by under- allocation of memory and a "kdump broken on ALtix 350" problem reported by Bernhard Walle.

Simon, this patch replaces my previous patch I submitted on the underallocation issue.

The logic-bug portion of this fix I am fine with. As you pointed out previously & has higher precedence than += which makes the following bogus:

Quoted 6 lines
                        loaded_segments[loaded_segments_num].end +=
                                (phdr->p_memsz + ELF_PAGE_SIZE - 1) &
                                ~(ELF_PAGE_SIZE - 1)

The segment merging portion I am less clear about. Why are there gaps in the first place?

I think there is still a bug in the kernel.

Quoted 3 lines

If these
multiple PT_LOAD segments are really one segment, why aren't they being fed in as one segment?

I searched a little... My guess is that it was done so that the program headers provided more information for kexec and crash? And thus put it back to simulate a boot loader? I really do not know why it was done. Maybe Vivek or Eric would know?

The current comparison part of code is basically a no-op. I think we can simplify the code here.

Regards,
jay

Quoted 29 lines
Quoted 29 lines
---
 kexec/arch/ia64/crashdump-ia64.c |   15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

Index: kexec-tools/kexec/arch/ia64/crashdump-ia64.c
===================================================================

kexec/arch/ia64/crashdump-ia64.c +6 -6
@@ -86,19 +86,20 @@ static void add_loaded_segments_info(str
                 loaded_segments[loaded_segments_num].end =
 			loaded_segments[loaded_segments_num].start;
 
+		/* Consolidate consecutive PL_LOAD segments into one.
+		 * The end addr of the last PL_LOAD segment, calculated by
+		 * adding p_memsz to p_paddr & rounded up to ELF_PAGE_SIZE,
+		 * will be the end address of this loaded_segments entry.
+		 */
 		while (i < ehdr->e_phnum) {
 			phdr = &ehdr->e_phdr[i];
 	                if (phdr->p_type != PT_LOAD)
 	                        break;
-			if (loaded_segments[loaded_segments_num].end !=
-				phdr->p_paddr & ~(ELF_PAGE_SIZE-1))
-				break;
-			loaded_segments[loaded_segments_num].end +=
-				(phdr->p_memsz + ELF_PAGE_SIZE - 1) &
-				~(ELF_PAGE_SIZE - 1);
+			loaded_segments[loaded_segments_num].end =
Quoted 10 lines

kexec/arch/ia64/crashdump-ia64.c static void add_loaded_segments_info(str · line 105

Quoted 8 lines
kexec/arch/ia64/crashdump-ia64.c +2 -1
+				(phdr->p_paddr + phdr->p_memsz +
+				ELF_PAGE_SIZE - 1) & ~(ELF_PAGE_SIZE - 1);
 			i++;
 		}
-
 		loaded_segments_num++;
 	}
 }

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec