Date: Mon, 10 Jun 2013 00:56:16 +0200 From: Olivier Houchard <cognet@ci0.org> To: Andrew Turner <andrew@fubar.geek.nz> Cc: freebsd-arm@freebsd.org Subject: Re: RFC: Patches with AXP support and pmap&smp fixes. Message-ID: <20130609225615.GA42548@ci0.org> In-Reply-To: <20130602130713.70b0b9f0@bender.Home> References: <517E8610.5050204@semihalf.com> <20130430142701.5bbfec2b@bender.lan> <20130430143311.GA71966@ci0.org> <518775B4.1010308@semihalf.com> <20130506134129.GA60131@ci0.org> <20130602130713.70b0b9f0@bender.Home>
next in thread | previous in thread | raw e-mail | index | archive | help
--7JfCtLOvnd9MIVvH
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
On Sun, Jun 02, 2013 at 01:07:13PM +0100, Andrew Turner wrote:
> On Mon, 6 May 2013 15:41:29 +0200
> Olivier Houchard <cognet@ci0.org> wrote:
>
> > On Mon, May 06, 2013 at 11:19:48AM +0200, Grzegorz Bernacki wrote:
> > >
> > > Hi,
> > >
> > > Our patch fixes only initialization of pcpu in pcpu_init(). In 4
> > > cores setup it is possible the at least two cores simultaneously
> > > updating queue of pcpus which causes corruption.
> > > I am not aware of any other problems with SMP. Let us try setup
> > > with WITNESS enabled and we'll see if we have the same problem.
> > > Olivier, could you share your patch? In case we have the problem we
> > > would like to have it.
> > >
> >
> > Hi,
> >
> > Sure I will dust it off, and send it. Maybe not before Sunday though,
> > I'll be mostly MIA for the week. I'm quite interested in you guys
> > having a look at it.
>
> Did you manage to extract the patch? I don't remember seeing it.
>
> Andrew
Hi Andrew,
Sorry for the long delay. I finally took time to update it to the latest
-CURRENT, and check that it boots.
Regards,
Olivier
--7JfCtLOvnd9MIVvH
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment; filename="pcpu-patch.diff"
Index: arm/arm/cpufunc_asm_armv7.S
===================================================================
--- arm/arm/cpufunc_asm_armv7.S (revision 251584)
+++ arm/arm/cpufunc_asm_armv7.S (working copy)
@@ -57,9 +57,9 @@
#define PT_OUTER_WBWA (1 << 3)
#ifdef SMP
-#define PT_ATTR (PT_S|PT_INNER_WT|PT_OUTER_WT|PT_NOS)
+#define PT_ATTR (PT_S|PT_INNER_WBWA|PT_OUTER_WBWA|PT_NOS)
#else
-#define PT_ATTR (PT_INNER_WT|PT_OUTER_WT)
+#define PT_ATTR (PT_INNER_WBWA|PT_OUTER_WBWA)
#endif
ENTRY(armv7_setttb)
@@ -98,7 +98,7 @@
ldr r1, .Lpage_mask
bic r0, r0, r1
#ifdef SMP
- mcr p15, 0, r0, c8, c3, 1 /* flush D tlb single entry Inner Shareable*/
+ mcr p15, 0, r0, c8, c3, 3 /* flush D tlb single entry Inner Shareable*/
mcr p15, 0, r0, c7, c1, 6 /* flush BTB Inner Shareable */
#else
mcr p15, 0, r0, c8, c7, 1 /* flush D tlb single entry */
@@ -113,6 +113,7 @@
ENTRY(armv7_dcache_wbinv_all)
stmdb sp!, {r4, r5, r6, r7, r8, r9}
+ dsb
/* Get cache level */
ldr r0, .Lcoherency_level
ldr r3, [r0]
@@ -188,6 +189,7 @@
and r2, r0, r3
add r1, r1, r2
bic r0, r0, r3
+ dsb
.Larmv7_wb_next:
mcr p15, 0, r0, c7, c10, 1 /* Clean D cache SE with VA */
add r0, r0, ip
@@ -203,6 +205,7 @@
and r2, r0, r3
add r1, r1, r2
bic r0, r0, r3
+ dsb
.Larmv7_wbinv_next:
mcr p15, 0, r0, c7, c14, 1 /* Purge D cache SE with VA */
add r0, r0, ip
@@ -222,6 +225,7 @@
and r2, r0, r3
add r1, r1, r2
bic r0, r0, r3
+ dsb
.Larmv7_inv_next:
mcr p15, 0, r0, c7, c6, 1 /* Invalidate D cache SE with VA */
add r0, r0, ip
@@ -237,6 +241,7 @@
and r2, r0, r3
add r1, r1, r2
bic r0, r0, r3
+ dsb
.Larmv7_id_wbinv_next:
mcr p15, 0, r0, c7, c5, 1 /* Invalidate I cache SE with VA */
mcr p15, 0, r0, c7, c14, 1 /* Purge D cache SE with VA */
Index: arm/arm/copystr.S
===================================================================
--- arm/arm/copystr.S (revision 251584)
+++ arm/arm/copystr.S (working copy)
@@ -51,14 +51,12 @@
.align 0
#ifdef _ARM_ARCH_6
-#define GET_PCB(tmp) \
- mrc p15, 0, tmp, c13, c0, 4; \
- add tmp, tmp, #(PC_CURPCB)
+KSTACK_LOCALS
#else
.Lpcb:
.word _C_LABEL(__pcpu) + PC_CURPCB
-#define GET_PCB(tmp) \
+#define GET_PCB(tmp, tmp2) \
ldr tmp, .Lpcb
#endif
@@ -114,8 +112,7 @@
moveq r0, #ENAMETOOLONG
beq 2f
- GET_PCB(r4)
- ldr r4, [r4]
+ GET_PCB(r4, r5)
#ifdef DIAGNOSTIC
teq r4, #0x00000000
@@ -162,8 +159,7 @@
moveq r0, #ENAMETOOLONG
beq 2f
- GET_PCB(r4)
- ldr r4, [r4]
+ GET_PCB(r4, r5)
#ifdef DIAGNOSTIC
teq r4, #0x00000000
Index: arm/arm/genassym.c
===================================================================
--- arm/arm/genassym.c (revision 251584)
+++ arm/arm/genassym.c (working copy)
@@ -25,6 +25,8 @@
*
*/
+#include "opt_kstack_max_pages.h"
+
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
@@ -69,6 +71,7 @@
ASSYM(PCB_R12, offsetof(struct pcb, un_32.pcb32_r12));
ASSYM(PCB_PC, offsetof(struct pcb, un_32.pcb32_pc));
ASSYM(PCB_SP, offsetof(struct pcb, un_32.pcb32_sp));
+ASSYM(PCB_CURTHREAD, offsetof(struct pcb, un_32.pcb32_curthread));
ASSYM(PC_CURPCB, offsetof(struct pcpu, pc_curpcb));
ASSYM(PC_CURTHREAD, offsetof(struct pcpu, pc_curthread));
@@ -138,3 +141,6 @@
ASSYM(MAXCOMLEN, MAXCOMLEN);
ASSYM(NIRQ, NIRQ);
+
+ASSYM(KSTACK_ADDR_MASK, KSTACK_MAX_PAGES * PAGE_SIZE - 1);
+ASSYM(KSTACK_PCB_ADDR, KSTACK_MAX_PAGES * PAGE_SIZE - sizeof(struct pcb));
Index: arm/arm/mp_machdep.c
===================================================================
--- arm/arm/mp_machdep.c (revision 251584)
+++ arm/arm/mp_machdep.c (working copy)
@@ -160,12 +160,13 @@
}
extern vm_paddr_t pmap_pa;
+
+void init_secondary_finish(void);
+
void
init_secondary(int cpu)
{
struct pcpu *pc;
- uint32_t loop_counter;
- int start = 0, end = 0;
cpu_setup(NULL);
setttb(pmap_pa);
@@ -195,7 +196,6 @@
;
/* Initialize curthread */
- KASSERT(PCPU_GET(idlethread) != NULL, ("no idle thread"));
pc->pc_curthread = pc->pc_idlethread;
pc->pc_curpcb = pc->pc_idlethread->td_pcb;
#ifdef ARM_VFP_SUPPORT
@@ -203,7 +203,16 @@
vfp_init();
#endif
+ __asm __volatile("mov sp, %0\n"
+ "b init_secondary_finish\n" : : "r" (pc->pc_idlethread->td_pcb->un_32.pcb32_sp));
+}
+void
+init_secondary_finish(void)
+{
+ int start = 0, end = 0;
+ uint32_t loop_counter;
+
mtx_lock_spin(&ap_boot_mtx);
atomic_add_rel_32(&smp_cpus, 1);
Index: arm/arm/machdep.c
===================================================================
--- arm/arm/machdep.c (revision 251584)
+++ arm/arm/machdep.c (working copy)
@@ -42,6 +42,7 @@
* Updated : 18/04/01 updated for new wscons
*/
+#include "opt_kstack_max_pages.h"
#include "opt_compat.h"
#include "opt_ddb.h"
#include "opt_platform.h"
@@ -106,6 +107,7 @@
#include <dev/ofw/openfirm.h>
#endif
+#define DEBUG
#ifdef DEBUG
#define debugf(fmt, args...) printf(fmt, ##args)
#else
@@ -398,6 +400,7 @@
USPACE_UNDEF_STACK_TOP;
pcb->un_32.pcb32_sp = (u_int)thread0.td_kstack +
USPACE_SVC_STACK_TOP;
+ pcb->un_32.pcb32_curthread = &thread0;
vector_page_setprot(VM_PROT_READ);
pmap_set_pcb_pagedir(pmap_kernel(), pcb);
pmap_postinit();
@@ -874,9 +877,9 @@
set_pcpu(pcpup);
#endif
pcpu_init(pcpup, 0, sizeof(struct pcpu));
- PCPU_SET(curthread, &thread0);
+ pcpup->pc_curthread = &thread0;
#ifdef ARM_VFP_SUPPORT
- PCPU_SET(cpu, 0);
+ pcpup->pc_cpu = 0;
#endif
}
@@ -1028,6 +1031,7 @@
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;
thread0.td_pcb->pcb_flags = 0;
+ thread0.td_pcb->un_32.pcb32_curthread = &thread0;
thread0.td_frame = &proc0_tf;
pcpup->pc_curpcb = thread0.td_pcb;
}
@@ -1181,6 +1185,12 @@
phys_avail[j + 1] = 0;
}
+void do_putc(char);
+void do_putc(char c)
+{
+ *(volatile char *)0x48020000 = c;
+}
+
void *
initarm(struct arm_boot_params *abp)
{
@@ -1232,8 +1242,10 @@
/* Grab physical memory regions information from device tree. */
if (fdt_get_mem_regions(memory_regions, &memory_regions_sz,
- &memsize) != 0)
+ &memsize) != 0) {
+ do_putc('a');
while(1);
+ }
/* Grab physical memory regions information from device tree. */
if (fdt_get_reserved_regions(reserved_regions, &reserved_regions_sz) != 0)
@@ -1364,8 +1376,20 @@
valloc_pages(irqstack, IRQ_STACK_SIZE * MAXCPU);
valloc_pages(abtstack, ABT_STACK_SIZE * MAXCPU);
valloc_pages(undstack, UND_STACK_SIZE * MAXCPU);
- valloc_pages(kernelstack, KSTACK_PAGES * MAXCPU);
valloc_pages(msgbufpv, round_page(msgbufsize) / PAGE_SIZE);
+ if ((KSTACK_MAX_PAGES * PAGE_SIZE) -
+ (freemempos & (KSTACK_MAX_MASK)) >= KSTACK_PAGES)
+ freemempos = (freemempos &~ KSTACK_MAX_MASK) +
+ (KSTACK_MAX_PAGES * PAGE_SIZE) -
+ (KSTACK_PAGES * PAGE_SIZE);
+
+ else
+ freemempos = (freemempos &~ KSTACK_MAX_MASK) +
+ (KSTACK_MAX_PAGES * PAGE_SIZE) +
+ 2 * (KSTACK_MAX_PAGES * PAGE_SIZE) -
+ (KSTACK_PAGES * PAGE_SIZE);
+
+ valloc_pages(kernelstack, KSTACK_PAGES);
/*
* Now we start construction of the L1 page table
@@ -1486,12 +1510,14 @@
init_proc0(kernelstack.pv_va);
arm_intrnames_init();
+ printf("hoho\n");
arm_vector_init(ARM_VECTORS_HIGH, ARM_VEC_ALL);
+ printf("hehe\n");
arm_dump_avail_init(memsize, sizeof(dump_avail) / sizeof(dump_avail[0]));
+ printf("plop\n");
pmap_bootstrap(freemempos, &kernel_l1pt);
+ printf("unplop\n");
msgbufp = (void *)msgbufpv.pv_va;
- msgbufinit(msgbufp, msgbufsize);
- mutex_init();
/*
* Prepare map of physical memory regions available to vm subsystem.
@@ -1500,6 +1526,7 @@
init_param2(physmem);
kdb_init();
+ msgbufinit(msgbufp, msgbufsize);
return ((void *)(kernelstack.pv_va + USPACE_SVC_STACK_TOP -
sizeof(struct pcb)));
Index: arm/arm/fusu.S
===================================================================
--- arm/arm/fusu.S (revision 251584)
+++ arm/arm/fusu.S (working copy)
@@ -40,14 +40,13 @@
__FBSDID("$FreeBSD$");
#ifdef _ARM_ARCH_6
-#define GET_PCB(tmp) \
- mrc p15, 0, tmp, c13, c0, 4; \
- add tmp, tmp, #(PC_CURPCB)
+KSTACK_LOCALS
#else
.Lcurpcb:
.word _C_LABEL(__pcpu) + PC_CURPCB
-#define GET_PCB(tmp) \
- ldr tmp, .Lcurpcb
+#define GET_PCB(tmp, tmp2) \
+ ldr tmp, .Lcurpcb; \
+ ldr tmp, [tmp]
#endif
/*
@@ -57,21 +56,28 @@
ENTRY_NP(casuword32)
ENTRY(casuword)
- GET_PCB(r3)
- ldr r3, [r3]
-
-#ifdef DIAGNOSTIC
- teq r3, #0x00000000
- beq .Lfusupcbfault
-#endif
stmfd sp!, {r4, r5}
+ GET_PCB(r3, r4)
+
adr r4, .Lcasuwordfault
str r4, [r3, #PCB_ONFAULT]
+#ifdef _ARM_ARCH_6
+1:
+ ldrex r5, [r0]
+ cmp r5, r1
+ movne r0, r5
+ bne 2f
+ strex r5, r2, [r0]
+ cmp r5, #0
+ bne 1b
+#else
ldrt r5, [r0]
cmp r5, r1
movne r0, r5
streqt r2, [r0]
+#endif
moveq r0, r1
+2:
ldmfd sp!, {r4, r5}
mov r1, #0x00000000
str r1, [r3, #PCB_ONFAULT]
@@ -97,14 +103,8 @@
ENTRY_NP(fuword32)
ENTRY(fuword)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r1, .Lfusufault
str r1, [r2, #PCB_ONFAULT]
@@ -123,14 +123,8 @@
*/
ENTRY(fusword)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r1, .Lfusufault
str r1, [r2, #PCB_ONFAULT]
@@ -159,14 +153,8 @@
mvnne r0, #0x00000000
RETne
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r1, _C_LABEL(fusubailout)
str r1, [r2, #PCB_ONFAULT]
@@ -199,14 +187,8 @@
*/
ENTRY(fubyte)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r1, .Lfusufault
str r1, [r2, #PCB_ONFAULT]
@@ -242,36 +224,15 @@
mvn r0, #0x00000000
RET
-#ifdef DIAGNOSTIC
/*
- * Handle earlier faults from [fs]u*(), due to no pcb
- */
-
-.Lfusupcbfault:
- mov r1, r0
- adr r0, fusupcbfaulttext
- b _C_LABEL(panic)
-
-fusupcbfaulttext:
- .asciz "Yikes - no valid PCB during fusuxxx() addr=%08x\n"
- .align 0
-#endif
-
-/*
* suword(caddr_t uaddr, int x);
* Store an int in the user's address space.
*/
ENTRY_NP(suword32)
ENTRY(suword)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r3, .Lfusufault
str r3, [r2, #PCB_ONFAULT]
@@ -296,14 +257,8 @@
mvnne r0, #0x00000000
RETne
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r3, _C_LABEL(fusubailout)
str r3, [r2, #PCB_ONFAULT]
@@ -327,14 +282,8 @@
*/
ENTRY(susword)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r3, .Lfusufault
str r3, [r2, #PCB_ONFAULT]
@@ -358,15 +307,8 @@
*/
ENTRY(subyte)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
-
-#ifdef DIAGNOSTIC
- teq r2, #0x00000000
- beq .Lfusupcbfault
-#endif
-
adr r3, .Lfusufault
str r3, [r2, #PCB_ONFAULT]
Index: arm/arm/gic.c
===================================================================
--- arm/arm/gic.c (revision 251584)
+++ arm/arm/gic.c (working copy)
@@ -253,7 +253,7 @@
active_irq &= 0x3FF;
if (active_irq == 0x3FF) {
- if (last_irq == -1)
+ if (last_irq == -1 && 0)
printf("Spurious interrupt detected [0x%08x]\n", active_irq);
return -1;
}
Index: arm/arm/swtch.S
===================================================================
--- arm/arm/swtch.S (revision 251584)
+++ arm/arm/swtch.S (working copy)
@@ -192,7 +192,7 @@
#endif
/* We have a new curthread now so make a note it */
- GET_CURTHREAD_PTR(r6)
+ GET_PCPU_CURTHREAD_PTR(r6)
str r5, [r6]
/* Set the new tp */
@@ -226,7 +226,7 @@
/* Process is now on a processor. */
/* We have a new curthread now so make a note it */
- GET_CURTHREAD_PTR(r7)
+ GET_PCPU_CURTHREAD_PTR(r7)
str r1, [r7]
/* Hook in a new pcb */
@@ -434,7 +434,11 @@
str r6, [r4, #TD_LOCK]
#if defined(SCHED_ULE) && defined(SMP)
ldr r6, .Lblocked_lock
- GET_CURTHREAD_PTR(r3)
+ /*
+ * It's fine to use GET_PCPU_CURTHREAD_PTR here, because interrupts
+ * are disabled, so it will be atomic.
+ */
+ GET_PCPU_CURTHREAD_PTR(r3)
1:
ldr r4, [r3, #TD_LOCK]
Index: arm/arm/bcopyinout.S
===================================================================
--- arm/arm/bcopyinout.S (revision 251584)
+++ arm/arm/bcopyinout.S (working copy)
@@ -39,6 +39,7 @@
#include "assym.s"
#include <machine/asm.h>
+#include <machine/asmacros.h>
#include <sys/errno.h>
.L_arm_memcpy:
@@ -55,14 +56,12 @@
.align 0
#ifdef _ARM_ARCH_6
-#define GET_PCB(tmp) \
- mrc p15, 0, tmp, c13, c0, 4; \
- add tmp, tmp, #(PC_CURPCB)
+KSTACK_LOCALS
#else
.Lcurpcb:
.word _C_LABEL(__pcpu) + PC_CURPCB
-#define GET_PCB(tmp) \
+#define GET_PCB(tmp, tmp2) \
ldr tmp, .Lcurpcb
#endif
@@ -116,8 +115,7 @@
.Lnormal:
SAVE_REGS
- GET_PCB(r4)
- ldr r4, [r4]
+ GET_PCB(r4, r5)
ldr r5, [r4, #PCB_ONFAULT]
@@ -354,8 +352,7 @@
.Lnormale:
SAVE_REGS
- GET_PCB(r4)
- ldr r4, [r4]
+ GET_PCB(r4, r5)
ldr r5, [r4, #PCB_ONFAULT]
adr r3, .Lcopyfault
@@ -549,8 +546,7 @@
* else EFAULT if a page fault occurred.
*/
ENTRY(badaddr_read_1)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
ldr ip, [r2, #PCB_ONFAULT]
adr r3, 1f
@@ -575,8 +571,7 @@
* else EFAULT if a page fault occurred.
*/
ENTRY(badaddr_read_2)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
ldr ip, [r2, #PCB_ONFAULT]
adr r3, 1f
@@ -601,8 +596,7 @@
* else EFAULT if a page fault occurred.
*/
ENTRY(badaddr_read_4)
- GET_PCB(r2)
- ldr r2, [r2]
+ GET_PCB(r2, r3)
ldr ip, [r2, #PCB_ONFAULT]
adr r3, 1f
Index: arm/arm/pmap-v6.c
===================================================================
--- arm/arm/pmap-v6.c (revision 251584)
+++ arm/arm/pmap-v6.c (working copy)
@@ -1818,7 +1818,7 @@
cpu_tlb_flushID();
cpu_cpwait();
if (vector_page < KERNBASE) {
- struct pcb *curpcb = PCPU_GET(curpcb);
+ struct pcb *_curpcb = curpcb;
pcb = thread0.td_pcb;
if (pmap_is_current(pmap)) {
/*
@@ -1838,10 +1838,10 @@
* Make sure cpu_switch(), et al, DTRT. This is safe to do
* since this process has no remaining mappings of its own.
*/
- curpcb->pcb_pl1vec = pcb->pcb_pl1vec;
- curpcb->pcb_l1vec = pcb->pcb_l1vec;
- curpcb->pcb_dacr = pcb->pcb_dacr;
- curpcb->pcb_pagedir = pcb->pcb_pagedir;
+ _curpcb->pcb_pl1vec = pcb->pcb_pl1vec;
+ _curpcb->pcb_l1vec = pcb->pcb_l1vec;
+ _curpcb->pcb_dacr = pcb->pcb_dacr;
+ _curpcb->pcb_pagedir = pcb->pcb_pagedir;
}
pmap_free_l1(pmap);
@@ -2229,6 +2229,8 @@
pmap_kextract(vm_offset_t va)
{
+ if (kernel_vm_end == 0)
+ return (0);
return (pmap_extract_locked(kernel_pmap, va));
}
Index: arm/arm/vm_machdep.c
===================================================================
--- arm/arm/vm_machdep.c (revision 251584)
+++ arm/arm/vm_machdep.c (working copy)
@@ -131,6 +131,7 @@
#endif
td2->td_pcb = pcb2;
bcopy(td1->td_pcb, pcb2, sizeof(*pcb2));
+ pcb2->un_32.pcb32_curthread = td2;
mdp2 = &p2->p_md;
bcopy(&td1->td_proc->p_md, mdp2, sizeof(*mdp2));
pcb2->un_32.pcb32_und_sp = td2->td_kstack + USPACE_UNDEF_STACK_TOP;
@@ -344,6 +345,7 @@
bcopy(td0->td_frame, td->td_frame, sizeof(struct trapframe));
bcopy(td0->td_pcb, td->td_pcb, sizeof(struct pcb));
+ td->td_pcb->un_32.pcb32_curthread = td;
tf = td->td_frame;
sf = (struct switchframe *)tf - 1;
sf->sf_r4 = (u_int)fork_return;
@@ -406,6 +408,7 @@
{
td->td_pcb = (struct pcb *)(td->td_kstack + td->td_kstack_pages *
PAGE_SIZE) - 1;
+ td->td_pcb->un_32.pcb32_curthread = td;
/*
* Ensure td_frame is aligned to an 8 byte boundary as it will be
* placed into the stack pointer which must be 8 byte aligned in
Index: arm/arm/cpufunc.c
===================================================================
--- arm/arm/cpufunc.c (revision 251584)
+++ arm/arm/cpufunc.c (working copy)
@@ -1116,10 +1116,10 @@
armv7_tlb_flushID, /* tlb_flushID */
armv7_tlb_flushID_SE, /* tlb_flushID_SE */
- arm11_tlb_flushI, /* tlb_flushI */
- arm11_tlb_flushI_SE, /* tlb_flushI_SE */
- arm11_tlb_flushD, /* tlb_flushD */
- arm11_tlb_flushD_SE, /* tlb_flushD_SE */
+ armv7_tlb_flushID, /* tlb_flushI */
+ armv7_tlb_flushID_SE, /* tlb_flushI_SE */
+ armv7_tlb_flushID, /* tlb_flushD */
+ armv7_tlb_flushID_SE, /* tlb_flushD_SE */
/* Cache operations */
Index: arm/arm/pmap.c
===================================================================
--- arm/arm/pmap.c (revision 251584)
+++ arm/arm/pmap.c (working copy)
@@ -2452,7 +2452,7 @@
pmap_tlb_flushID(pmap);
cpu_cpwait();
if (vector_page < KERNBASE) {
- struct pcb *curpcb = PCPU_GET(curpcb);
+ struct pcb *_curpcb = curpcb;
pcb = thread0.td_pcb;
if (pmap_is_current(pmap)) {
/*
@@ -2472,10 +2472,10 @@
* Make sure cpu_switch(), et al, DTRT. This is safe to do
* since this process has no remaining mappings of its own.
*/
- curpcb->pcb_pl1vec = pcb->pcb_pl1vec;
- curpcb->pcb_l1vec = pcb->pcb_l1vec;
- curpcb->pcb_dacr = pcb->pcb_dacr;
- curpcb->pcb_pagedir = pcb->pcb_pagedir;
+ _curpcb->pcb_pl1vec = pcb->pcb_pl1vec;
+ _curpcb->pcb_l1vec = pcb->pcb_l1vec;
+ _curpcb->pcb_dacr = pcb->pcb_dacr;
+ _curpcb->pcb_pagedir = pcb->pcb_pagedir;
}
pmap_free_l1(pmap);
Index: arm/arm/locore.S
===================================================================
--- arm/arm/locore.S (revision 251584)
+++ arm/arm/locore.S (working copy)
@@ -225,6 +225,7 @@
/* init arm will return the new stack pointer. */
mov sp, r0
+ bl _C_LABEL(mutex_init)
bl _C_LABEL(mi_startup) /* call mi_startup()! */
@@ -261,6 +262,7 @@
MMU_INIT(PHYSADDR, PHYSADDR , 64, L1_TYPE_S|L1_S_C|L1_S_AP(AP_KRW))
/* map VA 0xc0000000..0xc3ffffff to PA */
MMU_INIT(KERNBASE, PHYSADDR, 64, L1_TYPE_S|L1_S_C|L1_S_AP(AP_KRW))
+ MMU_INIT(0x48000000, 0x48000000, 1, L1_TYPE_S|L1_SHARED|L1_S_C|L1_S_AP(AP_KRW))
#else
MMU_INIT(PHYSADDR, PHYSADDR , 64, L1_TYPE_S|L1_SHARED|L1_S_C|L1_S_AP(AP_KRW))
/* map VA 0xc0000000..0xc3ffffff to PA */
Index: arm/arm/bcopyinout_xscale.S
===================================================================
--- arm/arm/bcopyinout_xscale.S (revision 251584)
+++ arm/arm/bcopyinout_xscale.S (working copy)
@@ -42,13 +42,11 @@
.align 0
#ifdef _ARM_ARCH_6
-#define GET_PCB(tmp) \
- mrc p15, 0, tmp, c13, c0, 4; \
- add tmp, tmp, #(PC_CURPCB)
+KSTACK_LOCALS
#else
.Lcurpcb:
.word _C_LABEL(__pcpu) + PC_CURPCB
-#define GET_PCB(tmp) \
+#define GET_PCB(tmp, tmp2) \
ldr tmp, .Lcurpcb
#endif
@@ -88,8 +86,7 @@
.Lnormal:
stmfd sp!, {r10-r11, lr}
- GET_PCB(r10)
- ldr r10, [r10]
+ GET_PCB(r10, r11)
mov r3, #0x00
adr ip, .Lcopyin_fault
@@ -530,8 +527,7 @@
.Lnormale:
stmfd sp!, {r10-r11, lr}
- GET_PCB(r10)
- ldr r10, [r10]
+ GET_PCB(r10, r11)
mov r3, #0x00
adr ip, .Lcopyout_fault
Index: arm/include/param.h
===================================================================
--- arm/include/param.h (revision 251584)
+++ arm/include/param.h (working copy)
@@ -124,6 +124,12 @@
#define KSTACK_GUARD_PAGES 1
#endif /* !KSTACK_GUARD_PAGES */
+#ifndef KSTACK_MAX_PAGES
+#define KSTACK_MAX_PAGES 32
+#endif /* !KSTACK_MAX_PAGES */
+
+#define KSTACK_MAX_MASK ((KSTACK_MAX_PAGES * PAGE_SIZE) - 1)
+
#define USPACE_SVC_STACK_TOP KSTACK_PAGES * PAGE_SIZE
#define USPACE_SVC_STACK_BOTTOM (USPACE_SVC_STACK_TOP - 0x1000)
#define USPACE_UNDEF_STACK_TOP (USPACE_SVC_STACK_BOTTOM - 0x10)
Index: arm/include/pcb.h
===================================================================
--- arm/include/pcb.h (revision 251584)
+++ arm/include/pcb.h (working copy)
@@ -63,6 +63,7 @@
u_int pcb32_lr;
u_int pcb32_pc;
u_int pcb32_und_sp;
+ void * pcb32_curthread;
};
#define pcb_pagedir un_32.pcb32_pagedir
#define pcb_pl1vec un_32.pcb32_pl1vec
Index: arm/include/asmacros.h
===================================================================
--- arm/include/asmacros.h (revision 251584)
+++ arm/include/asmacros.h (working copy)
@@ -216,8 +216,23 @@
name:
#ifdef _ARM_ARCH_6
-#define AST_LOCALS
-#define GET_CURTHREAD_PTR(tmp) \
+#define KSTACK_LOCALS \
+.Lkstack_consts: \
+ .word KSTACK_ADDR_MASK; \
+ .word KSTACK_PCB_ADDR;
+#define AST_LOCALS KSTACK_LOCALS
+
+#define GET_PCB(tmp, tmp2) \
+ ldr tmp2, .Lkstack_consts; \
+ bic tmp, sp, tmp2; \
+ ldr tmp2, .Lkstack_consts + 4; \
+ add tmp, tmp, tmp2
+
+#define GET_CURTHREAD(tmp, tmp2) \
+ GET_PCB(tmp, tmp2); \
+ ldr tmp, [tmp, #PCB_CURTHREAD]
+
+#define GET_PCPU_CURTHREAD_PTR(tmp) \
mrc p15, 0, tmp, c13, c0, 4; \
add tmp, tmp, #(PC_CURTHREAD)
#else
@@ -225,8 +240,12 @@
.Lcurthread: ;\
.word _C_LABEL(__pcpu) + PC_CURTHREAD
-#define GET_CURTHREAD_PTR(tmp) \
+#define GET_PCPU_CURTHREAD_PTR(tmp) \
ldr tmp, .Lcurthread
+
+#define GET_CURTHREAD(tmp, tmp2) \
+ ldr tmp, .Lcurthread; \
+ ldr tmp, [tmp]
#endif
#define DO_AST \
@@ -238,8 +257,7 @@
teq r0, #(PSR_USR32_MODE) ;\
bne 2f /* Nope, get out now */ ;\
bic r4, r4, #(I32_bit|F32_bit) ;\
-1: GET_CURTHREAD_PTR(r5) ;\
- ldr r5, [r5] ;\
+1: GET_CURTHREAD(r5, r6) ;\
ldr r1, [r5, #(TD_FLAGS)] ;\
and r1, r1, #(TDF_ASTPENDING|TDF_NEEDRESCHED) ;\
teq r1, #0x00000000 ;\
Index: arm/include/pcpu.h
===================================================================
--- arm/include/pcpu.h (revision 251584)
+++ arm/include/pcpu.h (working copy)
@@ -33,8 +33,14 @@
#ifdef _KERNEL
#include <machine/cpuconf.h>
+#include <machine/cpufunc.h>
#include <machine/frame.h>
+#include <machine/pcb.h>
+#include <machine/atomic.h>
+#include <sys/types.h>
+#include <sys/systm.h>
+
#define ALT_STACK_SIZE 128
struct vmspace;
@@ -94,15 +100,97 @@
__asm __volatile("mcr p15, 0, %0, c13, c0, 3" : : "r" (tls));
}
+
+/*
+ * Evaluates to the byte offset of the per-cpu variable name.
+ */
+#define __pcpu_offset(name) \
+ __offsetof(struct pcpu, name)
+
+/*
+ * Evaluates to the type of the per-cpu variable name.
+ */
+#define __pcpu_type(name) \
+ __typeof(((struct pcpu *)0)->name)
+
+/*
+ * Evaluates to the address of the per-cpu variable name.
+ */
+#define __PCPU_PTR(name) __extension__ ({ \
+ __pcpu_type(name) *__p; \
+ int __s; \
+ \
+ __s = disable_interrupts(I32_bit | F32_bit); \
+ __p = &get_pcpu()->name; \
+ restore_interrupts(__s); \
+ __p; \
+})
+
+/*
+ * Evaluates to the value of the per-cpu variable name.
+ */
+#define __PCPU_GET(name) __extension__ ({ \
+ int __s; \
+ __pcpu_type(name) __res; \
+ __s = disable_interrupts(I32_bit | F32_bit); \
+ __res = get_pcpu()->name; \
+ restore_interrupts(__s); \
+ __res; \
+})
+
+/*
+ * Adds the value to the per-cpu counter name. The implementation
+ * must be atomic with respect to interrupts.
+ */
+#define __PCPU_ADD(name, val) do { \
+ int __s ; \
+ __s = disable_interrupts(I32_bit | F32_bit); \
+ get_pcpu()->name += (val); \
+ restore_interrupts(__s); \
+} while (0)
+
+/*
+ * Sets the value of the per-cpu variable name to value val.
+ */
+#define __PCPU_SET(name, val) { \
+ int __s; \
+ __s = disable_interrupts(I32_bit | F32_bit); \
+ get_pcpu()->name = val; \
+ restore_interrupts(__s); \
+}
+
+#define PCPU_GET(member) __PCPU_GET(pc_ ## member)
+#define PCPU_ADD(member, val) __PCPU_ADD(pc_ ## member, val)
+#define PCPU_INC(member) PCPU_ADD(member, 1)
+#define PCPU_PTR(member) __PCPU_PTR(pc_ ## member)
+#define PCPU_SET(member, val) __PCPU_SET(pc_ ## member, val)
+
+static __inline struct pcb *
+__curpcb(void)
+{
+ register_t sp;
+
+ __asm __volatile("mov %0, sp" : "=r" (sp));
+ return ((struct pcb *)((sp &~ KSTACK_MAX_MASK) +
+ (KSTACK_MAX_PAGES * PAGE_SIZE - sizeof(struct pcb))));
+}
+
+
+#define curpcb (__curpcb())
+#if 0
+#define curthread ((struct thread *)(curpcb->un_32.pcb32_curthread))
+#endif
+
+
#else
#define get_pcpu() pcpup
-#endif
#define PCPU_GET(member) (get_pcpu()->pc_ ## member)
#define PCPU_ADD(member, value) (get_pcpu()->pc_ ## member += (value))
#define PCPU_INC(member) PCPU_ADD(member, 1)
#define PCPU_PTR(member) (&get_pcpu()->pc_ ## member)
#define PCPU_SET(member,value) (get_pcpu()->pc_ ## member = (value))
+#endif
void pcpu0_init(void);
#endif /* _KERNEL */
Index: arm/include/pmap.h
===================================================================
--- arm/include/pmap.h (revision 251584)
+++ arm/include/pmap.h (working copy)
@@ -63,7 +63,7 @@
#endif
#define PTE_CACHE 6
#define PTE_DEVICE 2
-#define PTE_PAGETABLE 4
+#define PTE_PAGETABLE 6
#else
#define PTE_NOCACHE 1
#define PTE_CACHE 2
@@ -463,7 +463,7 @@
#if (ARM_MMU_SA1 == 1) && (ARM_NMMUS == 1)
#define PMAP_NEEDS_PTE_SYNC 1
#define PMAP_INCLUDE_PTE_SYNC
-#elif defined(CPU_XSCALE_81342)
+#elif defined(CPU_XSCALE_81342) || defined(ARM_ARCH_7A)
#define PMAP_NEEDS_PTE_SYNC 1
#define PMAP_INCLUDE_PTE_SYNC
#elif (ARM_MMU_SA1 == 0)
@@ -529,11 +529,17 @@
#define PMAP_INCLUDE_PTE_SYNC
#endif
+#ifdef ARM_L2_PIPT
+#define _sync_l2(pte, size) cpu_l2cache_wb_range(vtophys(pte), size)
+#else
+#define _sync_l2(pte, size) cpu_l2_cache_wb_range(pte, size)
+#endif
+
#define PTE_SYNC(pte) \
do { \
if (PMAP_NEEDS_PTE_SYNC) { \
cpu_dcache_wb_range((vm_offset_t)(pte), sizeof(pt_entry_t));\
- cpu_l2cache_wb_range((vm_offset_t)(pte), sizeof(pt_entry_t));\
+ _sync_l2((vm_offset_t)(pte), sizeof(pt_entry_t));\
} else \
cpu_drain_writebuf(); \
} while (/*CONSTCOND*/0)
@@ -543,7 +549,7 @@
if (PMAP_NEEDS_PTE_SYNC) { \
cpu_dcache_wb_range((vm_offset_t)(pte), \
(cnt) << 2); /* * sizeof(pt_entry_t) */ \
- cpu_l2cache_wb_range((vm_offset_t)(pte), \
+ _sync_l2((vm_offset_t)(pte), \
(cnt) << 2); /* * sizeof(pt_entry_t) */ \
} else \
cpu_drain_writebuf(); \
@@ -703,6 +709,21 @@
void arm_unmap_nocache(void *, vm_size_t);
extern vm_paddr_t dump_avail[];
+
+/*
+ * We want to be able to determine the end address of the kstack, to be able
+ * to get the PCB address just using the SP.
+ */
+
+static __inline void
+pmap_align_stack(vm_offset_t *addr, vm_size_t length)
+{
+ if (((*addr + length) & (KSTACK_MAX_PAGES * PAGE_SIZE - 1)) == 0)
+ return;
+ *addr += (KSTACK_MAX_PAGES * PAGE_SIZE) - ((*addr + length) &
+ (KSTACK_MAX_PAGES * PAGE_SIZE - 1));
+}
+
#endif /* _KERNEL */
#endif /* !LOCORE */
Index: arm/ti/omap4/omap4_mp.c
===================================================================
--- arm/ti/omap4/omap4_mp.c (revision 251584)
+++ arm/ti/omap4/omap4_mp.c (working copy)
@@ -71,7 +71,7 @@
panic("Couldn't map the SCU\n");
/* Enable the SCU */
*(volatile unsigned int *)scu_addr |= 1;
- //*(volatile unsigned int *)(scu_addr + 0x30) |= 1;
+ *(volatile unsigned int *)(scu_addr + 0x30) |= 1;
cpu_idcache_wbinv_all();
cpu_l2cache_wbinv_all();
ti_smc0(0x200, 0xfffffdff, MODIFY_AUX_CORE_0);
Index: vm/vm_map.c
===================================================================
--- vm/vm_map.c (revision 251584)
+++ vm/vm_map.c (working copy)
@@ -1465,6 +1465,11 @@
pmap_align_tlb(addr);
break;
#endif
+#ifdef VMFS_STACK_ALIGNED_SPACE
+ case VMFS_STACK_ALIGNED_SPACE:
+ pmap_align_stack(addr, length);
+ break;
+#endif
default:
break;
}
@@ -1477,6 +1482,9 @@
#ifdef VMFS_TLB_ALIGNED_SPACE
|| find_space == VMFS_TLB_ALIGNED_SPACE
#endif
+#ifdef VMFS_STACK_ALIGNED_SPACE
+ || find_space == VMFS_STACK_ALIGNED_SPACE
+#endif
));
vm_map_unlock(map);
return (result);
Index: vm/vm_map.h
===================================================================
--- vm/vm_map.h (revision 251584)
+++ vm/vm_map.h (working copy)
@@ -346,6 +346,9 @@
#if defined(__mips__)
#define VMFS_TLB_ALIGNED_SPACE 3 /* find a TLB entry aligned range */
#endif
+#if defined(__arm__)
+#define VMFS_STACK_ALIGNED_SPACE 4 /* Align stack so that the last addres is predictable */
+#endif
/*
* vm_map_wire and vm_map_unwire option flags
Index: vm/vm_glue.c
===================================================================
--- vm/vm_glue.c (revision 251584)
+++ vm/vm_glue.c (working copy)
@@ -371,6 +371,9 @@
*/
ks = kmem_alloc_nofault_space(kernel_map,
(pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_TLB_ALIGNED_SPACE);
+#elif defined(__arm__)
+ ks = kmem_alloc_nofault_space(kernel_map,
+ (pages + KSTACK_GUARD_PAGES) * PAGE_SIZE, VMFS_STACK_ALIGNED_SPACE);
#else
ks = kmem_alloc_nofault(kernel_map,
(pages + KSTACK_GUARD_PAGES) * PAGE_SIZE);
--7JfCtLOvnd9MIVvH--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130609225615.GA42548>