Itsuro ODA wrote: > Hi, > > This patch is for xen hypervisor analysis function of the > crash command to apply to the xen-3.3.0 (the newest version of xen). > > * PERCPU_SHIFT becomes 13 (from 12) in the xen-3.3.0. > This value is calculated from "__per_cpu_start" and "__per_cpu_data_end". > * "jiffies" does not exist in the xen-3.3.0. > It was used to show the uptime. I found there is no altanernative > (i.e. the xen hypervisor does not have the uptime.). > Then if "jiffies" does not exist, "--:--:--" is showed as UPTIME in > the sys command. > (Is it better to eliminate the whole UPTIME line ?) > --- example --- > crash> sys > KERNEL: xen-syms > DUMPFILE: vmcore > CPUS: 4 > DOMAINS: 5 > UPTIME: --:--:-- > MACHINE: Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz (2660 Mhz) > MEMORY: 2 GB > ---------------- > > This patch is for crash-4.0-7.2. > > Thanks > Itsuro Oda The patch looks OK. But just for sanity's sake, is it guaranteed that the per_cpu data section will be greater than 4k on both architectures? Or could there be some combination of xen CONFIG options that could reduce the i386 per_cpu data section contents to less than 4K even though PERCPU_SHIFT is 13? Dave > > --- > --- xen_hyper_defs.h.org 2008-10-06 13:45:39.000000000 +0900 > +++ xen_hyper_defs.h 2008-10-06 13:44:44.000000000 +0900 > @@ -134,9 +134,8 @@ > #endif > > #if defined(X86) || defined(X86_64) > -#define XEN_HYPER_PERCPU_SHIFT 12 > #define xen_hyper_per_cpu(var, cpu) \ > - ((ulong)(var) + (((ulong)(cpu))<<XEN_HYPER_PERCPU_SHIFT)) > + ((ulong)(var) + (((ulong)(cpu))<<xht->percpu_shift)) > #elif defined(IA64) > #define xen_hyper_per_cpu(var, cpu) \ > ((xht->flags & XEN_HYPER_SMP) ? \ > @@ -404,6 +403,7 @@ > ulong *cpumask; > uint *cpu_idxs; > ulong *__per_cpu_offset; > + int percpu_shift; > }; > > struct xen_hyper_dumpinfo_context { > --- xen_hyper.c.org 2008-10-06 13:41:14.000000000 +0900 > +++ xen_hyper.c 2008-10-06 14:15:03.000000000 +0900 > @@ -71,6 +71,8 @@ > #endif > > #if defined(X86) || defined(X86_64) > + xht->percpu_shift = > + (symbol_value("__per_cpu_data_end") - symbol_value("__per_cpu_start") > 4096) ? 13: 12; > member_offset = MEMBER_OFFSET("cpuinfo_x86", "x86_model_id"); > buf = GETBUF(XEN_HYPER_SIZE(cpuinfo_x86)); > if (xen_hyper_test_pcpu_id(XEN_HYPER_CRASHING_CPU())) { > @@ -1746,9 +1748,11 @@ > tmp2 = (ulong)jiffies_64; > jiffies_64 = (ulonglong)(tmp2 - tmp1); > } > - } else { > + } else if (symbol_exists("jiffies")) { > get_symbol_data("jiffies", sizeof(long), &jiffies); > jiffies_64 = (ulonglong)jiffies; > + } else { > + jiffies_64 = 0; /* hypervisor does not have uptime */ > } > > return jiffies_64; > --- xen_hyper_command.c.org 2008-10-07 08:05:37.000000000 +0900 > +++ xen_hyper_command.c 2008-10-07 08:24:29.000000000 +0900 > @@ -1022,7 +1022,8 @@ > (buf1, "%d\n", XEN_HYPER_NR_DOMAINS())); > /* !!!Display a date here if it can be found. */ > XEN_HYPER_PRI(fp, len, "UPTIME: ", buf1, flag, > - (buf1, "%s\n", convert_time(xen_hyper_get_uptime_hyper(), buf2))); > + (buf1, "%s\n", (xen_hyper_get_uptime_hyper() ? > + convert_time(xen_hyper_get_uptime_hyper(), buf2) : "--:--:--"))); > /* !!!Display a version here if it can be found. */ > XEN_HYPER_PRI_CONST(fp, len, "MACHINE: ", flag); > if (strlen(uts->machine)) { > --- -- Crash-utility mailing list Crash-utility(a)redhat.com https://www.redhat.com/mailman/listinfo/crash-utility

----- "Itsuro ODA" <oda(a)valinux.co.jp&gt; wrote: > Hi, > > It is a good question. > I checked about i386. __per_cpu_data_end - __per_cpu_start is smaller > than 4K, but PERCPU_SHIFT is 13 (it is common both x86_32 and > x86_64). > Oops. > > I will consider more. Ok, I'll wait for a new patch set. Another direction that you might consider is to check whether a new symbol was introduced in the same patch-set that incremented the PERCPU_SHIFT values?

Or take Keir up on his offer to compile-in an absolute value that can be accessed like a symbol?

As far as the UPTIME display, it will probably raise fewer questions in the future if you would just remove it entirely. Thanks, Dave

Itsuro ODA wrote: > Hi, > > I decided to use the version number of xen which is included in the > XEN_ELFNOTE_CRASH_INFO after all. > The defect is that vmcores taken on the xen-unstable during the > xen-3.3 development before the changset 17831 (2008/6/12) can not be > analized with the newer crash. I think it is not a big problem since > it is rare case to analyze such vmcores right now. OK, that's fine with me. However, can you re-send me the new patch as an *attachment* to your email? Both this patch and the last one that you sent have whitespace problems because all tabs in the file have been turned into spaces: $ patch -p0 --dry-run < $HOME/xen-3.3.0.patch patching file defs.h patching file netdump.h Hunk #1 FAILED at 110. 1 out of 1 hunk FAILED -- saving rejects to file netdump.h.rej patching file netdump.c Hunk #1 FAILED at 1525. Hunk #2 FAILED at 1732. Hunk #3 succeeded at 2347 with fuzz 2. 2 out of 3 hunks FAILED -- saving rejects to file netdump.c.rej patching file xen_hyper_defs.h Hunk #1 FAILED at 134. Hunk #2 FAILED at 403. 2 out of 2 hunks FAILED -- saving rejects to file xen_hyper_defs.h.rej patching file xen_hyper.c Hunk #1 FAILED at 71. Hunk #2 FAILED at 1753. 2 out of 2 hunks FAILED -- saving rejects to file xen_hyper.c.rej patching file xen_hyper_command.c Hunk #1 FAILED at 1022. 1 out of 1 hunk FAILED -- saving rejects to file xen_hyper_command.c.rej $ Thanks, Dave > > On Wed, 8 Oct 2008 08:56:48 -0400 (EDT) > Dave Anderson <anderson(a)redhat.com&gt; wrote: > >> ----- "Itsuro ODA" <oda(a)valinux.co.jp&gt; wrote: >> >>> Hi, >>> >>> It is a good question. >>> I checked about i386. __per_cpu_data_end - __per_cpu_start is smaller >>> than 4K, but PERCPU_SHIFT is 13 (it is common both x86_32 and >>> x86_64). >>> Oops. >>> >>> I will consider more. >> Ok, I'll wait for a new patch set. >> >> Another direction that you might consider is to check whether >> a new symbol was introduced in the same patch-set that >> incremented the PERCPU_SHIFT values? > > I found this is equivalent as using "__per_cpu_data_end - __per_cpu_start > 4K" > since the change of the PERCPU_SHIFT was for x86_64 only and it was > not helpful for i386. > >> Or take Keir up on his offer to compile-in an absolute >> value that can be accessed like a symbol? > > It is checked first in the patch. > >> As far as the UPTIME display, it will probably raise >> fewer questions in the future if you would just remove >> it entirely. >> >> Thanks, >> Dave >> > > * PERCPU_SHIFT becomes 13 (from 12) in the xen-3.3.0. > This value is determined by xen version number. > (If the symbol "__per_cpu_shift" exists, it is used first for the future.) > * "jiffies" does not exist in the xen-3.3.0. > It was used to show the uptime. I found there is no altanernative > (i.e. the xen hypervisor does not have the uptime.). > Then if "jiffies" does not exist, "--:--:--" is showed as UPTIME in > the sys command. > --- example --- > crash> sys > KERNEL: xen-syms > DUMPFILE: vmcore > CPUS: 4 > DOMAINS: 5 > UPTIME: --:--:-- > MACHINE: Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz (2660 Mhz) > MEMORY: 2 GB > ---------------- > > This patch is for crash-4.0-7.2. > > Thanks > Itsuro Oda > > --- > --- defs.h.org 2008-10-09 09:25:11.000000000 +0900 > +++ defs.h 2008-10-09 09:53:53.000000000 +0900 > @@ -4118,6 +4118,8 @@ > int is_sadump_xen(void); > void set_xen_phys_start(char *); > ulong xen_phys_start(void); > +int xen_major_version(void); > +int xen_minor_version(void); > > /* > * diskdump.c > --- netdump.h.org 2008-10-09 09:25:03.000000000 +0900 > +++ netdump.h 2008-10-09 09:49:26.000000000 +0900 > @@ -110,6 +110,8 @@ > int p2m_frames; > ulong *p2m_mfn_frame_list; > ulong xen_phys_start; > + int xen_major_version; > + int xen_minor_version; > }; > > #define KDUMP_P2M_INIT (0x1) > --- netdump.c.org 2008-10-09 09:24:54.000000000 +0900 > +++ netdump.c 2008-10-09 09:53:20.000000000 +0900 > @@ -1525,6 +1525,8 @@ > nd->xen_kdump_data->p2m_mfn = *(uptr+(words-1)); > if (words > 9 && !nd->xen_kdump_data->xen_phys_start) > nd->xen_kdump_data->xen_phys_start = *(uptr+(words-2)); > + nd->xen_kdump_data->xen_major_version = *uptr; > + nd->xen_kdump_data->xen_minor_version = *(uptr+1); > } > } > break; > @@ -1730,6 +1732,8 @@ > nd->xen_kdump_data->p2m_mfn = *(up+(words-1)); > if (words > 9 && !nd->xen_kdump_data->xen_phys_start) > nd->xen_kdump_data->xen_phys_start = *(up+(words-2)); > + nd->xen_kdump_data->xen_major_version = *up; > + nd->xen_kdump_data->xen_minor_version = *(up+1); > } > } > break; > @@ -2343,3 +2347,15 @@ > { > return nd->xen_kdump_data->xen_phys_start; > } > + > +int > +xen_major_version(void) > +{ > + return nd->xen_kdump_data->xen_major_version; > +} > + > +int > +xen_minor_version(void) > +{ > + return nd->xen_kdump_data->xen_minor_version; > +} > --- xen_hyper_defs.h.org 2008-10-06 13:45:39.000000000 +0900 > +++ xen_hyper_defs.h 2008-10-06 13:44:44.000000000 +0900 > @@ -134,9 +134,8 @@ > #endif > > #if defined(X86) || defined(X86_64) > -#define XEN_HYPER_PERCPU_SHIFT 12 > #define xen_hyper_per_cpu(var, cpu) \ > - ((ulong)(var) + (((ulong)(cpu))<<XEN_HYPER_PERCPU_SHIFT)) > + ((ulong)(var) + (((ulong)(cpu))<<xht->percpu_shift)) > #elif defined(IA64) > #define xen_hyper_per_cpu(var, cpu) \ > ((xht->flags & XEN_HYPER_SMP) ? \ > @@ -404,6 +403,7 @@ > ulong *cpumask; > uint *cpu_idxs; > ulong *__per_cpu_offset; > + int percpu_shift; > }; > > struct xen_hyper_dumpinfo_context { > --- xen_hyper.c.org 2008-10-06 13:41:14.000000000 +0900 > +++ xen_hyper.c 2008-10-09 11:31:24.000000000 +0900 > @@ -71,6 +71,13 @@ > #endif > > #if defined(X86) || defined(X86_64) > + if (symbol_exists("__per_cpu_shift")) { > + xht->percpu_shift = (int)symbol_value("__per_cpu_shift"); > + } else if (xen_major_version() >= 3 && xen_minor_version() >= 3) { > + xht->percpu_shift = 13; > + } else { > + xht->percpu_shift = 12; > + } > member_offset = MEMBER_OFFSET("cpuinfo_x86", "x86_model_id"); > buf = GETBUF(XEN_HYPER_SIZE(cpuinfo_x86)); > if (xen_hyper_test_pcpu_id(XEN_HYPER_CRASHING_CPU())) { > @@ -1746,9 +1753,11 @@ > tmp2 = (ulong)jiffies_64; > jiffies_64 = (ulonglong)(tmp2 - tmp1); > } > - } else { > + } else if (symbol_exists("jiffies")) { > get_symbol_data("jiffies", sizeof(long), &jiffies); > jiffies_64 = (ulonglong)jiffies; > + } else { > + jiffies_64 = 0; /* hypervisor does not have uptime */ > } > > return jiffies_64; > --- xen_hyper_command.c.org 2008-10-07 08:05:37.000000000 +0900 > +++ xen_hyper_command.c 2008-10-07 08:24:29.000000000 +0900 > @@ -1022,7 +1022,8 @@ > (buf1, "%d\n", XEN_HYPER_NR_DOMAINS())); > /* !!!Display a date here if it can be found. */ > XEN_HYPER_PRI(fp, len, "UPTIME: ", buf1, flag, > - (buf1, "%s\n", convert_time(xen_hyper_get_uptime_hyper(), buf2))); > + (buf1, "%s\n", (xen_hyper_get_uptime_hyper() ? > + convert_time(xen_hyper_get_uptime_hyper(), buf2) : "--:--:--"))); > /* !!!Display a version here if it can be found. */ > XEN_HYPER_PRI_CONST(fp, len, "MACHINE: ", flag); > if (strlen(uts->machine)) { > ---