Skip to contentSkip to conversation

x86/mm: merge ptwr and mmio_ro page fault handlers

PATCH In review Wei Liu 2 participants 4 messages Sep 1, 2017

lore
Share & tools
lore

Wei Liu

author
#0 lore raw

v3 cover x86/mm: merge ptwr and mmio_ro page fault handlers

Can be found at my xenbits/xen.git wip.merge-mmio-ptwr-emul-v3

Wei Liu (2):
x86/mm: don't wrap x86_emulate_ctxt in ptwr_emulate_ctxt x86/mm: merge ptwr and mmio_ro page fault handlers

 xen/arch/x86/mm.c        | 222 ++++++++++++++++++++---------------------------
 xen/arch/x86/traps.c     |  20 +++--
 xen/include/asm-x86/mm.h |   5 +-
 3 files changed, 104 insertions(+), 143 deletions(-)
Signature · 7 lines
2.11.0


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xen.org/xen-devel
author
#1 lore raw

PATCH v3 1/2 x86/mm: don't wrap x86_emulate_ctxt in ptwr_emulate_ctxt

Rewrite the code so that it has the same structure as mmio_ro_emualte_ctxt. x86_emulate_ctxt now points to ptwr_emulate_ctxt via its data pointer.

This patch will help unify mmio_ro and ptwr code paths later.

---
v3: drop unnecessary hunk, but kept Andrew's Rb because the patch is
not essentially changed.

v2: do away with pointer in ptwr_emulate_ctxt

---
 xen/arch/x86/mm.c | 42 +++++++++++++++++++-----------------------
 1 file changed, 19 insertions(+), 23 deletions(-)
xen/arch/x86/mm.c +19 -23
index f4b9747aba..8831f06605 100644
@@ -4964,7 +4964,6 @@ long arch_memory_op(unsigned long cmd, XEN_GUEST_HANDLE_PARAM(void) arg)
  */
 
 struct ptwr_emulate_ctxt {
-    struct x86_emulate_ctxt ctxt;
     unsigned long cr2;
     l1_pgentry_t  pte;
 };
@@ -4995,7 +4994,7 @@ static int ptwr_emulated_update(
     paddr_t val,
     unsigned int bytes,
     unsigned int do_cmpxchg,
-    struct ptwr_emulate_ctxt *ptwr_ctxt)
+    struct x86_emulate_ctxt *ctxt)
 {
     unsigned long mfn;
     unsigned long unaligned_addr = addr;
@@ -5003,6 +5002,7 @@ static int ptwr_emulated_update(
     l1_pgentry_t pte, ol1e, nl1e, *pl1e;
     struct vcpu *v = current;
     struct domain *d = v->domain;
+    struct ptwr_emulate_ctxt *ptwr_ctxt = ctxt->data;
     int ret;
 
     /* Only allow naturally-aligned stores within the original %cr2 page. */
@@ -5026,7 +5026,7 @@ static int ptwr_emulated_update(
         {
             x86_emul_pagefault(0, /* Read fault. */
                                addr + sizeof(paddr_t) - rc,
-                               &ptwr_ctxt->ctxt);
+                               ctxt);
             return X86EMUL_EXCEPTION;
         }
         /* Mask out bits provided by caller. */
@@ -5141,9 +5141,7 @@ static int ptwr_emulated_write(
 
     memcpy(&val, p_data, bytes);
 
-    return ptwr_emulated_update(
-        offset, 0, val, bytes, 0,
-        container_of(ctxt, struct ptwr_emulate_ctxt, ctxt));
+    return ptwr_emulated_update(offset, 0, val, bytes, 0, ctxt);
 }
 
 static int ptwr_emulated_cmpxchg(
@@ -5166,9 +5164,7 @@ static int ptwr_emulated_cmpxchg(
     memcpy(&old, p_old, bytes);
     memcpy(&new, p_new, bytes);
 
-    return ptwr_emulated_update(
-        offset, old, new, bytes, 1,
-        container_of(ctxt, struct ptwr_emulate_ctxt, ctxt));
+    return ptwr_emulated_update(offset, old, new, bytes, 1, ctxt);
 }
 
 static const struct x86_emulate_ops ptwr_emulate_ops = {
@@ -5187,14 +5183,14 @@ int ptwr_do_page_fault(struct vcpu *v, unsigned long addr,
     struct domain *d = v->domain;
     struct page_info *page;
     l1_pgentry_t      pte;
-    struct ptwr_emulate_ctxt ptwr_ctxt = {
-        .ctxt = {
-            .regs = regs,
-            .vendor = d->arch.cpuid->x86_vendor,
-            .addr_size = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
-            .sp_size   = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
-            .lma       = !is_pv_32bit_domain(d),
-        },
+    struct ptwr_emulate_ctxt ptwr_ctxt;
+    struct x86_emulate_ctxt ctxt = {
+       .regs = regs,
+       .vendor = d->arch.cpuid->x86_vendor,
+       .addr_size = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
+       .sp_size   = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
+       .lma       = !is_pv_32bit_domain(d),
+       .data      = &ptwr_ctxt,
     };
     int rc;
 
@@ -5224,7 +5220,7 @@ int ptwr_do_page_fault(struct vcpu *v, unsigned long addr,
     ptwr_ctxt.cr2 = addr;
     ptwr_ctxt.pte = pte;
 
-    rc = x86_emulate(&ptwr_ctxt.ctxt, &ptwr_emulate_ops);
+    rc = x86_emulate(&ctxt, &ptwr_emulate_ops);
 
     page_unlock(page);
     put_page(page);
@@ -5239,18 +5235,18 @@ int ptwr_do_page_fault(struct vcpu *v, unsigned long addr,
          * emulation bug, or a guest playing with the instruction stream under
          * Xen's feet.
          */
-        if ( ptwr_ctxt.ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
-             ptwr_ctxt.ctxt.event.vector == TRAP_page_fault )
-            pv_inject_event(&ptwr_ctxt.ctxt.event);
+        if ( ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
+             ctxt.event.vector == TRAP_page_fault )
+            pv_inject_event(&ctxt.event);
         else
             gdprintk(XENLOG_WARNING,
                      "Unexpected event (type %u, vector %#x) from emulation\n",
-                     ptwr_ctxt.ctxt.event.type, ptwr_ctxt.ctxt.event.vector);
+                     ctxt.event.type, ctxt.event.vector);
 
         /* Fallthrough */
     case X86EMUL_OKAY:
 
-        if ( ptwr_ctxt.ctxt.retire.singlestep )
+        if ( ctxt.retire.singlestep )
             pv_inject_hw_exception(TRAP_debug, X86_EVENT_NO_EC);
 
         /* Fallthrough */
Signature · 7 lines
2.11.0


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xen.org/xen-devel
author
#2 lore raw

PATCH v3 2/2 x86/mm: merge ptwr and mmio_ro page fault handlers

In reply to Wei Liu

Provide a unified entry to avoid going through pte look-up, decode and emulation cycle more than necessary. The path taken is determined by the faulting address.

Note that the order of checks is changed in the new function, but the order of the checks is performed shouldn't matter.

The sole caller is changed to use the new function.

No functional change.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>

v3:
1. place new function in the original location
2. remove goto out in ptwr_do_page_fault
3. add const to struct domain in mmio_ro_do_page_fault
4. remove the vcpu parameter in pv_ro_page_fault and use current->domain

v2: rebase and (sorta) merge Andrew's changes

Dom0 boots fine. XTF is happy with this change. Let me know if more tests can be done.

---
 xen/arch/x86/mm.c        | 210 ++++++++++++++++++++---------------------------
 xen/arch/x86/traps.c     |  20 +++--
 xen/include/asm-x86/mm.h |   5 +-
 3 files changed, 100 insertions(+), 135 deletions(-)
xen/arch/x86/mm.c +88 -122
index 8831f06605..4d76319170 100644
@@ -5177,86 +5177,40 @@ static const struct x86_emulate_ops ptwr_emulate_ops = {
 };
 
 /* Write page fault handler: check if guest is trying to modify a PTE. */
-int ptwr_do_page_fault(struct vcpu *v, unsigned long addr,
-                       struct cpu_user_regs *regs)
+static int ptwr_do_page_fault(struct x86_emulate_ctxt *ctxt,
+                              unsigned long addr, l1_pgentry_t pte)
 {
-    struct domain *d = v->domain;
-    struct page_info *page;
-    l1_pgentry_t      pte;
-    struct ptwr_emulate_ctxt ptwr_ctxt;
-    struct x86_emulate_ctxt ctxt = {
-       .regs = regs,
-       .vendor = d->arch.cpuid->x86_vendor,
-       .addr_size = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
-       .sp_size   = is_pv_32bit_domain(d) ? 32 : BITS_PER_LONG,
-       .lma       = !is_pv_32bit_domain(d),
-       .data      = &ptwr_ctxt,
+    struct ptwr_emulate_ctxt ptwr_ctxt = {
+        .cr2 = addr,
+        .pte = pte,
     };
+    struct page_info *page;
     int rc;
 
-    /* Attempt to read the PTE that maps the VA being accessed. */
-    pte = guest_get_eff_l1e(addr);
-
-    /* We are looking only for read-only mappings of p.t. pages. */
-    if ( ((l1e_get_flags(pte) & (_PAGE_PRESENT|_PAGE_RW)) != _PAGE_PRESENT) ||
-         rangeset_contains_singleton(mmio_ro_ranges, l1e_get_pfn(pte)) ||
-         !get_page_from_mfn(l1e_get_mfn(pte), d) )
-        goto bail;
+    if ( !get_page_from_mfn(l1e_get_mfn(pte), current->domain) )
+        return X86EMUL_UNHANDLEABLE;
 
     page = l1e_get_page(pte);
     if ( !page_lock(page) )
     {
         put_page(page);
-        goto bail;
+        return X86EMUL_UNHANDLEABLE;
     }
 
     if ( (page->u.inuse.type_info & PGT_type_mask) != PGT_l1_page_table )
     {
         page_unlock(page);
         put_page(page);
-        goto bail;
+        return X86EMUL_UNHANDLEABLE;
     }
 
-    ptwr_ctxt.cr2 = addr;
-    ptwr_ctxt.pte = pte;
-
-    rc = x86_emulate(&ctxt, &ptwr_emulate_ops);
+    ctxt->data = &ptwr_ctxt;
+    rc = x86_emulate(ctxt, &ptwr_emulate_ops);
 
     page_unlock(page);
     put_page(page);
 
-    switch ( rc )
-    {
-    case X86EMUL_EXCEPTION:
-        /*
-         * This emulation only covers writes to pagetables which are marked
-         * read-only by Xen.  We tolerate #PF (in case a concurrent pagetable
-         * update has succeeded on a different vcpu).  Anything else is an
-         * emulation bug, or a guest playing with the instruction stream under
-         * Xen's feet.
-         */
-        if ( ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
-             ctxt.event.vector == TRAP_page_fault )
-            pv_inject_event(&ctxt.event);
-        else
-            gdprintk(XENLOG_WARNING,
-                     "Unexpected event (type %u, vector %#x) from emulation\n",
-                     ctxt.event.type, ctxt.event.vector);
-
-        /* Fallthrough */
-    case X86EMUL_OKAY:
-
-        if ( ctxt.retire.singlestep )
-            pv_inject_hw_exception(TRAP_debug, X86_EVENT_NO_EC);
-
-        /* Fallthrough */
-    case X86EMUL_RETRY:
-        perfc_incr(ptwr_emulations);
-        return EXCRET_fault_fixed;
-    }
-
- bail:
-    return 0;
+    return rc;
 }
 
 /*************************
@@ -5332,80 +5286,28 @@ static const struct x86_emulate_ops mmcfg_intercept_ops = {
 };
 
 /* Check if guest is trying to modify a r/o MMIO page. */
-int mmio_ro_do_page_fault(struct vcpu *v, unsigned long addr,
-                          struct cpu_user_regs *regs)
+static int mmio_ro_do_page_fault(struct x86_emulate_ctxt *ctxt,
+                                 unsigned long addr, l1_pgentry_t pte)
 {
-    l1_pgentry_t pte;
-    unsigned long mfn;
-    unsigned int addr_size = is_pv_32bit_vcpu(v) ? 32 : BITS_PER_LONG;
     struct mmio_ro_emulate_ctxt mmio_ro_ctxt = { .cr2 = addr };
-    struct x86_emulate_ctxt ctxt = {
-        .regs = regs,
-        .vendor = v->domain->arch.cpuid->x86_vendor,
-        .addr_size = addr_size,
-        .sp_size = addr_size,
-        .lma = !is_pv_32bit_vcpu(v),
-        .data = &mmio_ro_ctxt,
-    };
-    int rc;
+    mfn_t mfn = l1e_get_mfn(pte);
 
-    /* Attempt to read the PTE that maps the VA being accessed. */
-    pte = guest_get_eff_l1e(addr);
-
-    /* We are looking only for read-only mappings of MMIO pages. */
-    if ( ((l1e_get_flags(pte) & (_PAGE_PRESENT|_PAGE_RW)) != _PAGE_PRESENT) )
-        return 0;
-
-    mfn = l1e_get_pfn(pte);
-    if ( mfn_valid(_mfn(mfn)) )
+    if ( mfn_valid(mfn) )
     {
-        struct page_info *page = mfn_to_page(_mfn(mfn));
-        struct domain *owner = page_get_owner_and_reference(page);
+        struct page_info *page = mfn_to_page(mfn);
+        const struct domain *owner = page_get_owner_and_reference(page);
 
         if ( owner )
             put_page(page);
         if ( owner != dom_io )
-            return 0;
+            return X86EMUL_UNHANDLEABLE;
     }
 
-    if ( !rangeset_contains_singleton(mmio_ro_ranges, mfn) )
-        return 0;
-
-    if ( pci_ro_mmcfg_decode(mfn, &mmio_ro_ctxt.seg, &mmio_ro_ctxt.bdf) )
-        rc = x86_emulate(&ctxt, &mmcfg_intercept_ops);
+    ctxt->data = &mmio_ro_ctxt;
+    if ( pci_ro_mmcfg_decode(mfn_x(mfn), &mmio_ro_ctxt.seg, &mmio_ro_ctxt.bdf) )
+        return x86_emulate(ctxt, &mmcfg_intercept_ops);
     else
-        rc = x86_emulate(&ctxt, &mmio_ro_emulate_ops);
-
-    switch ( rc )
-    {
-    case X86EMUL_EXCEPTION:
-        /*
-         * This emulation only covers writes to MMCFG space or read-only MFNs.
-         * We tolerate #PF (from hitting an adjacent page or a successful
-         * concurrent pagetable update).  Anything else is an emulation bug,
-         * or a guest playing with the instruction stream under Xen's feet.
-         */
-        if ( ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
-             ctxt.event.vector == TRAP_page_fault )
-            pv_inject_event(&ctxt.event);
-        else
-            gdprintk(XENLOG_WARNING,
-                     "Unexpected event (type %u, vector %#x) from emulation\n",
-                     ctxt.event.type, ctxt.event.vector);
-
-        /* Fallthrough */
-    case X86EMUL_OKAY:
-
-        if ( ctxt.retire.singlestep )
-            pv_inject_hw_exception(TRAP_debug, X86_EVENT_NO_EC);
-
-        /* Fallthrough */
-    case X86EMUL_RETRY:
-        perfc_incr(mmio_ro_emulations);
-        return EXCRET_fault_fixed;
-    }
-
-    return 0;
+        return x86_emulate(ctxt, &mmio_ro_emulate_ops);
 }
 
 void *alloc_xen_pagetable(void)
@@ -6439,6 +6341,70 @@ void write_32bit_pse_identmap(uint32_t *l2)
                  _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
 }
 
+int pv_ro_page_fault(unsigned long addr, struct cpu_user_regs *regs)
+{
+    l1_pgentry_t pte;
+    const struct domain *currd = current->domain;
+    unsigned int addr_size = is_pv_32bit_domain(currd) ? 32 : BITS_PER_LONG;
+    struct x86_emulate_ctxt ctxt = {
+        .regs      = regs,
+        .vendor    = currd->arch.cpuid->x86_vendor,
+        .addr_size = addr_size,
+        .sp_size   = addr_size,
+        .lma       = addr_size > 32,
+    };
+    int rc;
+    bool mmio_ro;
+
+    /* Attempt to read the PTE that maps the VA being accessed. */
+    pte = guest_get_eff_l1e(addr);
+
+    /* We are only looking for read-only mappings */
+    if ( ((l1e_get_flags(pte) & (_PAGE_PRESENT | _PAGE_RW)) != _PAGE_PRESENT) )
+        return 0;
+
+    mmio_ro = rangeset_contains_singleton(mmio_ro_ranges, l1e_get_pfn(pte));
+    if ( mmio_ro )
+        rc = mmio_ro_do_page_fault(&ctxt, addr, pte);
+    else
+        rc = ptwr_do_page_fault(&ctxt, addr, pte);
+
+    switch ( rc )
+    {
+    case X86EMUL_EXCEPTION:
+        /*
+         * This emulation covers writes to:
+         *  - L1 pagetables.
+         *  - MMCFG space or read-only MFNs.
+         * We tolerate #PF (from hitting an adjacent page or a successful
+         * concurrent pagetable update).  Anything else is an emulation bug,
+         * or a guest playing with the instruction stream under Xen's feet.
+         */
+        if ( ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
+             ctxt.event.vector == TRAP_page_fault )
+            pv_inject_event(&ctxt.event);
+        else
+            gdprintk(XENLOG_WARNING,
+                     "Unexpected event (type %u, vector %#x) from emulation\n",
+                     ctxt.event.type, ctxt.event.vector);
+
+        /* Fallthrough */
+    case X86EMUL_OKAY:
+        if ( ctxt.retire.singlestep )
+            pv_inject_hw_exception(TRAP_debug, X86_EVENT_NO_EC);
+
+        /* Fallthrough */
+    case X86EMUL_RETRY:
+        if ( mmio_ro )
+            perfc_incr(mmio_ro_emulations);
+        else
+            perfc_incr(ptwr_emulations);
+        return EXCRET_fault_fixed;
+    }
+
+    return 0;
+}
+
 /*
  * Local variables:
  * mode: C
xen/arch/x86/traps.c +11 -9
index f525fa28d3..6091f239ce 100644
@@ -1308,16 +1308,18 @@ static int fixup_page_fault(unsigned long addr, struct cpu_user_regs *regs)
          !(regs->error_code & (PFEC_reserved_bit | PFEC_insn_fetch)) &&
          (regs->error_code & PFEC_write_access) )
     {
-        if ( VM_ASSIST(d, writable_pagetables) &&
-             /* Do not check if access-protection fault since the page may
-                legitimately be not present in shadow page tables */
-             (paging_mode_enabled(d) ||
-              (regs->error_code & PFEC_page_present)) &&
-             ptwr_do_page_fault(v, addr, regs) )
-            return EXCRET_fault_fixed;
+        bool ptwr, mmio_ro;
+
+        ptwr = VM_ASSIST(d, writable_pagetables) &&
+               /* Do not check if access-protection fault since the page may
+                  legitimately be not present in shadow page tables */
+               (paging_mode_enabled(d) ||
+                (regs->error_code & PFEC_page_present));
+
+        mmio_ro = is_hardware_domain(d) &&
+                  (regs->error_code & PFEC_page_present);
 
-        if ( is_hardware_domain(d) && (regs->error_code & PFEC_page_present) &&
-             mmio_ro_do_page_fault(v, addr, regs) )
+        if ( (ptwr || mmio_ro) && pv_ro_page_fault(addr, regs) )
             return EXCRET_fault_fixed;
     }
 
xen/include/asm-x86/mm.h +1 -4
index 4c03a33e79..97647dc401 100644
@@ -511,10 +511,7 @@ extern int mmcfg_intercept_write(enum x86_segment seg,
 int pv_emul_cpuid(uint32_t leaf, uint32_t subleaf,
                   struct cpuid_leaf *res, struct x86_emulate_ctxt *ctxt);
 
-int  ptwr_do_page_fault(struct vcpu *, unsigned long,
-                        struct cpu_user_regs *);
-int  mmio_ro_do_page_fault(struct vcpu *, unsigned long,
-                           struct cpu_user_regs *);
+int pv_ro_page_fault(unsigned long, struct cpu_user_regs *);
 
 int audit_adjust_pgtables(struct domain *d, int dir, int noisy);
 
Signature · 7 lines
2.11.0


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xen.org/xen-devel
Reviewed-by on v3 2/2
#3 lore raw
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 01.09.17 at 16:35, <[email protected]> wrote:

Provide a unified entry to avoid going through pte look-up, decode and emulation cycle more than necessary. The path taken is determined by the faulting address.

Quoted 8 lines

Note that the order of checks is changed in the new function, but the order of the checks is performed shouldn't matter.

The sole caller is changed to use the new function.

No functional change.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xen.org/xen-devel