On Thu, Dec 19, 2013 at 07:54:49PM +0100, Roger Pau Monne wrote:
> AP startup on PVH follows the PV method, so we need to add a hook in
> order to diverge from bare metal.
> ---
> sys/amd64/amd64/mp_machdep.c | 16 ++++---
> sys/amd64/include/cpu.h | 1 +
> sys/x86/xen/hvm.c | 17 +++++++-
> sys/x86/xen/pv.c | 90
> ++++++++++++++++++++++++++++++++++++++++++
> sys/xen/pv.h | 1 +
> 5 files changed, 117 insertions(+), 8 deletions(-)
>
> diff --git a/sys/amd64/amd64/mp_machdep.c b/sys/amd64/amd64/mp_machdep.c
> index 4ef4b3d..e302886 100644
> --- a/sys/amd64/amd64/mp_machdep.c
> +++ b/sys/amd64/amd64/mp_machdep.c
> @@ -90,7 +90,7 @@ extern struct pcpu __pcpu[];
>
> /* AP uses this during bootstrap. Do not staticize. */
> char *bootSTK;
> -static int bootAP;
> +int bootAP;
>
> /* Free these after use */
> void *bootstacks[MAXCPU];
> @@ -122,9 +122,12 @@ u_long *ipi_rendezvous_counts[MAXCPU];
> static u_long *ipi_hardclock_counts[MAXCPU];
> #endif
>
> +int native_start_all_aps(void);
> +
> /* Default cpu_ops implementation. */
> struct cpu_ops cpu_ops = {
> - .ipi_vectored = lapic_ipi_vectored
> + .ipi_vectored = lapic_ipi_vectored,
> + .start_all_aps = native_start_all_aps,
> };
>
> extern inthand_t IDTVEC(fast_syscall), IDTVEC(fast_syscall32);
> @@ -138,7 +141,7 @@ extern int pmap_pcid_enabled;
> static volatile cpuset_t ipi_nmi_pending;
>
> /* used to hold the AP's until we are ready to release them */
> -static struct mtx ap_boot_mtx;
> +struct mtx ap_boot_mtx;
>
> /* Set to 1 once we're ready to let the APs out of the pen. */
> static volatile int aps_ready = 0;
> @@ -165,7 +168,6 @@ static int cpu_cores; /* cores per
> package */
>
> static void assign_cpu_ids(void);
> static void set_interrupt_apic_ids(void);
> -static int start_all_aps(void);
> static int start_ap(int apic_id);
> static void release_aps(void *dummy);
>
> @@ -569,7 +571,7 @@ cpu_mp_start(void)
> assign_cpu_ids();
>
> /* Start each Application Processor */
> - start_all_aps();
> + cpu_ops.start_all_aps();
>
> set_interrupt_apic_ids();
> }
> @@ -908,8 +910,8 @@ assign_cpu_ids(void)
> /*
> * start each AP in our list
> */
> -static int
> -start_all_aps(void)
> +int
> +native_start_all_aps(void)
> {
> vm_offset_t va = boot_address + KERNBASE;
> u_int64_t *pt4, *pt3, *pt2;
> diff --git a/sys/amd64/include/cpu.h b/sys/amd64/include/cpu.h
> index 3d9ff531..ed9f1db 100644
> --- a/sys/amd64/include/cpu.h
> +++ b/sys/amd64/include/cpu.h
> @@ -64,6 +64,7 @@ struct cpu_ops {
> void (*cpu_init)(void);
> void (*cpu_resume)(void);
> void (*ipi_vectored)(u_int, int);
> + int (*start_all_aps)(void);
> };
>
> extern struct cpu_ops cpu_ops;
> diff --git a/sys/x86/xen/hvm.c b/sys/x86/xen/hvm.c
> index fb1ed79..5ec9f3a 100644
> --- a/sys/x86/xen/hvm.c
> +++ b/sys/x86/xen/hvm.c
> @@ -53,6 +53,9 @@ __FBSDID("$FreeBSD$");
> #include <xen/hypervisor.h>
> #include <xen/hvm.h>
> #include <xen/xen_intr.h>
> +#ifdef __amd64__
> +#include <xen/pv.h>
> +#endif
>
> #include <xen/interface/hvm/params.h>
> #include <xen/interface/vcpu.h>
> @@ -97,6 +100,11 @@ extern void pmap_lazyfix_action(void);
> /* Variables used by mp_machdep to perform the bitmap IPI */
> extern volatile u_int cpu_ipi_pending[MAXCPU];
>
> +#ifdef __amd64__
> +/* Native AP start used on PVHVM */
> +extern int native_start_all_aps(void);
> +#endif
> +
> /*---------------------------------- Macros
> ----------------------------------*/
> #define IPI_TO_IDX(ipi) ((ipi) - APIC_IPI_INTS)
>
> @@ -119,7 +127,10 @@ enum xen_domain_type xen_domain_type = XEN_NATIVE;
> struct cpu_ops xen_hvm_cpu_ops = {
> .ipi_vectored = lapic_ipi_vectored,
> .cpu_init = xen_hvm_cpu_init,
> - .cpu_resume = xen_hvm_cpu_resume
> + .cpu_resume = xen_hvm_cpu_resume,
> +#ifdef __amd64__
> + .start_all_aps = native_start_all_aps,
> +#endif
> };
>
> static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support");
> @@ -698,6 +709,10 @@ xen_hvm_init(enum xen_hvm_init_type init_type)
> setup_xen_features();
> cpu_ops = xen_hvm_cpu_ops;
> vm_guest = VM_GUEST_XEN;
> +#ifdef __amd64__
> + if (xen_pv_domain())
> + cpu_ops.start_all_aps = xen_pv_start_all_aps;
> +#endif
> break;
> case XEN_HVM_INIT_RESUME:
> if (error != 0)
> diff --git a/sys/x86/xen/pv.c b/sys/x86/xen/pv.c
> index db576e0..7e45a83 100644
> --- a/sys/x86/xen/pv.c
> +++ b/sys/x86/xen/pv.c
> @@ -34,21 +34,43 @@ __FBSDID("$FreeBSD$");
> #include <sys/kernel.h>
> #include <sys/reboot.h>
> #include <sys/systm.h>
> +#include <sys/malloc.h>
> +#include <sys/lock.h>
> +#include <sys/mutex.h>
> +#include <sys/smp.h>
> +
> +#include <vm/vm.h>
> +#include <vm/pmap.h>
> +#include <vm/vm_extern.h>
> +#include <vm/vm_kern.h>
>
> #include <machine/sysarch.h>
> #include <machine/clock.h>
> #include <machine/pc/bios.h>
> +#include <machine/smp.h>
>
> #include <xen/xen-os.h>
> #include <xen/pv.h>
> #include <xen/hypervisor.h>
>
> +#include <xen/interface/vcpu.h>
> +
> #define MAX_E820_ENTRIES 128
>
> /*--------------------------- Forward Declarations
> ---------------------------*/
> static caddr_t xen_pv_parse_preload_data(u_int64_t);
> static void xen_pv_parse_memmap(caddr_t, vm_paddr_t *, int *);
>
> +/*---------------------------- Extern Declarations
> ---------------------------*/
> +/* Variables used by amd64 mp_machdep to start APs */
> +extern struct mtx ap_boot_mtx;
> +extern void *bootstacks[];
> +extern char *doublefault_stack;
> +extern char *nmi_stack;
> +extern void *dpcpu;
> +extern int bootAP;
> +extern char *bootSTK;
> +
> /*-------------------------------- Global Data
> -------------------------------*/
> /* Xen init_ops implementation. */
> struct init_ops xen_init_ops = {
> @@ -78,6 +100,74 @@ static struct
>
> static struct bios_smap xen_smap[MAX_E820_ENTRIES];
>
> +static int
> +start_xen_ap(int cpu)
> +{
> + struct vcpu_guest_context *ctxt;
> + int ms, cpus = mp_naps;
> +
> + ctxt = malloc(sizeof(*ctxt), M_TEMP, M_NOWAIT | M_ZERO);
> + if (ctxt == NULL)
> + panic("unable to allocate memory");
> +
> + ctxt->flags = VGCF_IN_KERNEL;
> + ctxt->user_regs.rip = (unsigned long) init_secondary;
> + ctxt->user_regs.rsp = (unsigned long) bootSTK;
> +
> + /* Set the AP to use the same page tables */
> + ctxt->ctrlreg[3] = KPML4phys;
> +
> + if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
> + panic("unable to initialize AP#%d\n", cpu);
> +
> + free(ctxt, M_TEMP);
> +
> + /* Launch the vCPU */
> + if (HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
> + panic("unable to start AP#%d\n", cpu);
Why the panic? Why not just return 0 (which I presume
would mean we failed to init the CPU?)
> +
> + /* Wait up to 5 seconds for it to start. */
> + for (ms = 0; ms < 5000; ms++) {
> + if (mp_naps > cpus)
> + return (1); /* return SUCCESS */
> + DELAY(1000);
> + }
> +
Wow. So much simpler than the PV path.
> + return (0);
There isn't a doorbell mechanism for the booted processor to tell
the BSP it is up?
> +}
> +
> +int
> +xen_pv_start_all_aps(void)
> +{
> + int cpu;
> +
> + mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
> +
> + for (cpu = 1; cpu < mp_ncpus; cpu++) {
> +
> + /* allocate and set up an idle stack data page */
> + bootstacks[cpu] = (void *)kmem_malloc(kernel_arena,
> + KSTACK_PAGES * PAGE_SIZE, M_WAITOK | M_ZERO);
> + doublefault_stack = (char *)kmem_malloc(kernel_arena,
> + PAGE_SIZE, M_WAITOK | M_ZERO);
> + nmi_stack = (char *)kmem_malloc(kernel_arena, PAGE_SIZE,
> + M_WAITOK | M_ZERO);
> + dpcpu = (void *)kmem_malloc(kernel_arena, DPCPU_SIZE,
> + M_WAITOK | M_ZERO);
> +
> + bootSTK = (char *)bootstacks[cpu] + KSTACK_PAGES * PAGE_SIZE -
> 8;
> + bootAP = cpu;
> +
> + /* attempt to start the Application Processor */
> + if (!start_xen_ap(cpu))
> + panic("AP #%d failed to start!", cpu);
> +
> + CPU_SET(cpu, &all_cpus); /* record AP in CPU map */
> + }
> +
> + return (mp_naps);
> +}
> +
> /*
> * Functions to convert the "extra" parameters passed by Xen
> * into FreeBSD boot options (from the i386 Xen port).
> diff --git a/sys/xen/pv.h b/sys/xen/pv.h
> index 71b8776..60d9def 100644
> --- a/sys/xen/pv.h
> +++ b/sys/xen/pv.h
> @@ -24,5 +24,6 @@
> #define __XEN_PV_H__
>
> void xen_pv_set_init_ops(void);
> +int xen_pv_start_all_aps(void);
>
> #endif /* __XEN_PV_H__ */
> --
> 1.7.7.5 (Apple Git-26)
>
>
> _______________________________________________
> Xen-devel mailing list
> xen-de...@lists.xen.org
> http://lists.xen.org/xen-devel
_______________________________________________
freebsd-xen@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-xen
To unsubscribe, send any mail to "freebsd-xen-unsubscr...@freebsd.org"
