kexec: Improve & fix crash_exclude_mem_range() to handle overlapping ranges

The crash_exclude_mem_range() function can only handle one memory region a time.

It will fail in the case in which the passed in area covers several memory
regions. In this case, it will only exclude the first region, then return,
but leave the later regions unsolved.

E.g in a NEC system with two usable RAM regions inside the low 1M:

  ...
  BIOS-e820: [mem 0x0000000000000000-0x000000000003efff] usable
  BIOS-e820: [mem 0x000000000003f000-0x000000000003ffff] reserved
  BIOS-e820: [mem 0x0000000000040000-0x000000000009ffff] usable

It will only exclude the memory region [0, 0x3efff], the memory region
[0x40000, 0x9ffff] will still be added into /proc/vmcore, which may cause
the following failure when dumping vmcore:

 ioremap on RAM at 0x0000000000040000 - 0x0000000000040fff
 WARNING: CPU: 0 PID: 665 at arch/x86/mm/ioremap.c:186 __ioremap_caller+0x2c7/0x2e0
 ...
 RIP: 0010:__ioremap_caller+0x2c7/0x2e0
 ...
 cp: error reading '/proc/vmcore': Cannot allocate memory
 kdump: saving vmcore failed

In order to fix this bug, let's extend the crash_exclude_mem_range()
to handle the overlapping ranges.

[ mingo: Amended the changelog. ]

Signed-off-by: Lianbo Jiang <lijiang@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Dave Young <dyoung@redhat.com>
Link: https://lore.kernel.org/r/20200804044933.1973-3-lijiang@redhat.com

1 files changed, 23 insertions, 12 deletions

diff --git a/kernel/kexec_file.c b/kernel/kexec_file.c
index 94661d2d13ad..97fa68267197 100644
--- a/kernel/kexec_file.c
+++ b/kernel/kexec_file.c
@@ -1157,24 +1157,26 @@ int crash_exclude_mem_range(struct crash_mem *mem,
 			    unsigned long long mstart, unsigned long long mend)
 {
 	int i, j;
-	unsigned long long start, end;
+	unsigned long long start, end, p_start, p_end;
 	struct crash_mem_range temp_range = {0, 0};
 
 	for (i = 0; i < mem->nr_ranges; i++) {
 		start = mem->ranges[i].start;
 		end = mem->ranges[i].end;
+		p_start = mstart;
+		p_end = mend;
 
 		if (mstart > end || mend < start)
 			continue;
 
 		/* Truncate any area outside of range */
 		if (mstart < start)
-			mstart = start;
+			p_start = start;
 		if (mend > end)
-			mend = end;
+			p_end = end;
 
 		/* Found completely overlapping range */
-		if (mstart == start && mend == end) {
+		if (p_start == start && p_end == end) {
 			mem->ranges[i].start = 0;
 			mem->ranges[i].end = 0;
 			if (i < mem->nr_ranges - 1) {
@@ -1185,20 +1187,29 @@ int crash_exclude_mem_range(struct crash_mem *mem,
 					mem->ranges[j].end =
 							mem->ranges[j+1].end;
 				}
+
+				/*
+				 * Continue to check if there are another overlapping ranges
+				 * from the current position because of shifting the above
+				 * mem ranges.
+				 */
+				i--;
+				mem->nr_ranges--;
+				continue;
 			}
 			mem->nr_ranges--;
 			return 0;
 		}
 
-		if (mstart > start && mend < end) {
+		if (p_start > start && p_end < end) {
 			/* Split original range */
-			mem->ranges[i].end = mstart - 1;
-			temp_range.start = mend + 1;
+			mem->ranges[i].end = p_start - 1;
+			temp_range.start = p_end + 1;
 			temp_range.end = end;
-		} else if (mstart != start)
-			mem->ranges[i].end = mstart - 1;
+		} else if (p_start != start)
+			mem->ranges[i].end = p_start - 1;
 		else
-			mem->ranges[i].start = mend + 1;
+			mem->ranges[i].start = p_end + 1;
 		break;
 	}
 
@@ -1243,7 +1254,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
 	 * I think this is required by tools like gdb. So same physical
-	 * memory will be mapped in two elf headers. One will contain kernel
+	 * memory will be mapped in two elf  headers. One will contain kernel
 	 * text virtual addresses and other will have __va(physical) addresses.
 	 */
 
@@ -1270,7 +1281,7 @@ int crash_prepare_elf64_headers(struct crash_mem *mem, int kernel_map,
 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
 
-	/* Prepare one phdr of type PT_NOTE for each present cpu */
+	/* Prepare one phdr of type PT_NOTE for each present CPU */
 	for_each_present_cpu(cpu) {
 		phdr->p_type = PT_NOTE;
 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));