amd64: stop doing special allocation for the AP startup trampoline

There is no reason now why do we need to allocate trampoline page very
early in the boot process.  The only requirement for the page is that
it is below 1M to be usable by the real mode during init.  This can be
handled by vm_alloc_contig() when we do the startup.

Also assert that startup trampoline fits into single page.  In principle
we can do multi-page allocation if needed, but it is not.

Move the alloc_ap_trampoline() function and the boot_address variable to
i386/mp_machdep.c.  Keep existing mechanism of early alloc on i386.

Reviewed by:	markj
Sponsored by:	The FreeBSD Foundation
MFC after:	2 weeks
Differential revision:	https://reviews.freebsd.org/D31343
This commit is contained in:
Konstantin Belousov 2021-07-29 03:22:35 +03:00
parent 59b83c47e2
commit b27fe1c3ba
5 changed files with 77 additions and 71 deletions

View File

@ -1272,16 +1272,6 @@ getmemsize(caddr_t kmdp, u_int64_t first)
(boothowto & RB_VERBOSE))
printf("Physical memory use set to %ldK\n", Maxmem * 4);
/*
* Make hole for "AP -> long mode" bootstrap code. The
* mp_bootaddress vector is only available when the kernel
* is configured to support APs and APs for the system start
* in real mode mode (e.g. SMP bare metal).
*/
#ifdef SMP
alloc_ap_trampoline(physmap, &physmap_idx);
#endif
/* call pmap initialization to make new kernel address space */
pmap_bootstrap(&first);

View File

@ -112,7 +112,7 @@ extern u_int mptramp_nx;
* Local data and functions.
*/
static int start_ap(int apic_id);
static int start_ap(int apic_id, vm_paddr_t boot_address);
/*
* Initialize the IPI handlers and start up the AP's.
@ -322,17 +322,25 @@ mp_realloc_pcpu(int cpuid, int domain)
int
start_all_aps(void)
{
vm_page_t m_pml4, m_pdp, m_pd[4];
vm_page_t m_boottramp, m_pml4, m_pdp, m_pd[4];
pml5_entry_t old_pml45;
pml4_entry_t *v_pml4;
pdp_entry_t *v_pdp;
pd_entry_t *v_pd;
vm_paddr_t boot_address;
u_int32_t mpbioswarmvec;
int apic_id, cpu, domain, i;
u_char mpbiosreason;
mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN);
MPASS(bootMP_size <= PAGE_SIZE);
m_boottramp = vm_page_alloc_contig(NULL, 0, VM_ALLOC_NORMAL |
VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ, 1, 0,
(1ULL << 20), /* Trampoline should be below 1M for real mode */
PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
boot_address = VM_PAGE_TO_PHYS(m_boottramp);
/* Create a transient 1:1 mapping of low 4G */
if (la57) {
m_pml4 = pmap_page_alloc_below_4g(true);
@ -382,7 +390,7 @@ start_all_aps(void)
/* copy the AP 1st level boot code */
bcopy(mptramp_start, (void *)PHYS_TO_DMAP(boot_address), bootMP_size);
if (bootverbose)
printf("AP boot address %#x\n", boot_address);
printf("AP boot address %#lx\n", boot_address);
/* save the current value of the warm-start vector */
if (!efi_boot)
@ -436,7 +444,7 @@ start_all_aps(void)
bootAP = cpu;
/* attempt to start the Application Processor */
if (!start_ap(apic_id)) {
if (!start_ap(apic_id, boot_address)) {
/* restore the warmstart vector */
*(u_int32_t *) WARMBOOT_OFF = mpbioswarmvec;
panic("AP #%d (PHY# %d) failed!", cpu, apic_id);
@ -461,6 +469,7 @@ start_all_aps(void)
vm_page_free(m_pd[1]);
vm_page_free(m_pd[0]);
vm_page_free(m_pdp);
vm_page_free(m_boottramp);
/* number of APs actually started */
return (mp_naps);
@ -474,7 +483,7 @@ start_all_aps(void)
* but it seems to work.
*/
static int
start_ap(int apic_id)
start_ap(int apic_id, vm_paddr_t boot_address)
{
int vector, ms;
int cpus;

View File

@ -147,6 +147,61 @@ static int start_ap(int apic_id);
static char *ap_copyout_buf;
static char *ap_tramp_stack_base;
unsigned int boot_address;
#define MiB(v) (v ## ULL << 20)
/* Allocate memory for the AP trampoline. */
void
alloc_ap_trampoline(vm_paddr_t *physmap, unsigned int *physmap_idx)
{
unsigned int i;
bool allocated;
allocated = false;
for (i = *physmap_idx; i <= *physmap_idx; i -= 2) {
/*
* Find a memory region big enough and below the 1MB boundary
* for the trampoline code.
* NB: needs to be page aligned.
*/
if (physmap[i] >= MiB(1) ||
(trunc_page(physmap[i + 1]) - round_page(physmap[i])) <
round_page(bootMP_size))
continue;
allocated = true;
/*
* Try to steal from the end of the region to mimic previous
* behaviour, else fallback to steal from the start.
*/
if (physmap[i + 1] < MiB(1)) {
boot_address = trunc_page(physmap[i + 1]);
if ((physmap[i + 1] - boot_address) < bootMP_size)
boot_address -= round_page(bootMP_size);
physmap[i + 1] = boot_address;
} else {
boot_address = round_page(physmap[i]);
physmap[i] = boot_address + round_page(bootMP_size);
}
if (physmap[i] == physmap[i + 1] && *physmap_idx != 0) {
memmove(&physmap[i], &physmap[i + 2],
sizeof(*physmap) * (*physmap_idx - i + 2));
*physmap_idx -= 2;
}
break;
}
if (!allocated) {
boot_address = basemem * 1024 - bootMP_size;
if (bootverbose)
printf(
"Cannot find enough space for the boot trampoline, placing it at %#x",
boot_address);
}
}
/*
* Initialize the IPI handlers and start up the AP's.
*/

View File

@ -23,6 +23,10 @@
struct pmap;
#ifdef __i386__
extern unsigned int boot_address;
#endif
/* global data in mp_x86.c */
extern int mp_naps;
extern int boot_cpu_id;
@ -32,7 +36,6 @@ extern int bootAP;
extern void *dpcpu;
extern char *bootSTK;
extern void *bootstacks[];
extern unsigned int boot_address;
extern unsigned int bootMP_size;
extern volatile int aps_ready;
extern struct mtx ap_boot_mtx;
@ -84,12 +87,15 @@ inthand_t
typedef void (*smp_invl_cb_t)(struct pmap *, vm_offset_t addr1,
vm_offset_t addr2);
#ifdef __i386__
void alloc_ap_trampoline(vm_paddr_t *physmap, unsigned int *physmap_idx);
#endif
/* functions in x86_mp.c */
void assign_cpu_ids(void);
void cpu_add(u_int apic_id, char boot_cpu);
void cpustop_handler(void);
void cpususpend_handler(void);
void alloc_ap_trampoline(vm_paddr_t *physmap, unsigned int *physmap_idx);
void init_secondary_tail(void);
void init_secondary(void);
void ipi_startup(int apic_id, int vector);

View File

@ -168,14 +168,10 @@ struct cache_info {
int present;
} static caches[MAX_CACHE_LEVELS];
unsigned int boot_address;
static bool stop_mwait = false;
SYSCTL_BOOL(_machdep, OID_AUTO, stop_mwait, CTLFLAG_RWTUN, &stop_mwait, 0,
"Use MONITOR/MWAIT when stopping CPU, if available");
#define MiB(v) (v ## ULL << 20)
void
mem_range_AP_init(void)
{
@ -939,56 +935,6 @@ cpu_mp_probe(void)
return (mp_ncpus > 1);
}
/* Allocate memory for the AP trampoline. */
void
alloc_ap_trampoline(vm_paddr_t *physmap, unsigned int *physmap_idx)
{
unsigned int i;
bool allocated;
allocated = false;
for (i = *physmap_idx; i <= *physmap_idx; i -= 2) {
/*
* Find a memory region big enough and below the 1MB boundary
* for the trampoline code.
* NB: needs to be page aligned.
*/
if (physmap[i] >= MiB(1) ||
(trunc_page(physmap[i + 1]) - round_page(physmap[i])) <
round_page(bootMP_size))
continue;
allocated = true;
/*
* Try to steal from the end of the region to mimic previous
* behaviour, else fallback to steal from the start.
*/
if (physmap[i + 1] < MiB(1)) {
boot_address = trunc_page(physmap[i + 1]);
if ((physmap[i + 1] - boot_address) < bootMP_size)
boot_address -= round_page(bootMP_size);
physmap[i + 1] = boot_address;
} else {
boot_address = round_page(physmap[i]);
physmap[i] = boot_address + round_page(bootMP_size);
}
if (physmap[i] == physmap[i + 1] && *physmap_idx != 0) {
memmove(&physmap[i], &physmap[i + 2],
sizeof(*physmap) * (*physmap_idx - i + 2));
*physmap_idx -= 2;
}
break;
}
if (!allocated) {
boot_address = basemem * 1024 - bootMP_size;
if (bootverbose)
printf(
"Cannot find enough space for the boot trampoline, placing it at %#x",
boot_address);
}
}
/*
* AP CPU's call this to initialize themselves.
*/