]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
powerpc: Remove ksp_limit on ppc64
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>
Tue, 24 Sep 2013 05:17:21 +0000 (15:17 +1000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Wed, 25 Sep 2013 04:15:51 +0000 (14:15 +1000)
We've been keeping that field in thread_struct for a while, it contains
the "limit" of the current stack pointer and is meant to be used for
detecting stack overflows.

It has a few problems however:

 - First, it was never actually *used* on 64-bit. Set and updated but
not actually exploited

 - When switching stack to/from irq and softirq stacks, it's update
is racy unless we hard disable interrupts, which is costly. This
is fine on 32-bit as we don't soft-disable there but not on 64-bit.

Thus rather than fixing 2 in order to implement 1 in some hypothetical
future, let's remove the code completely from 64-bit. In order to avoid
a clutter of ifdef's, we remove the updates from C code completely
during interrupt stack switching, and instead maintain it from the
asm helper that is used to do the stack switching in the first place.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/processor.h
arch/powerpc/kernel/asm-offsets.c
arch/powerpc/kernel/irq.c
arch/powerpc/kernel/misc_32.S
arch/powerpc/kernel/process.c
arch/powerpc/lib/sstep.c

index e378cccfca55bb20db094f12a2009ad3ba7bf1a4..ce4de5aed7b5c302b292bd38c69f21bd2f059038 100644 (file)
@@ -149,8 +149,6 @@ typedef struct {
 
 struct thread_struct {
        unsigned long   ksp;            /* Kernel stack pointer */
-       unsigned long   ksp_limit;      /* if ksp <= ksp_limit stack overflow */
-
 #ifdef CONFIG_PPC64
        unsigned long   ksp_vsid;
 #endif
@@ -162,6 +160,7 @@ struct thread_struct {
 #endif
 #ifdef CONFIG_PPC32
        void            *pgdir;         /* root of page-table tree */
+       unsigned long   ksp_limit;      /* if ksp <= ksp_limit stack overflow */
 #endif
 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
        /*
@@ -321,7 +320,6 @@ struct thread_struct {
 #else
 #define INIT_THREAD  { \
        .ksp = INIT_SP, \
-       .ksp_limit = INIT_SP_LIMIT, \
        .regs = (struct pt_regs *)INIT_SP - 1, /* XXX bogus, I think */ \
        .fs = KERNEL_DS, \
        .fpr = {{0}}, \
index d8958be5f31a18b7c0bae16509749067aae75265..502c7a4e73f70dc1008754b7f55e5ae164f25a76 100644 (file)
@@ -80,10 +80,11 @@ int main(void)
        DEFINE(TASKTHREADPPR, offsetof(struct task_struct, thread.ppr));
 #else
        DEFINE(THREAD_INFO, offsetof(struct task_struct, stack));
+       DEFINE(THREAD_INFO_GAP, _ALIGN_UP(sizeof(struct thread_info), 16));
+       DEFINE(KSP_LIMIT, offsetof(struct thread_struct, ksp_limit));
 #endif /* CONFIG_PPC64 */
 
        DEFINE(KSP, offsetof(struct thread_struct, ksp));
-       DEFINE(KSP_LIMIT, offsetof(struct thread_struct, ksp_limit));
        DEFINE(PT_REGS, offsetof(struct thread_struct, regs));
 #ifdef CONFIG_BOOKE
        DEFINE(THREAD_NORMSAVES, offsetof(struct thread_struct, normsave[0]));
index 2234a1276a77f5d80f6c125bd4409064f4abeaa9..57d286a78f86f6ff1231c695b9a10a591af30710 100644 (file)
@@ -496,7 +496,6 @@ void do_IRQ(struct pt_regs *regs)
 {
        struct pt_regs *old_regs = set_irq_regs(regs);
        struct thread_info *curtp, *irqtp;
-       unsigned long saved_sp_limit;
 
        /* Switch to the irq stack to handle this */
        curtp = current_thread_info();
@@ -509,12 +508,6 @@ void do_IRQ(struct pt_regs *regs)
                return;
        }
 
-       /* Adjust the stack limit */
-       saved_sp_limit = current->thread.ksp_limit;
-       current->thread.ksp_limit = (unsigned long)irqtp +
-               _ALIGN_UP(sizeof(struct thread_info), 16);
-
-
        /* Prepare the thread_info in the irq stack */
        irqtp->task = curtp->task;
        irqtp->flags = 0;
@@ -526,7 +519,6 @@ void do_IRQ(struct pt_regs *regs)
        call_do_irq(regs, irqtp);
 
        /* Restore stack limit */
-       current->thread.ksp_limit = saved_sp_limit;
        irqtp->task = NULL;
 
        /* Copy back updates to the thread_info */
@@ -604,16 +596,12 @@ void irq_ctx_init(void)
 static inline void do_softirq_onstack(void)
 {
        struct thread_info *curtp, *irqtp;
-       unsigned long saved_sp_limit = current->thread.ksp_limit;
 
        curtp = current_thread_info();
        irqtp = softirq_ctx[smp_processor_id()];
        irqtp->task = curtp->task;
        irqtp->flags = 0;
-       current->thread.ksp_limit = (unsigned long)irqtp +
-                                   _ALIGN_UP(sizeof(struct thread_info), 16);
        call_do_softirq(irqtp);
-       current->thread.ksp_limit = saved_sp_limit;
        irqtp->task = NULL;
 
        /* Set any flag that may have been set on the
index 7da3882a362255fcf758a28a774ba7aa3ecd705a..2b0ad984536333d7a15a3b39e6c9234c11b4b669 100644 (file)
 
        .text
 
+/*
+ * We store the saved ksp_limit in the unused part
+ * of the STACK_FRAME_OVERHEAD
+ */
 _GLOBAL(call_do_softirq)
        mflr    r0
        stw     r0,4(r1)
+       lwz     r10,THREAD+KSP_LIMIT(r2)
+       addi    r11,r3,THREAD_INFO_GAP
        stwu    r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r3)
        mr      r1,r3
+       stw     r10,8(r1)
+       stw     r11,THREAD+KSP_LIMIT(r2)
        bl      __do_softirq
+       lwz     r10,8(r1)
        lwz     r1,0(r1)
        lwz     r0,4(r1)
+       stw     r10,THREAD+KSP_LIMIT(r2)
        mtlr    r0
        blr
 
 _GLOBAL(call_do_irq)
        mflr    r0
        stw     r0,4(r1)
+       lwz     r10,THREAD+KSP_LIMIT(r2)
+       addi    r11,r3,THREAD_INFO_GAP
        stwu    r1,THREAD_SIZE-STACK_FRAME_OVERHEAD(r4)
        mr      r1,r4
+       stw     r10,8(r1)
+       stw     r11,THREAD+KSP_LIMIT(r2)
        bl      __do_irq
+       lwz     r10,8(r1)
        lwz     r1,0(r1)
        lwz     r0,4(r1)
+       stw     r10,THREAD+KSP_LIMIT(r2)
        mtlr    r0
        blr
 
index 6f428da53e2085b877334270286069e2ba54da37..96d2fdf3aa9ebe3bba547fd567c5a232be20ec9a 100644 (file)
@@ -1000,9 +1000,10 @@ int copy_thread(unsigned long clone_flags, unsigned long usp,
        kregs = (struct pt_regs *) sp;
        sp -= STACK_FRAME_OVERHEAD;
        p->thread.ksp = sp;
+#ifdef CONFIG_PPC32
        p->thread.ksp_limit = (unsigned long)task_stack_page(p) +
                                _ALIGN_UP(sizeof(struct thread_info), 16);
-
+#endif
 #ifdef CONFIG_HAVE_HW_BREAKPOINT
        p->thread.ptrace_bps[0] = NULL;
 #endif
index a7ee978fb860c9ffd237d9fbafb716b724e3afec..b1faa1593c9067e68995e1cdc54dbb8465dce587 100644 (file)
@@ -1505,6 +1505,7 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                 */
                if ((ra == 1) && !(regs->msr & MSR_PR) \
                        && (val3 >= (regs->gpr[1] - STACK_INT_FRAME_SIZE))) {
+#ifdef CONFIG_PPC32
                        /*
                         * Check if we will touch kernel sack overflow
                         */
@@ -1513,7 +1514,7 @@ int __kprobes emulate_step(struct pt_regs *regs, unsigned int instr)
                                err = -EINVAL;
                                break;
                        }
-
+#endif /* CONFIG_PPC32 */
                        /*
                         * Check if we already set since that means we'll
                         * lose the previous value.