Skip to contentSkip to conversation

can read/write to mprotect(PROT_NONE) region with 2.6.14 on au1550

PATCH New Clem Taylor 3 participants 3 messages Dec 7, 2005 – Apr 7, 2006 MIPS

lore
Share & tools
lore
#0 lore raw

Hi,

I was trying to use mprotect(PROT_NONE) to help debug a problem, and it seems that mprotect() isn't actually doing anything with my 2.6.14 linux-mips kernel on an au1550. Attached is a simple test program that segfaults as expected on x86 (2.6.12), but does not segfault on mips (2.6.14). I can both read and write PROT_NONE memory without problem, which should result in a segfault. Originally, I was trying to mprotect() a mmaped GFP_DMA region which wasn't working and then I tried a simpler test that also wasn't working.

Shouldn't mprotect() work? Could I be missing a config option, or is this just broken?

Thanks,
Clem Taylor

#include <stdio.h>
#include <stdlib.h>
#include <malloc.h>
#include <sys/mman.h>
#include <string.h>
#include <errno.h>

int main ( int argc, char *argv [ ] )

{
    int size = 65536, i, ret;
    unsigned char *buffer;

    buffer = memalign ( size, size );
    if ( buffer == NULL )
    {
        fprintf ( stderr, "memalign() failed.\n" );
        return 1;
    }

    fprintf ( stderr, "buffer=%p size=%d\n", buffer, size );

    /* write and read buffer */
    memset ( buffer, 0xAA, size );
    for ( i = 0; i < 2; i++ )
        fprintf ( stderr, "buffer [ %d ] = 0x%02X\n", i, buffer [ i ] );

    /* disable reading and writing */
    ret = mprotect ( buffer, size, PROT_NONE );
    if ( ret != 0 )
    {
        fprintf ( stderr, "mprotect(%p,%d,PROT_NONE) failed: %s\n",
            buffer, size, strerror ( errno ) );
        return 1;
    }

    /* write and read buffer, should segfault */
    memset ( buffer, 0x55, size );
    for ( i = 0; i < 2; i++ )
        fprintf ( stderr, "buffer [ %d ] = 0x%02X\n", i, buffer [ i ] );

    return 0;
}
#1 lore raw
Quoted 11 lines · Clem Taylor

I was trying to use mprotect(PROT_NONE) to help debug a problem, and it seems that mprotect() isn't actually doing anything with my 2.6.14 linux-mips kernel on an au1550. Attached is a simple test program that segfaults as expected on x86 (2.6.12), but does not segfault on mips (2.6.14). I can both read and write PROT_NONE memory without problem, which should result in a segfault. Originally, I was trying to mprotect() a mmaped GFP_DMA region which wasn't working and then I tried a simpler test that also wasn't working.

Shouldn't mprotect() work? Could I be missing a config option, or is this just broken?

That's a defect in CONFIG_64BIT_PHYS_ADDR which unfortunately is need on Alchemy SOCs due to the silly address space layout.

Ralf

#2 lore raw

PATCH 1/1 FIx mprotect() syscall for MIPS32 w/36-bit physical address support

Hello.

Fix mprotect() syscall for MIPS32 CPUs with 36-bit physical address
support: pte_modify() macro didn't clear the hardware page protection bits before modifying...

WBR, Sergei

include/asm-mips/pgtable.h +3 -2
index 702a28f..80b3605 100644
@@ -335,8 +335,9 @@ static inline pgprot_t pgprot_noncached(
 #if defined(CONFIG_64BIT_PHYS_ADDR) && defined(CONFIG_CPU_MIPS32_R1)
 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
 {
-	pte.pte_low &= _PAGE_CHG_MASK;
-	pte.pte_low |= pgprot_val(newprot);
+	pte.pte_low  &= _PAGE_CHG_MASK;
+	pte.pte_high &= ~0x3f;
+	pte.pte_low  |= pgprot_val(newprot);
 	pte.pte_high |= pgprot_val(newprot) & 0x3f;
 	return pte;
 }