Re: [OpenBIOS] Back at it again( PCI Passthrough )
by Jd Lyons
Ok, I made those changes to openbios, here is the output.
> On May 8, 2018, at 6:50 AM, Joe van Tunen <joevt(a)shaw.ca> wrote:
>
> : (debug-feval) ( fcode# -- fcode# )
> \ Address
> fcode-stream 1 - . here u. ." : "
>
> \ Indicate if word is compiled
> state @ 0<> if
> ." (compile) "
> then
> dup fcode>xt cell - lfa2name type
> dup ." [ 0x" . ." ]" cr
> ;
3 months, 1 week
- 7
- 54
[PATCH 0/4] Add virtio-blk driver support
by Mark Cave-Ayland
These patches add support for booting from virtio-blk (PCI) devices as
used by QEMU.
Patches 1 and 2 add bootindex support which is the new way for passing
boot order information from QEMU to OpenBIOS. Note that there are also
corresponding patches for QEMU required, without which the generated
device paths will be incorrect and cause virtio boot to fail.
Patch 3 adds the legacy virtio-blk driver (it follows the 0.9 rather
than the 1.0 specification).
Patch 4 enables the new driver for both PPC and SPARC64 architectures
so that virtio-blk devices will be usable out-of-the-box with QEMU.
Once these patches have been applied to OpenBIOS (along with the
corresponding fw path fixes for QEMU) it is possible to boot from
virtio-blk devices like this:
PPC:
./qemu-system-ppc -drive file=debian-9.0-powerpc-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,drive=cd,bootindex=0 -m 256 -boot d
SPARC:
./qemu-system-sparc64 -drive debian-9.0-sparc64-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,bus=pciB,drive=cd,bootindex=0 -m 256 -boot d -nographic
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
Mark Cave-Ayland (4):
ppc: add bootindex support
SPARC64: add bootindex support
drivers: add legacy virtio-blk driver
config: enable virtio-blk driver for default PPC and SPARC64 builds
arch/ppc/qemu/init.c | 55 ++++-
arch/sparc64/boot.c | 2 +-
arch/sparc64/boot.h | 2 +-
arch/sparc64/openbios.c | 38 +++-
config/examples/ppc_config.xml | 1 +
config/examples/sparc64_config.xml | 1 +
drivers/build.xml | 1 +
drivers/pci.c | 18 ++
drivers/pci_database.c | 4 +-
drivers/pci_database.h | 1 +
drivers/virtio.c | 424 +++++++++++++++++++++++++++++++++++++
drivers/virtio.h | 331 +++++++++++++++++++++++++++++
include/drivers/drivers.h | 4 +
13 files changed, 862 insertions(+), 20 deletions(-)
create mode 100644 drivers/virtio.c
create mode 100644 drivers/virtio.h
--
2.11.0
1 year
- 3
- 8
[PATCH 0/2] 40p minor fixes
by Mark Cave-Ayland
Here are a couple of 40p fixes I noticed whilst debugging the interrupt
routing in QEMU's 40p PReP machine.
Note that the IRQ changes in patch 2 are dependent upon my fix for the
40p PCI IRQ routing posted at
https://lists.gnu.org/archive/html/qemu-devel/2018-08/msg05705.html.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
Mark Cave-Ayland (2):
40p: fix up generation of interrupt properties
40p: update PReP configuration to match 40p machine
arch/ppc/qemu/init.c | 4 ++--
drivers/pci.c | 15 +++++++--------
2 files changed, 9 insertions(+), 10 deletions(-)
--
2.11.0
1 year, 1 month
- 3
- 6
[PATCH v3 0/4] Add virtio-blk driver support
by Mark Cave-Ayland
These patches add support for booting from virtio-blk (PCI) devices as
used by QEMU.
Patches 1 and 2 add bootindex support which is the new way for passing
boot order information from QEMU to OpenBIOS. Note that there are also
corresponding patches for QEMU required, without which the generated
device paths will be incorrect and cause virtio boot to fail.
Patch 3 adds the virtio-1.0 compliant virtio-blk PCI driver.
Patch 4 enables the new driver for both PPC and SPARC64 architectures
so that virtio-blk-pci devices will be usable out-of-the-box with QEMU.
Once these patches have been applied to OpenBIOS (along with the
corresponding fw path fixes for QEMU) it is possible to boot from
virtio-blk-pci devices like this:
PPC:
./qemu-system-ppc -drive file=debian-9.0-powerpc-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,drive=cd,bootindex=0 -m 256 -boot d
SPARC:
./qemu-system-sparc64 -drive debian-9.0-sparc64-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,bus=pciB,drive=cd,bootindex=0 -m 256 -boot d -nographic
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
v3:
- Rebase onto master
- Convert from legacy driver to virtio-1.0 driver
v2:
- Switch over to using dma-* words to allocate rings and descriptors
- Fix up some minor whitespace issues
Mark Cave-Ayland (4):
ppc: add bootindex support
SPARC64: add bootindex support
drivers: add virtio-1.0 virtio-blk driver
config: enable virtio-blk driver for default PPC and SPARC64 builds
arch/ppc/qemu/init.c | 55 +++-
arch/sparc64/boot.c | 2 +-
arch/sparc64/boot.h | 2 +-
arch/sparc64/openbios.c | 38 ++-
config/examples/ppc_config.xml | 1 +
config/examples/sparc64_config.xml | 1 +
drivers/build.xml | 1 +
drivers/pci.c | 63 +++++
drivers/pci_database.c | 14 +-
drivers/pci_database.h | 1 +
drivers/virtio.c | 555 +++++++++++++++++++++++++++++++++++++
drivers/virtio.h | 370 +++++++++++++++++++++++++
include/drivers/drivers.h | 5 +
include/drivers/pci.h | 6 +
14 files changed, 1093 insertions(+), 21 deletions(-)
create mode 100644 drivers/virtio.c
create mode 100644 drivers/virtio.h
--
2.11.0
1 year, 1 month
- 1
- 5
[PATCH] ppc: make sure that we correctly map loader RAM at load-base
by Mark Cave-Ayland
An error in the logic related to FREE_BASE meant that instead of mapping the
RAM for the loader at load-base, a small section was being mapped at the
bottom of RAM instead.
Fix this by deferring the mapping of the load-base RAM to arch_init() when
we can access the load-base variable and map 8MB RAM with a 1:1 phys to
virt mapping.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
---
arch/ppc/qemu/init.c | 8 ++++++++
arch/ppc/qemu/ofmem.c | 12 ------------
2 files changed, 8 insertions(+), 12 deletions(-)
diff --git a/arch/ppc/qemu/init.c b/arch/ppc/qemu/init.c
index af15682..f5acf87 100644
--- a/arch/ppc/qemu/init.c
+++ b/arch/ppc/qemu/init.c
@@ -844,6 +844,7 @@ arch_of_init(void)
uint32_t temp = 0;
char *boot_device;
ofmem_t *ofmem = ofmem_arch_get_private();
+ ucell load_base;
openbios_init();
modules_init();
@@ -1105,4 +1106,11 @@ arch_of_init(void)
bind_func("platform-boot", boot);
bind_func("(arch-go)", arch_go);
+
+ /* Allocate 8MB memory at load-base */
+ fword("load-base");
+ load_base = POP();
+ ofmem_claim_phys(load_base, 0x800000, 0);
+ ofmem_claim_virt(load_base, 0x800000, 0);
+ ofmem_map(load_base, load_base, 0x800000, 0);
}
diff --git a/arch/ppc/qemu/ofmem.c b/arch/ppc/qemu/ofmem.c
index 7b8ced0..7a78a1e 100644
--- a/arch/ppc/qemu/ofmem.c
+++ b/arch/ppc/qemu/ofmem.c
@@ -46,7 +46,6 @@ extern void setup_mmu(unsigned long code_base);
*
*/
-#define FREE_BASE 0x00004000UL
#define OF_CODE_START 0xfff00000UL
#define OF_CODE_SIZE 0x00100000
#define IO_BASE 0x80000000UL
@@ -81,12 +80,6 @@ get_ram_top(void)
return get_hash_base() - (32 + 64 + 64) * 1024 - OFMEM_SIZE;
}
-static unsigned long
-get_ram_bottom(void)
-{
- return FREE_BASE;
-}
-
static unsigned long get_heap_top(void)
{
return get_hash_base() - (32 + 64 + 64) * 1024;
@@ -578,11 +571,6 @@ ofmem_init(void)
{
ofmem_t *ofmem = ofmem_arch_get_private();
- /* Map the memory (don't map page 0 to allow catching of NULL dereferences) */
- ofmem_claim_phys(PAGE_SIZE, get_ram_bottom() - PAGE_SIZE, 0);
- ofmem_claim_virt(PAGE_SIZE, get_ram_bottom() - PAGE_SIZE, 0);
- ofmem_map(PAGE_SIZE, PAGE_SIZE, get_ram_bottom() - PAGE_SIZE, 0);
-
/* Mark the first page as non-free */
ofmem_claim_phys(0, PAGE_SIZE, 0);
ofmem_claim_virt(0, PAGE_SIZE, 0);
--
2.11.0
1 year, 1 month
- 1
- 3
[PATCH] SPARC64: fix up dma-* words
by Mark Cave-Ayland
Commit 8584d42 "SPARC64: implement dma-* words" introduced the dma-* words
for SPARC64 but accidentally introduced an older prototype implementation
(probably introduced during a rebase) rather than using the correct defer
words.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
---
arch/sparc64/tree.fs | 13 +++++--------
1 file changed, 5 insertions(+), 8 deletions(-)
diff --git a/arch/sparc64/tree.fs b/arch/sparc64/tree.fs
index a1d5620..af8948d 100644
--- a/arch/sparc64/tree.fs
+++ b/arch/sparc64/tree.fs
@@ -52,26 +52,23 @@ include config.fs
: decode-unit decode-unit-upa ;
: dma-sync ( virt devaddr size -- )
- s" (dma-sync)" $find if execute then
+ (dma-sync)
;
: dma-alloc ( size -- virt )
- \ OpenBIOS doesn't enable the sun4u IOMMU so we can fall back to using
- \ alloc-mem
- h# 2000 + alloc-mem dup
- h# 2000 1 - and - \ align to 8K page size
+ (dma-alloc)
;
: dma-free ( virt size -- )
- 2drop
+ (dma-free)
;
: dma-map-in ( virt size cacheable? -- devaddr )
- 2drop
+ (dma-map-in)
;
: dma-map-out ( virt devaddr size -- )
- dma-sync
+ (dma-map-out)
;
new-device
--
2.11.0
1 year, 1 month
- 1
- 1
[PATCH] ppc: move init-program client stack to 1:1 virt to phys mapped memory
by Mark Cave-Ayland
The default context used by init-program is allocated from the OpenBIOS
heap which uses a different virt to phys mapping to avoid having to
physically relocate the PROM on startup for PPC machines.
It seems that the OpenBSD bootloader isn't happy with this change
introduced by the context rework, most likely because there is an
assumption that the client stack is mapped 1:1 virt to phys.
Fix this by overriding the default context stack during init-program
to a location just under the MMU hash table (SDR1) which was its
previous hard-coded location in 1:1 virt to phys mapped memory, and
allow OpenBSD to boot.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
---
arch/ppc/qemu/context.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/arch/ppc/qemu/context.c b/arch/ppc/qemu/context.c
index 03ca173..06e0122 100644
--- a/arch/ppc/qemu/context.c
+++ b/arch/ppc/qemu/context.c
@@ -16,6 +16,7 @@
#include "libopenbios/ofmem.h"
#include "libopenbios/initprogram.h"
#include "libopenbios/sys_info.h"
+#include "arch/ppc/processor.h"
#define MAIN_STACK_SIZE 16384
#define IMAGE_STACK_SIZE 4096*2
@@ -232,6 +233,14 @@ arch_init_program(void)
ctx->regs[REG_R6] = 0;
ctx->regs[REG_R7] = 0;
+ /* Override the stack in the default context: the OpenBSD bootloader
+ fails soon after setting up virt to phys mappings with the default
+ stack. My best guess is that this is because the malloc() heap
+ doesn't have a 1:1 virt to phys mapping. So for the moment we use
+ the original (pre-context) location just under the MMU hash table
+ (SDR1) which is mapped 1:1 and makes the bootloader happy. */
+ ctx->sp = mfsdr1() - 32768 - 65536;
+
/* Set param */
feval("load-state >ls.param @");
param = POP();
--
2.11.0
1 year, 1 month
- 1
- 1
[PATCH v2 0/4] Add virtio-blk driver support
by Mark Cave-Ayland
These patches add support for booting from virtio-blk (PCI) devices as
used by QEMU.
Patches 1 and 2 add bootindex support which is the new way for passing
boot order information from QEMU to OpenBIOS. Note that there are also
corresponding patches for QEMU required, without which the generated
device paths will be incorrect and cause virtio boot to fail.
Patch 3 adds the legacy virtio-blk driver (it follows the 0.9 rather
than the 1.0 specification).
Patch 4 enables the new driver for both PPC and SPARC64 architectures
so that virtio-blk devices will be usable out-of-the-box with QEMU.
Once these patches have been applied to OpenBIOS (along with the
corresponding fw path fixes for QEMU) it is possible to boot from
virtio-blk devices like this:
PPC:
./qemu-system-ppc -drive file=debian-9.0-powerpc-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,drive=cd,bootindex=0 -m 256 -boot d
SPARC:
./qemu-system-sparc64 -drive debian-9.0-sparc64-NETINST.iso,if=none,index=0,id=cd,media=cdrom \
-device virtio-blk-pci,bus=pciB,drive=cd,bootindex=0 -m 256 -boot d -nographic
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
v2:
- Switch over to using dma-* words to allocate rings and descriptors
- Fix up some minor whitespace issues
Mark Cave-Ayland (4):
ppc: add bootindex support
SPARC64: add bootindex support
drivers: add legacy virtio-blk driver
config: enable virtio-blk driver for default PPC and SPARC64 builds
arch/ppc/qemu/init.c | 55 +++-
arch/sparc64/boot.c | 2 +-
arch/sparc64/boot.h | 2 +-
arch/sparc64/openbios.c | 38 ++-
config/examples/ppc_config.xml | 1 +
config/examples/sparc64_config.xml | 1 +
drivers/build.xml | 1 +
drivers/pci.c | 18 ++
drivers/pci_database.c | 4 +-
drivers/pci_database.h | 1 +
drivers/virtio.c | 516 +++++++++++++++++++++++++++++++++++++
drivers/virtio.h | 330 ++++++++++++++++++++++++
include/drivers/drivers.h | 4 +
13 files changed, 953 insertions(+), 20 deletions(-)
create mode 100644 drivers/virtio.c
create mode 100644 drivers/virtio.h
--
2.11.0
1 year, 2 months
- 1
- 4
[PATCH 0/3] Miscellaneous fixes
by Mark Cave-Ayland
Here is a set of fixes discovered during my work on adding virtio-blk
support to OpenBIOS.
Patch 1 changes the way in which the IDE controller DT nodes are generated
so that they use the controller index as the address, and not the value
programmed in the PCI BAR. Otherwise it is impossible for QEMU to determine
the fw path for IDE boot devices, since the PCI BARs are programmed after
the machine has been initialised.
Patches 2 and 3 contain fixes for bugs that were discovered whilst trying
to initialise the virtio-blk device for both PPC and SPARC64 architectures.
Signed-off-by: Mark Cave-Ayland <mark.cave-ayland(a)ilande.co.uk>
Mark Cave-Ayland (3):
ide: don't use ioports as part of the controller node address
libopenbios: don't allow find_package_method() to push a NULL for
empty strings
pci: use absolute PCI IO addresses when programming bridge IO limits
config/examples/ppc64_config.xml | 2 +-
config/examples/ppc_config.xml | 2 +-
drivers/ide.c | 18 ++++++++----------
drivers/pci.c | 2 +-
libopenbios/bindings.c | 7 ++++++-
5 files changed, 17 insertions(+), 14 deletions(-)
--
2.11.0
1 year, 2 months
- 2
- 6