Re: [v4.17-rcx] Lost IBPB, IBRS_FW support for spectre_v2 mitigation.

[Thomas Gleixner <tglx@linutronix.de>]

[-- Attachment #1: Type: text/plain, Size: 2839 bytes --]

On Wed, 2 May 2018, Jörg Otte wrote:
> With revert:
> 
> jojo@fichte:~$ dmesg | grep -i -e spec -e micro -e "Linux version"
> 
> [    0.000000] microcode: microcode updated early to revision 0x24,
> date = 2018-01-21
> [    0.000000] Linux version 4.17.0-rc3-revert-00001-gcb1069f
> (jojo@fichte) (gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubu
> 
> dmesg | grep -i -e spec -e micro -e "Linux version"
> 
> [    0.000000] microcode: microcode updated early to revision 0x24,
> date = 2018-01-21
> [    0.000000] Linux version 4.17.0-rc3-patch-00001-gdc10603
> (jojo@fichte) (gcc version 5.4.0 20160609 (Ubuntu
> 5.4.0-6ubuntu1~16.04.9)) #20 SMP Wed May 2 09:08:07 CEST 2018
> [    0.028417] Spectre V2 : Mitigation: Full generic retpoline
> [    0.491803] microcode: sig=0x306c3, pf=0x10, revision=0x24
> [    0.491831] microcode: Microcode Update Driver: v2.2.ntu1~16.04.9))
> #21 SMP Wed May 2 09:14:29 CEST 2018
> [    0.028414] Spectre V2 : Mitigation: Full generic retpoline
> [    0.028415] Spectre V2 : Spectre v2 mitigation: Enabling Indirect
> Branch Prediction Barrier
> [    0.028415] Spectre V2 : Enabling Restricted Speculation for firmware calls
> [    0.500157] microcode: sig=0x306c3, pf=0x10, revision=0x24
> [    0.500183] microcode: Microcode Update Driver: v2.2.
> 
> 
> With patch:
> 
> dmesg | grep -i -e spec -e micro -e "Linux version"
> 
> [    0.000000] microcode: microcode updated early to revision 0x24,
> date = 2018-01-21
> [    0.000000] Linux version 4.17.0-rc3-patch-00001-gdc10603
> (jojo@fichte) (gcc version 5.4.0 20160609 (Ubuntu
> 5.4.0-6ubuntu1~16.04.9)) #20 SMP Wed May 2 09:08:07 CEST 2018
> [    0.028417] Spectre V2 : Mitigation: Full generic retpoline
> [    0.491803] microcode: sig=0x306c3, pf=0x10, revision=0x24
> [    0.491831] microcode: Microcode Update Driver: v2.2.

Ok, I think I know what's going wrong in that steaming pile of horrors of
CPUID detection. I need to analyze it down to the roots, but if you have
cycles, can you please test the patch below?

It's a hack and even if it fixes the problem I'm going to do it differently.

Thanks,

	tglx

8<-------------------
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -848,6 +848,11 @@ void get_cpu_cap(struct cpuinfo_x86 *c)
 		c->x86_power = edx;
 	}
 
+	if (c->extended_cpuid_level >= 0x80000008) {
+		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
+		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
+	}
+
 	if (c->extended_cpuid_level >= 0x8000000a)
 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
 
@@ -871,7 +876,6 @@ static void get_cpu_address_sizes(struct
 
 		c->x86_virt_bits = (eax >> 8) & 0xff;
 		c->x86_phys_bits = eax & 0xff;
-		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
 	}
 #ifdef CONFIG_X86_32
 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))