Hi Nan hai, On 11/23/06, Zou, Nanhai <[EMAIL PROTECTED]> wrote:> > -----Original Message-----> > From: Magnus Damm [mailto:[EMAIL PROTECTED]> > Sent: 2006年11月22日 12:53> > To: [EMAIL PROTECTED]> > Cc: Magnus Damm; fastboot@lists.osdl.org; Horms; [EMAIL PROTECTED]; Zou,> > Nanhai> > Subject: Re: [PATCH 005/013] kexec-tools: Make use of> > crash_create_elf64_headers() (x86_64)> >> > Hi Vivek and Nan hai,> >> > On 11/22/06, Vivek Goyal <[EMAIL PROTECTED]> wrote:> > > On Tue, Nov 21, 2006 at 02:56:23PM +0900, Magnus Damm wrote:> > > [..]> > > > >>> > > > >> +static struct crash_elf_info elf_info => > > > >> +{> > > > >> + class: ELFCLASS64,> > > > >> + data: ELFDATA2LSB,> > > > >> + machine: EM_X86_64,> > > > >> + alignment: 1024,> > > > >> > > > >This alignment thing is confusing me. Can you give some more details> > > > >please?> > > >> > > > Well, I only put it in because the original code had some alignment code:> > > >> > > > @@ -734,7 +615,6 @@ int load_c! rashdump_segments(struct kexec> > > > void *tmp;> > > > unsigned long sz, elfcorehdr;> > > > int nr_ranges, align = 1024, i;> > > > - long int nr_cpus = 0;> > > > struct memory_range *mem_range, *memmap_p;> > > >> > > > if (get_kernel_paddr(info))> > > > @@ -764,20 +644,9 @@ int load_crashdump_segments(struct kexec> > > > return -1;> > > >> > > > /* Create elf header segment and store crash image data. */> > > > - nr_cpus = sysconf(_SC_NPROCESSORS_CONF);> > > > - if (nr_cpus < 0) {> > > > - fprintf(stderr,"kexec_load (elf header segment)"> > > > - " failed: %s\n", strerror(errno));> > > > - return -1;> > > > - }> > > > - sz = sizeof(Elf64_Ehdr) + nr_cpus * sizeof(Elf64_Phdr) +> > > > - nr_ranges * sizeof(Elf64_Phdr);> > > > - sz = (sz + align - 1) & ~(align -1);> > > > - tmp = xmalloc(sz);> > > > - memset! (tmp, 0, sz);> > > > -> > > > - /* Prepare ELF64 core heaers. */> > > > - if (prepare_crash_memory_elf64_headers(info, tmp, sz) < 0)> > > > + if (crash_create_elf64_headers(info, &elf_info,> > > > + crash_memory_range, nr_ranges,> > > > + &tmp, &sz) < 0)> > > > return -1;> > > >> > > > Have a look at the align variable above? It is set to 1024. So I just> > > > broke out the alignment number and put it into the elf_info structure> > > > to make it per-arch.> > > >> > > > I don't know why the number of bytes allocated are rounded to an even> > > > 1024. It is probably because the data that follows the headers should> > > > be aligned. Word-size aligment I understand, but 1024 I don't. 1024> > > > was probably a handy value that happened to work regardless of 32 or> > > > 64 bit elf format...> > > >> > >> > >> > > Ok. Now I remember. I had put this alignment to make sure e! lf header> > > segment is atleast 1K aligned. And the reason behind that is when we> > > prepare the memmap= options to pass to second kernel, we can pass the> > > memory only with a granularity of KB and not less than that.> > >> > > So I need to remove the elf core header segment from the list of memory> > > regions which second kernel can use. I used delete_memmap() function to> > > do that and create a new list. If segments/memory regions I am handling> > > are at least 1K aligned, it becomes easier to handle them. Otherwise> > > its chaos.> >> > That explains the alignment. Thank you! Always aligning to 1K makes sense then.> >> > Nan hai, you align to EFI_PAGE_SIZE instead of 1K on ia64. Do you have> > any special reason for doing so?> >> > Right now I assume it is a "copy-and-paste-o" so my patches use> > 1K-alignment instead.> >> ELF spec says, loadable process segments should be aligned to page size.> But for core dump data, I guess 1k might be ok. But why not al! ign them to page size? The notes need to be 1k aligned because the kernel command line optionmemmap= only passes even kilobytes to the secondary kernel. See thereply from Vivek above for more information. I'll use 1k alignment on ia64 too then. Thanks! / magnus
_______________________________________________ fastboot mailing list fastboot@lists.osdl.org https://lists.osdl.org/mailman/listinfo/fastboot
