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

Re: [RFC PATCH] xen/arm: Replace vreg_emulate_{sysreg/cp}32 with vreg_emulate_{sysreg/cp}

On 28/07/2021 15:59, Julien Grall wrote:
Hi Michal,

On 27/07/2021 10:50, Michal Orzel wrote:
According to ARMv8A architecture, AArch64 registers
are 64bit wide even though in many cases the upper
32bit is reserved. Therefore there is no need for
function vreg_emulate_sysreg32 on arm64.

Ideally on arm64 there should be only one function
vreg_emulate_sysreg(using register_t) or
vreg_emulate_sysreg64(using uint64_t) but in the Xen code
there is a lot of functions passed both to the
vreg_emulate_cp* and vreg_emulate_sysreg*.
This would require to duplicate them which is not
a good solution.

I think you can drop vreg_emulate_sysreg64() completely. On arm64, register_t is an alias to uint64_t so you could interchangeably use the type in the callback.

For arm32, we would still need to keep vreg_emulate_cp64.

The easiest/minimal solution to fix this issue is
to replace vreg_emulate_{sysreg/cp}32 with
vreg_emulate_{sysreg/cp}. The modifed functions
are now taking function pointer:
-typedef bool (*vreg_reg_fn_t)(struct cpu_user_regs *regs,
                               register_t *r, bool read);
instead of:
-typedef bool (*vreg_reg32_fn_t)(struct cpu_user_regs *regs,
                                 uint32_t *r, bool read);

This change allows to properly use 64bit registers on arm64
and in case of 32bit guest the cast is done by the hardware
due to the 32bit registers being the lower part of 64bit ones.

The HW doesn't do any cast. From the Arm Arm (D1.19.1 in ARM DDI 0487F.c):

"Any modifications made to AArch32 System registers affects only those parts of those AArch64 registers that are mapped to the AArch32 System registers. Bits[63:32] of AArch64 registers, where they are not mapped to AArch32
registers, are unchanged by AArch32 state execution."

Actually, I quoted the paragraph for the system registers. Sorry :/

I should have quoted the section "Mapping of the general-purpose registers between the Execution states". But this is not really a cast, it would be a zero extend. Although, it is not 100% whether this is what happen because just above the section it is written:

"In AArch32 state, the upper 32 bits of AArch64 registers are inaccessible and are ignored.

The registers can be set by:
  * The toolstack (via XEN_DOMCTL_set_vcpucontext). We rely on the top bits to always be 0. Ideally, we should 0 it in vcpu_regs_user_to_hyp() just for safety.
   * The PSCI CPU ON call: They should always be 0.

For the rest of Xen, we expect that the top 32-bit will either be untouched or never be changed.

Signed-off-by: Michal Orzel <michal.orzel@xxxxxxx>
The reason for this change is to clean up the mess related to types.
This patch achieves that but it does not reduce the code size.
I'm not sure whether we want such change hence it is pushed as RFC.
  xen/arch/arm/vcpreg.c      | 16 +++++++++++-----
  xen/arch/arm/vtimer.c      | 18 +++++++++---------
  xen/include/asm-arm/vreg.h | 14 +++++++-------
  3 files changed, 27 insertions(+), 21 deletions(-)

diff --git a/xen/arch/arm/vcpreg.c b/xen/arch/arm/vcpreg.c
index e3ce56d875..376a1ceee2 100644
--- a/xen/arch/arm/vcpreg.c
+++ b/xen/arch/arm/vcpreg.c
@@ -57,9 +57,12 @@
  #define WRITE_SYSREG_SZ(sz, val, sysreg...)  WRITE_SYSREG##sz(val, sysreg)
+#define type32_t register_t
+#define type64_t uint64_t

Please use typedef rather than define for type. Also, please add a comment explaining why type32_t is defined as register_t.

  /* The name is passed from the upper macro to workaround macro expansion. */   #define TVM_REG(sz, func, reg...)                                           \ -static bool func(struct cpu_user_regs *regs, uint##sz##_t *r, bool read)    \ +static bool func(struct cpu_user_regs *regs, type##sz##_t *r, bool read)    \ { \       struct vcpu *v = current;                                               \       bool cache_enabled = vcpu_has_cache_enabled(v);                         \ @@ -83,7 +86,7 @@ static bool func(struct cpu_user_regs *regs, uint##sz##_t *r, bool read)    \
  #else /* CONFIG_ARM_64 */
  #define TVM_REG32_COMBINED(lowreg, hireg, xreg)                             \ -static bool vreg_emulate_##xreg(struct cpu_user_regs *regs, uint32_t *r,    \ +static bool vreg_emulate_##xreg(struct cpu_user_regs *regs, register_t *r,  \                                   bool read, bool hi)                         \ { \       struct vcpu *v = current;                                               \ @@ -108,13 +111,13 @@ static bool vreg_emulate_##xreg(struct cpu_user_regs *regs, uint32_t *r,    \       return true;                                                            \ } \ \ -static bool vreg_emulate_##lowreg(struct cpu_user_regs *regs, uint32_t *r,  \ +static bool vreg_emulate_##lowreg(struct cpu_user_regs *regs, register_t *r,\                                     bool read)                                \ { \       return vreg_emulate_##xreg(regs, r, read, false);                       \ } \ \ -static bool vreg_emulate_##hireg(struct cpu_user_regs *regs, uint32_t *r,   \ +static bool vreg_emulate_##hireg(struct cpu_user_regs *regs, register_t *r, \                                    bool read)                                 \ { \       return vreg_emulate_##xreg(regs, r, read, true);                        \
@@ -154,13 +157,16 @@ TVM_REG32_COMBINED(MAIR0, MAIR1, MAIR_EL1)
+#define VREG_EMULATE_CP32(regs, hsr, fn)  vreg_emulate_cp(regs, hsr, fn)
+#define VREG_EMULATE_CP64(regs, hsr, fn)  vreg_emulate_cp64(regs, hsr, fn)
  /* Macro to generate easily case for co-processor emulation. */
  #define GENERATE_CASE(reg, sz)                                      \
      case HSR_CPREG##sz(reg):                                        \
      {                                                               \
          bool res;                                                   \
-        res = vreg_emulate_cp##sz(regs, hsr, vreg_emulate_##reg);   \
+        res = VREG_EMULATE_CP##sz(regs, hsr, vreg_emulate_##reg);   \
          ASSERT(res);                                                \
          break;                                                      \
diff --git a/xen/arch/arm/vtimer.c b/xen/arch/arm/vtimer.c
index 167fc6127a..17b5649a05 100644
--- a/xen/arch/arm/vtimer.c
+++ b/xen/arch/arm/vtimer.c
@@ -162,7 +162,7 @@ void virt_timer_restore(struct vcpu *v)
      WRITE_SYSREG(v->arch.virt_timer.ctl, CNTV_CTL_EL0);
-static bool vtimer_cntp_ctl(struct cpu_user_regs *regs, uint32_t *r, bool read) +static bool vtimer_cntp_ctl(struct cpu_user_regs *regs, register_t *r, bool read)
      struct vcpu *v = current;
      s_time_t expires;
@@ -176,7 +176,7 @@ static bool vtimer_cntp_ctl(struct cpu_user_regs *regs, uint32_t *r, bool read)
-        uint32_t ctl = *r & ~CNTx_CTL_PENDING;
+        register_t ctl = *r & ~CNTx_CTL_PENDING;
You will still end up to mask the top 32-bit because CTx_CTL_PENDING is an unsigned 32-bit. I think we should not touch them top 32-bit at all so CNTx_CTL_PENDING (and probably CNT_x_CTL_ENABLE) should be defined as 1UL << X.

          if ( ctl & CNTx_CTL_ENABLE )
              ctl |= v->arch.phys_timer.ctl & CNTx_CTL_PENDING;
          v->arch.phys_timer.ctl = ctl;
@@ -197,7 +197,7 @@ static bool vtimer_cntp_ctl(struct cpu_user_regs *regs, uint32_t *r, bool read)
      return true;
-static bool vtimer_cntp_tval(struct cpu_user_regs *regs, uint32_t *r,
+static bool vtimer_cntp_tval(struct cpu_user_regs *regs, register_t *r,
                               bool read)
      struct vcpu *v = current;
@@ -211,11 +211,11 @@ static bool vtimer_cntp_tval(struct cpu_user_regs *regs, uint32_t *r,
      if ( read )
-        *r = (uint32_t)((v->arch.phys_timer.cval - cntpct) & 0xffffffffull); +        *r = (register_t)((v->arch.phys_timer.cval - cntpct) & 0xffffffffull);

This is computing the TimerVal is held in the first 32-bit of the registers. So I think this should stick to (uint32_t) here.

-        v->arch.phys_timer.cval = cntpct + (uint64_t)(int32_t)*r;
+        v->arch.phys_timer.cval = cntpct + (uint64_t)(register_t)*r;

This is not quite the same as before. We were using the first 32-bit as a signed value. Now, you are using the full register as a unsigned value.

          if ( v->arch.phys_timer.ctl & CNTx_CTL_ENABLE )
              v->arch.phys_timer.ctl &= ~CNTx_CTL_PENDING;
@@ -274,10 +274,10 @@ static bool vtimer_emulate_cp32(struct cpu_user_regs *regs, union hsr hsr)
      switch ( hsr.bits & HSR_CP32_REGS_MASK )
      case HSR_CPREG32(CNTP_CTL):
-        return vreg_emulate_cp32(regs, hsr, vtimer_cntp_ctl);
+        return vreg_emulate_cp(regs, hsr, vtimer_cntp_ctl);
      case HSR_CPREG32(CNTP_TVAL):
-        return vreg_emulate_cp32(regs, hsr, vtimer_cntp_tval);
+        return vreg_emulate_cp(regs, hsr, vtimer_cntp_tval);
          return false;
@@ -316,9 +316,9 @@ static bool vtimer_emulate_sysreg(struct cpu_user_regs *regs, union hsr hsr)
      switch ( hsr.bits & HSR_SYSREG_REGS_MASK )
-        return vreg_emulate_sysreg32(regs, hsr, vtimer_cntp_ctl);
+        return vreg_emulate_sysreg(regs, hsr, vtimer_cntp_ctl);
-        return vreg_emulate_sysreg32(regs, hsr, vtimer_cntp_tval);
+        return vreg_emulate_sysreg(regs, hsr, vtimer_cntp_tval);
          return vreg_emulate_sysreg64(regs, hsr, vtimer_cntp_cval);
diff --git a/xen/include/asm-arm/vreg.h b/xen/include/asm-arm/vreg.h
index 1253753833..cef55aabea 100644
--- a/xen/include/asm-arm/vreg.h
+++ b/xen/include/asm-arm/vreg.h
@@ -4,13 +4,13 @@
  #ifndef __ASM_ARM_VREG__
  #define __ASM_ARM_VREG__
-typedef bool (*vreg_reg32_fn_t)(struct cpu_user_regs *regs, uint32_t *r,
+typedef bool (*vreg_reg_fn_t)(struct cpu_user_regs *regs, register_t *r,
                                     bool read);
  typedef bool (*vreg_reg64_fn_t)(struct cpu_user_regs *regs, uint64_t *r,
                                     bool read);
-static inline bool vreg_emulate_cp32(struct cpu_user_regs *regs, union hsr hsr,
-                                     vreg_reg32_fn_t fn)
+static inline bool vreg_emulate_cp(struct cpu_user_regs *regs, union hsr hsr,
+                                     vreg_reg_fn_t fn)

The new name will add some confusion because now we have vreg_emulate_cp() (for 32-bit access) and vreg_emulate_c64() (for 64-bit access).

So I would rather keep the current naming.

      struct hsr_cp32 cp32 = hsr.cp32;
@@ -18,7 +18,7 @@ static inline bool vreg_emulate_cp32(struct cpu_user_regs *regs, union hsr hsr,
       * implementation error in the emulation (such as not correctly
       * setting r).
-    uint32_t r = 0;
+    register_t r = 0;
      bool ret;
      if ( !cp32.read )
@@ -64,11 +64,11 @@ static inline bool vreg_emulate_cp64(struct cpu_user_regs *regs, union hsr hsr,
  #ifdef CONFIG_ARM_64
-static inline bool vreg_emulate_sysreg32(struct cpu_user_regs *regs, union hsr hsr,
-                                         vreg_reg32_fn_t fn)
+static inline bool vreg_emulate_sysreg(struct cpu_user_regs *regs, union hsr hsr,
+                                         vreg_reg_fn_t fn)
      struct hsr_sysreg sysreg = hsr.sysreg;
-    uint32_t r = 0;
+    register_t r = 0;
      bool ret;
      if ( !sysreg.read )


Julien Grall



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