Re: [Qemu-devel] SPARC64: immediate segfault on startup with git mastervery

[Blue Swirl]

On Sun, Mar 18, 2012 at 09:44, Avi Kivity <address@hidden> wrote:
> On 03/18/2012 04:01 AM, Mark Cave-Ayland wrote:
>> Hi Avi/Blue,
>>
>> I've just updated to git master and found that SPARC64 is broken
>> again; a git bisect shows the following commit causes this:
>>
>>
>> commit f3705d53296d78b14f5823472ae2add16a25a0a5
>> Author: Avi Kivity <address@hidden>
>> Date:   Thu Mar 8 16:16:34 2012 +0200
>>
>>     memory: make phys_page_find() return an unadjusted section
>>
>>     We'd like to store the section index in the iotlb, so we can't
>>     adjust it before returning.  Return an unadjusted section and
>>     instead introduce section_addr(), which does the adjustment later.
>>
>>     Signed-off-by: Avi Kivity <address@hidden>
>>
>>
>> The symptom is that qemu-system-sparc64 segfaults immediately on
>> startup (note this is with an OpenBIOS image built from SVN r1048).
>> I've included a couple of backtraces below:
>>
>
> Please try the attached patch.

I tried this approach instead, seems to work (except Sparc32, Sparc64
and PPC displays are still not refreshed correctly). The problem is
that f3705d53296d78b14f5823472ae2add16a25a0a5 converted addr & mask +
section.offset to section_addr(), but in some cases the offset within
page was not wanted.

diff --git a/exec.c b/exec.c
index 8fd50a1..ad455be 100644
--- a/exec.c
+++ b/exec.c
@@ -1474,8 +1474,8 @@ static void breakpoint_invalidate(CPUArchState
*env, target_ulong pc)
           || (section->mr->rom_device && section->mr->readable))) {
         return;
     }
-    ram_addr = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK)
-        + section_addr(section, addr);
+    ram_addr = (memory_region_get_ram_addr(section->mr)
+                + section_addr(section, pc)) & TARGET_PAGE_MASK;
     tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0);
 }
 #endif
@@ -2205,14 +2205,14 @@ void tlb_set_page(CPUArchState *env, target_ulong vaddr,
     }
     if (is_ram_rom_romd(section)) {
         addend = (unsigned long)memory_region_get_ram_ptr(section->mr)
-                                 + section_addr(section, paddr);
+                 + section_addr(section, paddr & TARGET_PAGE_MASK);
     } else {
         addend = 0;
     }
     if (is_ram_rom(section)) {
         /* Normal RAM.  */
-        iotlb = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK)
-            + section_addr(section, paddr);
+        iotlb = (memory_region_get_ram_addr(section->mr)
+                 + section_addr(section, paddr)) & TARGET_PAGE_MASK;
         if (!section->readonly)
             iotlb |= phys_section_notdirty;
         else
@@ -2225,7 +2225,7 @@ void tlb_set_page(CPUArchState *env, target_ulong vaddr,
            We can't use the high bits of pd for this because
            IO_MEM_ROMD uses these as a ram address.  */
         iotlb = section - phys_sections;
-        iotlb += section_addr(section, paddr);
+        iotlb += section_addr(section, paddr & TARGET_PAGE_MASK);
     }

     code_address = address;
@@ -2265,7 +2265,7 @@ void tlb_set_page(CPUArchState *env, target_ulong vaddr,
         } else if (memory_region_is_ram(section->mr)
                    && !cpu_physical_memory_is_dirty(
                            section->mr->ram_addr
-                           + section_addr(section, paddr))) {
+                           + section_addr(section, paddr &
TARGET_PAGE_MASK))) {
             te->addr_write = address | TLB_NOTDIRTY;
         } else {
             te->addr_write = address;



> If it fails, can you upload bios and disk images somewhere?
>
> --
> error compiling committee.c: too many arguments to function
>