RE: [PATCH v5 00/34] x86: enable FRED for x86-64

From: Li, Xin3
Date: Sat Mar 11 2023 - 16:29:42 EST


> The following issues are seen in this round of test.
> +----------------+----------------+----------------+----------------+
> | | x86:test_ | bpf:test_progs | x86:sysret |
> | | vsyscall_32 | | _rip_64 |
> +----------------+----------------+----------------+----------------+
> | the 7th | | | |
> | Intel(R) | FAIL | FAIL | PASS |
> | Core(TM) CPU | | | |
> | 6.3.0-rc1+ | | | |
> +----------------+----------------+----------------+----------------+
> | the 7th | | | |
> | Intel(R) | | | |
> | Core(TM) CPU | FAIL | FAIL | PASS |
> | 6.3.0-rc1+ w/ | | | |
> | FRED patch set | | | |
> +----------------+----------------+----------------+----------------+
> | Intel Simics® | | | |
> | Simulator w/o | FAIL | FAIL | PASS |
> | FRED model | | | |
> | 6.3.0-rc1+ | | | |
> +----------------+----------------+----------------+----------------+
> | Intel Simics® | | | |
> | Simulator w/o | | | |
> | FRED model | FAIL | FAIL | PASS |
> | 6.3.0-rc1+ w/ | | | |
> | FRED patch set | | | |
> +----------------+----------------+----------------+----------------+
> | Intel Simics® | | | |
> | Simulator w/ | PASS | FAIL | PASS |
> | FRED model | | | |
> | 6.3.0-rc1+ | | | |
> +----------------+----------------+----------------+----------------+
> | Intel Simics® | | | |
> | Simulator w/ | | | |
> | FRED model | PASS | FAIL | PASS |
> | 6.3.0-rc1+ w/ | | | |
> | FRED patch set | | | |
> | FRED disabled | | | |
> +----------------+----------------+----------------+----------------+
> | Intel Simics® | | | |
> | Simulator w/ | | | |
> | FRED model | PASS | HANG | FAIL |
> | 6.3.0-rc1+ w/ | | | |
> | FRED patch set | | | |
> +----------------+----------------+----------------+----------------+
>
> The "x86:test_vsyscall_32" is a regression since the v3 FRED patch set.

The "x86:test_vsyscall_32" test passes on the Simics FRED model, no matter
whether FRED is enabled or not, because the Simics FRED model has the RDPID
instruction support.

While on a Simics non-FRED model or the test bare metal machine, which
don't have the RDPID instruction support, the test reads CPU ID from the
GDT_ENTRY_CPUNODE entry, thus it fails due to:
https://lore.kernel.org/lkml/20230311084824.2340-1-xin3.li@xxxxxxxxx/

Thanks!
Xin