Thanks for the patch. On 2022/12/06 19:00, Pavankumar Kondeti wrote: "if VA_BITS <= 48" ? What if kernels < 5.4? For such old kernels without vabits_actual introduced, probably we should not set VA_BITS_ACTUAL.. Thanks, Kazu > > Signed-off-by: Pavankumar Kondeti <quic_pkondeti(a)quicinc.com&gt; > --- > arm64.c | 33 +++++++++++++++++++++++++-------- > 1 file changed, 25 insertions(+), 8 deletions(-) > > diff --git a/arm64.c b/arm64.c > index c3e26a3..9d5d0bd 100644 > --- a/arm64.c > +++ b/arm64.c > @@ -4588,6 +4588,27 @@ arm64_IS_VMALLOC_ADDR(ulong vaddr) > > /* Return TRUE if we succeed, return FALSE on failure. */ > static int > +arm64_set_va_bits_fallback(void) > +{ > + if (machdep->machspec->VA_BITS_ACTUAL) { > + machdep->machspec->VA_BITS = machdep->machspec->VA_BITS_ACTUAL; > + machdep->machspec->VA_START = _VA_START(machdep->machspec->VA_BITS_ACTUAL); > + return TRUE; > + } > + > + if (machdep->machspec->CONFIG_ARM64_VA_BITS) { > + /* guess */ > + machdep->machspec->VA_BITS_ACTUAL = machdep->machspec->CONFIG_ARM64_VA_BITS; > + machdep->machspec->VA_BITS = machdep->machspec->CONFIG_ARM64_VA_BITS; > + machdep->machspec->VA_START = _VA_START(machdep->machspec->VA_BITS_ACTUAL); > + return TRUE; > + } > + > + return FALSE; > +} > + > +/* Return TRUE if we succeed, return FALSE on failure. */ > +static int > arm64_set_va_bits_by_tcr(void) > { > ulong value; > @@ -4648,14 +4669,8 @@ arm64_calc_VA_BITS(void) > else { > if (arm64_set_va_bits_by_tcr()) { > /* nothing */ > - } else if (machdep->machspec->VA_BITS_ACTUAL) { > - machdep->machspec->VA_BITS = machdep->machspec->VA_BITS_ACTUAL; > - machdep->machspec->VA_START = _VA_START(machdep->machspec->VA_BITS_ACTUAL); > - } else if (machdep->machspec->CONFIG_ARM64_VA_BITS) { > - /* guess */ > - machdep->machspec->VA_BITS_ACTUAL = machdep->machspec->CONFIG_ARM64_VA_BITS; > - machdep->machspec->VA_BITS = machdep->machspec->CONFIG_ARM64_VA_BITS; > - machdep->machspec->VA_START = _VA_START(machdep->machspec->VA_BITS_ACTUAL); > + } else if (arm64_set_va_bits_fallback()) { > + /* nothing */ > } else > error(FATAL, "cannot determine VA_BITS_ACTUAL\n"); > } > @@ -4671,6 +4686,8 @@ arm64_calc_VA_BITS(void) > return; > } else if (arm64_set_va_bits_by_tcr()) { > return; > + } else if (arm64_set_va_bits_fallback()) { > + return; > } > > if (!(sp = symbol_search("swapper_pg_dir")) &&

On 2022/12/07 13:25, Pavan Kondeti wrote: I think it's ok to fail, because they should not use that option for older kernels than 5.4, it's a wrong usage. + if (machdep->machspec->CONFIG_ARM64_VA_BITS) { + /* guess */ + machdep->machspec->VA_BITS_ACTUAL = machdep->machspec->CONFIG_ARM64_VA_BITS; This is my concern. Actually I found a 4.18.0 arm64 vmcore with NUMBER(VA_BITS)=48 in vmcoreinfo, crash fails to start with the patch. # crash-arm64 vmlinux vmcore ... crash-arm64: read error: kernel virtual address: fffff6003bcc0050 type: "IRQ stack pointer" crash-arm64: read error: kernel virtual address: fffff6003bcf0050 type: "IRQ stack pointer" crash-arm64: read error: kernel virtual address: fffff6003fffe400 type: "memory section root table" # hmm, currently THIS_KERNEL_VERSION cannot be used at the point. We need to fill it, but also need vmcoreinfo.. um, please let me know if I misunderstand, I don't have arm64 knowledge very much. Do ramdumps need "-m vabits_actual=xx" regardless of the existence of vabits_actual symbol on 5.4 and later? And it seems that recent arm64 vmcores have NUMBER(TCR_EL1_T1SZ) in vmcoreinfo, they are covered by arm64_set_va_bits_by_tcr(). so how about adding only this? --- a/arm64.c +++ b/arm64.c @@ -4671,6 +4671,10 @@ arm64_calc_VA_BITS(void) return; } else if (arm64_set_va_bits_by_tcr()) { return; + } else if (machdep->machspec->VA_BITS_ACTUAL) { + machdep->machspec->VA_BITS = machdep->machspec->VA_BITS_ACTUAL; + machdep->machspec->VA_START = _VA_START(machdep->machspec->VA_BITS_ACTUAL); + return; } if (!(sp = symbol_search("swapper_pg_dir")) && Thanks, Kazu