| Message ID | 27d828d2-fb4b-41fe-82e8-90720bd9ff8f@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v2] x86/CPU: convert vendor hook invocations to altcall | expand |
On 23/01/2024 11:00 am, Jan Beulich wrote: > While not performance critical, these hook invocations still want > converting: This way all pre-filled struct cpu_dev instances can become > __initconst_cf_clobber, thus allowing to eliminate further 8 ENDBR > during the 2nd phase of alternatives patching (besides moving previously > resident data to .init.*). > > Since all use sites need touching anyway, take the opportunity and also > address a Misra C:2012 Rule 5.5 violation: Rename the this_cpu static > variable. > > Signed-off-by: Jan Beulich <jbeulich@suse.com> Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> > --- > With LTO it might end up necessary to tag as __used more than just > "default_cpu". Why is it even needed here? LTO can't rid early_cpu_init() of the default clause, so can't make default_cpu unreferenced, I don't think. > Perhaps __used would better be integrated into __initconst_cf_clobber, > to be independent of the compiler potentially eliding structure > instances. Maybe. I guess the issue here is that the tools really can't see the connection between being in the clobber section, and alternatives going and making a modification. ~Andrew
On 02.02.2024 13:13, Andrew Cooper wrote: > On 23/01/2024 11:00 am, Jan Beulich wrote: >> While not performance critical, these hook invocations still want >> converting: This way all pre-filled struct cpu_dev instances can become >> __initconst_cf_clobber, thus allowing to eliminate further 8 ENDBR >> during the 2nd phase of alternatives patching (besides moving previously >> resident data to .init.*). >> >> Since all use sites need touching anyway, take the opportunity and also >> address a Misra C:2012 Rule 5.5 violation: Rename the this_cpu static >> variable. >> >> Signed-off-by: Jan Beulich <jbeulich@suse.com> > > Acked-by: Andrew Cooper <andrew.cooper3@citrix.com> Thanks. >> --- >> With LTO it might end up necessary to tag as __used more than just >> "default_cpu". > > Why is it even needed here? > > LTO can't rid early_cpu_init() of the default clause, so can't make > default_cpu unreferenced, I don't think. Even without LTO I've actually seen gcc eliminate default_cpu. The use in early_cpu_init() simply was expanded to assignments, rather then something memcpy()-like. >> Perhaps __used would better be integrated into __initconst_cf_clobber, >> to be independent of the compiler potentially eliding structure >> instances. > > Maybe. I guess the issue here is that the tools really can't see the > connection between being in the clobber section, and alternatives going > and making a modification. The const-ness is the issue, I expect. From that the compiler can infer that it may transform use of the variable into something which in the end doesn't require the variable anymore. And it can't know that we use all contributions to that section for a second purpose. Jan
--- a/xen/arch/x86/cpu/amd.c +++ b/xen/arch/x86/cpu/amd.c @@ -1302,7 +1302,7 @@ static void cf_check init_amd(struct cpu amd_log_freq(c); } -const struct cpu_dev amd_cpu_dev = { +const struct cpu_dev __initconst_cf_clobber amd_cpu_dev = { .c_early_init = early_init_amd, .c_init = init_amd, }; --- a/xen/arch/x86/cpu/centaur.c +++ b/xen/arch/x86/cpu/centaur.c @@ -54,6 +54,6 @@ static void cf_check init_centaur(struct init_c3(c); } -const struct cpu_dev centaur_cpu_dev = { +const struct cpu_dev __initconst_cf_clobber centaur_cpu_dev = { .c_init = init_centaur, }; --- a/xen/arch/x86/cpu/common.c +++ b/xen/arch/x86/cpu/common.c @@ -113,10 +113,10 @@ static void cf_check default_init(struct __clear_bit(X86_FEATURE_SEP, c->x86_capability); } -static const struct cpu_dev default_cpu = { +static const struct cpu_dev __initconst_cf_clobber __used default_cpu = { .c_init = default_init, }; -static const struct cpu_dev *this_cpu = &default_cpu; +static struct cpu_dev __ro_after_init actual_cpu; static DEFINE_PER_CPU(uint64_t, msr_misc_features); void (* __read_mostly ctxt_switch_masking)(const struct vcpu *next); @@ -336,12 +336,13 @@ void __init early_cpu_init(bool verbose) c->x86_vendor = x86_cpuid_lookup_vendor(ebx, ecx, edx); switch (c->x86_vendor) { - case X86_VENDOR_INTEL: this_cpu = &intel_cpu_dev; break; - case X86_VENDOR_AMD: this_cpu = &amd_cpu_dev; break; - case X86_VENDOR_CENTAUR: this_cpu = ¢aur_cpu_dev; break; - case X86_VENDOR_SHANGHAI: this_cpu = &shanghai_cpu_dev; break; - case X86_VENDOR_HYGON: this_cpu = &hygon_cpu_dev; break; + case X86_VENDOR_INTEL: actual_cpu = intel_cpu_dev; break; + case X86_VENDOR_AMD: actual_cpu = amd_cpu_dev; break; + case X86_VENDOR_CENTAUR: actual_cpu = centaur_cpu_dev; break; + case X86_VENDOR_SHANGHAI: actual_cpu = shanghai_cpu_dev; break; + case X86_VENDOR_HYGON: actual_cpu = hygon_cpu_dev; break; default: + actual_cpu = default_cpu; if (!verbose) break; printk(XENLOG_ERR @@ -448,8 +449,8 @@ static void generic_identify(struct cpui if (c->extended_cpuid_level >= 0x80000021) c->x86_capability[FEATURESET_e21a] = cpuid_eax(0x80000021); - if (this_cpu->c_early_init) - this_cpu->c_early_init(c); + if (actual_cpu.c_early_init) + alternative_vcall(actual_cpu.c_early_init, c); /* c_early_init() may have adjusted cpuid levels/features. Reread. */ c->cpuid_level = cpuid_eax(0); @@ -546,9 +547,8 @@ void identify_cpu(struct cpuinfo_x86 *c) * At the end of this section, c->x86_capability better * indicate the features this CPU genuinely supports! */ - if (this_cpu->c_init) - this_cpu->c_init(c); - + if (actual_cpu.c_init) + alternative_vcall(actual_cpu.c_init, c); /* * The vendor-specific functions might have changed features. Now --- a/xen/arch/x86/cpu/hygon.c +++ b/xen/arch/x86/cpu/hygon.c @@ -87,7 +87,7 @@ static void cf_check init_hygon(struct c amd_log_freq(c); } -const struct cpu_dev hygon_cpu_dev = { +const struct cpu_dev __initconst_cf_clobber hygon_cpu_dev = { .c_early_init = early_init_amd, .c_init = init_hygon, }; --- a/xen/arch/x86/cpu/intel.c +++ b/xen/arch/x86/cpu/intel.c @@ -598,7 +598,7 @@ static void cf_check init_intel(struct c setup_clear_cpu_cap(X86_FEATURE_CLWB); } -const struct cpu_dev intel_cpu_dev = { +const struct cpu_dev __initconst_cf_clobber intel_cpu_dev = { .c_early_init = early_init_intel, .c_init = init_intel, }; --- a/xen/arch/x86/cpu/shanghai.c +++ b/xen/arch/x86/cpu/shanghai.c @@ -15,6 +15,6 @@ static void cf_check init_shanghai(struc init_intel_cacheinfo(c); } -const struct cpu_dev shanghai_cpu_dev = { +const struct cpu_dev __initconst_cf_clobber shanghai_cpu_dev = { .c_init = init_shanghai, };
While not performance critical, these hook invocations still want converting: This way all pre-filled struct cpu_dev instances can become __initconst_cf_clobber, thus allowing to eliminate further 8 ENDBR during the 2nd phase of alternatives patching (besides moving previously resident data to .init.*). Since all use sites need touching anyway, take the opportunity and also address a Misra C:2012 Rule 5.5 violation: Rename the this_cpu static variable. Signed-off-by: Jan Beulich <jbeulich@suse.com> --- With LTO it might end up necessary to tag as __used more than just "default_cpu". Perhaps __used would better be integrated into __initconst_cf_clobber, to be independent of the compiler potentially eliding structure instances. --- v2: Add previously missing __used (due to sending out a stale patch).