Re: [PATCH 4/6] x86/trace: Reduce stack usage from HVMTRACE_ND()

[Jan Beulich <jbeulich@suse.com>]

On 20.09.2021 13:02, Andrew Cooper wrote:
> On 20/09/2021 10:05, Jan Beulich wrote:
>> On 17.09.2021 10:45, Andrew Cooper wrote:
>>> It is pointless to write all 6 entries and only consume the useful subset.
>>> bloat-o-meter shows quite how obscene the overhead is in vmx_vmexit_handler(),
>>> weighing in at 11% of the function arranging unread zeroes on the stack, and
>>> 8% for svm_vmexit_handler().
>>>
>>>   add/remove: 0/0 grow/shrink: 0/20 up/down: 0/-1867 (-1867)
>>>   Function                                     old     new   delta
>>>   hvm_msr_write_intercept                     1049    1033     -16
>>>   vmx_enable_intr_window                       238     214     -24
>>>   svm_enable_intr_window                       337     313     -24
>>>   hvmemul_write_xcr                            115      91     -24
>>>   hvmemul_write_cr                             350     326     -24
>>>   hvmemul_read_xcr                             115      91     -24
>>>   hvmemul_read_cr                              146     122     -24
>>>   hvm_mov_to_cr                                438     414     -24
>>>   hvm_mov_from_cr                              253     229     -24
>>>   vmx_intr_assist                             1150    1118     -32
>>>   svm_intr_assist                              459     427     -32
>>>   hvm_rdtsc_intercept                          138     106     -32
>>>   hvm_msr_read_intercept                       898     866     -32
>>>   vmx_vmenter_helper                          1142    1094     -48
>>>   vmx_inject_event                             813     765     -48
>>>   svm_vmenter_helper                           238     190     -48
>>>   hvm_hlt                                      197     146     -51
>>>   svm_inject_event                            1678    1614     -64
>>>   svm_vmexit_handler                          5880    5416    -464
>>>   vmx_vmexit_handler                          7281    6473    -808
>>>   Total: Before=3644184, After=3642317, chg -0.05%
>>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
>> Reviewed-by: Jan Beulich <jbeulich@suse.com>
> 
> Thanks, but this is buggy.  There are direct callers of HVMTRACE_ND()
> which need adjustments too.

Is this really "buggy" and "need" and not merely "incomplete" and
"want"? (Just for my own understanding in terms of what I may have
overlooked.)

Jan