[Qemu-devel] Re: 2.6.10-mm3: swsusp: out of memory on resume (was: Re: Ho ho ho - Linux v2.6.10)

[hugang]

On Sat, Jan 15, 2005 at 10:20:42PM +0100, Rafael J. Wysocki wrote:
> > > > > 
> > > > > 2.6.11-rc1-mm1 
> > > > >  -> 2005-1-14.core.diff       core patch              TEST PASSED
> > > > >   -> 2005-1-14.x86_64.diff    x86_64 patch    NOT TESTED
> > > > 
> > > > Unfortunately, on x86_64 it goes south on suspend, probably somwhere in 
> > > > write_pagedir(),
> > > > but I'm not quite sure as I can't make it print any useful stuff to the 
> > > > serial console
> > > > (everything is dumped to a virtual tty only).  Seemingly, it prints some
> > > > "write_pagedir: ..." debug messages and then starts to print garbage in
> > > > an infinite loop.
> 
> I have some good news for you. :-)
> 
> The patch actually works fine on my box.  What I thought was a result of an 
> infinite loop,
> turned out to be "only" a debug output from it, which is _really_ excessive.  
> After I had
> commented out the most of pr_debug()s in your code, it works nicely and I 
> like it very
> much.  Thanks a lot for porting it to x86_64!
> 
Cool, Current I making software suspend also works in Qemu X86_64
emulation, Here is a update patch to making copyback more safed and 
possible to improve copyback speed.

I change the swsusp_arch_resume to nosave section, the in memory copy
back it not touch this code. before not change that to nosave section,
I'm also geting a infinite loop in copy_one_page, From the qemu in_asm,
I sure that loop in copy_one_page, when I change it to nosave section,
that problem go away, I dont' sure tha't good idea to fixed it, but
current it works in my qemu, Can someone owner x86_64 test it.

I disable Flush TLB after copy page, It speedup the in qemu, But I can't
sure the right thing in real machine, can someone give me point.

-- 
Hu Gang       .-.
              /v\
             // \\ 
Linux User  /(   )\  [204016]
GPG Key ID   ^^-^^   http://soulinfo.com/~hugang/hugang.asc

2005-1-16.x86_64.diff

--- 2.6.11-rc1-mm1/arch/x86_64/kernel/suspend_asm.S     2004-12-30 
14:56:35.000000000 +0800
+++ 2.6.11-rc1-mm1-swsusp-x86_64/arch/x86_64/kernel/suspend_asm.S       
2005-01-16 13:38:25.000000000 +0800
@@ -35,6 +35,7 @@ ENTRY(swsusp_arch_suspend)
        call swsusp_save
        ret
 
+       .section    .data.nosave
 ENTRY(swsusp_arch_resume)
        /* set up cr3 */        
        leaq    init_level4_pgt(%rip),%rax
@@ -49,43 +50,47 @@ ENTRY(swsusp_arch_resume)
        movq    %rcx, %cr3;
        movq    %rax, %cr4;  # turn PGE back on
 
-       movl    nr_copy_pages(%rip), %eax
-       xorl    %ecx, %ecx
-       movq    $0, %r10
-       testl   %eax, %eax
-       jz      done
-.L105:
-       xorl    %esi, %esi
-       movq    $0, %r11
-       jmp     .L104
-       .p2align 4,,7
+       movq    pagedir_nosave(%rip), %rdi
+       testq   %rdi, %rdi
+       je              done
+
+copyback_page:
+       movq    24(%rdi), %r9
+       xorl    %r8d, %r8d
+
+copy_one_pgdir:
+       movq    8(%rdi), %rsi
+       testq   %rsi, %rsi
+       je      done
+       movq    (%rdi), %rcx
+       xorl    %edx, %edx
+
 copy_one_page:
-       movq    %r10, %rcx
-.L104:
-       movq    pagedir_nosave(%rip), %rdx
-       movq    %rcx, %rax
-       salq    $5, %rax
-       movq    8(%rdx,%rax), %rcx
-       movq    (%rdx,%rax), %rax
-       movzbl  (%rsi,%rax), %eax
-       movb    %al, (%rsi,%rcx)
-
-       movq    %cr3, %rax;  # flush TLB
-       movq    %rax, %cr3;
-
-       movq    %r11, %rax
-       incq    %rax
-       cmpq    $4095, %rax
-       movq    %rax, %rsi
-       movq    %rax, %r11
-       jbe     copy_one_page
-       movq    %r10, %rax
-       incq    %rax
-       movq    %rax, %rcx
-       movq    %rax, %r10
-       mov     nr_copy_pages(%rip), %eax
-       cmpq    %rax, %rcx
-       jb      .L105
+       movq    (%rcx,%rdx,8), %rax
+       movq    %rax, (%rsi,%rdx,8)
+       movq    8(%rcx,%rdx,8), %rax
+       movq    %rax, 8(%rsi,%rdx,8)
+       movq    16(%rcx,%rdx,8), %rax
+       movq    %rax, 16(%rsi,%rdx,8)
+       movq    24(%rcx,%rdx,8), %rax
+       movq    %rax, 24(%rsi,%rdx,8)
+#if 0 /* XXX speep up in qemu */
+       movq    %cr3, %rax;  # flush TLB
+       movq    %rax, %cr3;
+#endif
+       addq    $4, %rdx
+       cmpq    $511, %rdx
+       jbe     copy_one_page; # copy one page
+
+       incq    %r8
+       addq    $32, %rdi
+       cmpq    $127, %r8
+       jbe     copy_one_pgdir; # copy one pgdir
+
+       testq   %r9, %r9
+       movq    %r9, %rdi
+       jne     copyback_page
+
 done:
        movl    $24, %eax
        movl    %eax, %ds

Here is the patch for current qemu, with this it has no problem in
software suspend/resume, everything. :)

--- target-i386/helper.c~cvs    2005-01-16 14:06:22.000000000 -0500
+++ target-i386/helper.c        2005-01-16 14:07:21.000000000 -0500
@@ -1454,7 +1454,7 @@ void load_seg(int seg_reg, int selector)
     selector &= 0xffff;
     if ((selector & 0xfffc) == 0) {
         /* null selector case */
-        if (seg_reg == R_SS)
+        if (seg_reg == R_CS)
             raise_exception_err(EXCP0D_GPF, 0);
         cpu_x86_load_seg_cache(env, seg_reg, selector, 0, 0, 0);
     } else {