* Re: [PATCH] hw/arm/virt: use pflash image real size when mapping
2020-09-18 13:41 ` Philippe Mathieu-Daudé
@ 2020-09-19 10:03 ` haibinzhang(张海斌)
2020-09-21 5:34 ` haibinzhang(张海斌)
2020-09-22 6:57 ` Laszlo Ersek
2 siblings, 0 replies; 6+ messages in thread
From: haibinzhang(张海斌) @ 2020-09-19 10:03 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Peter Maydell, Stefan Hajnoczi, qemu-devel@nongnu.org,
qemu-arm@nongnu.org, Ard Biesheuvel, Leif Lindholm, Laszlo Ersek
[-- Attachment #1: Type: text/plain, Size: 3844 bytes --]
On Sep 18, 2020, at 9:41 PM, Philippe Mathieu-Daudé <philmd@redhat.com<mailto:philmd@redhat.com>> wrote:
Cc'ing firmware experts.
On 9/18/20 2:26 PM, haibinzhang(寮犳捣鏂? wrote:
Default size of arm-virt pflash image is 64MB which
will cost extra 128MB(64MBx2) memory per qemu process
and 12.5GB for 100 qemu processes. Host memory is
precious and it is valuable to reduce pflash image size.
For compatibility arm-virt uses real size when mapping.
Flash#0 is a device because eventually its model will handle
sector/block protection, so firmware can do CapsuleUpdate
(updating itself). Meanwhile you could treat flash#0 as a pure
ROM device... But this wouldn't be the 'arm-virt' machine anymore.
Example:
1. Get QEMU_EFI.fd and QEMU_VARS.fd from tianocore/edk2
2. Make pflash images
dd if=/dev/zero bs=1M count=64 of=flash0.img
dd if=/dev/zero bs=1M count=64 of=flash1.img
dd if=QEMU_EFI.fd bs=1M of=flash0.img conv=notrunc
dd if=QEMU_VARS.fd bs=1M of=flash1.img conv=notrunc
3. Start VM
qemu-system-aarch64 -machine virt,accel=kvm,gic-version=3 \
-drive file=flash0.img,if=pflash,format=raw,unit=0,readonly=on \
-drive file=flash1.img,if=pflash,format=raw,unit=1 \
…
In virt machine, VIRT_FLASH size is fixed 128MB (fixed 64MB for Flash0, fixed 64MB for Flash1).
Flash0 and Flash1 are two Block Backend devices and will be copied into two memory storages alloced
in virt_flash_map, not file mapping. Size of memory storage is fixed 64MB.
Since size checking, virt machine currently doesn’t support <64MB pflash images. So, get real
size of pflash image first, and use real size to alloc memory storage.
I always use 8MB Flash0 and Flash1, using following commands
dd if=/dev/zero bs=1M count=8 of=flash0.img
dd if=/dev/zero bs=1M count=8 of=flash1.img
dd if=QEMU_EFI.fd bs=1M of=flash0.img conv=notrunc
dd if=QEMU_VARS.fd bs=1M of=flash1.img conv=notrunc
Signed-off-by: Haibin Zhang <haibinzhang@tencent.com<mailto:haibinzhang@tencent.com>>
---
hw/arm/virt.c | 22 ++++++++++++++++++++--
1 file changed, 20 insertions(+), 2 deletions(-)
diff --git a/hw/arm/virt.c b/hw/arm/virt.c
index acf9bfbece..3545e12865 100644
--- a/hw/arm/virt.c
+++ b/hw/arm/virt.c
@@ -50,6 +50,7 @@
#include "sysemu/sysemu.h"
#include "sysemu/tpm.h"
#include "sysemu/kvm.h"
+#include "sysemu/block-backend.h"
#include "hw/loader.h"
#include "exec/address-spaces.h"
#include "qemu/bitops.h"
@@ -1001,10 +1002,27 @@ static void virt_flash_map(VirtMachineState *vms,
*/
hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
+ int64_t realsize;
+ BlockBackend *blk;
- virt_flash_map1(vms->flash[0], flashbase, flashsize,
+ realsize = flashsize;
+ blk = pflash_cfi01_get_blk(vms->flash[0]);
+ if (blk) {
+ realsize = blk_getlength(blk);
+ realsize = realsize < flashsize ? realsize : flashsize;
+ }
Stefan recently posted "nvdimm: read-only file support" which
might be a better way to achieve what you want:
https://www.mail-archive.com/qemu-devel@nongnu.org/msg741137.html
Flash0 is read-only and shared by all VMs, but Flash1 is not and private for each VM.
Btw., how to create a nvdimm device mapping to address range 0~128MB?
+
+ virt_flash_map1(vms->flash[0], flashbase, realsize,
secure_sysmem);
- virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
+
+ realsize = flashsize;
+ blk = pflash_cfi01_get_blk(vms->flash[1]);
+ if (blk) {
+ realsize = blk_getlength(blk);
+ realsize = realsize < flashsize ? realsize : flashsize;
+ }
+
+ virt_flash_map1(vms->flash[1], flashbase + flashsize, realsize,
sysmem);
}
--
2.23.0
[-- Attachment #2: Type: text/html, Size: 17584 bytes --]
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH] hw/arm/virt: use pflash image real size when mapping
2020-09-18 13:41 ` Philippe Mathieu-Daudé
2020-09-19 10:03 ` haibinzhang(张海斌)
@ 2020-09-21 5:34 ` haibinzhang(张海斌)
2020-09-22 6:57 ` Laszlo Ersek
2 siblings, 0 replies; 6+ messages in thread
From: haibinzhang(张海斌) @ 2020-09-21 5:34 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Peter Maydell, xtay.haibin@gmail.com, Stefan Hajnoczi,
qemu-devel@nongnu.org, qemu-arm@nongnu.org, Ard Biesheuvel,
Leif Lindholm, Laszlo Ersek
Resend the mail using plain text format
> On Sep 18, 2020, at 9:41 PM, Philippe Mathieu-Daudé <philmd@redhat.com> wrote:
>
> Cc'ing firmware experts.
>
> On 9/18/20 2:26 PM, haibinzhang(寮犳捣鏂? wrote:
>> Default size of arm-virt pflash image is 64MB which
>> will cost extra 128MB(64MBx2) memory per qemu process
>> and 12.5GB for 100 qemu processes. Host memory is
>> precious and it is valuable to reduce pflash image size.
>> For compatibility arm-virt uses real size when mapping.
>
> Flash#0 is a device because eventually its model will handle
> sector/block protection, so firmware can do CapsuleUpdate
> (updating itself). Meanwhile you could treat flash#0 as a pure
> ROM device... But this wouldn't be the 'arm-virt' machine anymore.
>
But Flash#0/#1 are corresponding to two rom MemoryRegions related to memory cost.
Example:
1. Get QEMU_EFI.fd and QEMU_VARS.fd from tianocore/edk2
2. Make pflash images
dd if=/dev/zero bs=1M count=64 of=flash0.img
dd if=/dev/zero bs=1M count=64 of=flash1.img
dd if=QEMU_EFI.fd bs=1M of=flash0.img conv=notrunc
dd if=QEMU_VARS.fd bs=1M of=flash1.img conv=notrunc
3. Start VM
qemu-system-aarch64 -machine virt,accel=kvm,gic-version=3 \
-drive file=flash0.img,if=pflash,format=raw,unit=0,readonly=on \
-drive file=flash1.img,if=pflash,format=raw,unit=1 \
…
In virt machine, VIRT_FLASH size is fixed 128MB (fixed 64MB for Flash0, fixed 64MB for Flash1).
Flash0 and Flash1 are two Block Backend devices and will be copied into two memory storages alloced
in virt_flash_map, not file mapping. Size of memory storage is fixed 64MB.
Since size checking, virt machine currently doesn’t support <64MB pflash images. So, get real
size of pflash image first, and use real size to alloc memory storage.
I always use 8MB Flash0 and Flash1, using following commands
dd if=/dev/zero bs=1M count=8 of=flash0.img
dd if=/dev/zero bs=1M count=8 of=flash1.img
dd if=QEMU_EFI.fd bs=1M of=flash0.img conv=notrunc
dd if=QEMU_VARS.fd bs=1M of=flash1.img conv=notrunc
>>
>> Signed-off-by: Haibin Zhang <haibinzhang@tencent.com>
>> ---
>> hw/arm/virt.c | 22 ++++++++++++++++++++--
>> 1 file changed, 20 insertions(+), 2 deletions(-)
>>
>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>> index acf9bfbece..3545e12865 100644
>> --- a/hw/arm/virt.c
>> +++ b/hw/arm/virt.c
>> @@ -50,6 +50,7 @@
>> #include "sysemu/sysemu.h"
>> #include "sysemu/tpm.h"
>> #include "sysemu/kvm.h"
>> +#include "sysemu/block-backend.h"
>> #include "hw/loader.h"
>> #include "exec/address-spaces.h"
>> #include "qemu/bitops.h"
>> @@ -1001,10 +1002,27 @@ static void virt_flash_map(VirtMachineState *vms,
>> */
>> hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
>> hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
>> + int64_t realsize;
>> + BlockBackend *blk;
>>
>> - virt_flash_map1(vms->flash[0], flashbase, flashsize,
>> + realsize = flashsize;
>> + blk = pflash_cfi01_get_blk(vms->flash[0]);
>> + if (blk) {
>> + realsize = blk_getlength(blk);
>> + realsize = realsize < flashsize ? realsize : flashsize;
>> + }
>
> Stefan recently posted "nvdimm: read-only file support" which
> might be a better way to achieve what you want:
> https://www.mail-archive.com/qemu-devel@nongnu.org/msg741137.html
Flash0 is read-only and shared by all VMs, but Flash1 is not and private for each VM.
Btw., how to create a nvdimm device mapping to address range 0~128MB?
>
>> +
>> + virt_flash_map1(vms->flash[0], flashbase, realsize,
>> secure_sysmem);
>> - virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
>> +
>> + realsize = flashsize;
>> + blk = pflash_cfi01_get_blk(vms->flash[1]);
>> + if (blk) {
>> + realsize = blk_getlength(blk);
>> + realsize = realsize < flashsize ? realsize : flashsize;
>> + }
>> +
>> + virt_flash_map1(vms->flash[1], flashbase + flashsize, realsize,
>> sysmem);
>> }
>>
>> --
>> 2.23.0
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH] hw/arm/virt: use pflash image real size when mapping
2020-09-18 13:41 ` Philippe Mathieu-Daudé
2020-09-19 10:03 ` haibinzhang(张海斌)
2020-09-21 5:34 ` haibinzhang(张海斌)
@ 2020-09-22 6:57 ` Laszlo Ersek
2 siblings, 0 replies; 6+ messages in thread
From: Laszlo Ersek @ 2020-09-22 6:57 UTC (permalink / raw)
To: Philippe Mathieu-Daudé,
haibinzhang(张海斌), Peter Maydell,
qemu-arm@nongnu.org
Cc: Leif Lindholm, Ard Biesheuvel, qemu-devel@nongnu.org,
Stefan Hajnoczi
On 09/18/20 15:41, Philippe Mathieu-Daudé wrote:
> Cc'ing firmware experts.
>
> On 9/18/20 2:26 PM, haibinzhang(寮犳捣鏂�) wrote:
>> Default size of arm-virt pflash image is 64MB which
>> will cost extra 128MB(64MBx2) memory per qemu process
>> and 12.5GB for 100 qemu processes. Host memory is
>> precious and it is valuable to reduce pflash image size.
>> For compatibility arm-virt uses real size when mapping.
>
> Flash#0 is a device because eventually its model will handle
> sector/block protection, so firmware can do CapsuleUpdate
> (updating itself). Meanwhile you could treat flash#0 as a pure
> ROM device... But this wouldn't be the 'arm-virt' machine anymore.
>
>>
>> Signed-off-by: Haibin Zhang <haibinzhang@tencent.com>
>> ---
>> hw/arm/virt.c | 22 ++++++++++++++++++++--
>> 1 file changed, 20 insertions(+), 2 deletions(-)
>>
>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>> index acf9bfbece..3545e12865 100644
>> --- a/hw/arm/virt.c
>> +++ b/hw/arm/virt.c
>> @@ -50,6 +50,7 @@
>> #include "sysemu/sysemu.h"
>> #include "sysemu/tpm.h"
>> #include "sysemu/kvm.h"
>> +#include "sysemu/block-backend.h"
>> #include "hw/loader.h"
>> #include "exec/address-spaces.h"
>> #include "qemu/bitops.h"
>> @@ -1001,10 +1002,27 @@ static void virt_flash_map(VirtMachineState *vms,
>> */
>> hwaddr flashsize = vms->memmap[VIRT_FLASH].size / 2;
>> hwaddr flashbase = vms->memmap[VIRT_FLASH].base;
>> + int64_t realsize;
>> + BlockBackend *blk;
>>
>> - virt_flash_map1(vms->flash[0], flashbase, flashsize,
>> + realsize = flashsize;
>> + blk = pflash_cfi01_get_blk(vms->flash[0]);
>> + if (blk) {
>> + realsize = blk_getlength(blk);
>> + realsize = realsize < flashsize ? realsize : flashsize;
>> + }
>
> Stefan recently posted "nvdimm: read-only file support" which
> might be a better way to achieve what you want:
> https://www.mail-archive.com/qemu-devel@nongnu.org/msg741137.html
>
>> +
>> + virt_flash_map1(vms->flash[0], flashbase, realsize,
>> secure_sysmem);
>> - virt_flash_map1(vms->flash[1], flashbase + flashsize, flashsize,
>> +
>> + realsize = flashsize;
>> + blk = pflash_cfi01_get_blk(vms->flash[1]);
>> + if (blk) {
>> + realsize = blk_getlength(blk);
>> + realsize = realsize < flashsize ? realsize : flashsize;
>> + }
>> +
>> + virt_flash_map1(vms->flash[1], flashbase + flashsize, realsize,
>> sysmem);
>> }
>>
>> --
>> 2.23.0
>>
>
We've been here before.
[Qemu-devel] [RFC PATCH] hw/arm/virt: use variable size of flash device to save memory
https://lists.gnu.org/archive/html/qemu-devel/2019-03/msg06773.html
http://mid.mail-archive.com/20190325125142.11628-1-zhengxiang9@huawei.com
I don't have anything to add.
Laszlo
^ permalink raw reply [flat|nested] 6+ messages in thread