On Friday, April 12, 2013 08:24:12 AM Eric W. Biederman wrote:
> Thomas Renninger <trenn at suse.de> writes:
> > Currently ranges are passed via kernel boot parameters:
> > memmap=exactmap memmap=X#Y memmap=
> >
> > Pass them via e820 table directly instead.
>
> Reading through this code I am not seeing us mark areas of memory that
> we may not use as reserved. Am I missing something?
They are added.
acpi and reserved (only like in the original kernel) come from
crash_memory_range[] array
(like in cmdline_add_memmap_acpi and cmdline_add_memmap_reserved
and the free to use for crash kernel memory is coming from the array
(memmap_p):
cmdline_add_memmap_acpi
I see that the merging function is not that nice to read, but I couldn't come
up with something better for now.
In fact it should work exactly the same way as before with memmap= passing.
There is something odd with not passing the hightest e820 entry.
It seem to be needed, but it was not passed via memmap= as well?, so my
version should also work and a test did:
The dumpfile is saved to /root/abuild/dumps/2013-04-15-12:59/vmcore
file size: 3251795465
free
total
Mem: 32633312
I answer more detailed on Yinghai's mail.
Thomas
> Those areas need to be marked reserved or else the pci resource
> allocator in the kernel will think it is ok to put pci memory there.
>
> Eric
>
> > CC: Simon Horman <horms at verge.net.au>
> > CC: kexec at lists.infradead.org
> > CC: H. Peter Anvin <hpa at zytor.com>
> > CC: Eric W. Biederman <ebiederm at xmission.com>
> > CC: vgoyal at redhat.com
> > CC: yinghai at kernel.org
> > CC: cpw at sgi.com
> >
> > Signed-off-by: Thomas Renninger <trenn at suse.de>
> > Signed-off-by: Thomas Renninger <Thomas Renninger" trenn at suse.de>
> > ---
> >
> > kexec/arch/i386/crashdump-x86.c | 221
> > ++++++++++++++++++------------------- kexec/arch/i386/x86-linux-setup.c
> > | 11 ++-
> > 2 files changed, 116 insertions(+), 116 deletions(-)
> >
> > diff --git a/kexec/arch/i386/crashdump-x86.c
> > b/kexec/arch/i386/crashdump-x86.c index f7821bc..8009efe 100644
> > --- a/kexec/arch/i386/crashdump-x86.c
> > +++ b/kexec/arch/i386/crashdump-x86.c
> > @@ -659,70 +659,6 @@ static void ultoa(unsigned long i, char *str)
> >
> > }
> >
> > }
> >
> > -static void cmdline_add_memmap_internal(char *cmdline, unsigned long
> > startk, - unsigned long endk, int type)
> > -{
> > - int cmdlen, len;
> > - char str_mmap[256], str_tmp[20];
> > -
> > - strcpy (str_mmap, " memmap=");
> > - ultoa((endk-startk), str_tmp);
> > - strcat (str_mmap, str_tmp);
> > -
> > - if (type == RANGE_RAM)
> > - strcat (str_mmap, "K@");
> > - else if (type == RANGE_RESERVED)
> > - strcat (str_mmap, "K$");
> > - else if (type == RANGE_ACPI || type == RANGE_ACPI_NVS)
> > - strcat (str_mmap, "K#");
> > -
> > - ultoa(startk, str_tmp);
> > - strcat (str_mmap, str_tmp);
> > - strcat (str_mmap, "K");
> > - len = strlen(str_mmap);
> > - cmdlen = strlen(cmdline) + len;
> > - if (cmdlen > (COMMAND_LINE_SIZE - 1))
> > - die("Command line overflow\n");
> > - strcat(cmdline, str_mmap);
> > -}
> > -
> > -/* Adds the appropriate memmap= options to command line, indicating the
> > - * memory regions the new kernel can use to boot into. */
> > -static int cmdline_add_memmap(char *cmdline, struct memory_range
> > *memmap_p) -{
> > - int i, cmdlen, len;
> > - unsigned long min_sizek = 100;
> > - char str_mmap[256];
> > -
> > - /* Exact map */
> > - strcpy(str_mmap, " memmap=exactmap");
> > - len = strlen(str_mmap);
> > - cmdlen = strlen(cmdline) + len;
> > - if (cmdlen > (COMMAND_LINE_SIZE - 1))
> > - die("Command line overflow\n");
> > - strcat(cmdline, str_mmap);
> > -
> > - for (i = 0; i < CRASH_MAX_MEMMAP_NR; i++) {
> > - unsigned long startk, endk;
> > - startk = (memmap_p[i].start/1024);
> > - endk = ((memmap_p[i].end + 1)/1024);
> > - if (!startk && !endk)
> > - /* All regions traversed. */
> > - break;
> > -
> > - /* A region is not worth adding if region size < 100K. It eats
> > - * up precious command line length. */
> > - if ((endk - startk) < min_sizek)
> > - continue;
> > - cmdline_add_memmap_internal(cmdline, startk, endk, RANGE_RAM);
> > - }
> > -
> > - dbgprintf("Command line after adding memmap\n");
> > - dbgprintf("%s\n", cmdline);
> > -
> > - return 0;
> > -}
> > -
> >
> > /* Adds the elfcorehdr= command line parameter to command line. */
> > static int cmdline_add_elfcorehdr(char *cmdline, unsigned long addr)
> > {
> >
> > @@ -803,26 +739,6 @@ static enum coretype get_core_type(struct
> > crash_elf_info *elf_info,>
> > }
> >
> > }
> >
> > -/* Appends memmap=X#Y commandline for ACPI to command line*/
> > -static int cmdline_add_memmap_acpi(char *cmdline, unsigned long start,
> > - unsigned long end)
> > -{
> > - int align = 1024;
> > - unsigned long startk, endk;
> > -
> > - if (!(end - start))
> > - return 0;
> > -
> > - startk = start/1024;
> > - endk = (end + align - 1)/1024;
> > - cmdline_add_memmap_internal(cmdline, startk, endk, RANGE_ACPI);
> > -
> > - dbgprintf("Command line after adding acpi memmap\n");
> > - dbgprintf("%s\n", cmdline);
> > -
> > - return 0;
> > -}
> > -
> >
> > /* Appends 'acpi_rsdp=' commandline for efi boot crash dump */
> > static void cmdline_add_efi(char *cmdline)
> > {
> >
> > @@ -881,24 +797,101 @@ static void get_backup_area(struct kexec_info
> > *info,
> >
> > info->backup_src_size = BACKUP_SRC_END - BACKUP_SRC_START + 1;
> >
> > }
> >
> > -/* Appends memmap=X$Y commandline for reserved memory to command line*/
> > -static int cmdline_add_memmap_reserved(char *cmdline, unsigned long
> > start,
> > - unsigned long end)
> > +/*
> > + * This function takes reserved (all kind of) memory from global
> > + * crash_memory_range[] memory ranges and takes memory the kdump/crash
> > + * kernel is allowed to use from the passed usable_mem memory ranges.
> > + * The passed usable_mem ranges are zero (!start && !end) terminated.
> > + *
> > + * The final memory map is again written into crash_memory_range[]
> > + * and intended to get passed as e820 table to the crash kernel
> > + */
> > +static int create_final_crash_map(struct memory_range *usable_mem)
> >
> > {
> >
> > - int align = 1024;
> > - unsigned long startk, endk;
> > + int i, m, c, tmp_map1_ranges, tmp_map2_ranges;
> > + unsigned long min_sizek = 100;
> > + /* crash_memory_map with usable memory ranges cut out */
> > + struct memory_range tmp_map1[MAX_MEMORY_RANGES];
> > + /* merge_map, but small ranges cut out */
> > + struct memory_range tmp_map2[MAX_MEMORY_RANGES];
> >
> > - if (!(end - start))
> > - return 0;
> > + /*
> > + * Ignore usable memory ranges for kdump kernel smaller
> > + * than 100k to avoid too much ranges passed
> > + * Save the new ranges (exluding lower than 100k ranges) in tmp_map
> > + * and store the number of elements in tmp_map_ranges
> > + */
> > + for (m = 0, i = 0; i < CRASH_MAX_MEMMAP_NR; i++) {
> > + unsigned long startk, endk;
> > + startk = (usable_mem[i].start/1024);
> > + endk = ((usable_mem[i].end + 1)/1024);
> > + if (!startk && !endk)
> > + /* All regions traversed. */
> > + break;
> > +
> > + /* A region is not worth adding if region size < 100K. It eats
> > + * up precious command line length. */
> > + if ((endk - startk) < min_sizek) {
> > + dbgprintf("Removing: %luk - %luk\n", startk, endk);
> > + continue;
> > + } else {
> > + tmp_map1[m].start = usable_mem[i].start;
> > + tmp_map1[m].end = usable_mem[i].end;
> > + tmp_map1[m].type = usable_mem[i].type;
> > + m++;
> > + }
> > + }
> > + /* No need to check for !start && !end anymore */
> > + tmp_map1_ranges = m;
> >
> > - startk = start/1024;
> > - endk = (end + align - 1)/1024;
> > - cmdline_add_memmap_internal(cmdline, startk, endk, RANGE_RESERVED);
> > + for(i = 0; i < tmp_map1_ranges; ++i)
> > + dbgprintf("%016Lx-%016Lx (%d)\n", tmp_map1[i].start,
> > + tmp_map1[i].end, tmp_map1[i].type);
> > +
> > + /*
> > + * Cut out RANGE_RAM regions from crash_memory_ranges and store
> > + * them in tmp_map2_ranges
> > + */
> > + for (c = 0, i = 0; i < crash_ranges; i++) {
> > + if (crash_memory_range[i].type == RANGE_RAM)
> > + continue;
> > + tmp_map2[c].start = crash_memory_range[i].start;
> > + tmp_map2[c].end = crash_memory_range[i].end;
> > + tmp_map2[c].type = crash_memory_range[i].type;
> > + c++;
> > + }
> > + tmp_map2_ranges = c;
> > +
> > + /*
> > + * TBD: Check that no ranges overlap?
> > + * Can this happen at all?
> > + */
> > + for (c = 0, m = 0, i = 0; i < MAX_MEMORY_RANGES; i++) {
> > + if (m < tmp_map1_ranges &&
> > + (c >= tmp_map2_ranges ||
> > + tmp_map2[c].start > tmp_map1[m].start)) {
> > + crash_memory_range[i].start = tmp_map1[m].start;
> > + crash_memory_range[i].end = tmp_map1[m].end;
> > + crash_memory_range[i].type = RANGE_RAM;
> > + m++;
> > + continue;
> > + } else if (c < tmp_map2_ranges) {
> > + crash_memory_range[i] = tmp_map2[c];
> > + c++;
> > + continue;
> > + } else
> > + break;
> > + }
> > + crash_ranges = i;
> > +
> > + /*
> > + * End address has to be exlusive for e820 map
> > + * x - 00010000
> > + * 00010000 - y
> > + */
> > + for(i = 0; i < crash_ranges; ++i)
> > + crash_memory_range[i].end++;
> >
> > -#ifdef DEBUG
> > - printf("Command line after adding reserved memmap\n");
> > - printf("%s\n", cmdline);
> > -#endif
> >
> > return 0;
> >
> > }
> >
> > @@ -944,6 +937,12 @@ int load_crashdump_segments(struct kexec_info *info,
> > char* mod_cmdline,>
> > return -1;
> >
> > }
> >
> > + /*
> > + * From now on the memory regions are stored in crash_memory_range[]
> > + * Currently the end address is inclusive at this point:
> > + * x - 0000ffff
> > + * 00010000 - y
> > + */
> >
> > if (xen_present()) {
> >
> > if (get_crash_memory_ranges_xen(&mem_range, &crash_ranges,
> >
> > elf_info.lowmem_limit) < 0)
> >
> > @@ -971,7 +970,7 @@ int load_crashdump_segments(struct kexec_info *info,
> > char* mod_cmdline,>
> > get_backup_area(info, mem_range, crash_ranges);
> >
> > - dbgprintf("CRASH MEMORY RANGES\n");
> > + dbgprintf("TEMPORARY CRASH MEMORY RANGES\n");
> >
> > for(i = 0; i < crash_ranges; ++i)
> >
> > dbgprintf("%016Lx-%016Lx (%d)\n", mem_range[i].start,
> >
> > @@ -1063,24 +1062,18 @@ int load_crashdump_segments(struct kexec_info
> > *info, char* mod_cmdline,>
> > dbgprintf("Created elf header segment at 0x%lx\n", elfcorehdr);
> > if (delete_memmap(memmap_p, elfcorehdr, memsz) < 0)
> >
> > return -1;
> >
> > - cmdline_add_memmap(mod_cmdline, memmap_p);
> >
> > cmdline_add_efi(mod_cmdline);
> > cmdline_add_elfcorehdr(mod_cmdline, elfcorehdr);
> >
> > - /* Inform second kernel about the presence of ACPI tables. */
> > - for (i = 0; i < MAX_MEMORY_RANGES; i++) {
> > - unsigned long start, end;
> > - if ( !( mem_range[i].type == RANGE_ACPI
> > - || mem_range[i].type == RANGE_ACPI_NVS
> > - || mem_range[i].type == RANGE_RESERVED) )
> > - continue;
> > - start = mem_range[i].start;
> > - end = mem_range[i].end;
> > - if (mem_range[i].type == RANGE_RESERVED)
> > - cmdline_add_memmap_reserved(mod_cmdline, start, end);
> > - else
> > - cmdline_add_memmap_acpi(mod_cmdline, start, end);
> > - }
> > + /*
> > + * Redo crash_memory_range so that it can get passed as e820 info
> > + */
> > + create_final_crash_map(memmap_p);
> > +
> > + dbgprintf("FINAL CRASH MEMORY RANGES\n");
> > + for(i = 0; i < crash_ranges; ++i)
> > + dbgprintf("%016Lx-%016Lx (%d)\n", mem_range[i].start,
> > + mem_range[i].end, mem_range[i].type);
> >
> > return 0;
> >
> > }
> >
> > diff --git a/kexec/arch/i386/x86-linux-setup.c
> > b/kexec/arch/i386/x86-linux-setup.c index c538897..82b4bb9 100644
> > --- a/kexec/arch/i386/x86-linux-setup.c
> > +++ b/kexec/arch/i386/x86-linux-setup.c
> > @@ -505,8 +505,15 @@ void setup_linux_system_parameters(struct kexec_info
> > *info,>
> > /* another safe default */
> > real_mode->aux_device_info = 0;
> >
> > - range = info->memory_range;
> > - ranges = info->memory_ranges;
> > + if (info->kexec_flags & KEXEC_ON_CRASH ||
> > + info->kexec_flags & KEXEC_PRESERVE_CONTEXT) {
> > + range = crash_memory_range;
> > + ranges = crash_ranges;
> > + } else {
> > + range = info->memory_range;
> > + ranges = info->memory_ranges;
> > + }
> > +
> >
> > if (ranges > E820MAX) {
> >
> > if (!(info->kexec_flags & KEXEC_ON_CRASH))
> >
> > /*
