[Qemu-devel] [PATCH] Increase max -kernel size

[Qemu-devel] [PATCH] Increase max -kernel size

[Gwenole Beauchesne]

Hi,

The following patch increases max kernel size to 8 MB when qemu is invoked 
with -kernel and -initrd. Otherwise, x86_64 kernel panics when loading the 
initrd (e.g. <Cooker>/x86_64/isolinux/alt0/{vmlinuz,all.rdz}).

Thanks,
Gwenole.

2007-03-27  Gwenole Beauchesne  <gbeauchesne@mandriva.com>

	* hw/pc.c (INITRD_LOAD_ADDR): Increase max kernel size with
	initrd.

--- qemu-0.9.0/hw/pc.c.increase-initrd-load-addr	2007-03-15 13:36:53.000000000 -0400
+++ qemu-0.9.0/hw/pc.c	2007-03-27 13:27:43.000000000 -0400
@@ -37,7 +37,7 @@ extern int kvm_allowed;
 #define LINUX_BOOT_FILENAME "linux_boot.bin"
 
 #define KERNEL_LOAD_ADDR     0x00100000
-#define INITRD_LOAD_ADDR     0x00600000
+#define INITRD_LOAD_ADDR     0x00900000
 #define KERNEL_PARAMS_ADDR   0x00090000
 #define KERNEL_CMDLINE_ADDR  0x00099000
 

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Daniel Jacobowitz]

On Tue, Mar 27, 2007 at 06:14:22PM +0200, Gwenole Beauchesne wrote:
> Hi,
> 
> The following patch increases max kernel size to 8 MB when qemu is invoked 
> with -kernel and -initrd. Otherwise, x86_64 kernel panics when loading the 
> initrd (e.g. <Cooker>/x86_64/isolinux/alt0/{vmlinuz,all.rdz}).

When I talked to Fabrice about this he recommended looking at what
other loaders (lilo, grub) do to avoid overwriting the kernel.  I've
been meaning to do that for months but never got around to it.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Thiemo Seufer]

Gwenole Beauchesne wrote:
> Hi,
> 
> The following patch increases max kernel size to 8 MB when qemu is invoked 
> with -kernel and -initrd. Otherwise, x86_64 kernel panics when loading the 
> initrd (e.g. <Cooker>/x86_64/isolinux/alt0/{vmlinuz,all.rdz}).

I would like a patch which adjusts the initrd load address dynamically,
depending on the kernel load size, instead of hardcoding just a
different magic number.


Thiemo

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Daniel Jacobowitz]

On Tue, Mar 27, 2007 at 08:16:40PM +0100, Thiemo Seufer wrote:
> Gwenole Beauchesne wrote:
> > Hi,
> > 
> > The following patch increases max kernel size to 8 MB when qemu is invoked 
> > with -kernel and -initrd. Otherwise, x86_64 kernel panics when loading the 
> > initrd (e.g. <Cooker>/x86_64/isolinux/alt0/{vmlinuz,all.rdz}).
> 
> I would like a patch which adjusts the initrd load address dynamically,
> depending on the kernel load size, instead of hardcoding just a
> different magic number.

Note, that does not readily work - this is where we load the
compressed kernel image and initrd, but what matters is the size it
gets uncompressed to.

I don't remember any more than that so I'm sorry if it's total
garbage, which it might be.  I did write a patch to stick the initrd
right after the loaded kernel; it didn't work and I convinced myself
that was the reason.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Thiemo Seufer]

Daniel Jacobowitz wrote:
> On Tue, Mar 27, 2007 at 08:16:40PM +0100, Thiemo Seufer wrote:
> > Gwenole Beauchesne wrote:
> > > Hi,
> > > 
> > > The following patch increases max kernel size to 8 MB when qemu is invoked 
> > > with -kernel and -initrd. Otherwise, x86_64 kernel panics when loading the 
> > > initrd (e.g. <Cooker>/x86_64/isolinux/alt0/{vmlinuz,all.rdz}).
> > 
> > I would like a patch which adjusts the initrd load address dynamically,
> > depending on the kernel load size, instead of hardcoding just a
> > different magic number.
> 
> Note, that does not readily work - this is where we load the
> compressed kernel image and initrd, but what matters is the size it
> gets uncompressed to.

I think what matters is the space taken by (uncompressed and loaded)
kernel ALLOC segments. Everything above that should be ok to load a
ramdisk.

IOW, the ELF loader would need to return the topmost address it came
across, including .bss-like parts. The up-aligned value would become
the initrd load address.


Thiemo

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Paul Brook]

> > Note, that does not readily work - this is where we load the
> > compressed kernel image and initrd, but what matters is the size it
> > gets uncompressed to.
>
> I think what matters is the space taken by (uncompressed and loaded)
> kernel ALLOC segments. Everything above that should be ok to load a
> ramdisk.
>
> IOW, the ELF loader would need to return the topmost address it came
> across, including .bss-like parts. The up-aligned value would become
> the initrd load address.

That only works if qemu is loading the uncomplressed ELF vmlinux image 
directly. In most cases it will be loading a binary blob (or ELF file) of the 
compressed zImage. In this case it has no way of telling how big the final 
kernel ALLOC sections are.

By my reading grub puts the inirtd at 0x38000000 (896Mb), or at the top of 
physical memory, whichever is lowest, and lilo puts it at the top of physical 
memory, give or take a bit of alignment.

Paul

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Thiemo Seufer]

Paul Brook wrote:
> > > Note, that does not readily work - this is where we load the
> > > compressed kernel image and initrd, but what matters is the size it
> > > gets uncompressed to.
> >
> > I think what matters is the space taken by (uncompressed and loaded)
> > kernel ALLOC segments. Everything above that should be ok to load a
> > ramdisk.
> >
> > IOW, the ELF loader would need to return the topmost address it came
> > across, including .bss-like parts. The up-aligned value would become
> > the initrd load address.
> 
> That only works if qemu is loading the uncomplressed ELF vmlinux image 
> directly. In most cases it will be loading a binary blob (or ELF file) of the 
> compressed zImage. In this case it has no way of telling how big the final 
> kernel ALLOC sections are.

Right, a piggyback-style loader would likely fail in that case.


Thiemo

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Daniel Jacobowitz]

On Wed, Mar 28, 2007 at 12:35:18AM +0100, Thiemo Seufer wrote:
> Right, a piggyback-style loader would likely fail in that case.

Which is exactly the interesting case for x86_64.  So, since Paul's
verified that grub and lilo just put them far apart, maybe bumping up
the size is the right thing to do.  Or always putting the initrd at
the top of emulated RAM.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Paul Brook]

On Wednesday 28 March 2007 03:21, Daniel Jacobowitz wrote:
> On Wed, Mar 28, 2007 at 12:35:18AM +0100, Thiemo Seufer wrote:
> > Right, a piggyback-style loader would likely fail in that case.
>
> Which is exactly the interesting case for x86_64.  So, since Paul's
> verified that grub and lilo just put them far apart, maybe bumping up
> the size is the right thing to do.  Or always putting the initrd at
> the top of emulated RAM.

I'd prefer the latter, so that both big and small machines work. Linux still 
(just about) boots on 8Mb machines.

Paul

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Daniel Jacobowitz]

On Wed, Mar 28, 2007 at 03:00:08PM +0100, Paul Brook wrote:
> On Wednesday 28 March 2007 03:21, Daniel Jacobowitz wrote:
> > On Wed, Mar 28, 2007 at 12:35:18AM +0100, Thiemo Seufer wrote:
> > > Right, a piggyback-style loader would likely fail in that case.
> >
> > Which is exactly the interesting case for x86_64.  So, since Paul's
> > verified that grub and lilo just put them far apart, maybe bumping up
> > the size is the right thing to do.  Or always putting the initrd at
> > the top of emulated RAM.
> 
> I'd prefer the latter, so that both big and small machines work. Linux still 
> (just about) boots on 8Mb machines.

Like this?  I can't reliably test that it fixes the problem - for some
bizarre reason, I can now load kernels without this patch.  But it
looks right and I can load kernels with it, too.

-- 
Daniel Jacobowitz
CodeSourcery

---
 hw/pc.c |   19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

Index: qemu/hw/pc.c
===================================================================
--- qemu.orig/hw/pc.c	2007-03-31 11:36:34.000000000 -0400
+++ qemu/hw/pc.c	2007-03-31 11:38:24.000000000 -0400
@@ -32,7 +32,6 @@
 #define LINUX_BOOT_FILENAME "linux_boot.bin"
 
 #define KERNEL_LOAD_ADDR     0x00100000
-#define INITRD_LOAD_ADDR     0x00600000
 #define KERNEL_PARAMS_ADDR   0x00090000
 #define KERNEL_CMDLINE_ADDR  0x00099000
 
@@ -452,6 +451,7 @@ static void pc_init1(int ram_size, int v
     char buf[1024];
     int ret, linux_boot, initrd_size, i;
     ram_addr_t ram_addr, vga_ram_addr, bios_offset, vga_bios_offset;
+    ram_addr_t initrd_offset;
     int bios_size, isa_bios_size, vga_bios_size;
     PCIBus *pci_bus;
     int piix3_devfn = -1;
@@ -599,8 +599,21 @@ static void pc_init1(int ram_size, int v
         
         /* load initrd */
         initrd_size = 0;
+        initrd_offset = 0;
         if (initrd_filename) {
-            initrd_size = load_image(initrd_filename, phys_ram_base + INITRD_LOAD_ADDR);
+            initrd_size = get_image_size (initrd_filename);
+            if (initrd_size > 0) {
+                initrd_offset = ram_size - initrd_size;
+                if (initrd_size > ram_size ||
+                    initrd_offset < KERNEL_LOAD_ADDR + ret) {
+                    fprintf(stderr,
+                            "qemu: memory too small for initial ram disk '%s'\n",
+                            initrd_filename);
+                    exit(1);
+                }
+                initrd_size = load_image(initrd_filename,
+                                         phys_ram_base + initrd_offset);
+            }
             if (initrd_size < 0) {
                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n", 
                         initrd_filename);
@@ -608,7 +621,7 @@ static void pc_init1(int ram_size, int v
             }
         }
         if (initrd_size > 0) {
-            stl_raw(phys_ram_base + KERNEL_PARAMS_ADDR + 0x218, INITRD_LOAD_ADDR);
+            stl_raw(phys_ram_base + KERNEL_PARAMS_ADDR + 0x218, initrd_offset);
             stl_raw(phys_ram_base + KERNEL_PARAMS_ADDR + 0x21c, initrd_size);
         }
         pstrcpy(phys_ram_base + KERNEL_CMDLINE_ADDR, 4096,

Re: [Qemu-devel] [PATCH] Increase max -kernel size

[Paul Brook]

> Like this?  I can't reliably test that it fixes the problem - for some
> bizarre reason, I can now load kernels without this patch.  But it
> looks right and I can load kernels with it, too.

Turns out there's a problem. qemu puts ACPI tables at the top of ram. We have 
to be careful to avoid stomping on them.

Paul