[PATCH 0/2] Fix could not boot vm image which converted from phy-partition

[PATCH 0/2] Fix could not boot vm image which converted from phy-partition

[Joe Jin]

Hi,

When booting up windows VM which converted by dd windows physical
partition, it failed with "Error Loading Operating System", the
partition build on scsi/raid controller. The root cause is qemu
and hvmloader could not calaulated C/H/S correctly, then ntloader
could not found the root partition.

Thanks,
Joe

[PATCH 1/2] QEMU: Don't set all to default C/H/S when vm heads>16

[Joe Jin]

Hi,

Current qemu set all C/H/S to default(16383/16/63), when image file converted
from physical scsi/raid disk, use default will cause failed to boot the VM.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>

---
 block.c |   10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/block.c b/block.c
index 05ff8cb..519f955 100644
--- a/block.c
+++ b/block.c
@@ -890,7 +890,15 @@ void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *pse
         *psecs = secs;
     } else {
         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
-            if (heads > 16) {
+            if(secs == 32){
+                /* if image file convert from physical disk to and
+                   physical partition on scsi, sector = 32, if set
+                   it to default will failed to boot up it */
+                *pcyls = cylinders;
+                *pheads = heads;
+                *psecs = secs;
+                bdrv_set_translation_hint(bs, BIOS_ATA_TRANSLATION_LBA);
+            } else if (heads > 16) {
                 /* if heads > 16, it means that a BIOS LBA
                    translation was active, so the default
                    hardware geometry is OK */

[PATCH 2/2] hvmloader->rombios use C/H/S, LBA when have been set by qemu

[Joe Jin]

Hi,

hvmloader->rombios use C/H/S, LBA when have been set by qemu.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@oracle.com>
Signed-off-by: Joe Jin <joe.jin@oracle.com>
---
 rombios.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff -r 3263d0ff9476 tools/firmware/rombios/rombios.c
--- a/tools/firmware/rombios/rombios.c	Thu Jul 29 16:53:40 2010 +0100
+++ b/tools/firmware/rombios/rombios.c	Fri Aug 06 07:09:45 2010 +0800
@@ -2738,8 +2738,7 @@
         case ATA_TRANSLATION_NONE:
           break;
         case ATA_TRANSLATION_LBA:
-          spt = 63;
-          sectors_low /= 63;
+          sectors_low /= spt;
           heads = sectors_low / 1024;
           if (heads>128) heads = 255;
           else if (heads>64) heads = 128;
@@ -5440,6 +5439,7 @@
   Bit16u npc, nph, npspt, nlc, nlh, nlspt;
   Bit16u size, count;
   Bit8u  device, status;
+  Bit8u  translation;
 
   BX_DEBUG_INT13_HD("int13_harddisk: AX=%04x BX=%04x CX=%04x DX=%04x ES=%04x\n", AX, BX, CX, DX, ES);
 
@@ -5509,9 +5509,10 @@
 
       nph   = read_word(ebda_seg, &EbdaData->ata.devices[device].pchs.heads);
       npspt = read_word(ebda_seg, &EbdaData->ata.devices[device].pchs.spt);
+      translation = read_byte(ebda_seg,&EbdaData->ata.devices[device].translation);
 
       // if needed, translate lchs to lba, and execute command
-      if ( (nph != nlh) || (npspt != nlspt)) {
+      if ( (nph != nlh) || (npspt != nlspt) || (translation == ATA_TRANSLATION_LBA)) {
         lba_low = ((((Bit32u)cylinder * (Bit32u)nlh) + (Bit32u)head) * (Bit32u)nlspt) + (Bit32u)sector - 1;
         lba_high = 0;
         sector = 0; // this forces the command to be lba

Re: [PATCH 0/2] Fix could not boot vm image which converted from phy-partition

[Ian Jackson]

Joe Jin writes ("[Xen-devel] [PATCH 0/2] Fix could not boot vm image which converted from phy-partition"):
> When booting up windows VM which converted by dd windows physical
> partition, it failed with "Error Loading Operating System", the
> partition build on scsi/raid controller. The root cause is qemu
> and hvmloader could not calaulated C/H/S correctly, then ntloader
> could not found the root partition.

Thanks for this, but I'm afraid I'm kind of missing the background
here.  I'm not very familiar with how the CHS-guessing works.  Perhaps
you could explain ?

Also, have you looked at upstream qemu to see what is done there ?
We're currently working on a merge of qemu-dm back into upstream qemu
and if possible we'd like to avoid adding more interface
incompatibilities.

Ian.

Re: [PATCH 0/2] Fix could not boot vm image which converted from phy-partition

[Joe Jin]

On 2010-08-10 16:50, Ian Jackson wrote:
> Joe Jin writes ("[Xen-devel] [PATCH 0/2] Fix could not boot vm image which converted from phy-partition"):
> > When booting up windows VM which converted by dd windows physical
> > partition, it failed with "Error Loading Operating System", the
> > partition build on scsi/raid controller. The root cause is qemu
> > and hvmloader could not calaulated C/H/S correctly, then ntloader
> > could not found the root partition.
> 
> Thanks for this, but I'm afraid I'm kind of missing the background
> here.  I'm not very familiar with how the CHS-guessing works.  Perhaps
> you could explain ?

The background is we dd physical MS-Windows partition, physical disk based
on RAID controller, we tried to boot it but failed with "Error Loading
Operation System".

The error message reported by rombios for it could not find the first
sector. qemu calculated CHS by read MBR(the first 512 bytes) of image file,
when H(eads)>16, qemu set it to default value heads(16). in fact disk on 
RAID heads is 32, this made qemu pass incorrect offset of image file to 
rombios, then rombios failed to find the boot loader.

> 
> Also, have you looked at upstream qemu to see what is done there ?
> We're currently working on a merge of qemu-dm back into upstream qemu
> and if possible we'd like to avoid adding more interface
> incompatibilities.

I did not check qemu codes but we have tried to boot up the image file
by qemu, same error, we did not applied the patch to qemu.

Thanks,
Joe

[PATCH 2/2] hvmloader->rombios use C/H/S, LBA when have been set by qemu

[DuanZhenzhong]

Hi,
Multiple guests rebooting, across multiple OVS nodes. Not all guest at 
the same time.
Error in /var/log/xend.log taking the form:
...
[2010-09-06 12:19:55 8816] WARNING (image:490) domain 234_sscgrantsd: 
device model failure: pid 16657: died due to signal 11; see
...

This is due to some ide state pointer in ide.c is NULL.
Patch from carnold@novell.com fixed this problem.
After applying, our system get steady.

--- xen-3.4.0.old/qemu-xen.git/hw/ide.c 2010-10-19 10:45:23.000000000 +0800
+++ xen-3.4.0/qemu-xen.git/hw/ide.c     2010-10-19 10:52:57.000000000 +0800
@@ -934,8 +934,9 @@ static inline void ide_dma_submit_check(

static inline void ide_set_irq(IDEState *s)
{
-    BMDMAState *bm = s->bmdma;
-    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
+    BMDMAState *bm;
+    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */
+    bm = s->bmdma;
     if (!(s->cmd & IDE_CMD_DISABLE_IRQ)) {
         if (bm) {
             bm->status |= BM_STATUS_INT;
@@ -1223,14 +1224,14 @@ static void ide_read_dma_cb(void *opaque
     int n;
     int64_t sector_num;

+    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */
+
     if (ret < 0) {
         dma_buf_commit(s, 1);
        ide_dma_error(s);
        return;
     }

-    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
-
     n = s->io_buffer_size >> 9;
     sector_num = ide_get_sector(s);
     if (n > 0) {
@@ -1334,6 +1335,8 @@ static void ide_write_flush_cb(void *opa
     BMDMAState *bm = opaque;
     IDEState *s = bm->ide_if;

+    if (!s) return;
+
     if (ret != 0) {
        ide_dma_error(s);
        return;
@@ -1365,6 +1368,8 @@ static void ide_write_dma_cb(void *opaqu
     int n;
     int64_t sector_num;

+    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */
+
     if (ret < 0) {
         if (ide_handle_write_error(s, -ret,  BM_STATUS_DMA_RETRY))
             return;
@@ -1375,8 +1380,6 @@ static void ide_write_dma_cb(void *opaqu
       return; /* ouch2 */
     }

-    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
-
     n = s->io_buffer_size >> 9;
     sector_num = ide_get_sector(s);
     if (n > 0) {
@@ -1433,7 +1436,7 @@ static void ide_flush_cb(void *opaque, i
{
     IDEState *s = opaque;

-    if (!s->bs) return; /* ouch! (see below) */
+    if (!s || !s->bs) return; /* ouch! (see below) */

     if (ret) {
         /* We are completely doomed.  The IDE spec does not permit us
@@ -1690,7 +1693,7 @@ static void ide_atapi_cmd_read_dma_cb(vo
     IDEState *s = bm->ide_if;
     int data_offset, n;

-    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
+    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */

     if (ret < 0) {
         ide_atapi_io_error(s, ret);
@@ -2368,7 +2371,7 @@ static void cdrom_change_cb(void *opaque
     IDEState *s = opaque;
     uint64_t nb_sectors;

-    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
+    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */

     bdrv_get_geometry(s->bs, &nb_sectors);
     s->nb_sectors = nb_sectors;

Re: [PATCH 2/2] hvmloader->rombios use C/H/S, LBA when have been set by qemu

[Ian Jackson]

DuanZhenzhong writes ("[Xen-devel] [PATCH 2/2] hvmloader->rombios use C/H/S, LBA when have been set by qemu"):
> This is due to some ide state pointer in ide.c is NULL.

How did the pointer become null ?  I think the bug has happened
earlier and that your patch is therefore not correct.

For example,

> static inline void ide_set_irq(IDEState *s)
> {
> -    BMDMAState *bm = s->bmdma;
> -    if (!s->bs) return; /* ouch! (see ide_flush_cb) */
> +    BMDMAState *bm;
> +    if (!s || !s->bs) return; /* ouch! (see ide_flush_cb) */

I think it is always wrong to call ide_set_irq(NULL).

Ian.