[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [RFC PATCH 02/10] arm64: Add hook to handle guest GICv3 sysreg accesses





On 02/26/2018 12:12 PM, Manish Jaggi wrote:


On 02/01/2018 04:24 PM, Julien Grall wrote:
Hi Manish,

On 01/02/18 08:51, Manish Jaggi wrote:
On 01/25/2018 11:37 PM, Julien Grall wrote:
Hi,

I forgot to mention one thing about the placement of do_fixup_vgic_errata.

On 16/01/18 15:42, mjaggi@xxxxxxxxxxxxxxxxxx wrote:
diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
index f6f6de3691..d4f0581d33 100644
--- a/xen/arch/arm/traps.c
+++ b/xen/arch/arm/traps.c
@@ -2103,6 +2103,17 @@ void do_trap_guest_sync(struct cpu_user_regs *regs)
  {
      const union hsr hsr = { .bits = regs->hsr };
  +#ifdef CONFIG_VGIC_ERRATA
+    int ret;
+
+    ret  = do_fixup_vgic_errata(regs,hsr);
+    if ( !ret )
+    {
+        advance_pc(regs, hsr);
+        return;

I am fully aware that I suggested this solution and still support that the vGIC errata should be fully separated. After all, it deals with hardware bug and the errata will just update the LRs as the hardware would do.

enter_hypervisor_head() will sync the LRs state to the internal vGIC state. leave_hypervisor_head() will process pending softirq and write/update the LRs based on the internal vGIC state.

As you rightfully did, the do_fixup_vgic_errata should be called before syncing the LRs. However, even if you return early here, you will still execute leave_hypervisor_tail(). This mean that pending softirqs will be processed and potentially the vCPU rescheduled. Because the LRs were not synced (enter_hypervisor_head()) was not called, then the vGIC state will not out-of-date and would lead to all sort of potential issues.

As the vGIC errata implies trapping the register such as IAR1 (reading interrupt), we want to get a fastpath for it (e.g not trying to execute softirq...). So I think we should bypass leave_hypervisor_tail(). I am not entirely sure how to do it nicely thought.


How about adding a check for group1_trap enable in leave_hypervisor_tail().

void leave_hypervisor_tail(void)
{
+if (group1_trap)
+   return;

I guess you mean the variable you introduced in patch #10. In that case, this would be totally wrong. You only want to skip leave_hypervisor_tail() when you are handling a ICV_* System registers.

What you want is adding a bool in the structure cpu_info to tell whether leave_hypervisor_tail() should be skipped.

Why would it be wrong? I don't follow our point.
Also adding a flag in cpu_info would be overkill, use of a global variable is simpler.
I am planning to remove group1_trap as a command line, and use it as

static void gicv3_hyp_init(void)
{
...
#ifdef CONFIG_CAVIUM_ERRATUM_30115
    if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_30115))
        group1_trap = 1;
#endif
...
}
Attaching test patch

diff --git a/xen/arch/arm/gic-v3.c b/xen/arch/arm/gic-v3.c
index 473e26111f..581c07b274 100644
--- a/xen/arch/arm/gic-v3.c
+++ b/xen/arch/arm/gic-v3.c
@@ -64,6 +64,7 @@ static DEFINE_PER_CPU(void __iomem*, rbase);
 #define GICD                   (gicv3.map_dbase)
 #define GICD_RDIST_BASE        (this_cpu(rbase))
 #define GICD_RDIST_SGI_BASE    (GICD_RDIST_BASE + SZ_64K)
+static bool group1_trap = 0;

 /*
  * Saves all 16(Max) LR registers. Though number of LRs implemented
@@ -825,7 +826,7 @@ static void gicv3_cpu_disable(void)

 static void gicv3_hyp_init(void)
 {
-    uint32_t vtr;
+    uint32_t vtr, reg32;

     vtr = READ_SYSREG32(ICH_VTR_EL2);
     gicv3_info.nr_lrs  = (vtr & GICH_VTR_NRLRGS) + 1;
@@ -837,6 +838,10 @@ static void gicv3_hyp_init(void)

     WRITE_SYSREG32(GICH_VMCR_EOI | GICH_VMCR_VENG1, ICH_VMCR_EL2);
     WRITE_SYSREG32(GICH_HCR_EN, ICH_HCR_EL2);
+
+    reg32 = GICH_HCR_EN;
+    reg32 |= (group1_trap) ? GICH_HCR_TALL1 : 0;
+    WRITE_SYSREG32(reg32, ICH_HCR_EL2);
 }

 /* Set up the per-CPU parts of the GIC for a secondary CPU */
@@ -1651,6 +1656,11 @@ static int __init gicv3_init(void)
         return -ENODEV;
     }

+#ifdef CONFIG_CAVIUM_ERRATUM_30115
+    if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_30115))
+        group1_trap = 1;
+#endif
+
     if ( acpi_disabled )
         gicv3_dt_init();
     else
diff --git a/xen/arch/arm/traps.c b/xen/arch/arm/traps.c
index f6f6de3691..dbee0c322f 100644
--- a/xen/arch/arm/traps.c
+++ b/xen/arch/arm/traps.c
@@ -2295,6 +2295,10 @@ void do_trap_fiq(struct cpu_user_regs *regs)

 void leave_hypervisor_tail(void)
 {
+#ifdef CONFIG_CAVIUM_ERRATUM_30115
+    if (cpus_have_const_cap(ARM64_WORKAROUND_CAVIUM_30115))
+        return;
+#endif
     while (1)
     {
         local_irq_disable();
diff --git a/xen/include/asm-arm/gic.h b/xen/include/asm-arm/gic.h
index d3d7bda50d..e4c77fefd6 100644
--- a/xen/include/asm-arm/gic.h
+++ b/xen/include/asm-arm/gic.h
@@ -117,6 +117,7 @@
 #define GICH_HCR_VGRP0DIE (1 << 5)
 #define GICH_HCR_VGRP1EIE (1 << 6)
 #define GICH_HCR_VGRP1DIE (1 << 7)
+#define GICH_HCR_TALL1    (1 << 12)

 #define GICH_MISR_EOI     (1 << 0)
 #define GICH_MISR_U       (1 << 1)


Cheers,




_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.