]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
KVM: s390: Optimize paths where get_vcpu_asce() is invoked
authorAlexander Yarygin <yarygin@linux.vnet.ibm.com>
Thu, 22 Jan 2015 09:44:11 +0000 (12:44 +0300)
committerChristian Borntraeger <borntraeger@de.ibm.com>
Tue, 17 Mar 2015 15:25:31 +0000 (16:25 +0100)
During dynamic address translation the get_vcpu_asce()
function can be invoked several times. It's ok for usual modes, but will
be slow if CPUs are in AR mode. Let's call the get_vcpu_asce() once and
pass the result to the called functions.

Signed-off-by: Alexander Yarygin <yarygin@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: David Hildenbrand <dahi@linux.vnet.ibm.com>
Acked-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
arch/s390/kvm/gaccess.c

index 494131eda8c16d70423992ef75faa2e1827d1196..c74462a12c6dee909b8d32b77c446c42033a020b 100644 (file)
@@ -330,6 +330,7 @@ static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
  * @vcpu: virtual cpu
  * @gva: guest virtual address
  * @gpa: points to where guest physical (absolute) address should be stored
+ * @asce: effective asce
  * @write: indicates if access is a write access
  *
  * Translate a guest virtual address into a guest absolute address by means
@@ -345,7 +346,8 @@ static int deref_table(struct kvm *kvm, unsigned long gpa, unsigned long *val)
  *           by the architecture
  */
 static unsigned long guest_translate(struct kvm_vcpu *vcpu, unsigned long gva,
-                                    unsigned long *gpa, int write)
+                                    unsigned long *gpa, const union asce asce,
+                                    int write)
 {
        union vaddress vaddr = {.addr = gva};
        union raddress raddr = {.addr = gva};
@@ -354,12 +356,10 @@ static unsigned long guest_translate(struct kvm_vcpu *vcpu, unsigned long gva,
        union ctlreg0 ctlreg0;
        unsigned long ptr;
        int edat1, edat2;
-       union asce asce;
 
        ctlreg0.val = vcpu->arch.sie_block->gcr[0];
        edat1 = ctlreg0.edat && test_kvm_facility(vcpu->kvm, 8);
        edat2 = edat1 && test_kvm_facility(vcpu->kvm, 78);
-       asce.val = get_vcpu_asce(vcpu);
        if (asce.r)
                goto real_address;
        ptr = asce.origin * 4096;
@@ -506,15 +506,14 @@ static inline int is_low_address(unsigned long ga)
        return (ga & ~0x11fful) == 0;
 }
 
-static int low_address_protection_enabled(struct kvm_vcpu *vcpu)
+static int low_address_protection_enabled(struct kvm_vcpu *vcpu,
+                                         const union asce asce)
 {
        union ctlreg0 ctlreg0 = {.val = vcpu->arch.sie_block->gcr[0]};
        psw_t *psw = &vcpu->arch.sie_block->gpsw;
-       union asce asce;
 
        if (!ctlreg0.lap)
                return 0;
-       asce.val = get_vcpu_asce(vcpu);
        if (psw_bits(*psw).t && asce.p)
                return 0;
        return 1;
@@ -536,7 +535,7 @@ enum {
 
 static int guest_page_range(struct kvm_vcpu *vcpu, unsigned long ga,
                            unsigned long *pages, unsigned long nr_pages,
-                           int write)
+                           const union asce asce, int write)
 {
        struct kvm_s390_pgm_info *pgm = &vcpu->arch.pgm;
        psw_t *psw = &vcpu->arch.sie_block->gpsw;
@@ -547,7 +546,7 @@ static int guest_page_range(struct kvm_vcpu *vcpu, unsigned long ga,
        tec_bits = (struct trans_exc_code_bits *)&pgm->trans_exc_code;
        tec_bits->fsi = write ? FSI_STORE : FSI_FETCH;
        tec_bits->as = psw_bits(*psw).as;
-       lap_enabled = low_address_protection_enabled(vcpu);
+       lap_enabled = low_address_protection_enabled(vcpu, asce);
        while (nr_pages) {
                ga = kvm_s390_logical_to_effective(vcpu, ga);
                tec_bits->addr = ga >> PAGE_SHIFT;
@@ -557,7 +556,7 @@ static int guest_page_range(struct kvm_vcpu *vcpu, unsigned long ga,
                }
                ga &= PAGE_MASK;
                if (psw_bits(*psw).t) {
-                       rc = guest_translate(vcpu, ga, pages, write);
+                       rc = guest_translate(vcpu, ga, pages, asce, write);
                        if (rc < 0)
                                return rc;
                        if (rc == PGM_PROTECTION)
@@ -604,7 +603,7 @@ int access_guest(struct kvm_vcpu *vcpu, unsigned long ga, ar_t ar, void *data,
        need_ipte_lock = psw_bits(*psw).t && !asce.r;
        if (need_ipte_lock)
                ipte_lock(vcpu);
-       rc = guest_page_range(vcpu, ga, pages, nr_pages, write);
+       rc = guest_page_range(vcpu, ga, pages, nr_pages, asce, write);
        for (idx = 0; idx < nr_pages && !rc; idx++) {
                gpa = *(pages + idx) + (ga & ~PAGE_MASK);
                _len = min(PAGE_SIZE - (gpa & ~PAGE_MASK), len);
@@ -671,16 +670,16 @@ int guest_translate_address(struct kvm_vcpu *vcpu, unsigned long gva, ar_t ar,
        tec->as = psw_bits(*psw).as;
        tec->fsi = write ? FSI_STORE : FSI_FETCH;
        tec->addr = gva >> PAGE_SHIFT;
-       if (is_low_address(gva) && low_address_protection_enabled(vcpu)) {
+       asce.val = get_vcpu_asce(vcpu);
+       if (is_low_address(gva) && low_address_protection_enabled(vcpu, asce)) {
                if (write) {
                        rc = pgm->code = PGM_PROTECTION;
                        return rc;
                }
        }
 
-       asce.val = get_vcpu_asce(vcpu);
        if (psw_bits(*psw).t && !asce.r) {      /* Use DAT? */
-               rc = guest_translate(vcpu, gva, gpa, write);
+               rc = guest_translate(vcpu, gva, gpa, asce, write);
                if (rc > 0) {
                        if (rc == PGM_PROTECTION)
                                tec->b61 = 1;