make IPI barrier a real function

Hensoldt-Cyber · Jun 20, 2024 · 83755b3 · 83755b3
1 parent 0931cf0
commit 83755b3
Show file tree

Hide file tree

Showing 8 changed files with 37 additions and 23 deletions.
diff --git a/include/arch/arm/arch/machine.h b/include/arch/arm/arch/machine.h
@@ -63,6 +63,17 @@ static inline void arch_pause(void)
 {
     /* TODO */
 }
+
+static inline void ipi_mem_barrier(void)
+{
+    /* For GICv2 systems a dmb() is sufficient, but it's not enough with GICv3
+     * due to the way IPIs is triggered (memory-mapped or MSR inst.). A dmb()
+     * does not prevent re-ordering to happen between memory accesses and
+     * instructions, this guarantee requires a dsb().
+     */
+    dsb_ishst();
+}
+
 #endif /* ENABLE_SMP_SUPPORT */
 
 /* Update the value of the actual regsiter to hold the expected value */

diff --git a/include/arch/arm/arch/machine/hardware.h b/include/arch/arm/arch/machine/hardware.h
@@ -19,17 +19,7 @@ typedef word_t vm_fault_type_t;
 #define PAGE_BASE(_p, _s)        ((_p) & ~MASK(pageBitsForSize((_s))))
 #define PAGE_OFFSET(_p, _s)      ((_p) & MASK(pageBitsForSize((_s))))
 
-#define IPI_MEM_BARRIER \
-  do { \
-     /* This can be relaxed for GICv2 but for GICv3 dmb() no longer works */ \
-     /* since the way IPI is triggered is different (memory-mapped or MSR inst.) */ \
-     /* and dmb() is not able to avoid re-ordering between memory accesses and */ \
-     /* instructions. In order to support both GICv2 and v3 dsb() is required. */ \
-     dsb_ishst(); \
-  } while (0)
-
 #endif /* __ASSEMBLER__ */
 
 #define L1_CACHE_LINE_SIZE_BITS CONFIG_L1_CACHE_LINE_SIZE_BITS
 #define L1_CACHE_LINE_SIZE BIT(L1_CACHE_LINE_SIZE_BITS)
-
diff --git a/include/arch/riscv/arch/machine.h b/include/arch/riscv/arch/machine.h
@@ -71,6 +71,10 @@ static inline void fence_rw_rw(void)
     asm volatile("fence rw, rw" ::: "memory");
 }
 
+static inline void ipi_mem_barrier(void) {
+    fence_rw_rw();
+}
+
 static inline void fence_w_rw(void)
 {
     asm volatile("fence w, rw" ::: "memory");

diff --git a/include/arch/riscv/arch/machine/hardware.h b/include/arch/riscv/arch/machine/hardware.h
@@ -122,9 +122,3 @@ static inline void arch_clean_invalidate_caches(void)
 
 #define LOAD_S STRINGIFY(LOAD)
 #define STORE_S STRINGIFY(STORE)
-
-#define IPI_MEM_BARRIER \
-    do { \
-        asm volatile("fence rw,rw" ::: "memory"); \
-    } while (0)
-
diff --git a/include/arch/x86/arch/kernel/x2apic.h b/include/arch/x86/arch/kernel/x2apic.h
@@ -65,6 +65,14 @@ static inline void apic_write_icr(word_t high, word_t low)
     x86_wrmsr(APIC_ICR, icr);
 }
 
-#define IPI_ICR_BARRIER  asm volatile("mfence" ::: "memory")
-#define IPI_MEM_BARRIER  IPI_ICR_BARRIER
+static inline void ipi_icr_barrier(void)
+{
+    asm volatile("mfence" ::: "memory");
+}
+
+static inline void ipi_mem_barrier(void)
+{
+    IPI_ICR_BARRIER;
+}
+
 #endif /* CONFIG_X2APIC */
diff --git a/include/arch/x86/arch/kernel/xapic.h b/include/arch/x86/arch/kernel/xapic.h
@@ -66,7 +66,14 @@ static inline void apic_write_icr(word_t high, word_t low)
     apic_write_reg(APIC_ICR1, low);
 }
 
-#define IPI_ICR_BARRIER  asm volatile("" ::: "memory")
-#define IPI_MEM_BARRIER IPI_ICR_BARRIER
-#endif  /* CONFIG_XAPIC */
+static inline void ipi_icr_barrier(void)
+{
+    asm volatile("" ::: "memory");
+}
 
+static inline void ipi_mem_barrier(void)
+{
+    ipi_icr_barrier();
+}
+
+#endif  /* CONFIG_XAPIC */
diff --git a/src/arch/x86/smp/ipi.c b/src/arch/x86/smp/ipi.c
@@ -110,7 +110,7 @@ static void x86_ipi_send_mask(interrupt_t ipi, word_t mask, bool_t isBlocking)
     } while (mask != 0);
 
     /* broadcast IPIs to clusters... */
-    IPI_ICR_BARRIER;
+    ipi_icr_barrier();
     for (int i = 0; i < nr_target_clusters; i++) {
         apic_send_ipi_cluster(ipi, target_clusters[i]);
     }

diff --git a/src/smp/ipi.c b/src/smp/ipi.c
@@ -139,7 +139,7 @@ void generic_ipi_send_mask(irq_t ipi, word_t mask, bool_t isBlocking)
 
     if (nr_target_cores > 0) {
         /* sending IPIs... */
-        IPI_MEM_BARRIER;
+        ipi_mem_barrier();
         for (int i = 0; i < nr_target_cores; i++) {
             ipi_send_target(ipi, cpuIndexToID(target_cores[i]));
         }