summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorSubrata Banik <subrata.banik@intel.com>2019-05-06 19:51:34 +0530
committerSubrata Banik <subrata.banik@intel.com>2019-05-13 02:06:46 +0000
commit095c931cf12924da9011b47aa64f4a6f11d89f13 (patch)
tree3e4da8c0a4aca79f1e7bcd686a905e5e85655533
parenta6d401c193c9bb53162df9523f9ba4ebb2a6f1a7 (diff)
src/arch/x86: Use core apic id to get cpu_index()
This cpu_index() implementation assumes that cpu_index() function might always getting called from coreboot context (ESP stack pointer will always refer to coreboot). This might not be true in case of proposed PI spec MP_SERVICES_PPI implementation, where FSP context (stack pointer refers to fsp) will request to get cpu_index(), natural alignment logic will use ESP and retrieve struct cpu_info *ci from (stack_top - 8 byte). This is not the place where cpu_index is actually stored by ramstage c_start.S Hence this patch tries to remove those dependencies while retrieving cpu_index(), rather it uses cpuid to fetch lapic id and matches with cpus_default_apic_id[] variable to return correct cpu_index(). BRANCH=none BUG=b:79562868 TEST=Ensures functions can be run on APs without any failure and cpu_index() also provides correct index number. Change-Id: I55023a3e0cf42f0496d45bc6af8ead447f402350 Signed-off-by: Subrata Banik <subrata.banik@intel.com> Reviewed-on: https://review.coreboot.org/c/coreboot/+/26346 Reviewed-by: Nico Huber <nico.h@gmx.de> Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
-rw-r--r--src/arch/x86/cpu.c24
-rw-r--r--src/arch/x86/include/arch/cpu.h21
2 files changed, 38 insertions, 7 deletions
diff --git a/src/arch/x86/cpu.c b/src/arch/x86/cpu.c
index f19b389441..fb4c7b6cfe 100644
--- a/src/arch/x86/cpu.c
+++ b/src/arch/x86/cpu.c
@@ -337,3 +337,27 @@ void arch_bootstate_coreboot_exit(void)
/* APs are waiting for work. Last thing to do is park them. */
mp_park_aps();
}
+
+/*
+ * Previously cpu_index() implementation assumes that cpu_index()
+ * function will always getting called from coreboot context
+ * (ESP stack pointer will always refer to coreboot).
+ *
+ * But with FSP_USES_MP_SERVICES_PPI implementation in coreboot this
+ * assumption might not be true, where FSP context (stack pointer refers
+ * to FSP) will request to get cpu_index().
+ *
+ * Hence new logic to use cpuid to fetch lapic id and matches with
+ * cpus_default_apic_id[] variable to return correct cpu_index().
+ */
+unsigned long cpu_index(void)
+{
+ int i;
+ int lapic_id = initial_lapicid();
+
+ for (i = 0; i < CONFIG_MAX_CPUS; i++) {
+ if (cpu_get_apic_id(i) == lapic_id)
+ return i;
+ }
+ return -1;
+}
diff --git a/src/arch/x86/include/arch/cpu.h b/src/arch/x86/include/arch/cpu.h
index 61b17a6d2a..481ee9d8e0 100644
--- a/src/arch/x86/include/arch/cpu.h
+++ b/src/arch/x86/include/arch/cpu.h
@@ -261,13 +261,6 @@ static inline struct cpu_info *cpu_info(void)
);
return ci;
}
-
-static inline unsigned long cpu_index(void)
-{
- struct cpu_info *ci;
- ci = cpu_info();
- return ci->index;
-}
#endif
#ifndef __ROMCC__ // romcc is segfaulting in some cases
@@ -374,4 +367,18 @@ uint32_t cpu_get_feature_flags_ecx(void);
*/
uint32_t cpu_get_feature_flags_edx(void);
+/*
+ * Previously cpu_index() implementation assumes that cpu_index()
+ * function will always getting called from coreboot context
+ * (ESP stack pointer will always refer to coreboot).
+ *
+ * But with FSP_USES_MP_SERVICES_PPI implementation in coreboot this
+ * assumption might not be true, where FSP context (stack pointer refers
+ * to FSP) will request to get cpu_index().
+ *
+ * Hence new logic to use cpuid to fetch lapic id and matches with
+ * cpus_default_apic_id[] variable to return correct cpu_index().
+ */
+unsigned long cpu_index(void);
+
#endif /* ARCH_CPU_H */