Skip to contentSkip to conversation

Add support for 32 bit physical address

PATCH Superseded Ayan Kumar Halder 7 participants 64 messages Jan 17, 2023 – Feb 10, 2023

Subsystems (2)

v2 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

v2 cover Add support for 32 bit physical address

Hi All,

Please have a look at https://lists.xenproject.org/archives/html/xen-devel/2022-11/msg01465.html
for the context.

The benefits of using 32 bit physical addresses are as follows :-

1. It helps to use Xen on platforms (for eg R52) which supports 32 bit physical addresses and has no support for large page address extension. On 32 bit MPU systems which supports flat-mapping (for eg R52), it helps to translate 32 bit VA into 32 bit PA.

2. It also helps in code optimization when the underlying platform does not use large page address extension.

The following points are to be noted :-
1. Device tree always use u64 for address and size. The caller needs to translate between u64 and u32 (when 32 bit physical addressing is used).
2. Currently, we have enabled this option for Arm_32 as the MMU for Arm_64 uses 48 bit physical addressing.
3. https://lists.xenproject.org/archives/html/xen-devel/2022-12/msg00117.html
has been added to this series.

Changes from :

v1 - 1. Reordered the patches such that the first three patches fixes issues in the existing codebase. These can be applied independent of the remaining patches in this serie,

2. Dropped translate_dt_address_size() for the address/size translation between paddr_t and u64 (as parsed from the device tree). Also, dropped the check for truncation (while converting u64 to paddr_t).
Instead now we have modified device_tree_get_reg() and typecasted the return for dt_read_number(), to obtain paddr_t. Also, introduced wrappers for fdt_get_mem_rsv() and dt_device_get_address() for the same purpose. These can be found in patch 4/11 and patch 6/11.

3. Split "Other adaptations required to support 32bit paddr" into the following individual patches for each adaptation :
xen/arm: smmu: Use writeq_relaxed_non_atomic() for writing to
SMMU_CBn_TTBR0
xen/arm: guest_walk: LPAE specific bits should be enclosed within

    "ifndef CONFIG_ARM_PA_32"

4. Introduced "xen/arm: p2m: Enable support for 32bit IPA".

Ayan Kumar Halder (11):
xen/ns16550: Remove unneeded truncation check in the DT init code xen/arm: Use the correct format specifier
xen/arm: domain_build: Replace use of paddr_t in find_domU_holes() xen/arm: Typecast the DT values into paddr_t
xen/arm: Use paddr_t instead of u64 for address/size xen/arm: Introduce a wrapper for dt_device_get_address() to handle
paddr_t
xen/arm: smmu: Use writeq_relaxed_non_atomic() for writing to
SMMU_CBn_TTBR0
xen/arm: guest_walk: LPAE specific bits should be enclosed within

    "ifndef CONFIG_ARM_PA_32"

xen/arm: Introduce ARM_PA_32 to support 32 bit physical address xen/arm: Restrict zeroeth_table_offset for ARM_64 xen/arm: p2m: Enable support for 32bit IPA

 xen/arch/arm/Kconfig                   |  9 ++++
 xen/arch/arm/bootfdt.c                 | 23 ++++++----
 xen/arch/arm/domain_build.c            | 32 +++++++-------
 xen/arch/arm/gic-v2.c                  | 17 ++++----
 xen/arch/arm/gic-v3.c                  | 11 ++---
 xen/arch/arm/guest_walk.c              |  2 +
 xen/arch/arm/include/asm/device_tree.h | 59 ++++++++++++++++++++++++++
 xen/arch/arm/include/asm/lpae.h        |  4 ++
 xen/arch/arm/include/asm/page-bits.h   |  2 +
 xen/arch/arm/include/asm/setup.h       |  2 +-
 xen/arch/arm/include/asm/types.h       |  7 +++
 xen/arch/arm/mm.c                      |  9 +---
 xen/arch/arm/p2m.c                     | 10 +++--
 xen/arch/arm/platforms/exynos5.c       | 33 +++++++-------
 xen/arch/arm/platforms/sunxi.c         |  3 +-
 xen/arch/arm/setup.c                   | 14 +++---
 xen/arch/arm/smpboot.c                 |  2 +-
 xen/drivers/char/exynos4210-uart.c     |  5 ++-
 xen/drivers/char/ns16550.c             | 16 +++----
 xen/drivers/char/omap-uart.c           |  5 ++-
 xen/drivers/char/pl011.c               |  7 +--
 xen/drivers/char/scif-uart.c           |  5 ++-
 xen/drivers/passthrough/arm/smmu.c     | 24 ++++++-----
 23 files changed, 199 insertions(+), 102 deletions(-)
 create mode 100644 xen/arch/arm/include/asm/device_tree.h
Signature · 1 line
2.17.1
#1 lore raw

PATCH v2 1/11 xen/ns16550: Remove unneeded truncation check in the DT init code

In an earlier commit (7c1de0038895), "ns16550.io_size" was u32 and "io_size" was u64. Thus, the ASSERT() was needed to check if the values are the same.
However, in a later commit (c9f8e0aee507), "ns16550.io_size" was changed to u64. Thus, the ASSERT() became redundant.

So, now as "io_size" and "uart->io_size" are both u64, there will be no truncation. Thus, one can remove the ASSERT() and extra assignment.

---

Changes from :

v1 - 1. Updated commit message/title.
2. Added Rb.

 xen/drivers/char/ns16550.c | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
xen/drivers/char/ns16550.c +1 -6
index 01a05c9aa8..58d0ccd889 100644
@@ -1747,7 +1747,6 @@ static int __init ns16550_uart_dt_init(struct dt_device_node *dev,
     struct ns16550 *uart;
     int res;
     u32 reg_shift, reg_width;
-    u64 io_size;
 
     uart = &ns16550_com[0];
 
@@ -1758,14 +1757,10 @@ static int __init ns16550_uart_dt_init(struct dt_device_node *dev,
     uart->parity    = UART_PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &uart->io_base, &io_size);
+    res = dt_device_get_address(dev, 0, &uart->io_base, &uart->io_size);
     if ( res )
         return res;
 
-    uart->io_size = io_size;
-
-    ASSERT(uart->io_size == io_size); /* Detect truncation */
-
     res = dt_property_read_u32(dev, "reg-shift", &reg_shift);
     if ( !res )
         uart->reg_shift = 0;
Signature · 1 line
2.17.1
#2 lore raw

PATCH v2 2/11 xen/arm: Use the correct format specifier

In reply to Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

---

Changes from -

v1 - 1. Moved the patch earlier.
2. Moved a part of change from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr" into this patch.

 xen/arch/arm/domain_build.c | 10 +++++-----
 xen/arch/arm/gic-v2.c       |  6 +++---
 xen/arch/arm/mm.c           |  2 +-
 3 files changed, 9 insertions(+), 9 deletions(-)
xen/arch/arm/domain_build.c +5 -5
index 829cea8de8..33a5945a2d 100644
@@ -1315,7 +1315,7 @@ static int __init make_memory_node(const struct domain *d,
     dt_dprintk("Create memory node\n");
 
     /* ePAPR 3.4 */
-    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[i].start);
+    snprintf(buf, sizeof(buf), "memory@%"PRIpaddr, mem->bank[i].start);
     res = fdt_begin_node(fdt, buf);
     if ( res )
         return res;
@@ -1383,7 +1383,7 @@ static int __init make_shm_memory_node(const struct domain *d,
         __be32 *cells;
         unsigned int len = (addrcells + sizecells) * sizeof(__be32);
 
-        snprintf(buf, sizeof(buf), "xen-shmem@%"PRIx64, mem->bank[i].start);
+        snprintf(buf, sizeof(buf), "xen-shmem@%"PRIpaddr, mem->bank[i].start);
         res = fdt_begin_node(fdt, buf);
         if ( res )
             return res;
@@ -2719,7 +2719,7 @@ static int __init make_gicv2_domU_node(struct kernel_info *kinfo)
     /* Placeholder for interrupt-controller@ + a 64-bit number + \0 */
     char buf[38];
 
-    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIx64,
+    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIpaddr,
              vgic_dist_base(&d->arch.vgic));
     res = fdt_begin_node(fdt, buf);
     if ( res )
@@ -2775,7 +2775,7 @@ static int __init make_gicv3_domU_node(struct kernel_info *kinfo)
     char buf[38];
     unsigned int i, len = 0;
 
-    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIx64,
+    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIpaddr,
              vgic_dist_base(&d->arch.vgic));
 
     res = fdt_begin_node(fdt, buf);
@@ -2861,7 +2861,7 @@ static int __init make_vpl011_uart_node(struct kernel_info *kinfo)
     /* Placeholder for sbsa-uart@ + a 64-bit number + \0 */
     char buf[27];
 
-    snprintf(buf, sizeof(buf), "sbsa-uart@%"PRIx64, d->arch.vpl011.base_addr);
+    snprintf(buf, sizeof(buf), "sbsa-uart@%"PRIpaddr, d->arch.vpl011.base_addr);
     res = fdt_begin_node(fdt, buf);
     if ( res )
         return res;
xen/arch/arm/gic-v2.c +3 -3
index 61802839cb..5d4d298b86 100644
@@ -1049,7 +1049,7 @@ static void __init gicv2_dt_init(void)
     if ( csize < SZ_8K )
     {
         printk(XENLOG_WARNING "GICv2: WARNING: "
-               "The GICC size is too small: %#"PRIx64" expected %#x\n",
+               "The GICC size is too small: %#"PRIpaddr" expected %#x\n",
                csize, SZ_8K);
         if ( platform_has_quirk(PLATFORM_QUIRK_GIC_64K_STRIDE) )
         {
@@ -1280,11 +1280,11 @@ static int __init gicv2_init(void)
         gicv2.map_cbase += aliased_offset;
 
         printk(XENLOG_WARNING
-               "GICv2: Adjusting CPU interface base to %#"PRIx64"\n",
+               "GICv2: Adjusting CPU interface base to %#"PRIpaddr"\n",
                cbase + aliased_offset);
     } else if ( csize == SZ_128K )
         printk(XENLOG_WARNING
-               "GICv2: GICC size=%#"PRIx64" but not aliased\n",
+               "GICv2: GICC size=%#"PRIpaddr" but not aliased\n",
                csize);
 
     gicv2.map_hbase = ioremap_nocache(hbase, PAGE_SIZE);
xen/arch/arm/mm.c +1 -1
index 0fc6f2992d..fab54618ab 100644
@@ -249,7 +249,7 @@ void dump_pt_walk(paddr_t ttbr, paddr_t addr,
 
         pte = mapping[offsets[level]];
 
-        printk("%s[0x%03x] = 0x%"PRIpaddr"\n",
+        printk("%s[0x%03x] = 0x%"PRIx64"\n",
                level_strs[level], offsets[level], pte.bits);
 
         if ( level == 3 || !pte.walk.valid || !pte.walk.table )
Signature · 1 line
2.17.1
Reviewed-by on v2 2/11
#3 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Quoted 6 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 105 lines

---

Changes from -

v1 - 1. Moved the patch earlier.
2. Moved a part of change from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr" into this patch.

 xen/arch/arm/domain_build.c | 10 +++++-----
 xen/arch/arm/gic-v2.c       |  6 +++---
 xen/arch/arm/mm.c           |  2 +-
 3 files changed, 9 insertions(+), 9 deletions(-)
xen/arch/arm/domain_build.c +5 -5
index 829cea8de8..33a5945a2d 100644
@@ -1315,7 +1315,7 @@ static int __init make_memory_node(const struct domain *d,
     dt_dprintk("Create memory node\n");
 
     /* ePAPR 3.4 */
-    snprintf(buf, sizeof(buf), "memory@%"PRIx64, mem->bank[i].start);
+    snprintf(buf, sizeof(buf), "memory@%"PRIpaddr, mem->bank[i].start);
     res = fdt_begin_node(fdt, buf);
     if ( res )
         return res;
@@ -1383,7 +1383,7 @@ static int __init make_shm_memory_node(const struct domain *d,
         __be32 *cells;
         unsigned int len = (addrcells + sizecells) * sizeof(__be32);
 
-        snprintf(buf, sizeof(buf), "xen-shmem@%"PRIx64, mem->bank[i].start);
+        snprintf(buf, sizeof(buf), "xen-shmem@%"PRIpaddr, mem->bank[i].start);
         res = fdt_begin_node(fdt, buf);
         if ( res )
             return res;
@@ -2719,7 +2719,7 @@ static int __init make_gicv2_domU_node(struct kernel_info *kinfo)
     /* Placeholder for interrupt-controller@ + a 64-bit number + \0 */
     char buf[38];
 
-    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIx64,
+    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIpaddr,
              vgic_dist_base(&d->arch.vgic));
     res = fdt_begin_node(fdt, buf);
     if ( res )
@@ -2775,7 +2775,7 @@ static int __init make_gicv3_domU_node(struct kernel_info *kinfo)
     char buf[38];
     unsigned int i, len = 0;
 
-    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIx64,
+    snprintf(buf, sizeof(buf), "interrupt-controller@%"PRIpaddr,
              vgic_dist_base(&d->arch.vgic));
 
     res = fdt_begin_node(fdt, buf);
@@ -2861,7 +2861,7 @@ static int __init make_vpl011_uart_node(struct kernel_info *kinfo)
     /* Placeholder for sbsa-uart@ + a 64-bit number + \0 */
     char buf[27];
 
-    snprintf(buf, sizeof(buf), "sbsa-uart@%"PRIx64, d->arch.vpl011.base_addr);
+    snprintf(buf, sizeof(buf), "sbsa-uart@%"PRIpaddr, d->arch.vpl011.base_addr);
     res = fdt_begin_node(fdt, buf);
     if ( res )
         return res;
xen/arch/arm/gic-v2.c +3 -3
index 61802839cb..5d4d298b86 100644
@@ -1049,7 +1049,7 @@ static void __init gicv2_dt_init(void)
     if ( csize < SZ_8K )
     {
         printk(XENLOG_WARNING "GICv2: WARNING: "
-               "The GICC size is too small: %#"PRIx64" expected %#x\n",
+               "The GICC size is too small: %#"PRIpaddr" expected %#x\n",
                csize, SZ_8K);
         if ( platform_has_quirk(PLATFORM_QUIRK_GIC_64K_STRIDE) )
         {
@@ -1280,11 +1280,11 @@ static int __init gicv2_init(void)
         gicv2.map_cbase += aliased_offset;
 
         printk(XENLOG_WARNING
-               "GICv2: Adjusting CPU interface base to %#"PRIx64"\n",
+               "GICv2: Adjusting CPU interface base to %#"PRIpaddr"\n",
                cbase + aliased_offset);
     } else if ( csize == SZ_128K )
         printk(XENLOG_WARNING
-               "GICv2: GICC size=%#"PRIx64" but not aliased\n",
+               "GICv2: GICC size=%#"PRIpaddr" but not aliased\n",
                csize);
 
     gicv2.map_hbase = ioremap_nocache(hbase, PAGE_SIZE);
xen/arch/arm/mm.c +1 -1
index 0fc6f2992d..fab54618ab 100644
@@ -249,7 +249,7 @@ void dump_pt_walk(paddr_t ttbr, paddr_t addr,
 
         pte = mapping[offsets[level]];
 
-        printk("%s[0x%03x] = 0x%"PRIpaddr"\n",
+        printk("%s[0x%03x] = 0x%"PRIx64"\n",
                level_strs[level], offsets[level], pte.bits);
 
         if ( level == 3 || !pte.walk.valid || !pte.walk.table )
Signature · 1 line
2.17.1
#4 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Hi,

Quoted 9 lines · Stefano Stabellini
Quoted 6 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

I have committed the patch.

Cheers,

Signature · 1 line
Julien Grall
#5 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Quoted 1 line · Julien Grall

Hi,

Hi Julien,

Quoted 8 lines · Stefano Stabellini
Quoted 6 lines · Ayan Kumar Halder
Quoted 5 lines

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

Thanks for the reviews and commit. :)

Did you miss "[XEN v2 01/11] xen/ns16550: Remove unneeded truncation check in the DT init code" ?

- Ayan

Quoted 3 lines

Cheers,

#6 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

In reply to Julien Grall

Hello,

Quoted 11 lines · Julien Grall

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

I took a look at it and this is because in the test script we
try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and we end up creating nodes like:

memory@0000000050000000

instead of:

memory@60000000

We could modify the script, but do we really want to create nodes with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

[1] https://elinux.org/Device_Tree_Linux

Quoted 6 lines

Cheers,

--
Julien Grall

~Michal

#7 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Quoted 1 line · Michal Orzel

Hello,

Hi,

Quoted 13 lines · Julien Grall
Quoted 11 lines

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 6 lines

I took a look at it and this is because in the test script we
try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Cheers,

[2] https://www.devicetree.org/

Signature · 1 line
Julien Grall
#8 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Hi Julien,

Quoted 56 lines · Julien Grall
Quoted 1 line · Michal Orzel

Hello,

Hi,

Quoted 13 lines · Julien Grall
Quoted 11 lines

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 6 lines

I took a look at it and this is because in the test script we
try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Quoted 6 lines

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Yes, the CI needs to be updated as well.
I'm in favor of removing leading zeroes because this is what Xen uses in all the other places when creating nodes (or copying them from the host dtb) including xl when creating dtb for domUs. Having a mismatch may be confusing and having a unit-address with leading zeroes looks unusual.

Quoted 7 lines

Cheers,

[2] https://www.devicetree.org/

--
Julien Grall

~Michal

#9 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Quoted 1 line · Michal Orzel

Hi Julien,

Hi Michal,

Quoted 58 lines · Julien Grall
Quoted 56 lines · Michal Orzel
Quoted 1 line

Hello,

Hi,

Quoted 13 lines · Julien Grall
Quoted 11 lines

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 6 lines

I took a look at it and this is because in the test script we
try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Quoted 7 lines
Quoted 6 lines

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Yes, the CI needs to be updated as well.

Can either you or Ayan look at it?

Quoted 4 lines

I'm in favor of removing leading zeroes because this is what Xen uses in all the other places when creating nodes (or copying them from the host dtb) including xl when creating dtb for domUs. Having a mismatch may be confusing and having a unit-address with leading zeroes looks unusual.

I decided to revert the patch mainly because it will be easier to review the fix if it is folded in this patch.

I would consider to create a wrapper on top of fdt_begin_node() that will take the name of the node and the unit. Something like:

/* XXX: Explain why the wrapper is needed */
static void domain_fdt_begin_node(void *fdt, const char *name, uint64_t unit) {
char buf[X];

snprintf(buf, sizeof(buf), "%s@%"PRIx64, ...);
/* XXX check the return of snprintf() */

return fdt_begin_node(buf);
}

X would be a value that is large enough to accommodate the existing name.

The reason I don't suggest a new PRI* is because I can't think of a name that is short and descriptive enough to understand the different with the existing PRIpaddr.

Cheers,

Signature · 1 line
Julien Grall
#10 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Hi Julien/Michal,

Quoted 69 lines · Julien Grall
Quoted 1 line · Michal Orzel

Hi Julien,

Hi Michal,

Quoted 62 lines · Julien Grall
Quoted 60 lines · Michal Orzel
Quoted 1 line

Hello,

Hi,

Quoted 13 lines · Julien Grall
Quoted 11 lines

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current use of 'PRIpaddr' for printing PTE is buggy as this is not a physical address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 6 lines

I took a look at it and this is because in the test script we
try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

Quoted 5 lines
Quoted 3 lines
Quoted 2 lines

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Yes, the CI needs to be updated as well.

Can either you or Ayan look at it?

Does this change match the expectation ?

automation/scripts/qemu-smoke-dom0less-arm64.sh +0 -0
index 2b59346fdc..9f5e700f0e 100755
@@ -20,7 +20,7 @@ if [[ "${test_variant}" == "static-mem" ]]; then

     domu_size="10000000"
     passed="${test_variant} test passed"
     domU_check="
-current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@${domu_base}/reg 2>/dev/null) +current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@$[0-9]*/reg 2>/dev/null) expected=$(printf \"%016x%016x\" 0x${domu_base} 0x${domu_size})
 if [[ \"\${expected}\" == \"\${current}\" ]]; then
        echo \"${passed}\"

Quoted 19 lines
Quoted 5 lines

I'm in favor of removing leading zeroes because this is what Xen uses in all the other places when creating nodes (or copying them from the host dtb) including xl when creating dtb for domUs. Having a mismatch may be confusing and having a unit-address with leading zeroes looks unusual.

I decided to revert the patch mainly because it will be easier to review the fix if it is folded in this patch.

I would consider to create a wrapper on top of fdt_begin_node() that will take the name of the node and the unit. Something like:

/* XXX: Explain why the wrapper is needed */
static void domain_fdt_begin_node(void *fdt, const char *name, uint64_t unit) {
   char buf[X];

   snprintf(buf, sizeof(buf), "%s@%"PRIx64, ...);
   /* XXX check the return of snprintf() */

Quoted 6 lines

   return fdt_begin_node(buf);
}

X would be a value that is large enough to accommodate the existing name.

The reason I don't suggest a new PRI* is because I can't think of a name that is short and descriptive enough to understand the different with the existing PRIpaddr.

Looks fine to me. I can incorporate the change in this existing patch.

- Ayan

Quoted 3 lines

Cheers,

#11 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Quoted 91 lines · Ayan Kumar Halder

Hi Julien/Michal,

Quoted 72 lines · Julien Grall
Quoted 1 line · Michal Orzel

Hi Julien,

Hi Michal,

Quoted 65 lines · Julien Grall
Quoted 63 lines · Michal Orzel
Quoted 1 line

Hello,

Hi,

Quoted 15 lines · Julien Grall
Quoted 13 lines

Hi,

Quoted 8 lines · Stefano Stabellini
Quoted 7 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current
use of 'PRIpaddr' for printing PTE is buggy as this is not a physical
address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 7 lines

I took a look at it and this is because in the test script we try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and
we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

Quoted 5 lines
Quoted 3 lines
Quoted 2 lines

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Yes, the CI needs to be updated as well.

Can either you or Ayan look at it?

Does this change match the expectation ?

automation/scripts/qemu-smoke-dom0less-arm64.sh +0 -0
index 2b59346fdc..9f5e700f0e 100755
@@ -20,7 +20,7 @@ if [[ "${test_variant}" == "static-mem" ]]; then

     domu_size="10000000"
     passed="${test_variant} test passed"

Quoted 8 lines

     domU_check="
-current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@${domu_base}/reg 2>/dev/null) +current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@$[0-9]*/reg 2>/dev/null)
 expected=$(printf \"%016x%016x\" 0x${domu_base} 0x${domu_size}) if [[ \"\${expected}\" == \"\${current}\" ]]; then
        echo \"${passed}\"

We need to check for ${domu_base} with or without leading zeroes:

current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@*(0)${domu_base}/reg 2>/dev/null)

#12 lore raw

Re: [XEN v2 02/11] xen/arm: Use the correct format specifier

Hi Stefano, Ayan, Julien

Quoted 105 lines · Stefano Stabellini
Quoted 98 lines · Ayan Kumar Halder

Hi Julien/Michal,

Quoted 72 lines · Julien Grall
Quoted 1 line · Michal Orzel

Hi Julien,

Hi Michal,

Quoted 65 lines · Julien Grall
Quoted 63 lines · Michal Orzel
Quoted 1 line

Hello,

Hi,

Quoted 15 lines · Julien Grall
Quoted 13 lines

Hi,

Quoted 8 lines · Stefano Stabellini
Quoted 7 lines · Ayan Kumar Halder

1. One should use 'PRIpaddr' to display 'paddr_t' variables.
2. One should use 'PRIx64' to display 'u64' in hex format. The current
use of 'PRIpaddr' for printing PTE is buggy as this is not a physical
address.

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

I have committed the patch.

The CI test jobs (static-mem) failed on this patch:
https://gitlab.com/xen-project/xen/-/pipelines/752911309

Thanks for the report.

Quoted 7 lines

I took a look at it and this is because in the test script we try to find a node whose unit-address does not have leading zeroes. However, after this patch, we switched to PRIpaddr which is defined as 016lx/016llx and
we end up creating nodes like:

Quoted 7 lines

memory@0000000050000000

instead of:

memory@60000000

We could modify the script,

TBH, I think it was a mistake for the script to rely on how Xen describe the memory banks in the Device-Tree.

For instance, from my understanding, it would be valid for Xen to create a single node for all the banks or even omitting the unit-address if there is only one bank.

Quoted 3 lines

but do we really want to create nodes
with leading zeroes? The dt spec does not mention it, although [1] specifies that the Linux convention is not to have leading zeroes.

Reading through the spec in [2], it suggested the current naming is fine. That said the example match the Linux convention (I guess that's not surprising...).

Quoted 5 lines
Quoted 3 lines
Quoted 2 lines

I am open to remove the leading. However, I think the CI also needs to be updated (see above why).

Yes, the CI needs to be updated as well.

Can either you or Ayan look at it?

Does this change match the expectation ?

automation/scripts/qemu-smoke-dom0less-arm64.sh +1 -1
index 2b59346fdc..9f5e700f0e 100755
@@ -20,7 +20,7 @@ if [[ "${test_variant}" == "static-mem" ]]; then
     domu_size="10000000"
     passed="${test_variant} test passed"
     domU_check="
-current=\$(hexdump -e '16/1 \"%02x\"'
 /proc/device-tree/memory@${domu_base}/reg 2>/dev/null)
+current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@$[0-9]*/reg
 2>/dev/null)
 expected=$(printf \"%016x%016x\" 0x${domu_base} 0x${domu_size})

if [[ \"\${expected}\" == \"\${current}\" ]]; then

Quoted 1 line
        echo \"${passed}\"

We need to check for ${domu_base} with or without leading zeroes:

Quoted 1 line

current=\$(hexdump -e '16/1 \"%02x\"' /proc/device-tree/memory@*(0)${domu_base}/reg 2>/dev/null)

This check is still bound to the way Xen exposes a memory node in a device tree which might change as Julien suggested. We need to have a check not relying on device-tree.
My proposal would be to use /proc/iomem which prints memory ranges in %08x format. This would look like as follows:

mem_range=$(printf \"%08x-%08x\" ${domu_base} $(( ${domu_base} + ${domu_size} - 1 ))) if grep -q \${mem_range} /proc/iomem; then

    echo ${passed}

fi

If you are ok with that, I will push a patch on Monday.

~Michal

#13 lore raw

PATCH v2 3/11 xen/arm: domain_build: Replace use of paddr_t in find_domU_holes()

In reply to Ayan Kumar Halder

bankbase, banksize and bankend are used to hold values of type 'unsigned long long'. This can be represented as 'uint64_t' instead of 'paddr_t'. This will ensure consistency with allocate_static_memory() (where we use 'uint64_t' for rambase and ramsize).

In future, paddr_t can be used for 'uin32_t' as well to represent 32bit physical addresses.

---

Changes from -

v1 - Modified the title/commit message from "[XEN v1 6/9] xen/arm: Use 'u64' to represent 'unsigned long long" and moved it earlier.

 xen/arch/arm/domain_build.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
xen/arch/arm/domain_build.c +3 -3
index 33a5945a2d..f904f12408 100644
@@ -1726,9 +1726,9 @@ static int __init find_domU_holes(const struct kernel_info *kinfo,
                                   struct meminfo *ext_regions)
 {
     unsigned int i;
-    paddr_t bankend;
-    const paddr_t bankbase[] = GUEST_RAM_BANK_BASES;
-    const paddr_t banksize[] = GUEST_RAM_BANK_SIZES;
+    uint64_t bankend;
+    const uint64_t bankbase[] = GUEST_RAM_BANK_BASES;
+    const uint64_t banksize[] = GUEST_RAM_BANK_SIZES;
     int res = -ENOENT;
 
     for ( i = 0; i < GUEST_RAM_BANKS; i++ )
Signature · 1 line
2.17.1
Acked-by on v2 3/11
#14 lore raw

Re: [XEN v2 03/11] xen/arm: domain_build: Replace use of paddr_t in find_domU_holes()

Quoted 9 lines · Ayan Kumar Halder

bankbase, banksize and bankend are used to hold values of type 'unsigned long long'. This can be represented as 'uint64_t' instead of 'paddr_t'. This will ensure consistency with allocate_static_memory() (where we use 'uint64_t' for rambase and ramsize).

In future, paddr_t can be used for 'uin32_t' as well to represent 32bit physical addresses.

I saw that Julien commented about using unsigned long long. To be honest, given that we typically use explicitly-sized integers (which is a good thing) and unsigned long long is always uint64_t on all architectures, I can see the benefits of using uint64_t here.

At the same time, I can see that the reason for change the type here is that we are dealing with ULL values, so it would make sense to use unsigned long long.

I cannot see any big problem/advantages either way, so I am OK with this patch. (Julien, if you prefer unsigned long long I am fine with that also.)

Quoted 30 lines

---

Changes from -

v1 - Modified the title/commit message from "[XEN v1 6/9] xen/arm: Use 'u64' to represent 'unsigned long long" and moved it earlier.

 xen/arch/arm/domain_build.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
xen/arch/arm/domain_build.c +3 -3
index 33a5945a2d..f904f12408 100644
@@ -1726,9 +1726,9 @@ static int __init find_domU_holes(const struct kernel_info *kinfo,
                                   struct meminfo *ext_regions)
 {
     unsigned int i;
-    paddr_t bankend;
-    const paddr_t bankbase[] = GUEST_RAM_BANK_BASES;
-    const paddr_t banksize[] = GUEST_RAM_BANK_SIZES;
+    uint64_t bankend;
+    const uint64_t bankbase[] = GUEST_RAM_BANK_BASES;
+    const uint64_t banksize[] = GUEST_RAM_BANK_SIZES;
     int res = -ENOENT;
 
     for ( i = 0; i < GUEST_RAM_BANKS; i++ )
Signature · 1 line
2.17.1
#15 lore raw

Re: [XEN v2 03/11] xen/arm: domain_build: Replace use of paddr_t in find_domU_holes()

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

bankbase, banksize and bankend are used to hold values of type 'unsigned long long'. This can be represented as 'uint64_t' instead of 'paddr_t'. This will ensure consistency with allocate_static_memory() (where we use 'uint64_t' for rambase and ramsize).

Quoted 8 lines
Quoted 4 lines

In future, paddr_t can be used for 'uin32_t' as well to represent 32bit physical addresses.

I saw that Julien commented about using unsigned long long. To be honest, given that we typically use explicitly-sized integers (which is a good thing)

From the CODING_STYLE:

"Fixed width types should only be used when a fixed width quantity is meant (which for example may be a value read from or to be written to a register)."

This is also how we used it in the Arm port so far.

Quoted 2 lines

and unsigned long long is always uint64_t on all architectures, I can see the benefits of using uint64_t here.

From my understanding of the C spec, the only requirement is that "unsigned long long" can fit a 2^64 - 1. So it may technically be bigger than 64-bit.

Quoted 7 lines

At the same time, I can see that the reason for change the type here is that we are dealing with ULL values, so it would make sense to use unsigned long long. >
I cannot see any big problem/advantages either way, so I am OK with this patch. (Julien, if you prefer unsigned long long I am fine with that also.)

I don't mind too much here.

Quoted 31 lines
Quoted 26 lines

---

Changes from -

v1 - Modified the title/commit message from "[XEN v1 6/9] xen/arm: Use 'u64' to represent 'unsigned long long" and moved it earlier.

  xen/arch/arm/domain_build.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
xen/arch/arm/domain_build.c +3 -3
index 33a5945a2d..f904f12408 100644
@@ -1726,9 +1726,9 @@ static int __init find_domU_holes(const struct kernel_info *kinfo,
                                    struct meminfo *ext_regions)
  {
      unsigned int i;
-    paddr_t bankend;
-    const paddr_t bankbase[] = GUEST_RAM_BANK_BASES;
-    const paddr_t banksize[] = GUEST_RAM_BANK_SIZES;
+    uint64_t bankend;
+    const uint64_t bankbase[] = GUEST_RAM_BANK_BASES;
+    const uint64_t banksize[] = GUEST_RAM_BANK_SIZES;
      int res = -ENOENT;
 
Quoted 4 lines
Quoted 4 lines
xen/arch/arm/domain_build.c +0 -0
      for ( i = 0; i < GUEST_RAM_BANKS; i++ )
Signature · 1 line
2.17.1

Cheers,

Signature · 1 line
Julien Grall
#16 lore raw

Re: [XEN v2 03/11] xen/arm: domain_build: Replace use of paddr_t in find_domU_holes()

Quoted 33 lines · Julien Grall

Hi,

Quoted 6 lines · Stefano Stabellini
Quoted 5 lines · Ayan Kumar Halder

bankbase, banksize and bankend are used to hold values of type 'unsigned long long'. This can be represented as 'uint64_t' instead of 'paddr_t'. This will ensure consistency with allocate_static_memory() (where we use 'uint64_t' for rambase and ramsize).

Quoted 8 lines
Quoted 4 lines

In future, paddr_t can be used for 'uin32_t' as well to represent 32bit physical addresses.

I saw that Julien commented about using unsigned long long. To be honest, given that we typically use explicitly-sized integers (which is a good thing)

From the CODING_STYLE:

"Fixed width types should only be used when a fixed width quantity is meant (which for example may be a value read from or to be written to a register)."

This is also how we used it in the Arm port so far.

Quoted 2 lines

and unsigned long long is always uint64_t on all architectures, I can see the benefits of using uint64_t here.

From my understanding of the C spec, the only requirement is that
"unsigned long long" can fit a 2^64 - 1. So it may technically be bigger than 64-bit.

Quoted 1 line
Quoted 11 lines
Quoted 6 lines

At the same time, I can see that the reason for change the type here is that we are dealing with ULL values, so it would make sense to use unsigned long long. >
I cannot see any big problem/advantages either way, so I am OK with this patch. (Julien, if you prefer unsigned long long I am fine with that also.)

I don't mind too much here.

Quoted 2 lines

Committed.

Cheers,

Signature · 1 line
Julien Grall
#17 lore raw

PATCH v2 4/11 xen/arm: Typecast the DT values into paddr_t

In reply to Ayan Kumar Halder

In future, we wish to support 32 bit physical address. However, one can only read u64 values from the DT. Thus, we need to typecast the values appropriately from u64 to paddr_t.

device_tree_get_reg() should now be able to return paddr_t. This is invoked by various callers to get DT address and size. Similarly, dt_read_number() is invoked as well to get DT address and size. The return value is typecasted to paddr_t. fdt_get_mem_rsv() can only accept u64 values. So, we provide a warpper for this called fdt_get_mem_rsv_paddr() which will do the necessary typecasting before invoking fdt_get_mem_rsv() and while returning.

---

Changes from

v1 - 1. Dropped "[XEN v1 2/9] xen/arm: Define translate_dt_address_size() for the translation between u64 and paddr_t" and "[XEN v1 4/9] xen/arm: Use translate_dt_address_size() to translate between device tree addr/size and paddr_t", instead this approach achieves the same purpose.

2. No need to check for truncation while converting values from u64 to paddr_t.

 xen/arch/arm/bootfdt.c                 | 23 +++++++++------
 xen/arch/arm/domain_build.c            |  2 +-
 xen/arch/arm/include/asm/device_tree.h | 40 ++++++++++++++++++++++++++
 xen/arch/arm/include/asm/setup.h       |  2 +-
 xen/arch/arm/setup.c                   | 14 ++++-----
 xen/arch/arm/smpboot.c                 |  2 +-
 6 files changed, 65 insertions(+), 18 deletions(-)
 create mode 100644 xen/arch/arm/include/asm/device_tree.h
xen/arch/arm/bootfdt.c +15 -8
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
 #include <xen/efi.h>
 #include <xen/device_tree.h>
 #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/sort.h>
 #include <xsm/xsm.h>
+#include <asm/device_tree.h>
 #include <asm/setup.h>
 
 static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init device_tree_node_compatible(const void *fdt, int node,
 }
 
 void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 *size)
+                                u32 size_cells, paddr_t *start, paddr_t *size)
 {
-    *start = dt_next_cell(address_cells, cell);
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell);
+    *size = (paddr_t) dt_next_cell(size_cells, cell);
 }
 
 static int __init device_tree_get_meminfo(const void *fdt, int node,
@@ -326,7 +331,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-start property has invalid length %d\n", len);
         return -EINVAL;
     }
-    start = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    start = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     prop = fdt_get_property(fdt, node, "linux,initrd-end", &len);
     if ( !prop )
@@ -339,7 +344,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-end property has invalid length %d\n", len);
         return -EINVAL;
     }
-    end = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    end = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     if ( start >= end )
     {
@@ -594,9 +599,11 @@ static void __init early_print_info(void)
     for ( i = 0; i < nr_rsvd; i++ )
     {
         paddr_t s, e;
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &s, &e) < 0 )
+
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &s, &e) < 0 )
             continue;
-        /* fdt_get_mem_rsv returns length */
+
+        /* fdt_get_mem_rsv_paddr returns length */
         e += s;
         printk(" RESVD[%u]: %"PRIpaddr" - %"PRIpaddr"\n", i, s, e);
     }
xen/arch/arm/domain_build.c +1 -1
index f904f12408..72b9afbb4c 100644
@@ -949,7 +949,7 @@ static int __init process_shm(struct domain *d, struct kernel_info *kinfo,
         BUG_ON(!prop);
         cells = (const __be32 *)prop->value;
         device_tree_get_reg(&cells, addr_cells, addr_cells, &pbase, &gbase);
-        psize = dt_read_number(cells, size_cells);
+        psize = (paddr_t) dt_read_number(cells, size_cells);
         if ( !IS_ALIGNED(pbase, PAGE_SIZE) || !IS_ALIGNED(gbase, PAGE_SIZE) )
         {
             printk("%pd: physical address 0x%"PRIpaddr", or guest address 0x%"PRIpaddr" is not suitably aligned.\n",
xen/arch/arm/include/asm/device_tree.h new file +40 -0
new file mode 100644
index 0000000000..51e0f0ae20
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * xen/arch/arm/include/asm/device_tree.h
+ * 
+ * Wrapper functions for device tree. This helps to convert dt values
+ * between u64 and paddr_t.
+ *
+ * Copyright (C) 2023, Advanced Micro Devices, Inc. All Rights Reserved.
+ */
+
+#ifndef __ARCH_ARM_DEVICE_TREE__
+#define __ARCH_ARM_DEVICE_TREE__
+
+#include <xen/libfdt/libfdt.h>
+
+inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
+                                 paddr_t *address,
+                                 paddr_t *size)
+{
+    uint64_t dt_addr;
+    uint64_t dt_size;
+    int ret = 0;
+
+    ret = fdt_get_mem_rsv(fdt, n, &dt_addr, &dt_size);
+
+    *address = (paddr_t) dt_addr;
+    *size = (paddr_t) dt_size;
+
+    return ret;
+}
+
+#endif /* __ARCH_ARM_DEVICE_TREE__ */
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
xen/arch/arm/include/asm/setup.h +1 -1
index a926f30a2b..6105e5cae3 100644
@@ -158,7 +158,7 @@ extern uint32_t hyp_traps_vector[];
 void init_traps(void);
 
 void device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                         u32 size_cells, u64 *start, u64 *size);
+                         u32 size_cells, paddr_t *start, paddr_t *size);
 
 u32 device_tree_get_u32(const void *fdt, int node,
                         const char *prop_name, u32 dflt);
xen/arch/arm/setup.c +7 -7
index 1f26f67b90..da13439e62 100644
@@ -29,7 +29,6 @@
 #include <xen/virtual_region.h>
 #include <xen/vmap.h>
 #include <xen/trace.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/acpi.h>
 #include <xen/warning.h>
 #include <asm/alternative.h>
@@ -39,6 +38,7 @@
 #include <asm/gic.h>
 #include <asm/cpuerrata.h>
 #include <asm/cpufeature.h>
+#include <asm/device_tree.h>
 #include <asm/platform.h>
 #include <asm/procinfo.h>
 #include <asm/setup.h>
@@ -222,11 +222,11 @@ static void __init dt_unreserved_regions(paddr_t s, paddr_t e,
     {
         paddr_t r_s, r_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &r_s, &r_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &r_s, &r_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        r_e += r_s; /* fdt_get_mem_rsv returns length */
+        r_e += r_s; /* fdt_get_mem_rsv_paddr returns length */
 
         if ( s < r_e && r_s < e )
         {
@@ -502,13 +502,13 @@ static paddr_t __init consider_modules(paddr_t s, paddr_t e,
     {
         paddr_t mod_s, mod_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened,
-                             i - mi->nr_mods,
-                             &mod_s, &mod_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened,
+                                   i - mi->nr_mods,
+                                   &mod_s, &mod_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        /* fdt_get_mem_rsv returns length */
+        /* fdt_get_mem_rsv_paddr returns length */
         mod_e += mod_s;
 
         if ( s < mod_e && mod_s < e )
xen/arch/arm/smpboot.c +1 -1
index 412ae22869..ee59b1d379 100644
@@ -159,7 +159,7 @@ static void __init dt_smp_init_cpus(void)
             continue;
         }
 
-        addr = dt_read_number(prop, dt_n_addr_cells(cpu));
+        addr = (paddr_t) dt_read_number(prop, dt_n_addr_cells(cpu));
 
         hwid = addr;
         if ( hwid != addr )
Signature · 1 line
2.17.1
#18 lore raw

Re: [XEN v2 04/11] xen/arm: Typecast the DT values into paddr_t

Quoted 12 lines · Ayan Kumar Halder

In future, we wish to support 32 bit physical address. However, one can only read u64 values from the DT. Thus, we need to typecast the values appropriately from u64 to paddr_t.

device_tree_get_reg() should now be able to return paddr_t. This is invoked by various callers to get DT address and size. Similarly, dt_read_number() is invoked as well to get DT address and size. The return value is typecasted to paddr_t. fdt_get_mem_rsv() can only accept u64 values. So, we provide a warpper for this called fdt_get_mem_rsv_paddr() which will do the necessary typecasting before invoking fdt_get_mem_rsv() and while returning.

Quoted 1 line

I know we discussed this before and you implemented exactly what we suggested, but now looking at this patch I think we should do the following:

- also add a wrapper for dt_read_number, something like dt_read_number_paddr that returns paddr
- add a check for the top 32-bit being zero in all the wrappers (dt_read_number_paddr, device_tree_get_reg, fdt_get_mem_rsv_paddr) when paddr!=uint64_t. In case the top 32-bit are != zero I think we should print an error

Julien, I remember you were concerned about BUG_ON/panic/ASSERT and I agree with you there (especially considering Vikram's device tree overlay series). So here I am only suggesting to check truncation and printk a message, not panic. Would you be OK with that?

Last comment, maybe we could add fdt_get_mem_rsv_paddr to setup.h instead of introducing xen/arch/arm/include/asm/device_tree.h, given that we already have device tree definitions there (device_tree_get_reg). I am OK either way.

Quoted 225 lines

---

Changes from

v1 - 1. Dropped "[XEN v1 2/9] xen/arm: Define translate_dt_address_size() for the translation between u64 and paddr_t" and "[XEN v1 4/9] xen/arm: Use translate_dt_address_size() to translate between device tree addr/size and paddr_t", instead this approach achieves the same purpose.

2. No need to check for truncation while converting values from u64 to paddr_t.

 xen/arch/arm/bootfdt.c                 | 23 +++++++++------
 xen/arch/arm/domain_build.c            |  2 +-
 xen/arch/arm/include/asm/device_tree.h | 40 ++++++++++++++++++++++++++
 xen/arch/arm/include/asm/setup.h       |  2 +-
 xen/arch/arm/setup.c                   | 14 ++++-----
 xen/arch/arm/smpboot.c                 |  2 +-
 6 files changed, 65 insertions(+), 18 deletions(-)
 create mode 100644 xen/arch/arm/include/asm/device_tree.h
xen/arch/arm/bootfdt.c +15 -8
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
 #include <xen/efi.h>
 #include <xen/device_tree.h>
 #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/sort.h>
 #include <xsm/xsm.h>
+#include <asm/device_tree.h>
 #include <asm/setup.h>
 
 static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init device_tree_node_compatible(const void *fdt, int node,
 }
 
 void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 *size)
+                                u32 size_cells, paddr_t *start, paddr_t *size)
 {
-    *start = dt_next_cell(address_cells, cell);
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell);
+    *size = (paddr_t) dt_next_cell(size_cells, cell);
 }
 
 static int __init device_tree_get_meminfo(const void *fdt, int node,
@@ -326,7 +331,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-start property has invalid length %d\n", len);
         return -EINVAL;
     }
-    start = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    start = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     prop = fdt_get_property(fdt, node, "linux,initrd-end", &len);
     if ( !prop )
@@ -339,7 +344,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-end property has invalid length %d\n", len);
         return -EINVAL;
     }
-    end = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    end = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     if ( start >= end )
     {
@@ -594,9 +599,11 @@ static void __init early_print_info(void)
     for ( i = 0; i < nr_rsvd; i++ )
     {
         paddr_t s, e;
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &s, &e) < 0 )
+
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &s, &e) < 0 )
             continue;
-        /* fdt_get_mem_rsv returns length */
+
+        /* fdt_get_mem_rsv_paddr returns length */
         e += s;
         printk(" RESVD[%u]: %"PRIpaddr" - %"PRIpaddr"\n", i, s, e);
     }
xen/arch/arm/domain_build.c +1 -1
index f904f12408..72b9afbb4c 100644
@@ -949,7 +949,7 @@ static int __init process_shm(struct domain *d, struct kernel_info *kinfo,
         BUG_ON(!prop);
         cells = (const __be32 *)prop->value;
         device_tree_get_reg(&cells, addr_cells, addr_cells, &pbase, &gbase);
-        psize = dt_read_number(cells, size_cells);
+        psize = (paddr_t) dt_read_number(cells, size_cells);
         if ( !IS_ALIGNED(pbase, PAGE_SIZE) || !IS_ALIGNED(gbase, PAGE_SIZE) )
         {
             printk("%pd: physical address 0x%"PRIpaddr", or guest address 0x%"PRIpaddr" is not suitably aligned.\n",
xen/arch/arm/include/asm/device_tree.h new file +40 -0
new file mode 100644
index 0000000000..51e0f0ae20
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * xen/arch/arm/include/asm/device_tree.h
+ * 
+ * Wrapper functions for device tree. This helps to convert dt values
+ * between u64 and paddr_t.
+ *
+ * Copyright (C) 2023, Advanced Micro Devices, Inc. All Rights Reserved.
+ */
+
+#ifndef __ARCH_ARM_DEVICE_TREE__
+#define __ARCH_ARM_DEVICE_TREE__
+
+#include <xen/libfdt/libfdt.h>
+
+inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
+                                 paddr_t *address,
+                                 paddr_t *size)
+{
+    uint64_t dt_addr;
+    uint64_t dt_size;
+    int ret = 0;
+
+    ret = fdt_get_mem_rsv(fdt, n, &dt_addr, &dt_size);
+
+    *address = (paddr_t) dt_addr;
+    *size = (paddr_t) dt_size;
+
+    return ret;
+}
+
+#endif /* __ARCH_ARM_DEVICE_TREE__ */
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
xen/arch/arm/include/asm/setup.h +1 -1
index a926f30a2b..6105e5cae3 100644
@@ -158,7 +158,7 @@ extern uint32_t hyp_traps_vector[];
 void init_traps(void);
 
 void device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                         u32 size_cells, u64 *start, u64 *size);
+                         u32 size_cells, paddr_t *start, paddr_t *size);
 
 u32 device_tree_get_u32(const void *fdt, int node,
                         const char *prop_name, u32 dflt);
xen/arch/arm/setup.c +7 -7
index 1f26f67b90..da13439e62 100644
@@ -29,7 +29,6 @@
 #include <xen/virtual_region.h>
 #include <xen/vmap.h>
 #include <xen/trace.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/acpi.h>
 #include <xen/warning.h>
 #include <asm/alternative.h>
@@ -39,6 +38,7 @@
 #include <asm/gic.h>
 #include <asm/cpuerrata.h>
 #include <asm/cpufeature.h>
+#include <asm/device_tree.h>
 #include <asm/platform.h>
 #include <asm/procinfo.h>
 #include <asm/setup.h>
@@ -222,11 +222,11 @@ static void __init dt_unreserved_regions(paddr_t s, paddr_t e,
     {
         paddr_t r_s, r_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &r_s, &r_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &r_s, &r_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        r_e += r_s; /* fdt_get_mem_rsv returns length */
+        r_e += r_s; /* fdt_get_mem_rsv_paddr returns length */
 
         if ( s < r_e && r_s < e )
         {
@@ -502,13 +502,13 @@ static paddr_t __init consider_modules(paddr_t s, paddr_t e,
     {
         paddr_t mod_s, mod_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened,
-                             i - mi->nr_mods,
-                             &mod_s, &mod_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened,
+                                   i - mi->nr_mods,
+                                   &mod_s, &mod_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        /* fdt_get_mem_rsv returns length */
+        /* fdt_get_mem_rsv_paddr returns length */
         mod_e += mod_s;
 
         if ( s < mod_e && mod_s < e )
xen/arch/arm/smpboot.c +1 -1
index 412ae22869..ee59b1d379 100644
@@ -159,7 +159,7 @@ static void __init dt_smp_init_cpus(void)
             continue;
         }
 
-        addr = dt_read_number(prop, dt_n_addr_cells(cpu));
+        addr = (paddr_t) dt_read_number(prop, dt_n_addr_cells(cpu));
 
         hwid = addr;
         if ( hwid != addr )
Signature · 1 line
2.17.1
#19 lore raw

Re: [XEN v2 04/11] xen/arm: Typecast the DT values into paddr_t

Quoted 31 lines · Stefano Stabellini
Quoted 12 lines · Ayan Kumar Halder

In future, we wish to support 32 bit physical address. However, one can only read u64 values from the DT. Thus, we need to typecast the values appropriately from u64 to paddr_t.

device_tree_get_reg() should now be able to return paddr_t. This is invoked by various callers to get DT address and size. Similarly, dt_read_number() is invoked as well to get DT address and size. The return value is typecasted to paddr_t. fdt_get_mem_rsv() can only accept u64 values. So, we provide a warpper for this called fdt_get_mem_rsv_paddr() which will do the necessary typecasting before invoking fdt_get_mem_rsv() and while returning.

Quoted 1 line

I know we discussed this before and you implemented exactly what we suggested, but now looking at this patch I think we should do the following:

- also add a wrapper for dt_read_number, something like dt_read_number_paddr that returns paddr
- add a check for the top 32-bit being zero in all the wrappers (dt_read_number_paddr, device_tree_get_reg, fdt_get_mem_rsv_paddr) when paddr!=uint64_t. In case the top 32-bit are != zero I think we should print an error

Julien, I remember you were concerned about BUG_ON/panic/ASSERT and I agree with you there (especially considering Vikram's device tree overlay series). So here I am only suggesting to check truncation and printk a message, not panic. Would you be OK with that?

Quoted 4 lines

Last comment, maybe we could add fdt_get_mem_rsv_paddr to setup.h instead of introducing xen/arch/arm/include/asm/device_tree.h, given that we already have device tree definitions there (device_tree_get_reg). I am OK either way.

Actually I noticed you also add dt_device_get_paddr to xen/arch/arm/include/asm/device_tree.h. So it sounds like it is a good idea to introduce xen/arch/arm/include/asm/device_tree.h, and we could also move the declarations of device_tree_get_reg, device_tree_get_u32 there.

Quoted 226 lines
Quoted 225 lines

---

Changes from

v1 - 1. Dropped "[XEN v1 2/9] xen/arm: Define translate_dt_address_size() for the translation between u64 and paddr_t" and "[XEN v1 4/9] xen/arm: Use translate_dt_address_size() to translate between device tree addr/size and paddr_t", instead this approach achieves the same purpose.

2. No need to check for truncation while converting values from u64 to paddr_t.

 xen/arch/arm/bootfdt.c                 | 23 +++++++++------
 xen/arch/arm/domain_build.c            |  2 +-
 xen/arch/arm/include/asm/device_tree.h | 40 ++++++++++++++++++++++++++
 xen/arch/arm/include/asm/setup.h       |  2 +-
 xen/arch/arm/setup.c                   | 14 ++++-----
 xen/arch/arm/smpboot.c                 |  2 +-
 6 files changed, 65 insertions(+), 18 deletions(-)
 create mode 100644 xen/arch/arm/include/asm/device_tree.h
xen/arch/arm/bootfdt.c +15 -8
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
 #include <xen/efi.h>
 #include <xen/device_tree.h>
 #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/sort.h>
 #include <xsm/xsm.h>
+#include <asm/device_tree.h>
 #include <asm/setup.h>
 
 static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init device_tree_node_compatible(const void *fdt, int node,
 }
 
 void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 *size)
+                                u32 size_cells, paddr_t *start, paddr_t *size)
 {
-    *start = dt_next_cell(address_cells, cell);
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell);
+    *size = (paddr_t) dt_next_cell(size_cells, cell);
 }
 
 static int __init device_tree_get_meminfo(const void *fdt, int node,
@@ -326,7 +331,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-start property has invalid length %d\n", len);
         return -EINVAL;
     }
-    start = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    start = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     prop = fdt_get_property(fdt, node, "linux,initrd-end", &len);
     if ( !prop )
@@ -339,7 +344,7 @@ static int __init process_chosen_node(const void *fdt, int node,
         printk("linux,initrd-end property has invalid length %d\n", len);
         return -EINVAL;
     }
-    end = dt_read_number((void *)&prop->data, dt_size_to_cells(len));
+    end = (paddr_t) dt_read_number((void *)&prop->data, dt_size_to_cells(len));
 
     if ( start >= end )
     {
@@ -594,9 +599,11 @@ static void __init early_print_info(void)
     for ( i = 0; i < nr_rsvd; i++ )
     {
         paddr_t s, e;
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &s, &e) < 0 )
+
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &s, &e) < 0 )
             continue;
-        /* fdt_get_mem_rsv returns length */
+
+        /* fdt_get_mem_rsv_paddr returns length */
         e += s;
         printk(" RESVD[%u]: %"PRIpaddr" - %"PRIpaddr"\n", i, s, e);
     }
xen/arch/arm/domain_build.c +1 -1
index f904f12408..72b9afbb4c 100644
@@ -949,7 +949,7 @@ static int __init process_shm(struct domain *d, struct kernel_info *kinfo,
         BUG_ON(!prop);
         cells = (const __be32 *)prop->value;
         device_tree_get_reg(&cells, addr_cells, addr_cells, &pbase, &gbase);
-        psize = dt_read_number(cells, size_cells);
+        psize = (paddr_t) dt_read_number(cells, size_cells);
         if ( !IS_ALIGNED(pbase, PAGE_SIZE) || !IS_ALIGNED(gbase, PAGE_SIZE) )
         {
             printk("%pd: physical address 0x%"PRIpaddr", or guest address 0x%"PRIpaddr" is not suitably aligned.\n",
xen/arch/arm/include/asm/device_tree.h new file +40 -0
new file mode 100644
index 0000000000..51e0f0ae20
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * xen/arch/arm/include/asm/device_tree.h
+ * 
+ * Wrapper functions for device tree. This helps to convert dt values
+ * between u64 and paddr_t.
+ *
+ * Copyright (C) 2023, Advanced Micro Devices, Inc. All Rights Reserved.
+ */
+
+#ifndef __ARCH_ARM_DEVICE_TREE__
+#define __ARCH_ARM_DEVICE_TREE__
+
+#include <xen/libfdt/libfdt.h>
+
+inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
+                                 paddr_t *address,
+                                 paddr_t *size)
+{
+    uint64_t dt_addr;
+    uint64_t dt_size;
+    int ret = 0;
+
+    ret = fdt_get_mem_rsv(fdt, n, &dt_addr, &dt_size);
+
+    *address = (paddr_t) dt_addr;
+    *size = (paddr_t) dt_size;
+
+    return ret;
+}
+
+#endif /* __ARCH_ARM_DEVICE_TREE__ */
+/*
+ * Local variables:
+ * mode: C
+ * c-file-style: "BSD"
+ * c-basic-offset: 4
+ * indent-tabs-mode: nil
+ * End:
+ */
xen/arch/arm/include/asm/setup.h +1 -1
index a926f30a2b..6105e5cae3 100644
@@ -158,7 +158,7 @@ extern uint32_t hyp_traps_vector[];
 void init_traps(void);
 
 void device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                         u32 size_cells, u64 *start, u64 *size);
+                         u32 size_cells, paddr_t *start, paddr_t *size);
 
 u32 device_tree_get_u32(const void *fdt, int node,
                         const char *prop_name, u32 dflt);
xen/arch/arm/setup.c +7 -7
index 1f26f67b90..da13439e62 100644
@@ -29,7 +29,6 @@
 #include <xen/virtual_region.h>
 #include <xen/vmap.h>
 #include <xen/trace.h>
-#include <xen/libfdt/libfdt.h>
 #include <xen/acpi.h>
 #include <xen/warning.h>
 #include <asm/alternative.h>
@@ -39,6 +38,7 @@
 #include <asm/gic.h>
 #include <asm/cpuerrata.h>
 #include <asm/cpufeature.h>
+#include <asm/device_tree.h>
 #include <asm/platform.h>
 #include <asm/procinfo.h>
 #include <asm/setup.h>
@@ -222,11 +222,11 @@ static void __init dt_unreserved_regions(paddr_t s, paddr_t e,
     {
         paddr_t r_s, r_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened, i, &r_s, &r_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened, i, &r_s, &r_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        r_e += r_s; /* fdt_get_mem_rsv returns length */
+        r_e += r_s; /* fdt_get_mem_rsv_paddr returns length */
 
         if ( s < r_e && r_s < e )
         {
@@ -502,13 +502,13 @@ static paddr_t __init consider_modules(paddr_t s, paddr_t e,
     {
         paddr_t mod_s, mod_e;
 
-        if ( fdt_get_mem_rsv(device_tree_flattened,
-                             i - mi->nr_mods,
-                             &mod_s, &mod_e ) < 0 )
+        if ( fdt_get_mem_rsv_paddr(device_tree_flattened,
+                                   i - mi->nr_mods,
+                                   &mod_s, &mod_e ) < 0 )
             /* If we can't read it, pretend it doesn't exist... */
             continue;
 
-        /* fdt_get_mem_rsv returns length */
+        /* fdt_get_mem_rsv_paddr returns length */
         mod_e += mod_s;
 
         if ( s < mod_e && mod_s < e )
xen/arch/arm/smpboot.c +1 -1
index 412ae22869..ee59b1d379 100644
@@ -159,7 +159,7 @@ static void __init dt_smp_init_cpus(void)
             continue;
         }
 
-        addr = dt_read_number(prop, dt_n_addr_cells(cpu));
+        addr = (paddr_t) dt_read_number(prop, dt_n_addr_cells(cpu));
 
         hwid = addr;
         if ( hwid != addr )
Signature · 1 line
2.17.1
#20 lore raw

Re: [XEN v2 04/11] xen/arm: Typecast the DT values into paddr_t

Hi,

I am answering to multiple e-mails at the same time.

Quoted 4 lines · Stefano Stabellini
Quoted 3 lines · Stefano Stabellini
Quoted 2 lines · Ayan Kumar Halder

In future, we wish to support 32 bit physical address. However, one can only read u64 values from the DT. Thus, we need to

A cell in the DT is a 32-bit value. So you could read 32-bit value (address-size could be #1). It is just that our wrapper return 64-bit values because this is how we use the most.

Quoted 1 line
Quoted 1 line
Quoted 1 line

typecast the values appropriately from u64 to paddr_t.

C will perfectly be able to truncate a 64-bit to 32-bit value. So this is not a very good argument to explain why all of this is necessary.

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

device_tree_get_reg() should now be able to return paddr_t. This is invoked by various callers to get DT address and size. Similarly, dt_read_number() is invoked as well to get DT address and size. The return value is typecasted to paddr_t. fdt_get_mem_rsv() can only accept u64 values. So, we provide a warpper

Typo: s/warpper/wrapper/

Quoted 11 lines
Quoted 11 lines
Quoted 4 lines

for this called fdt_get_mem_rsv_paddr() which will do the necessary typecasting before invoking fdt_get_mem_rsv() and while returning.

I know we discussed this before and you implemented exactly what we suggested, but now looking at this patch I think we should do the following:

- also add a wrapper for dt_read_number, something like dt_read_number_paddr that returns paddr

"number" and "paddr" pretty much means the same. I think it would be better to name it "dt_read_paddr".

Quoted 9 lines
Quoted 9 lines

- add a check for the top 32-bit being zero in all the wrappers (dt_read_number_paddr, device_tree_get_reg, fdt_get_mem_rsv_paddr) when paddr!=uint64_t. In case the top 32-bit are != zero I think we should print an error

Julien, I remember you were concerned about BUG_ON/panic/ASSERT and I agree with you there (especially considering Vikram's device tree overlay series). So here I am only suggesting to check truncation and printk a message, not panic. Would you be OK with that?

Aside dt_read_number(), I would expect that most of the helper can return an error. So if you want to check the truncation, then we should propagate the error.

Quoted 11 lines
Quoted 5 lines

Last comment, maybe we could add fdt_get_mem_rsv_paddr to setup.h instead of introducing xen/arch/arm/include/asm/device_tree.h, given that we already have device tree definitions there (device_tree_get_reg). I am OK either way.

Actually I noticed you also add dt_device_get_paddr to xen/arch/arm/include/asm/device_tree.h. So it sounds like it is a good idea to introduce xen/arch/arm/include/asm/device_tree.h, and we could also move the declarations of device_tree_get_reg, device_tree_get_u32 there.

None of the helpers you mention sounds arm specific. So why should the be move an arch specific headers?

[...]

Quoted 18 lines
Quoted 18 lines
Quoted 18 lines
xen/arch/arm/bootfdt.c +1 -1
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
  #include <xen/efi.h>
  #include <xen/device_tree.h>
  #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
  #include <xen/sort.h>
  #include <xsm/xsm.h>
+#include <asm/device_tree.h>
  #include <asm/setup.h>
 
  static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init device_tree_node_compatible(const void *fdt, int node,
  }
 
Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
xen/arch/arm/bootfdt.c +1 -2
  void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 *size)
+                                u32 size_cells, paddr_t *start, paddr_t *size)
  {
-    *start = dt_next_cell(address_cells, cell);
Quoted 8 lines

xen/arch/arm/bootfdt.c static bool __init device_tree_node_compatible(const void *fdt, int node, · line 64

Quoted 8 lines
xen/arch/arm/bootfdt.c +7 -1
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell);
+    *size = (paddr_t) dt_next_cell(size_cells, cell);

There is no need for explicit cast here.

Cheers,

Signature · 1 line
Julien Grall
#21 lore raw

Re: [XEN v2 04/11] xen/arm: Typecast the DT values into paddr_t

Quoted 1 line · Julien Grall

Hi,

Hi Julien/Stefano,

Quoted 10 lines

I am answering to multiple e-mails at the same time.

Quoted 4 lines · Stefano Stabellini
Quoted 3 lines · Stefano Stabellini
Quoted 2 lines · Ayan Kumar Halder

In future, we wish to support 32 bit physical address. However, one can only read u64 values from the DT. Thus, we need to

A cell in the DT is a 32-bit value. So you could read 32-bit value (address-size could be #1). It is just that our wrapper return 64-bit values because this is how we use the most.

ok

Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

typecast the values appropriately from u64 to paddr_t.

C will perfectly be able to truncate a 64-bit to 32-bit value. So this is not a very good argument to explain why all of this is necessary.

I can remove this line from the commit message. However, I have a related point below...

Quoted 10 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

device_tree_get_reg() should now be able to return paddr_t. This is invoked by various callers to get DT address and size. Similarly, dt_read_number() is invoked as well to get DT address and size. The return value is typecasted to paddr_t. fdt_get_mem_rsv() can only accept u64 values. So, we provide a warpper

Typo: s/warpper/wrapper/

ok

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines

for this called fdt_get_mem_rsv_paddr() which will do the necessary typecasting before invoking fdt_get_mem_rsv() and while returning.

Quoted 8 lines
Quoted 6 lines
Quoted 6 lines

I know we discussed this before and you implemented exactly what we suggested, but now looking at this patch I think we should do the following:

- also add a wrapper for dt_read_number, something like dt_read_number_paddr that returns paddr

"number" and "paddr" pretty much means the same. I think it would be better to name it "dt_read_paddr".

ok

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

- add a check for the top 32-bit being zero in all the wrappers (dt_read_number_paddr, device_tree_get_reg, fdt_get_mem_rsv_paddr) when paddr!=uint64_t. In case the top 32-bit are != zero I think we should print an error

Quoted 6 lines
Quoted 4 lines
Quoted 4 lines

Julien, I remember you were concerned about BUG_ON/panic/ASSERT and I agree with you there (especially considering Vikram's device tree overlay series). So here I am only suggesting to check truncation and printk a message, not panic. Would you be OK with that?

Aside dt_read_number(), I would expect that most of the helper can return an error. So if you want to check the truncation, then we should propagate the error.

ok

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Last comment, maybe we could add fdt_get_mem_rsv_paddr to setup.h instead of introducing xen/arch/arm/include/asm/device_tree.h, given that we already have device tree definitions there (device_tree_get_reg). I am OK either way.

Quoted 7 lines
Quoted 5 lines

  Actually I noticed you also add dt_device_get_paddr to xen/arch/arm/include/asm/device_tree.h. So it sounds like it is a good idea to introduce xen/arch/arm/include/asm/device_tree.h, and we could also move the declarations of device_tree_get_reg, device_tree_get_u32 there.

None of the helpers you mention sounds arm specific. So why should the be move an arch specific headers?

The functions (fdt_get_mem_rsv_paddr(), device_tree_get_reg(), device_tree_get_u32()) are currently used by Arm specific code only.

So, I thought of implementing fdt_get_mem_rsv_paddr() in xen/arch/arm/include/asm/device_tree.h.

However, I see your point as well. So the alternative approach would be :-

Approach 1) Implement fdt_get_mem_rsv_paddr() in ./xen/include/xen/libfdt/libfdt.h.

However libfdt is imported from external sources, so I am not sure if this is the  best approach.

Approach 2) Rename fdt_get_mem_rsv_paddr() to dt_get_mem_rsv_paddr() and implement it in ./xen/include/xen/device_tree.h.

However, this will be an odd one as it invokes fdt_get_mem_rsv() for which we will have to include libfdt.h in device_tree.h.

So, I am biased towards having xen/arch/arm/include/asm/device_tree.h in which we can implement all the non-static fdt and dt functions (which are required by xen/arch/arm/*).

And then (as Stefano suggested), we can move the definitions of the following to xen/arch/arm/include/asm/device_tree.h.

device_tree_get_reg()

device_tree_get_u32()

device_tree_for_each_node()

Please let me know your thoughts.

Quoted 32 lines

[...]

Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
xen/arch/arm/bootfdt.c +4 -4
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
   #include <xen/efi.h>
   #include <xen/device_tree.h>
   #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
   #include <xen/sort.h>
   #include <xsm/xsm.h>
+#include <asm/device_tree.h>
   #include <asm/setup.h>
     static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init 
 device_tree_node_compatible(const void *fdt, int node,
   }
     void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 
 *size)
+                                u32 size_cells, paddr_t *start, 
 paddr_t *size)
   {
-    *start = dt_next_cell(address_cells, cell);
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or 

u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for

Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell); +    *size = (paddr_t) dt_next_cell(size_cells, cell);

There is no need for explicit cast here.

Should we have it for the sake of documentation (that it casts u64 to paddr_t) ?

- Ayan

Quoted 3 lines

Cheers,

#22 lore raw

Re: [XEN v2 04/11] xen/arm: Typecast the DT values into paddr_t

Hi Ayan,

Quoted 15 lines · Ayan Kumar Halder
Quoted 11 lines · Julien Grall
Quoted 9 lines
Quoted 4 lines

Last comment, maybe we could add fdt_get_mem_rsv_paddr to setup.h instead of introducing xen/arch/arm/include/asm/device_tree.h, given that we already have device tree definitions there (device_tree_get_reg). I am OK either way.

  Actually I noticed you also add dt_device_get_paddr to xen/arch/arm/include/asm/device_tree.h. So it sounds like it is a good idea to introduce xen/arch/arm/include/asm/device_tree.h, and we could also move the declarations of device_tree_get_reg, device_tree_get_u32 there.

None of the helpers you mention sounds arm specific. So why should the be move an arch specific headers?

The functions (fdt_get_mem_rsv_paddr(), device_tree_get_reg(), device_tree_get_u32()) are currently used by Arm specific code only.

Quoted 7 lines

So, I thought of implementing fdt_get_mem_rsv_paddr() in xen/arch/arm/include/asm/device_tree.h.

However, I see your point as well. So the alternative approach would be :-

Approach 1) Implement fdt_get_mem_rsv_paddr() in ./xen/include/xen/libfdt/libfdt.h.

However libfdt is imported from external sources, so I am not sure if this is the  best approach.

One way would be to introduce "libfdt_xen.h" which would contain all the implementation specific to Xen.

Quoted 4 lines

Approach 2) Rename fdt_get_mem_rsv_paddr() to dt_get_mem_rsv_paddr() and implement it in ./xen/include/xen/device_tree.h.

However, this will be an odd one as it invokes fdt_get_mem_rsv() for which we will have to include libfdt.h in device_tree.h.

You could implement the functions in the device_tree.c but see below.

Quoted 11 lines

So, I am biased towards having xen/arch/arm/include/asm/device_tree.h in which we can implement all the non-static fdt and dt functions (which are required by xen/arch/arm/*).

And then (as Stefano suggested), we can move the definitions of the following to xen/arch/arm/include/asm/device_tree.h.

device_tree_get_reg()

device_tree_get_u32()

device_tree_for_each_node()

Well none of them are truly arch specific as well. I could imagine RISC-v to use it in the future if they also care about checking the truncation.

I have a slight preference to introduce libfdt_xen.h over adding the implementation in device_tree.h so we can keep the unflatten API (device_tree.h) separated from the flatten API (libfdt.h).

But this is not a strong preference between the two. That said, I would strongly argue against adding the helper in asm/*.h because there is nothing Arm specific in them.

Quoted 41 lines
Quoted 32 lines

[...]

Quoted 29 lines
Quoted 29 lines
Quoted 29 lines
xen/arch/arm/bootfdt.c +4 -4
index 0085c28d74..f536a3f3ab 100644
@@ -11,9 +11,9 @@
   #include <xen/efi.h>
   #include <xen/device_tree.h>
   #include <xen/lib.h>
-#include <xen/libfdt/libfdt.h>
   #include <xen/sort.h>
   #include <xsm/xsm.h>
+#include <asm/device_tree.h>
   #include <asm/setup.h>
     static bool __init device_tree_node_matches(const void *fdt, int node,
@@ -53,10 +53,15 @@ static bool __init 
 device_tree_node_compatible(const void *fdt, int node,
   }
     void __init device_tree_get_reg(const __be32 **cell, u32 address_cells,
-                                u32 size_cells, u64 *start, u64 
 *size)
+                                u32 size_cells, paddr_t *start, 
 paddr_t *size)
   {
-    *start = dt_next_cell(address_cells, cell);
-    *size = dt_next_cell(size_cells, cell);
+    /*
+     * dt_next_cell will return u64 whereas paddr_t may be u64 or 

u32. Thus, one
+     * needs to cast paddr_t to u32. Note that we do not check for

Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

truncation as
+     * it is user's responsibility to provide the correct values in the DT.
+     */
+    *start = (paddr_t) dt_next_cell(address_cells, cell); +    *size = (paddr_t) dt_next_cell(size_cells, cell);

There is no need for explicit cast here.

Quoted 1 line

Should we have it for the sake of documentation (that it casts u64 to paddr_t) ?

You already have a comment on top of dt_next_cell() to explain the typecast. So I would rather not add the explicit casts.

Cheers,

Signature · 1 line
Julien Grall
#23 lore raw

PATCH v2 5/11 xen/arm: Use paddr_t instead of u64 for address/size

In reply to Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

 xen/arch/arm/domain_build.c        |  9 +++++----
 xen/arch/arm/gic-v3.c              |  2 +-
 xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
 xen/drivers/char/exynos4210-uart.c |  2 +-
 xen/drivers/char/ns16550.c         |  8 ++++----
 xen/drivers/char/omap-uart.c       |  2 +-
 xen/drivers/char/pl011.c           |  4 ++--
 xen/drivers/char/scif-uart.c       |  2 +-
 xen/drivers/passthrough/arm/smmu.c |  6 +++---
 9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +5 -4
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
     dt_for_each_device_node( dt_host, np )
     {
         unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
 
         naddr = dt_number_of_address(np);
 
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain *d, struct dt_device_node *dev,
     unsigned int naddr;
     unsigned int i;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
     bool own_device = !dt_device_for_passthrough(dev);
     /*
      * We want to avoid mapping the MMIO in dom0 for the following cases:
@@ -2941,9 +2941,10 @@ static int __init handle_passthrough_prop(struct kernel_info *kinfo,
         if ( res )
         {
             printk(XENLOG_ERR "Unable to permit to dom%d access to"
-                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n",
                    kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1);
+                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));
             return res;
         }
 
xen/arch/arm/gic-v3.c +1 -1
index bb59ea94cd..391dfa53d7 100644
@@ -1393,7 +1393,7 @@ static void __init gicv3_dt_init(void)
 
     for ( i = 0; i < gicv3.rdist_count; i++ )
     {
-        uint64_t rdist_base, rdist_size;
+        paddr_t rdist_base, rdist_size;
 
         res = dt_device_get_address(node, 1 + i, &rdist_base, &rdist_size);
         if ( res )
xen/arch/arm/platforms/exynos5.c +13 -13
index 6560507092..f79fad9957 100644
@@ -42,8 +42,8 @@ static int exynos5_init_time(void)
     void __iomem *mct;
     int rc;
     struct dt_device_node *node;
-    u64 mct_base_addr;
-    u64 size;
+    paddr_t mct_base_addr;
+    paddr_t size;
 
     node = dt_find_compatible_node(NULL, NULL, "samsung,exynos4210-mct");
     if ( !node )
@@ -59,7 +59,7 @@ static int exynos5_init_time(void)
         return -ENXIO;
     }
 
-    dprintk(XENLOG_INFO, "mct_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_INFO, "mct_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
             mct_base_addr, size);
 
     mct = ioremap_nocache(mct_base_addr, size);
@@ -97,9 +97,9 @@ static int __init exynos5_smp_init(void)
     struct dt_device_node *node;
     void __iomem *sysram;
     char *compatible;
-    u64 sysram_addr;
-    u64 size;
-    u64 sysram_offset;
+    paddr_t sysram_addr;
+    paddr_t size;
+    paddr_t sysram_offset;
     int rc;
 
     node = dt_find_compatible_node(NULL, NULL, "samsung,secure-firmware");
@@ -131,7 +131,7 @@ static int __init exynos5_smp_init(void)
         dprintk(XENLOG_ERR, "Error in %s\n", compatible);
         return -ENXIO;
     }
-    dprintk(XENLOG_INFO, "sysram_addr: %016llx size: %016llx offset: %016llx\n",
+    dprintk(XENLOG_INFO,"sysram_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"offset: 0x%"PRIpaddr"\n",
             sysram_addr, size, sysram_offset);
 
     sysram = ioremap_nocache(sysram_addr, size);
@@ -189,7 +189,7 @@ static int exynos5_cpu_power_up(void __iomem *power, int cpu)
     return 0;
 }
 
-static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
+static int exynos5_get_pmu_baseandsize(paddr_t *power_base_addr, paddr_t *size)
 {
     struct dt_device_node *node;
     int rc;
@@ -215,7 +215,7 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
         return -ENXIO;
     }
 
-    dprintk(XENLOG_DEBUG, "power_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_DEBUG, "power_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
             *power_base_addr, *size);
 
     return 0;
@@ -223,8 +223,8 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
 
 static int exynos5_cpu_up(int cpu)
 {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
     void __iomem *power;
     int rc;
 
@@ -256,8 +256,8 @@ static int exynos5_cpu_up(int cpu)
 
 static void exynos5_reset(void)
 {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
     void __iomem *pmu;
     int rc;
 
xen/drivers/char/exynos4210-uart.c +1 -1
index 43aaf02e18..32cc8c78b5 100644
@@ -303,7 +303,7 @@ static int __init exynos4210_uart_init(struct dt_device_node *dev,
     const char *config = data;
     struct exynos4210_uart *uart;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/ns16550.c +4 -4
index 58d0ccd889..8ef895a2bb 100644
@@ -42,8 +42,8 @@
 
 static struct ns16550 {
     int baud, clock_hz, data_bits, parity, stop_bits, fifo_size, irq;
-    u64 io_base;   /* I/O port or memory-mapped I/O address. */
-    u64 io_size;
+    paddr_t io_base;   /* I/O port or memory-mapped I/O address. */
+    paddr_t io_size;
     int reg_shift; /* Bits to shift register offset by */
     int reg_width; /* Size of access to use, the registers
                     * themselves are still bytes */
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
 static int __init
 pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
 {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
     unsigned int b, d, f, nextf, i;
 
     /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                     else
                         size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                     (bar & PCI_BASE_ADDRESS_MEM_MASK);
                 }
                 /* IO based */
xen/drivers/char/omap-uart.c +1 -1
index d6a5d59aa2..3b53e1909a 100644
@@ -324,7 +324,7 @@ static int __init omap_uart_init(struct dt_device_node *dev,
     struct omap_uart *uart;
     u32 clkspec;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/pl011.c +2 -2
index be67242bc0..256ec11e3f 100644
@@ -222,7 +222,7 @@ static struct uart_driver __read_mostly pl011_driver = {
     .vuart_info   = pl011_vuart,
 };
 
-static int __init pl011_uart_init(int irq, u64 addr, u64 size, bool sbsa)
+static int __init pl011_uart_init(int irq, paddr_t addr, paddr_t size, bool sbsa)
 {
     struct pl011 *uart;
 
@@ -258,7 +258,7 @@ static int __init pl011_dt_uart_init(struct dt_device_node *dev,
 {
     const char *config = data;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
     {
xen/drivers/char/scif-uart.c +1 -1
index 2fccafe340..b425881d06 100644
@@ -311,7 +311,7 @@ static int __init scif_uart_init(struct dt_device_node *dev,
     const char *config = data;
     struct scif_uart *uart;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/passthrough/arm/smmu.c +3 -3
index 0a514821b3..490d253d44 100644
@@ -73,8 +73,8 @@
 /* Xen: Helpers to get device MMIO and IRQs */
 struct resource
 {
-	u64 addr;
-	u64 size;
+	paddr_t addr;
+	paddr_t size;
 	unsigned int type;
 };
 
@@ -169,7 +169,7 @@ static void __iomem *devm_ioremap_resource(struct device *dev,
 	ptr = ioremap_nocache(res->addr, res->size);
 	if (!ptr) {
 		dev_err(dev,
-			"ioremap failed (addr 0x%"PRIx64" size 0x%"PRIx64")\n",
+			"ioremap failed (addr 0x%"PRIpaddr" size 0x%"PRIpaddr")\n",
 			res->addr, res->size);
 		return ERR_PTR(-ENOMEM);
 	}
Signature · 1 line
2.17.1
#24 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 8 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

Quoted 7 lines

v1 - 1. Rebased the patch.

 xen/arch/arm/domain_build.c        |  9 +++++----
 xen/arch/arm/gic-v3.c              |  2 +-
 xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
 xen/drivers/char/exynos4210-uart.c |  2 +-
 xen/drivers/char/ns16550.c         |  8 ++++----

Please make sure you Cc all maintainers.

Quoted 10 lines
xen/drivers/char/ns16550.c +1 -1
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
 static int __init
 pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
 {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
     unsigned int b, d, f, nextf, i;
 
     /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx) · line 1264

Quoted 7 lines
xen/drivers/char/ns16550.c +1 -1
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                     else
                         size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                     (bar & PCI_BASE_ADDRESS_MEM_MASK);
                 }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You need to refuse acting on 64-bit BARs with the upper address bits non-zero.

If already you correct logic even in code not used on Arm (which I appreciate), then there's actually also related command line handling which needs adjustment. The use of simple_strtoul() to obtain ->io_base is bogus - this then needs to be simple_strtoull() (perhaps in a separate prereq patch), and in the 32-bit- paddr case you'd again need to check for truncation (in the patch here).

While doing the review I've noticed this

    uart->io_size = spcr->serial_port.bit_width;

in ns16550_acpi_uart_init(). This was introduced in 17b516196c55 ("ns16550:
add ACPI support for ARM only"), so Wei, Julien: Doesn't the right hand value need DIV_ROUND_UP(, 8) to convert from bit count to byte count?

Jan

#25 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Hi Jan,

Quoted 9 lines · Jan Beulich
Quoted 8 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

Quoted 8 lines
Quoted 7 lines

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++----
  xen/arch/arm/gic-v3.c              |  2 +-
  xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
  xen/drivers/char/exynos4210-uart.c |  2 +-
  xen/drivers/char/ns16550.c         |  8 ++++----

Please make sure you Cc all maintainers.

Ack.

Quoted 14 lines
Quoted 14 lines
xen/drivers/char/ns16550.c +1 -1
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
Quoted 6 lines

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx) · line 1264

Quoted 4 lines
xen/drivers/char/ns16550.c +1 -1
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
                  }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You need to refuse acting on 64-bit BARs with the upper address bits non-zero.

Yes, I was treating this like others (where Xen does not detect for truncation while getting the address/size from device-tree and typecasting it to paddr_t).

However in this case, as Xen is reading from PCI registers, so it needs to check for truncation.

I think the following change should do good.

@@ -1180,6 +1180,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx) unsigned int bar_idx = 0, port_idx = idx;
                 uint32_t bar, bar_64 = 0, len, len_64;
                 u64 size = 0;
+                uint64_t io_base = 0;
                 const struct ns16550_config_param *param = uart_param;

                 nextf = (f || (pci_conf_read16(PCI_SBDF(0, b, d, f),
@@ -1260,8 +1261,11 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx) else
                         size = len & PCI_BASE_ADDRESS_MEM_MASK;

-                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) | +                    io_base = ((u64)bar_64 << 32) |
                                     (bar & PCI_BASE_ADDRESS_MEM_MASK); +
+                    uart->io_base = (paddr_t) io_base;
+                    ASSERT(uart->io_base == io_base); /* Detect truncation */ }
                 /* IO based */
                 else if ( !param->mmio && (bar & PCI_BASE_ADDRESS_SPACE_IO) )

Quoted 6 lines

If already you correct logic even in code not used on Arm (which I appreciate), then there's actually also related command line handling which needs adjustment. The use of simple_strtoul() to obtain ->io_base is bogus - this then needs to be simple_strtoull() (perhaps in a separate prereq patch), and in the 32-bit- paddr case you'd again need to check for truncation (in the patch here).

Agreed this needs to be done in a separate prereq patch.

Quoted 8 lines

While doing the review I've noticed this

     uart->io_size = spcr->serial_port.bit_width;

in ns16550_acpi_uart_init(). This was introduced in 17b516196c55 ("ns16550:
add ACPI support for ARM only"), so Wei, Julien: Doesn't the right hand value need DIV_ROUND_UP(, 8) to convert from bit count to byte count?

Yes, I think it should be

uart->io_size = DIV_ROUND_UP(spcr->serial_port.bit_width, BITS_PER_BYTE);

However, Julien/Wei can confirm this.

- Ayan

Quoted 2 lines

Jan

#26 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 51 lines · Ayan Kumar Halder
Quoted 14 lines · Jan Beulich
Quoted 13 lines · Ayan Kumar Halder
xen/drivers/char/ns16550.c +1 -1
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
Quoted 6 lines

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx) · line 1264

Quoted 4 lines
xen/drivers/char/ns16550.c +1 -1
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
                  }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You need to refuse acting on 64-bit BARs with the upper address bits non-zero.

Yes, I was treating this like others (where Xen does not detect for truncation while getting the address/size from device-tree and typecasting it to paddr_t).

However in this case, as Xen is reading from PCI registers, so it needs to check for truncation.

I think the following change should do good.

(quoted diff) +1 -0
@@ -1180,6 +1180,7 @@ pci_uart_config(struct ns16550 *uart, bool_t 
 skip_amt, unsigned int idx)
                  unsigned int bar_idx = 0, port_idx = idx;
                  uint32_t bar, bar_64 = 0, len, len_64;
                  u64 size = 0;
+                uint64_t io_base = 0;
                  const struct ns16550_config_param *param = uart_param;
 

                 nextf = (f || (pci_conf_read16(PCI_SBDF(0, b, d, f),

(quoted diff) +4 -1
@@ -1260,8 +1261,11 @@ pci_uart_config(struct ns16550 *uart, bool_t 
 skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
+                    io_base = ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
+
+                    uart->io_base = (paddr_t) io_base;
+                    ASSERT(uart->io_base == io_base); /* Detect 
Quoted 5 lines

truncation */
                 }
                 /* IO based */
                 else if ( !param->mmio && (bar &
PCI_BASE_ADDRESS_SPACE_IO) )

An assertion can only ever check assumption on behavior elsewhere in Xen. Anything external needs handling properly, including in non-debug builds.

Jan

#27 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 52 lines · Jan Beulich
Quoted 3 lines · Ayan Kumar Halder
Quoted 2 lines · Jan Beulich

Review static const struct ns16550_config __initconst

Quoted 1 line · Ayan Kumar Halder
(quoted diff) +0 -0
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst

uart_config[] =

Quoted 2 lines
Quoted 2 lines

Review static const struct ns16550_config __initconst · line 1167

Quoted 2 lines
(quoted diff) +0 -0
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int

idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c static const struct ns16550_config __initconst uart_config[] =

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on

bus 0. */

Quoted 1 line
Quoted 1 line

Review pci_uart_config(struct ns16550 *uart, bool_t

Quoted 1 line
(quoted diff) +0 -0
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t

skip_amt, unsigned int idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar &

PCI_BASE_ADDRESS_MEM_MASK);

Quoted 2 lines
Quoted 2 lines

Review pci_uart_config(struct ns16550 *uart, bool_t · line 1264

Quoted 1 line
(quoted diff) +0 -0
                  }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You

need

Quoted 25 lines
Quoted 1 line

to refuse acting on 64-bit BARs with the upper address bits non-zero.

Yes, I was treating this like others (where Xen does not detect for truncation while getting the address/size from device-tree and typecasting it to paddr_t).

However in this case, as Xen is reading from PCI registers, so it needs to check for truncation.

I think the following change should do good.

(quoted diff) +1 -0
@@ -1180,6 +1180,7 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                  unsigned int bar_idx = 0, port_idx = idx;
                  uint32_t bar, bar_64 = 0, len, len_64;
                  u64 size = 0;
+                uint64_t io_base = 0;
                  const struct ns16550_config_param *param = uart_param;
 
                  nextf = (f || (pci_conf_read16(PCI_SBDF(0, b, d, f),
xen/drivers/char/ns16550.c +0 -0
@@ -1260,8 +1261,11 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
Quoted 15 lines
Quoted 11 lines
(quoted diff) +4 -1
-                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
+                    io_base = ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
+
+                    uart->io_base = (paddr_t) io_base;
+                    ASSERT(uart->io_base == io_base); /* Detect

truncation */

                  }
                  /* IO based */
                  else if ( !param->mmio && (bar &

PCI_BASE_ADDRESS_SPACE_IO) )

An assertion can only ever check assumption on behavior elsewhere in Xen. Anything external needs handling properly, including in non-debug builds.

Except in this case, it's detecting the result of the compiler cast just above it, isn't it? In which case it seems like it should be a BUILD_BUG_ON() of some sort.

-George

#28 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 70 lines · George Dunlap
Quoted 52 lines · Jan Beulich
Quoted 3 lines · Ayan Kumar Halder
Quoted 2 lines · Jan Beulich

Review static const struct ns16550_config __initconst

Quoted 1 line · Ayan Kumar Halder
(quoted diff) +0 -0
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst

uart_config[] =

Quoted 2 lines
Quoted 2 lines

Review static const struct ns16550_config __initconst · line 1167

Quoted 2 lines
(quoted diff) +0 -0
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int

idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c static const struct ns16550_config __initconst uart_config[] =

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on

bus 0. */

Quoted 1 line
Quoted 1 line

Review pci_uart_config(struct ns16550 *uart, bool_t

Quoted 1 line
(quoted diff) +0 -0
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t

skip_amt, unsigned int idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar &

PCI_BASE_ADDRESS_MEM_MASK);

Quoted 2 lines
Quoted 2 lines

Review pci_uart_config(struct ns16550 *uart, bool_t · line 1264

Quoted 1 line
(quoted diff) +0 -0
                  }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You

need

Quoted 25 lines
Quoted 1 line

to refuse acting on 64-bit BARs with the upper address bits non-zero.

Yes, I was treating this like others (where Xen does not detect for truncation while getting the address/size from device-tree and typecasting it to paddr_t).

However in this case, as Xen is reading from PCI registers, so it needs to check for truncation.

I think the following change should do good.

(quoted diff) +1 -0
@@ -1180,6 +1180,7 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                  unsigned int bar_idx = 0, port_idx = idx;
                  uint32_t bar, bar_64 = 0, len, len_64;
                  u64 size = 0;
+                uint64_t io_base = 0;
                  const struct ns16550_config_param *param = uart_param;
 
                  nextf = (f || (pci_conf_read16(PCI_SBDF(0, b, d, f),
xen/drivers/char/ns16550.c +0 -0
@@ -1260,8 +1261,11 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
Quoted 15 lines
Quoted 11 lines
(quoted diff) +4 -1
-                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
+                    io_base = ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
+
+                    uart->io_base = (paddr_t) io_base;
+                    ASSERT(uart->io_base == io_base); /* Detect

truncation */

                  }
                  /* IO based */
                  else if ( !param->mmio && (bar &

PCI_BASE_ADDRESS_SPACE_IO) )

An assertion can only ever check assumption on behavior elsewhere in Xen. Anything external needs handling properly, including in non-debug builds.

Quoted 2 lines

Except in this case, it's detecting the result of the compiler cast just above it, isn't it?

Not really, no - it checks whether there was truncation, but the absence (or presence) thereof is still a property of the underlying system, not one in Xen.

Quoted 1 line

In which case it seems like it should be a BUILD_BUG_ON() of some sort.

Such a check would then be to make sure paddr_t == uint64_t, which is precisely an equivalence Ayan wants to do away with.

Jan

#29 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 75 lines · Jan Beulich
Quoted 40 lines · George Dunlap
Quoted 38 lines · Jan Beulich
Quoted 3 lines · Ayan Kumar Halder
Quoted 2 lines · Jan Beulich

Review static const struct ns16550_config __initconst

Quoted 1 line · Ayan Kumar Halder
(quoted diff) +0 -0
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst

uart_config[] =

Quoted 2 lines
Quoted 2 lines

Review static const struct ns16550_config __initconst · line 1167

Quoted 2 lines
(quoted diff) +0 -0
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int

idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c static const struct ns16550_config __initconst uart_config[] =

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on

bus 0. */

Quoted 1 line
Quoted 1 line

Review pci_uart_config(struct ns16550 *uart, bool_t

Quoted 1 line
(quoted diff) +0 -0
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t

skip_amt, unsigned int idx)

Quoted 6 lines
Quoted 6 lines

xen/drivers/char/ns16550.c pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)

Quoted 6 lines
xen/drivers/char/ns16550.c +1 -1
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar &

PCI_BASE_ADDRESS_MEM_MASK);

Quoted 2 lines
Quoted 2 lines

Review pci_uart_config(struct ns16550 *uart, bool_t · line 1264

Quoted 1 line
(quoted diff) +0 -0
                  }

This looks wrong to me: You shouldn't blindly truncate to 32 bits. You

need

Quoted 11 lines
Quoted 1 line

to refuse acting on 64-bit BARs with the upper address bits non-zero.

Yes, I was treating this like others (where Xen does not detect for truncation while getting the address/size from device-tree and typecasting it to paddr_t).

However in this case, as Xen is reading from PCI registers, so it needs to check for truncation.

I think the following change should do good.

Quoted 7 lines
Quoted 7 lines

Review pci_uart_config(struct ns16550 *uart, bool_t · line 1185

Quoted 7 lines
(quoted diff) +1 -0
@@ -1180,6 +1180,7 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                  unsigned int bar_idx = 0, port_idx = idx;
                  uint32_t bar, bar_64 = 0, len, len_64;
                  u64 size = 0;
+                uint64_t io_base = 0;
                  const struct ns16550_config_param *param =

uart_param;

Quoted 10 lines
Quoted 10 lines

Review pci_uart_config(struct ns16550 *uart, bool_t · line 1266

Quoted 10 lines
(quoted diff) +1 -1
                  nextf = (f || (pci_conf_read16(PCI_SBDF(0, b, d, f),
@@ -1260,8 +1261,11 @@ pci_uart_config(struct ns16550 *uart, bool_t
 skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
+                    io_base = ((u64)bar_64 << 32) |
                                      (bar &

PCI_BASE_ADDRESS_MEM_MASK);

Quoted 10 lines
Quoted 10 lines
Quoted 8 lines
(quoted diff) +3 -0
+
+                    uart->io_base = (paddr_t) io_base;
+                    ASSERT(uart->io_base == io_base); /* Detect

truncation */

                  }
                  /* IO based */
                  else if ( !param->mmio && (bar &

PCI_BASE_ADDRESS_SPACE_IO) )

An assertion can only ever check assumption on behavior elsewhere in

Xen.

Quoted 1 line
Quoted 1 line

Anything external needs handling properly, including in non-debug

builds.

Quoted 2 lines
Quoted 1 line
Quoted 7 lines
Quoted 2 lines

Except in this case, it's detecting the result of the compiler cast just above it, isn't it?

Not really, no - it checks whether there was truncation, but the absence (or presence) thereof is still a property of the underlying system, not one in Xen.

Ah, gotcha. Ayan, it might be helpful to take a look at the 'Handling unexpected conditions' section of our CODING_STYLE [1] for a discussion of when (and when not) to use ASSERT().

-George

[1] https://github.com/xen-project/xen/blob/master/CODING_STYLE

#30 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

In reply to Ayan Kumar Halder
Quoted 44 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

 xen/arch/arm/domain_build.c        |  9 +++++----
 xen/arch/arm/gic-v3.c              |  2 +-
 xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
 xen/drivers/char/exynos4210-uart.c |  2 +-
 xen/drivers/char/ns16550.c         |  8 ++++----
 xen/drivers/char/omap-uart.c       |  2 +-
 xen/drivers/char/pl011.c           |  4 ++--
 xen/drivers/char/scif-uart.c       |  2 +-
 xen/drivers/passthrough/arm/smmu.c |  6 +++---
 9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +2 -2
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
     dt_for_each_device_node( dt_host, np )
     {
         unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
 
         naddr = dt_number_of_address(np);
 
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain *d, struct dt_device_node *dev,
     unsigned int naddr;
     unsigned int i;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
     bool own_device = !dt_device_for_passthrough(dev);
     /*
      * We want to avoid mapping the MMIO in dom0 for the following cases:

xen/arch/arm/domain_build.c static int __init handle_passthrough_prop(struct kernel_info *kinfo, · line 2947

Quoted 9 lines
xen/arch/arm/domain_build.c +3 -2
@@ -2941,9 +2941,10 @@ static int __init handle_passthrough_prop(struct kernel_info *kinfo,
         if ( res )
         {
             printk(XENLOG_ERR "Unable to permit to dom%d access to"
-                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n",
                    kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1);
+                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));

Why do you need the casts here? mstart is already defined as paddr_t

Quoted 223 lines
xen/arch/arm/domain_build.c +0 -0
             return res;
         }
 
xen/arch/arm/gic-v3.c +1 -1
index bb59ea94cd..391dfa53d7 100644
@@ -1393,7 +1393,7 @@ static void __init gicv3_dt_init(void)
 
     for ( i = 0; i < gicv3.rdist_count; i++ )
     {
-        uint64_t rdist_base, rdist_size;
+        paddr_t rdist_base, rdist_size;
 
         res = dt_device_get_address(node, 1 + i, &rdist_base, &rdist_size);
         if ( res )
xen/arch/arm/platforms/exynos5.c +13 -13
index 6560507092..f79fad9957 100644
@@ -42,8 +42,8 @@ static int exynos5_init_time(void)
     void __iomem *mct;
     int rc;
     struct dt_device_node *node;
-    u64 mct_base_addr;
-    u64 size;
+    paddr_t mct_base_addr;
+    paddr_t size;
 
     node = dt_find_compatible_node(NULL, NULL, "samsung,exynos4210-mct");
     if ( !node )
@@ -59,7 +59,7 @@ static int exynos5_init_time(void)
         return -ENXIO;
     }
 
-    dprintk(XENLOG_INFO, "mct_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_INFO, "mct_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
             mct_base_addr, size);
 
     mct = ioremap_nocache(mct_base_addr, size);
@@ -97,9 +97,9 @@ static int __init exynos5_smp_init(void)
     struct dt_device_node *node;
     void __iomem *sysram;
     char *compatible;
-    u64 sysram_addr;
-    u64 size;
-    u64 sysram_offset;
+    paddr_t sysram_addr;
+    paddr_t size;
+    paddr_t sysram_offset;
     int rc;
 
     node = dt_find_compatible_node(NULL, NULL, "samsung,secure-firmware");
@@ -131,7 +131,7 @@ static int __init exynos5_smp_init(void)
         dprintk(XENLOG_ERR, "Error in %s\n", compatible);
         return -ENXIO;
     }
-    dprintk(XENLOG_INFO, "sysram_addr: %016llx size: %016llx offset: %016llx\n",
+    dprintk(XENLOG_INFO,"sysram_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"offset: 0x%"PRIpaddr"\n",
             sysram_addr, size, sysram_offset);
 
     sysram = ioremap_nocache(sysram_addr, size);
@@ -189,7 +189,7 @@ static int exynos5_cpu_power_up(void __iomem *power, int cpu)
     return 0;
 }
 
-static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
+static int exynos5_get_pmu_baseandsize(paddr_t *power_base_addr, paddr_t *size)
 {
     struct dt_device_node *node;
     int rc;
@@ -215,7 +215,7 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
         return -ENXIO;
     }
 
-    dprintk(XENLOG_DEBUG, "power_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_DEBUG, "power_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
             *power_base_addr, *size);
 
     return 0;
@@ -223,8 +223,8 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
 
 static int exynos5_cpu_up(int cpu)
 {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
     void __iomem *power;
     int rc;
 
@@ -256,8 +256,8 @@ static int exynos5_cpu_up(int cpu)
 
 static void exynos5_reset(void)
 {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
     void __iomem *pmu;
     int rc;
 
xen/drivers/char/exynos4210-uart.c +1 -1
index 43aaf02e18..32cc8c78b5 100644
@@ -303,7 +303,7 @@ static int __init exynos4210_uart_init(struct dt_device_node *dev,
     const char *config = data;
     struct exynos4210_uart *uart;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/ns16550.c +4 -4
index 58d0ccd889..8ef895a2bb 100644
@@ -42,8 +42,8 @@
 
 static struct ns16550 {
     int baud, clock_hz, data_bits, parity, stop_bits, fifo_size, irq;
-    u64 io_base;   /* I/O port or memory-mapped I/O address. */
-    u64 io_size;
+    paddr_t io_base;   /* I/O port or memory-mapped I/O address. */
+    paddr_t io_size;
     int reg_shift; /* Bits to shift register offset by */
     int reg_width; /* Size of access to use, the registers
                     * themselves are still bytes */
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
 static int __init
 pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
 {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
     unsigned int b, d, f, nextf, i;
 
     /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                     else
                         size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                     (bar & PCI_BASE_ADDRESS_MEM_MASK);
                 }
                 /* IO based */
xen/drivers/char/omap-uart.c +1 -1
index d6a5d59aa2..3b53e1909a 100644
@@ -324,7 +324,7 @@ static int __init omap_uart_init(struct dt_device_node *dev,
     struct omap_uart *uart;
     u32 clkspec;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/pl011.c +2 -2
index be67242bc0..256ec11e3f 100644
@@ -222,7 +222,7 @@ static struct uart_driver __read_mostly pl011_driver = {
     .vuart_info   = pl011_vuart,
 };
 
-static int __init pl011_uart_init(int irq, u64 addr, u64 size, bool sbsa)
+static int __init pl011_uart_init(int irq, paddr_t addr, paddr_t size, bool sbsa)
 {
     struct pl011 *uart;
 
@@ -258,7 +258,7 @@ static int __init pl011_dt_uart_init(struct dt_device_node *dev,
 {
     const char *config = data;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
     {
xen/drivers/char/scif-uart.c +1 -1
index 2fccafe340..b425881d06 100644
@@ -311,7 +311,7 @@ static int __init scif_uart_init(struct dt_device_node *dev,
     const char *config = data;
     struct scif_uart *uart;
     int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
     if ( strcmp(config, "") )
         printk("WARNING: UART configuration is not supported\n");
xen/drivers/passthrough/arm/smmu.c +3 -3
index 0a514821b3..490d253d44 100644
@@ -73,8 +73,8 @@
 /* Xen: Helpers to get device MMIO and IRQs */
 struct resource
 {
-	u64 addr;
-	u64 size;
+	paddr_t addr;
+	paddr_t size;
 	unsigned int type;
 };
 
@@ -169,7 +169,7 @@ static void __iomem *devm_ioremap_resource(struct device *dev,
 	ptr = ioremap_nocache(res->addr, res->size);
 	if (!ptr) {
 		dev_err(dev,
-			"ioremap failed (addr 0x%"PRIx64" size 0x%"PRIx64")\n",
+			"ioremap failed (addr 0x%"PRIpaddr" size 0x%"PRIpaddr")\n",
 			res->addr, res->size);
 		return ERR_PTR(-ENOMEM);
 	}
Signature · 1 line
2.17.1
#31 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Hi Stefano,

Quoted 47 lines · Stefano Stabellini
Quoted 46 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++----
  xen/arch/arm/gic-v3.c              |  2 +-
  xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
  xen/drivers/char/exynos4210-uart.c |  2 +-
  xen/drivers/char/ns16550.c         |  8 ++++----
  xen/drivers/char/omap-uart.c       |  2 +-
  xen/drivers/char/pl011.c           |  4 ++--
  xen/drivers/char/scif-uart.c       |  2 +-
  xen/drivers/passthrough/arm/smmu.c |  6 +++---
  9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +2 -2
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
      dt_for_each_device_node( dt_host, np )
      {
          unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
 
          naddr = dt_number_of_address(np);
 
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain *d, struct dt_device_node *dev,
      unsigned int naddr;
      unsigned int i;
      int res;
-    u64 addr, size;
+    paddr_t addr, size;
      bool own_device = !dt_device_for_passthrough(dev);
      /*
       * We want to avoid mapping the MMIO in dom0 for the following cases:
@@ -2941,9 +2941,10 @@ static int __init handle_passthrough_prop(struct kernel_info *kinfo,
          if ( res )
          {
Quoted 8 lines

xen/arch/arm/domain_build.c static int __init handle_passthrough_prop(struct kernel_info *kinfo, · line 2947

Quoted 7 lines
xen/arch/arm/domain_build.c +3 -2
              printk(XENLOG_ERR "Unable to permit to dom%d access to"
-                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n",
                     kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1);
+                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));

Why do you need the casts here? mstart is already defined as paddr_t

Actually, this is because the PAGE_MASK is defined as 'long'. See xen/include/xen/page-size.h :-

#define PAGE_SIZE           (_AC(1,L) << PAGE_SHIFT) #define PAGE_MASK           (~(PAGE_SIZE-1))

So, the resultant type inferred is 'long unsigned int'. Thus, we need to add an explicit cast.

- Ayan

Quoted 225 lines
Quoted 225 lines
xen/arch/arm/domain_build.c +0 -0
              return res;
          }
 
xen/arch/arm/gic-v3.c +1 -1
index bb59ea94cd..391dfa53d7 100644
@@ -1393,7 +1393,7 @@ static void __init gicv3_dt_init(void)
 
      for ( i = 0; i < gicv3.rdist_count; i++ )
      {
-        uint64_t rdist_base, rdist_size;
+        paddr_t rdist_base, rdist_size;
 
          res = dt_device_get_address(node, 1 + i, &rdist_base, &rdist_size);
          if ( res )
xen/arch/arm/platforms/exynos5.c +13 -13
index 6560507092..f79fad9957 100644
@@ -42,8 +42,8 @@ static int exynos5_init_time(void)
      void __iomem *mct;
      int rc;
      struct dt_device_node *node;
-    u64 mct_base_addr;
-    u64 size;
+    paddr_t mct_base_addr;
+    paddr_t size;
 
      node = dt_find_compatible_node(NULL, NULL, "samsung,exynos4210-mct");
      if ( !node )
@@ -59,7 +59,7 @@ static int exynos5_init_time(void)
          return -ENXIO;
      }
 
-    dprintk(XENLOG_INFO, "mct_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_INFO, "mct_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
              mct_base_addr, size);
 
      mct = ioremap_nocache(mct_base_addr, size);
@@ -97,9 +97,9 @@ static int __init exynos5_smp_init(void)
      struct dt_device_node *node;
      void __iomem *sysram;
      char *compatible;
-    u64 sysram_addr;
-    u64 size;
-    u64 sysram_offset;
+    paddr_t sysram_addr;
+    paddr_t size;
+    paddr_t sysram_offset;
      int rc;
 
      node = dt_find_compatible_node(NULL, NULL, "samsung,secure-firmware");
@@ -131,7 +131,7 @@ static int __init exynos5_smp_init(void)
          dprintk(XENLOG_ERR, "Error in %s\n", compatible);
          return -ENXIO;
      }
-    dprintk(XENLOG_INFO, "sysram_addr: %016llx size: %016llx offset: %016llx\n",
+    dprintk(XENLOG_INFO,"sysram_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"offset: 0x%"PRIpaddr"\n",
              sysram_addr, size, sysram_offset);
 
      sysram = ioremap_nocache(sysram_addr, size);
@@ -189,7 +189,7 @@ static int exynos5_cpu_power_up(void __iomem *power, int cpu)
      return 0;
  }
 
-static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
+static int exynos5_get_pmu_baseandsize(paddr_t *power_base_addr, paddr_t *size)
  {
      struct dt_device_node *node;
      int rc;
@@ -215,7 +215,7 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
          return -ENXIO;
      }
 
-    dprintk(XENLOG_DEBUG, "power_base_addr: %016llx size: %016llx\n",
+    dprintk(XENLOG_DEBUG, "power_base_addr: 0x%"PRIpaddr" size: 0x%"PRIpaddr"\n",
              *power_base_addr, *size);
 
      return 0;
@@ -223,8 +223,8 @@ static int exynos5_get_pmu_baseandsize(u64 *power_base_addr, u64 *size)
 
  static int exynos5_cpu_up(int cpu)
  {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
      void __iomem *power;
      int rc;
 
@@ -256,8 +256,8 @@ static int exynos5_cpu_up(int cpu)
 
  static void exynos5_reset(void)
  {
-    u64 power_base_addr;
-    u64 size;
+    paddr_t power_base_addr;
+    paddr_t size;
      void __iomem *pmu;
      int rc;
 
xen/drivers/char/exynos4210-uart.c +1 -1
index 43aaf02e18..32cc8c78b5 100644
@@ -303,7 +303,7 @@ static int __init exynos4210_uart_init(struct dt_device_node *dev,
      const char *config = data;
      struct exynos4210_uart *uart;
      int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
      if ( strcmp(config, "") )
          printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/ns16550.c +4 -4
index 58d0ccd889..8ef895a2bb 100644
@@ -42,8 +42,8 @@
 
  static struct ns16550 {
      int baud, clock_hz, data_bits, parity, stop_bits, fifo_size, irq;
-    u64 io_base;   /* I/O port or memory-mapped I/O address. */
-    u64 io_size;
+    paddr_t io_base;   /* I/O port or memory-mapped I/O address. */
+    paddr_t io_size;
      int reg_shift; /* Bits to shift register offset by */
      int reg_width; /* Size of access to use, the registers
                      * themselves are still bytes */
@@ -1166,7 +1166,7 @@ static const struct ns16550_config __initconst uart_config[] =
  static int __init
  pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
  {
-    u64 orig_base = uart->io_base;
+    paddr_t orig_base = uart->io_base;
      unsigned int b, d, f, nextf, i;
 
      /* NB. Start at bus 1 to avoid AMT: a plug-in card cannot be on bus 0. */
@@ -1259,7 +1259,7 @@ pci_uart_config(struct ns16550 *uart, bool_t skip_amt, unsigned int idx)
                      else
                          size = len & PCI_BASE_ADDRESS_MEM_MASK;
 
-                    uart->io_base = ((u64)bar_64 << 32) |
+                    uart->io_base = (paddr_t) ((u64)bar_64 << 32) |
                                      (bar & PCI_BASE_ADDRESS_MEM_MASK);
                  }
                  /* IO based */
xen/drivers/char/omap-uart.c +1 -1
index d6a5d59aa2..3b53e1909a 100644
@@ -324,7 +324,7 @@ static int __init omap_uart_init(struct dt_device_node *dev,
      struct omap_uart *uart;
      u32 clkspec;
      int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
      if ( strcmp(config, "") )
          printk("WARNING: UART configuration is not supported\n");
xen/drivers/char/pl011.c +2 -2
index be67242bc0..256ec11e3f 100644
@@ -222,7 +222,7 @@ static struct uart_driver __read_mostly pl011_driver = {
      .vuart_info   = pl011_vuart,
  };
 
-static int __init pl011_uart_init(int irq, u64 addr, u64 size, bool sbsa)
+static int __init pl011_uart_init(int irq, paddr_t addr, paddr_t size, bool sbsa)
  {
      struct pl011 *uart;
 
@@ -258,7 +258,7 @@ static int __init pl011_dt_uart_init(struct dt_device_node *dev,
  {
      const char *config = data;
      int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
      if ( strcmp(config, "") )
      {
xen/drivers/char/scif-uart.c +1 -1
index 2fccafe340..b425881d06 100644
@@ -311,7 +311,7 @@ static int __init scif_uart_init(struct dt_device_node *dev,
      const char *config = data;
      struct scif_uart *uart;
      int res;
-    u64 addr, size;
+    paddr_t addr, size;
 
      if ( strcmp(config, "") )
          printk("WARNING: UART configuration is not supported\n");
xen/drivers/passthrough/arm/smmu.c +3 -3
index 0a514821b3..490d253d44 100644
@@ -73,8 +73,8 @@
  /* Xen: Helpers to get device MMIO and IRQs */
  struct resource
  {
-	u64 addr;
-	u64 size;
+	paddr_t addr;
+	paddr_t size;
  	unsigned int type;
  };
 
@@ -169,7 +169,7 @@ static void __iomem *devm_ioremap_resource(struct device *dev,
  	ptr = ioremap_nocache(res->addr, res->size);
  	if (!ptr) {
  		dev_err(dev,
-			"ioremap failed (addr 0x%"PRIx64" size 0x%"PRIx64")\n",
+			"ioremap failed (addr 0x%"PRIpaddr" size 0x%"PRIpaddr")\n",
  			res->addr, res->size);
  		return ERR_PTR(-ENOMEM);
  	}
Signature · 1 line
2.17.1
#32 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Quoted 60 lines · Ayan Kumar Halder

Hi Stefano,

Quoted 48 lines · Stefano Stabellini
Quoted 47 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++---- xen/arch/arm/gic-v3.c              |  2 +- xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++------------- xen/drivers/char/exynos4210-uart.c |  2 +- xen/drivers/char/ns16550.c         |  8 ++++---- xen/drivers/char/omap-uart.c       |  2 +- xen/drivers/char/pl011.c           |  4 ++-- xen/drivers/char/scif-uart.c       |  2 +- xen/drivers/passthrough/arm/smmu.c |  6 +++---

  9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +2 -2
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const 
 struct kernel_info *kinfo,
       dt_for_each_device_node( dt_host, np )
       {
           unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
           naddr = dt_number_of_address(np);
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain 
 *d, struct dt_device_node *dev,
       unsigned int naddr;
       unsigned int i;
       int res;
-    u64 addr, size;
+    paddr_t addr, size;
       bool own_device = !dt_device_for_passthrough(dev);
       /*

       * We want to avoid mapping the MMIO in dom0 for the following cases:

xen/arch/arm/domain_build.c +0 -0
@@ -2941,9 +2941,10 @@ static int __init 
 handle_passthrough_prop(struct kernel_info *kinfo,
           if ( res )
           {
Quoted 8 lines
Quoted 7 lines

              printk(XENLOG_ERR "Unable to permit to dom%d access to" -                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n", kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1); +                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));

Why do you need the casts here? mstart is already defined as paddr_t

Quoted 6 lines

Actually, this is because the PAGE_MASK is defined as 'long'. See xen/include/xen/page-size.h :-

#define PAGE_SIZE           (_AC(1,L) << PAGE_SHIFT) #define PAGE_MASK           (~(PAGE_SIZE-1))

So, the resultant type inferred is 'long unsigned int'. Thus, we need to add an explicit cast.

Hmmm... I am a bit confused with this statement. If the resultant type inferred is actually 'long unsigned int', then why was the current specifier worked before ('long unsigned int' is 32-bit on Arm32)?

Cheers,

Signature · 1 line
Julien Grall
#33 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Hi Julien,

Quoted 64 lines · Julien Grall
Quoted 61 lines · Ayan Kumar Halder

Hi Stefano,

Quoted 57 lines · Stefano Stabellini
Quoted 47 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++---- xen/arch/arm/gic-v3.c              |  2 +- xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++------------- xen/drivers/char/exynos4210-uart.c |  2 +- xen/drivers/char/ns16550.c         |  8 ++++---- xen/drivers/char/omap-uart.c       |  2 +- xen/drivers/char/pl011.c           |  4 ++-- xen/drivers/char/scif-uart.c       |  2 +- xen/drivers/passthrough/arm/smmu.c |  6 +++---

  9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +2 -2
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const 
 struct kernel_info *kinfo,
       dt_for_each_device_node( dt_host, np )
       {
           unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
           naddr = dt_number_of_address(np);
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain 
 *d, struct dt_device_node *dev,
       unsigned int naddr;
       unsigned int i;
       int res;
-    u64 addr, size;
+    paddr_t addr, size;
       bool own_device = !dt_device_for_passthrough(dev);
       /*

       * We want to avoid mapping the MMIO in dom0 for the following cases:

xen/arch/arm/domain_build.c +0 -0
@@ -2941,9 +2941,10 @@ static int __init 
 handle_passthrough_prop(struct kernel_info *kinfo,
           if ( res )
           {
Quoted 8 lines

              printk(XENLOG_ERR "Unable to permit to dom%d access to" -                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n", kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1);
+                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));

Why do you need the casts here? mstart is already defined as paddr_t

Quoted 8 lines
Quoted 6 lines

Actually, this is because the PAGE_MASK is defined as 'long'. See xen/include/xen/page-size.h :-

#define PAGE_SIZE           (_AC(1,L) << PAGE_SHIFT) #define PAGE_MASK           (~(PAGE_SIZE-1))

So, the resultant type inferred is 'long unsigned int'. Thus, we need to add an explicit cast.

Hmmm... I am a bit confused with this statement. If the resultant type inferred is actually 'long unsigned int', then why was the current specifier worked before ('long unsigned int' is 32-bit on Arm32)?

Before this change, mstart was of type paddr_t (ie u64 ie unsigned long long on Arm_32). When 'unsigned long long' was operated with 'long' (ie PAGE_SIZE), then the resultant type is 'unsigned long long'. The rule from 'C Standard 2011' (https://web.cs.dal.ca/~vlado/pl/C_Standard_2011-n1570.pdf) , page 53 (Section 6.3.1.8 Usual arithmetic conversions) is :-

"Otherwise, if the operand that has unsigned integer type has rank greater or equal to the rank of the type of the other operand, then the operand with signed integer type is converted to the type of the operand with unsigned integer type."

And from 6.3.1.1

"The rank of long long int shall be greater than the rank of long int, which shall be greater than the rank of int, which shall be greater than the rank of short int, which shall be greater than the rank of signed char.

The rank of any unsigned integer type shall equal the rank of the corresponding signed integer type, if any."

And using 'PRIx64' to print 'unsigned long long' will not require any cast.

Now with the change,
mstart is of paddr_t (ie 'unsigned int'). 'unsigned int' operated with 'long', then  the result is 'unsigned long int'. As the rank of 'unsigned long int' > 'int', thus it cannot be printed using PRIx32 (integer format specifier) without an explicit cast.

Please let me know if this makes sense.

-Ayan

Quoted 3 lines

Cheers,

#34 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

Hi Ayan,

Quoted 82 lines · Ayan Kumar Halder
Quoted 64 lines · Julien Grall
Quoted 61 lines · Ayan Kumar Halder

Hi Stefano,

Quoted 57 lines · Stefano Stabellini
Quoted 47 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++---- xen/arch/arm/gic-v3.c              |  2 +- xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++------------- xen/drivers/char/exynos4210-uart.c |  2 +- xen/drivers/char/ns16550.c         |  8 ++++---- xen/drivers/char/omap-uart.c       |  2 +- xen/drivers/char/pl011.c           |  4 ++-- xen/drivers/char/scif-uart.c       |  2 +- xen/drivers/passthrough/arm/smmu.c |  6 +++---

  9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +2 -2
index 72b9afbb4c..cf8ae37a14 100644
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const 
 struct kernel_info *kinfo,
       dt_for_each_device_node( dt_host, np )
       {
           unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;
           naddr = dt_number_of_address(np);
@@ -2445,7 +2445,7 @@ static int __init handle_device(struct domain 
 *d, struct dt_device_node *dev,
       unsigned int naddr;
       unsigned int i;
       int res;
-    u64 addr, size;
+    paddr_t addr, size;
       bool own_device = !dt_device_for_passthrough(dev);
       /*

       * We want to avoid mapping the MMIO in dom0 for the following cases:

xen/arch/arm/domain_build.c +0 -0
@@ -2941,9 +2941,10 @@ static int __init 
 handle_passthrough_prop(struct kernel_info *kinfo,
           if ( res )
           {
Quoted 8 lines

              printk(XENLOG_ERR "Unable to permit to dom%d access to" -                   " 0x%"PRIx64" - 0x%"PRIx64"\n",
+                   " 0x%"PRIpaddr" - 0x%"PRIpaddr"\n", kinfo->d->domain_id,
-                   mstart & PAGE_MASK, PAGE_ALIGN(mstart + size) - 1);
+                   (paddr_t) (mstart & PAGE_MASK),
+                   (paddr_t) (PAGE_ALIGN(mstart + size) - 1));

Why do you need the casts here? mstart is already defined as paddr_t

Quoted 8 lines
Quoted 6 lines

Actually, this is because the PAGE_MASK is defined as 'long'. See xen/include/xen/page-size.h :-

#define PAGE_SIZE           (_AC(1,L) << PAGE_SHIFT) #define PAGE_MASK           (~(PAGE_SIZE-1))

So, the resultant type inferred is 'long unsigned int'. Thus, we need to add an explicit cast.

Hmmm... I am a bit confused with this statement. If the resultant type inferred is actually 'long unsigned int', then why was the current specifier worked before ('long unsigned int' is 32-bit on Arm32)?

Before this change, mstart was of type paddr_t (ie u64 ie unsigned long long on Arm_32). When 'unsigned long long' was operated with 'long' (ie PAGE_SIZE), then the resultant type is 'unsigned long long'. The rule from 'C Standard 2011' (https://web.cs.dal.ca/~vlado/pl/C_Standard_2011-n1570.pdf) , page 53 (Section 6.3.1.8 Usual arithmetic conversions) is :-

"Otherwise, if the operand that has unsigned integer type has rank greater or equal to the rank of the type of the other operand, then the operand with signed integer type is converted to the type of the operand with unsigned integer type."

And from 6.3.1.1

"The rank of long long int shall be greater than the rank of long int, which shall be greater than the rank of int, which shall be greater than the rank of short int, which shall be greater than the rank of signed char.

Quoted 8 lines

The rank of any unsigned integer type shall equal the rank of the corresponding signed integer type, if any."

And using 'PRIx64' to print 'unsigned long long' will not require any cast.

Now with the change,
mstart is of paddr_t (ie 'unsigned int'). 'unsigned int' operated with 'long', then  the result is 'unsigned long int'. As the rank of 'unsigned long int' > 'int', thus it cannot be printed using PRIx32 (integer format specifier) without an explicit cast.

Please let me know if this makes sense.

Thanks for the explanation. I would expect that there will be several places in Xen where we would need such cast.

So it sounds like we want to define paddr_t as ``unsigned long`` and update PRIpaddr accordingly.

Cheers,

Signature · 1 line
Julien Grall
#35 lore raw

Re: [XEN v2 05/11] xen/arm: Use paddr_t instead of u64 for address/size

In reply to Ayan Kumar Halder

Hi,

Quoted 26 lines · Ayan Kumar Halder

One should now be able to use 'paddr_t' to represent address and size. Consequently, one should use 'PRIpaddr' as a format specifier for paddr_t.

---

Changes from -

v1 - 1. Rebased the patch.

  xen/arch/arm/domain_build.c        |  9 +++++----
  xen/arch/arm/gic-v3.c              |  2 +-
  xen/arch/arm/platforms/exynos5.c   | 26 +++++++++++++-------------
  xen/drivers/char/exynos4210-uart.c |  2 +-
  xen/drivers/char/ns16550.c         |  8 ++++----
  xen/drivers/char/omap-uart.c       |  2 +-
  xen/drivers/char/pl011.c           |  4 ++--
  xen/drivers/char/scif-uart.c       |  2 +-
  xen/drivers/passthrough/arm/smmu.c |  6 +++---
  9 files changed, 31 insertions(+), 30 deletions(-)
xen/arch/arm/domain_build.c +0 -0
index 72b9afbb4c..cf8ae37a14 100644

xen/arch/arm/domain_build.c static int __init find_memory_holes(const struct kernel_info *kinfo, · line 1669

Quoted 5 lines
xen/arch/arm/domain_build.c +1 -1
@@ -1666,7 +1666,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
      dt_for_each_device_node( dt_host, np )
      {
          unsigned int naddr;
-        u64 addr, size;
+        paddr_t addr, size;

Without the next patch, this change is incorrect because dt_device_get_address() expects a 64-bit value rather than paddr_t.

So the type change wants to be moved in the next patch. The same goes for any variable you modifed in this patch used by dt_device_get_address().

Cheers,

Signature · 1 line
Julien Grall
#36 lore raw

PATCH v2 6/11 xen/arm: Introduce a wrapper for dt_device_get_address() to handle paddr_t

In reply to Ayan Kumar Halder

dt_device_get_address() can accept u64 only for address and size. The various callers will use 'paddr_t' datatype for address and size. 'paddr_t' is currently defined as u64, but we may support u32 as well. Thus, we need an appropriate wrapper which can handle this type conversion.

The callers will now invoke dt_device_get_paddr(). This inturn invokes dt_device_get_address() with u64 address/size. And then it typecasts the u64 address/size to paddr_t address/size.

---

Changes from -

v1 - 1. New patch introduced.

 xen/arch/arm/domain_build.c            |  5 +++--
 xen/arch/arm/gic-v2.c                  | 11 ++++++-----
 xen/arch/arm/gic-v3.c                  |  9 +++++----
 xen/arch/arm/include/asm/device_tree.h | 19 +++++++++++++++++++
 xen/arch/arm/platforms/exynos5.c       |  7 ++++---
 xen/arch/arm/platforms/sunxi.c         |  3 ++-
 xen/drivers/char/exynos4210-uart.c     |  3 ++-
 xen/drivers/char/ns16550.c             |  3 ++-
 xen/drivers/char/omap-uart.c           |  3 ++-
 xen/drivers/char/pl011.c               |  3 ++-
 xen/drivers/char/scif-uart.c           |  3 ++-
 xen/drivers/passthrough/arm/smmu.c     |  3 ++-
 12 files changed, 51 insertions(+), 21 deletions(-)
xen/arch/arm/domain_build.c +3 -2
index cf8ae37a14..21199b624b 100644
@@ -7,6 +7,7 @@
 #include <xen/domain_page.h>
 #include <xen/sched.h>
 #include <xen/sizes.h>
+#include <asm/device_tree.h>
 #include <asm/irq.h>
 #include <asm/regs.h>
 #include <xen/errno.h>
@@ -1672,7 +1673,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
 
         for ( i = 0; i < naddr; i++ )
         {
-            res = dt_device_get_address(np, i, &addr, &size);
+            res = dt_device_get_paddr(np, i, &addr, &size);
             if ( res )
             {
                 printk(XENLOG_ERR "Unable to retrieve address %u for %s\n",
@@ -2500,7 +2501,7 @@ static int __init handle_device(struct domain *d, struct dt_device_node *dev,
     /* Give permission and map MMIOs */
     for ( i = 0; i < naddr; i++ )
     {
-        res = dt_device_get_address(dev, i, &addr, &size);
+        res = dt_device_get_paddr(dev, i, &addr, &size);
         if ( res )
         {
             printk(XENLOG_ERR "Unable to retrieve address %u for %s\n",
xen/arch/arm/gic-v2.c +6 -5
index 5d4d298b86..5230c4ebaf 100644
@@ -24,6 +24,7 @@
 #include <xen/acpi.h>
 #include <acpi/actables.h>
 #include <asm/p2m.h>
+#include <asm/device_tree.h>
 #include <asm/domain.h>
 #include <asm/platform.h>
 #include <asm/device.h>
@@ -993,7 +994,7 @@ static void gicv2_extension_dt_init(const struct dt_device_node *node)
             continue;
 
         /* Get register frame resource from DT. */
-        if ( dt_device_get_address(v2m, 0, &addr, &size) )
+        if ( dt_device_get_paddr(v2m, 0, &addr, &size) )
             panic("GICv2: Cannot find a valid v2m frame address\n");
 
         /*
@@ -1018,19 +1019,19 @@ static void __init gicv2_dt_init(void)
     paddr_t vsize;
     const struct dt_device_node *node = gicv2_info.node;
 
-    res = dt_device_get_address(node, 0, &dbase, NULL);
+    res = dt_device_get_paddr(node, 0, &dbase, NULL);
     if ( res )
         panic("GICv2: Cannot find a valid address for the distributor\n");
 
-    res = dt_device_get_address(node, 1, &cbase, &csize);
+    res = dt_device_get_paddr(node, 1, &cbase, &csize);
     if ( res )
         panic("GICv2: Cannot find a valid address for the CPU\n");
 
-    res = dt_device_get_address(node, 2, &hbase, NULL);
+    res = dt_device_get_paddr(node, 2, &hbase, NULL);
     if ( res )
         panic("GICv2: Cannot find a valid address for the hypervisor\n");
 
-    res = dt_device_get_address(node, 3, &vbase, &vsize);
+    res = dt_device_get_paddr(node, 3, &vbase, &vsize);
     if ( res )
         panic("GICv2: Cannot find a valid address for the virtual CPU\n");
 
xen/arch/arm/gic-v3.c +5 -4
index 391dfa53d7..58d2eb0690 100644
@@ -29,6 +29,7 @@
 
 #include <asm/cpufeature.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/gic.h>
 #include <asm/gic_v3_defs.h>
 #include <asm/gic_v3_its.h>
@@ -1377,7 +1378,7 @@ static void __init gicv3_dt_init(void)
     int res, i;
     const struct dt_device_node *node = gicv3_info.node;
 
-    res = dt_device_get_address(node, 0, &dbase, NULL);
+    res = dt_device_get_paddr(node, 0, &dbase, NULL);
     if ( res )
         panic("GICv3: Cannot find a valid distributor address\n");
 
@@ -1395,7 +1396,7 @@ static void __init gicv3_dt_init(void)
     {
         paddr_t rdist_base, rdist_size;
 
-        res = dt_device_get_address(node, 1 + i, &rdist_base, &rdist_size);
+        res = dt_device_get_paddr(node, 1 + i, &rdist_base, &rdist_size);
         if ( res )
             panic("GICv3: No rdist base found for region %d\n", i);
 
@@ -1417,10 +1418,10 @@ static void __init gicv3_dt_init(void)
      * For GICv3 supporting GICv2, GICC and GICV base address will be
      * provided.
      */
-    res = dt_device_get_address(node, 1 + gicv3.rdist_count,
+    res = dt_device_get_paddr(node, 1 + gicv3.rdist_count,
                                 &cbase, &csize);
     if ( !res )
-        dt_device_get_address(node, 1 + gicv3.rdist_count + 2,
+        dt_device_get_paddr(node, 1 + gicv3.rdist_count + 2,
                               &vbase, &vsize);
 }
 
xen/arch/arm/include/asm/device_tree.h +19 -0
index 51e0f0ae20..7f58f1f278 100644
@@ -11,6 +11,7 @@
 #ifndef __ARCH_ARM_DEVICE_TREE__
 #define __ARCH_ARM_DEVICE_TREE__
 
+#include <xen/device_tree.h>
 #include <xen/libfdt/libfdt.h>
 
 inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
@@ -29,6 +30,24 @@ inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
     return ret;
 }
 
+inline int dt_device_get_paddr(const struct dt_device_node *dev,
+                               unsigned int index, paddr_t *addr,
+                               paddr_t *size)
+{
+    u64 dt_addr, dt_size;
+    int ret;
+
+    ret = dt_device_get_address(dev, index, &dt_addr, &dt_size);
+
+    if ( addr )
+        *addr = dt_addr;
+
+    if ( size )
+        *size = dt_size;
+
+    return ret;
+}
+
 #endif /* __ARCH_ARM_DEVICE_TREE__ */
 /*
  * Local variables:
xen/arch/arm/platforms/exynos5.c +4 -3
index f79fad9957..55b6ac1e7e 100644
@@ -22,6 +22,7 @@
 #include <xen/mm.h>
 #include <xen/vmap.h>
 #include <xen/delay.h>
+#include <asm/device_tree.h>
 #include <asm/platforms/exynos5.h>
 #include <asm/platform.h>
 #include <asm/io.h>
@@ -52,7 +53,7 @@ static int exynos5_init_time(void)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, &mct_base_addr, &size);
+    rc = dt_device_get_paddr(node, 0, &mct_base_addr, &size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in \"samsung,exynos4210-mct\"\n");
@@ -125,7 +126,7 @@ static int __init exynos5_smp_init(void)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, &sysram_addr, &size);
+    rc = dt_device_get_paddr(node, 0, &sysram_addr, &size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in %s\n", compatible);
@@ -208,7 +209,7 @@ static int exynos5_get_pmu_baseandsize(paddr_t *power_base_addr, paddr_t *size)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, power_base_addr, size);
+    rc = dt_device_get_paddr(node, 0, power_base_addr, size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in \"samsung,exynos5XXX-pmu\"\n");
xen/arch/arm/platforms/sunxi.c +2 -1
index e8e4d88bef..ce47f97507 100644
@@ -18,6 +18,7 @@
 
 #include <xen/mm.h>
 #include <xen/vmap.h>
+#include <asm/device_tree.h>
 #include <asm/platform.h>
 #include <asm/io.h>
 
@@ -50,7 +51,7 @@ static void __iomem *sunxi_map_watchdog(bool *new_wdt)
         return NULL;
     }
 
-    ret = dt_device_get_address(node, 0, &wdt_start, &wdt_len);
+    ret = dt_device_get_paddr(node, 0, &wdt_start, &wdt_len);
     if ( ret )
     {
         dprintk(XENLOG_ERR, "Cannot read watchdog register address\n");
xen/drivers/char/exynos4210-uart.c +2 -1
index 32cc8c78b5..6d2008c44f 100644
@@ -24,6 +24,7 @@
 #include <xen/irq.h>
 #include <xen/mm.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/exynos4210-uart.h>
 #include <asm/io.h>
 
@@ -316,7 +317,7 @@ static int __init exynos4210_uart_init(struct dt_device_node *dev,
     uart->parity    = PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("exynos4210: Unable to retrieve the base"
xen/drivers/char/ns16550.c +2 -1
index 8ef895a2bb..7226f3c2f7 100644
@@ -35,6 +35,7 @@
 #include <asm/io.h>
 #ifdef CONFIG_HAS_DEVICE_TREE
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #endif
 #ifdef CONFIG_X86
 #include <asm/fixmap.h>
@@ -1757,7 +1758,7 @@ static int __init ns16550_uart_dt_init(struct dt_device_node *dev,
     uart->parity    = UART_PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &uart->io_base, &uart->io_size);
+    res = dt_device_get_paddr(dev, 0, &uart->io_base, &uart->io_size);
     if ( res )
         return res;
 
xen/drivers/char/omap-uart.c +2 -1
index 3b53e1909a..06200bc9f1 100644
@@ -15,6 +15,7 @@
 #include <xen/init.h>
 #include <xen/irq.h>
 #include <xen/device_tree.h>
+#include <asm/device_tree.h>
 #include <asm/device.h>
 #include <xen/errno.h>
 #include <xen/mm.h>
@@ -344,7 +345,7 @@ static int __init omap_uart_init(struct dt_device_node *dev,
     uart->parity = UART_PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("omap-uart: Unable to retrieve the base"
xen/drivers/char/pl011.c +2 -1
index 256ec11e3f..b4c1d9d592 100644
@@ -26,6 +26,7 @@
 #include <asm/device.h>
 #include <xen/mm.h>
 #include <xen/vmap.h>
+#include <asm/device_tree.h>
 #include <asm/pl011-uart.h>
 #include <asm/io.h>
 
@@ -265,7 +266,7 @@ static int __init pl011_dt_uart_init(struct dt_device_node *dev,
         printk("WARNING: UART configuration is not supported\n");
     }
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("pl011: Unable to retrieve the base"
xen/drivers/char/scif-uart.c +2 -1
index b425881d06..af14388f70 100644
@@ -26,6 +26,7 @@
 #include <xen/mm.h>
 #include <xen/delay.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/scif-uart.h>
 #include <asm/io.h>
 
@@ -318,7 +319,7 @@ static int __init scif_uart_init(struct dt_device_node *dev,
 
     uart = &scif_com;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("scif-uart: Unable to retrieve the base"
xen/drivers/passthrough/arm/smmu.c +2 -1
index 490d253d44..0c89cb644e 100644
@@ -51,6 +51,7 @@
 #include <xen/sizes.h>
 #include <asm/atomic.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/io.h>
 #include <asm/iommu_fwspec.h>
 #include <asm/platform.h>
@@ -101,7 +102,7 @@ static struct resource *platform_get_resource(struct platform_device *pdev,
 
 	switch (type) {
 	case IORESOURCE_MEM:
-		ret = dt_device_get_address(pdev, num, &res.addr, &res.size);
+		ret = dt_device_get_paddr(pdev, num, &res.addr, &res.size);
 
 		return ((ret) ? NULL : &res);
 
Signature · 1 line
2.17.1
#37 lore raw

Re: [XEN v2 06/11] xen/arm: Introduce a wrapper for dt_device_get_address() to handle paddr_t

Quoted 169 lines · Ayan Kumar Halder

dt_device_get_address() can accept u64 only for address and size. The various callers will use 'paddr_t' datatype for address and size. 'paddr_t' is currently defined as u64, but we may support u32 as well. Thus, we need an appropriate wrapper which can handle this type conversion.

The callers will now invoke dt_device_get_paddr(). This inturn invokes dt_device_get_address() with u64 address/size. And then it typecasts the u64 address/size to paddr_t address/size.

---

Changes from -

v1 - 1. New patch introduced.

 xen/arch/arm/domain_build.c            |  5 +++--
 xen/arch/arm/gic-v2.c                  | 11 ++++++-----
 xen/arch/arm/gic-v3.c                  |  9 +++++----
 xen/arch/arm/include/asm/device_tree.h | 19 +++++++++++++++++++
 xen/arch/arm/platforms/exynos5.c       |  7 ++++---
 xen/arch/arm/platforms/sunxi.c         |  3 ++-
 xen/drivers/char/exynos4210-uart.c     |  3 ++-
 xen/drivers/char/ns16550.c             |  3 ++-
 xen/drivers/char/omap-uart.c           |  3 ++-
 xen/drivers/char/pl011.c               |  3 ++-
 xen/drivers/char/scif-uart.c           |  3 ++-
 xen/drivers/passthrough/arm/smmu.c     |  3 ++-
 12 files changed, 51 insertions(+), 21 deletions(-)
xen/arch/arm/domain_build.c +3 -2
index cf8ae37a14..21199b624b 100644
@@ -7,6 +7,7 @@
 #include <xen/domain_page.h>
 #include <xen/sched.h>
 #include <xen/sizes.h>
+#include <asm/device_tree.h>
 #include <asm/irq.h>
 #include <asm/regs.h>
 #include <xen/errno.h>
@@ -1672,7 +1673,7 @@ static int __init find_memory_holes(const struct kernel_info *kinfo,
 
         for ( i = 0; i < naddr; i++ )
         {
-            res = dt_device_get_address(np, i, &addr, &size);
+            res = dt_device_get_paddr(np, i, &addr, &size);
             if ( res )
             {
                 printk(XENLOG_ERR "Unable to retrieve address %u for %s\n",
@@ -2500,7 +2501,7 @@ static int __init handle_device(struct domain *d, struct dt_device_node *dev,
     /* Give permission and map MMIOs */
     for ( i = 0; i < naddr; i++ )
     {
-        res = dt_device_get_address(dev, i, &addr, &size);
+        res = dt_device_get_paddr(dev, i, &addr, &size);
         if ( res )
         {
             printk(XENLOG_ERR "Unable to retrieve address %u for %s\n",
xen/arch/arm/gic-v2.c +6 -5
index 5d4d298b86..5230c4ebaf 100644
@@ -24,6 +24,7 @@
 #include <xen/acpi.h>
 #include <acpi/actables.h>
 #include <asm/p2m.h>
+#include <asm/device_tree.h>
 #include <asm/domain.h>
 #include <asm/platform.h>
 #include <asm/device.h>
@@ -993,7 +994,7 @@ static void gicv2_extension_dt_init(const struct dt_device_node *node)
             continue;
 
         /* Get register frame resource from DT. */
-        if ( dt_device_get_address(v2m, 0, &addr, &size) )
+        if ( dt_device_get_paddr(v2m, 0, &addr, &size) )
             panic("GICv2: Cannot find a valid v2m frame address\n");
 
         /*
@@ -1018,19 +1019,19 @@ static void __init gicv2_dt_init(void)
     paddr_t vsize;
     const struct dt_device_node *node = gicv2_info.node;
 
-    res = dt_device_get_address(node, 0, &dbase, NULL);
+    res = dt_device_get_paddr(node, 0, &dbase, NULL);
     if ( res )
         panic("GICv2: Cannot find a valid address for the distributor\n");
 
-    res = dt_device_get_address(node, 1, &cbase, &csize);
+    res = dt_device_get_paddr(node, 1, &cbase, &csize);
     if ( res )
         panic("GICv2: Cannot find a valid address for the CPU\n");
 
-    res = dt_device_get_address(node, 2, &hbase, NULL);
+    res = dt_device_get_paddr(node, 2, &hbase, NULL);
     if ( res )
         panic("GICv2: Cannot find a valid address for the hypervisor\n");
 
-    res = dt_device_get_address(node, 3, &vbase, &vsize);
+    res = dt_device_get_paddr(node, 3, &vbase, &vsize);
     if ( res )
         panic("GICv2: Cannot find a valid address for the virtual CPU\n");
 
xen/arch/arm/gic-v3.c +5 -4
index 391dfa53d7..58d2eb0690 100644
@@ -29,6 +29,7 @@
 
 #include <asm/cpufeature.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/gic.h>
 #include <asm/gic_v3_defs.h>
 #include <asm/gic_v3_its.h>
@@ -1377,7 +1378,7 @@ static void __init gicv3_dt_init(void)
     int res, i;
     const struct dt_device_node *node = gicv3_info.node;
 
-    res = dt_device_get_address(node, 0, &dbase, NULL);
+    res = dt_device_get_paddr(node, 0, &dbase, NULL);
     if ( res )
         panic("GICv3: Cannot find a valid distributor address\n");
 
@@ -1395,7 +1396,7 @@ static void __init gicv3_dt_init(void)
     {
         paddr_t rdist_base, rdist_size;
 
-        res = dt_device_get_address(node, 1 + i, &rdist_base, &rdist_size);
+        res = dt_device_get_paddr(node, 1 + i, &rdist_base, &rdist_size);
         if ( res )
             panic("GICv3: No rdist base found for region %d\n", i);
 
@@ -1417,10 +1418,10 @@ static void __init gicv3_dt_init(void)
      * For GICv3 supporting GICv2, GICC and GICV base address will be
      * provided.
      */
-    res = dt_device_get_address(node, 1 + gicv3.rdist_count,
+    res = dt_device_get_paddr(node, 1 + gicv3.rdist_count,
                                 &cbase, &csize);
     if ( !res )
-        dt_device_get_address(node, 1 + gicv3.rdist_count + 2,
+        dt_device_get_paddr(node, 1 + gicv3.rdist_count + 2,
                               &vbase, &vsize);
 }
 
xen/arch/arm/include/asm/device_tree.h +5 -0
index 51e0f0ae20..7f58f1f278 100644
@@ -11,6 +11,7 @@
 #ifndef __ARCH_ARM_DEVICE_TREE__
 #define __ARCH_ARM_DEVICE_TREE__
 
+#include <xen/device_tree.h>
 #include <xen/libfdt/libfdt.h>
 
 inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
@@ -29,6 +30,24 @@ inline int fdt_get_mem_rsv_paddr(const void *fdt, int n,
     return ret;
 }
 
+inline int dt_device_get_paddr(const struct dt_device_node *dev,
+                               unsigned int index, paddr_t *addr,
+                               paddr_t *size)
+{

xen/arch/arm/include/asm/device_tree.h inline int fdt_get_mem_rsv_paddr(const void *fdt, int n, · line 46

Quoted 10 lines
xen/arch/arm/include/asm/device_tree.h +10 -0
+    u64 dt_addr, dt_size;
+    int ret;
+
+    ret = dt_device_get_address(dev, index, &dt_addr, &dt_size);
+
+    if ( addr )
+        *addr = dt_addr;
+
+    if ( size )
+        *size = dt_size;

I think we should check for truncation (top 32-bit != 0) and return an error in that case.

Quoted 195 lines
xen/arch/arm/include/asm/device_tree.h +3 -0
+    return ret;
+}
+
 #endif /* __ARCH_ARM_DEVICE_TREE__ */
 /*
  * Local variables:
xen/arch/arm/platforms/exynos5.c +4 -3
index f79fad9957..55b6ac1e7e 100644
@@ -22,6 +22,7 @@
 #include <xen/mm.h>
 #include <xen/vmap.h>
 #include <xen/delay.h>
+#include <asm/device_tree.h>
 #include <asm/platforms/exynos5.h>
 #include <asm/platform.h>
 #include <asm/io.h>
@@ -52,7 +53,7 @@ static int exynos5_init_time(void)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, &mct_base_addr, &size);
+    rc = dt_device_get_paddr(node, 0, &mct_base_addr, &size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in \"samsung,exynos4210-mct\"\n");
@@ -125,7 +126,7 @@ static int __init exynos5_smp_init(void)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, &sysram_addr, &size);
+    rc = dt_device_get_paddr(node, 0, &sysram_addr, &size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in %s\n", compatible);
@@ -208,7 +209,7 @@ static int exynos5_get_pmu_baseandsize(paddr_t *power_base_addr, paddr_t *size)
         return -ENXIO;
     }
 
-    rc = dt_device_get_address(node, 0, power_base_addr, size);
+    rc = dt_device_get_paddr(node, 0, power_base_addr, size);
     if ( rc )
     {
         dprintk(XENLOG_ERR, "Error in \"samsung,exynos5XXX-pmu\"\n");
xen/arch/arm/platforms/sunxi.c +2 -1
index e8e4d88bef..ce47f97507 100644
@@ -18,6 +18,7 @@
 
 #include <xen/mm.h>
 #include <xen/vmap.h>
+#include <asm/device_tree.h>
 #include <asm/platform.h>
 #include <asm/io.h>
 
@@ -50,7 +51,7 @@ static void __iomem *sunxi_map_watchdog(bool *new_wdt)
         return NULL;
     }
 
-    ret = dt_device_get_address(node, 0, &wdt_start, &wdt_len);
+    ret = dt_device_get_paddr(node, 0, &wdt_start, &wdt_len);
     if ( ret )
     {
         dprintk(XENLOG_ERR, "Cannot read watchdog register address\n");
xen/drivers/char/exynos4210-uart.c +2 -1
index 32cc8c78b5..6d2008c44f 100644
@@ -24,6 +24,7 @@
 #include <xen/irq.h>
 #include <xen/mm.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/exynos4210-uart.h>
 #include <asm/io.h>
 
@@ -316,7 +317,7 @@ static int __init exynos4210_uart_init(struct dt_device_node *dev,
     uart->parity    = PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("exynos4210: Unable to retrieve the base"
xen/drivers/char/ns16550.c +2 -1
index 8ef895a2bb..7226f3c2f7 100644
@@ -35,6 +35,7 @@
 #include <asm/io.h>
 #ifdef CONFIG_HAS_DEVICE_TREE
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #endif
 #ifdef CONFIG_X86
 #include <asm/fixmap.h>
@@ -1757,7 +1758,7 @@ static int __init ns16550_uart_dt_init(struct dt_device_node *dev,
     uart->parity    = UART_PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &uart->io_base, &uart->io_size);
+    res = dt_device_get_paddr(dev, 0, &uart->io_base, &uart->io_size);
     if ( res )
         return res;
 
xen/drivers/char/omap-uart.c +2 -1
index 3b53e1909a..06200bc9f1 100644
@@ -15,6 +15,7 @@
 #include <xen/init.h>
 #include <xen/irq.h>
 #include <xen/device_tree.h>
+#include <asm/device_tree.h>
 #include <asm/device.h>
 #include <xen/errno.h>
 #include <xen/mm.h>
@@ -344,7 +345,7 @@ static int __init omap_uart_init(struct dt_device_node *dev,
     uart->parity = UART_PARITY_NONE;
     uart->stop_bits = 1;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("omap-uart: Unable to retrieve the base"
xen/drivers/char/pl011.c +2 -1
index 256ec11e3f..b4c1d9d592 100644
@@ -26,6 +26,7 @@
 #include <asm/device.h>
 #include <xen/mm.h>
 #include <xen/vmap.h>
+#include <asm/device_tree.h>
 #include <asm/pl011-uart.h>
 #include <asm/io.h>
 
@@ -265,7 +266,7 @@ static int __init pl011_dt_uart_init(struct dt_device_node *dev,
         printk("WARNING: UART configuration is not supported\n");
     }
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("pl011: Unable to retrieve the base"
xen/drivers/char/scif-uart.c +2 -1
index b425881d06..af14388f70 100644
@@ -26,6 +26,7 @@
 #include <xen/mm.h>
 #include <xen/delay.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/scif-uart.h>
 #include <asm/io.h>
 
@@ -318,7 +319,7 @@ static int __init scif_uart_init(struct dt_device_node *dev,
 
     uart = &scif_com;
 
-    res = dt_device_get_address(dev, 0, &addr, &size);
+    res = dt_device_get_paddr(dev, 0, &addr, &size);
     if ( res )
     {
         printk("scif-uart: Unable to retrieve the base"
xen/drivers/passthrough/arm/smmu.c +2 -1
index 490d253d44..0c89cb644e 100644
@@ -51,6 +51,7 @@
 #include <xen/sizes.h>
 #include <asm/atomic.h>
 #include <asm/device.h>
+#include <asm/device_tree.h>
 #include <asm/io.h>
 #include <asm/iommu_fwspec.h>
 #include <asm/platform.h>
@@ -101,7 +102,7 @@ static struct resource *platform_get_resource(struct platform_device *pdev,
 
 	switch (type) {
 	case IORESOURCE_MEM:
-		ret = dt_device_get_address(pdev, num, &res.addr, &res.size);
+		ret = dt_device_get_paddr(pdev, num, &res.addr, &res.size);
 
 		return ((ret) ? NULL : &res);
 
Signature · 1 line
2.17.1
#38 lore raw

PATCH v2 7/11 xen/arm: smmu: Use writeq_relaxed_non_atomic() for writing to SMMU_CBn_TTBR0

In reply to Ayan Kumar Halder

Refer ARM IHI 0062D.c ID070116 (SMMU 2.0 spec), 17-360, 17.3.9, SMMU_CBn_TTBR0 is a 64 bit register. Thus, one can use writeq_relaxed_non_atomic() to write to it instead of invoking writel_relaxed() twice for lower half and upper half of the register.

This also helps us as p2maddr is 'paddr_t' (which may be u32 in future). Thus, one can assign p2maddr to a 64 bit register and do the bit manipulations on it, to generate the value for SMMU_CBn_TTBR0.

---
Changes from -

v1 - 1. Extracted the patch from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr". Use writeq_relaxed_non_atomic() to write u64 register in a non-atomic
fashion.

 xen/drivers/passthrough/arm/smmu.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
xen/drivers/passthrough/arm/smmu.c +8 -7
index 0c89cb644e..84b6803b4e 100644
@@ -500,8 +500,7 @@ enum arm_smmu_s2cr_privcfg {
 #define ARM_SMMU_CB_SCTLR		0x0
 #define ARM_SMMU_CB_RESUME		0x8
 #define ARM_SMMU_CB_TTBCR2		0x10
-#define ARM_SMMU_CB_TTBR0_LO		0x20
-#define ARM_SMMU_CB_TTBR0_HI		0x24
+#define ARM_SMMU_CB_TTBR0		0x20
 #define ARM_SMMU_CB_TTBCR		0x30
 #define ARM_SMMU_CB_S1_MAIR0		0x38
 #define ARM_SMMU_CB_FSR			0x58
@@ -1084,6 +1083,7 @@ static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
 {
 	u32 reg;
+	u64 reg64;
 	bool stage1;
 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 	struct arm_smmu_device *smmu = smmu_domain->smmu;
@@ -1178,12 +1178,13 @@ static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
 	dev_notice(smmu->dev, "d%u: p2maddr 0x%"PRIpaddr"\n",
 		   smmu_domain->cfg.domain->domain_id, p2maddr);
 
-	reg = (p2maddr & ((1ULL << 32) - 1));
-	writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
-	reg = (p2maddr >> 32);
+	reg64 = p2maddr;
+
 	if (stage1)
-		reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
-	writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
+		reg64 |= (((uint64_t) (ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT))
+		         << 32);
+
+	writeq_relaxed_non_atomic(reg64, cb_base + ARM_SMMU_CB_TTBR0);
 
 	/*
 	 * TTBCR
Signature · 1 line
2.17.1
Reviewed-by on v2 7/11
#39 lore raw

Re: [XEN v2 07/11] xen/arm: smmu: Use writeq_relaxed_non_atomic() for writing to SMMU_CBn_TTBR0

Quoted 10 lines · Ayan Kumar Halder

Refer ARM IHI 0062D.c ID070116 (SMMU 2.0 spec), 17-360, 17.3.9, SMMU_CBn_TTBR0 is a 64 bit register. Thus, one can use writeq_relaxed_non_atomic() to write to it instead of invoking writel_relaxed() twice for lower half and upper half of the register.

This also helps us as p2maddr is 'paddr_t' (which may be u32 in future). Thus, one can assign p2maddr to a 64 bit register and do the bit manipulations on it, to generate the value for SMMU_CBn_TTBR0.

Quoted 54 lines

---
Changes from -

v1 - 1. Extracted the patch from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr". Use writeq_relaxed_non_atomic() to write u64 register in a non-atomic
fashion.

 xen/drivers/passthrough/arm/smmu.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)
xen/drivers/passthrough/arm/smmu.c +8 -7
index 0c89cb644e..84b6803b4e 100644
@@ -500,8 +500,7 @@ enum arm_smmu_s2cr_privcfg {
 #define ARM_SMMU_CB_SCTLR		0x0
 #define ARM_SMMU_CB_RESUME		0x8
 #define ARM_SMMU_CB_TTBCR2		0x10
-#define ARM_SMMU_CB_TTBR0_LO		0x20
-#define ARM_SMMU_CB_TTBR0_HI		0x24
+#define ARM_SMMU_CB_TTBR0		0x20
 #define ARM_SMMU_CB_TTBCR		0x30
 #define ARM_SMMU_CB_S1_MAIR0		0x38
 #define ARM_SMMU_CB_FSR			0x58
@@ -1084,6 +1083,7 @@ static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
 {
 	u32 reg;
+	u64 reg64;
 	bool stage1;
 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 	struct arm_smmu_device *smmu = smmu_domain->smmu;
@@ -1178,12 +1178,13 @@ static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
 	dev_notice(smmu->dev, "d%u: p2maddr 0x%"PRIpaddr"\n",
 		   smmu_domain->cfg.domain->domain_id, p2maddr);
 
-	reg = (p2maddr & ((1ULL << 32) - 1));
-	writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
-	reg = (p2maddr >> 32);
+	reg64 = p2maddr;
+
 	if (stage1)
-		reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
-	writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
+		reg64 |= (((uint64_t) (ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT))
+		         << 32);
+
+	writeq_relaxed_non_atomic(reg64, cb_base + ARM_SMMU_CB_TTBR0);
 
 	/*
 	 * TTBCR
Signature · 1 line
2.17.1
#40 lore raw

PATCH v2 8/11 xen/arm: guest_walk: LPAE specific bits should be enclosed within "ifndef CONFIG_ARM_PA_32"

In reply to Ayan Kumar Halder

In the subsequent patch, we introduce "CONFIG_ARM_PA_32" to support 32 bit physical addresses. Thus, the code specific to "Large Page Address Extension" (ie LPAE) should be enclosed within "ifndef CONFIG_ARM_PA_32".

Refer xen/arch/arm/include/asm/short-desc.h, "short_desc_l1_supersec_t"
unsigned int extbase1:4; /* Extended base address, PA[35:32] */
unsigned int extbase2:4; /* Extended base address, PA[39:36] */

Thus, extbase1 and extbase2 are not valid when 32 bit physical addresses are supported.

---

Changes from -
v1 - 1. Extracted from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr".

 xen/arch/arm/guest_walk.c | 2 ++
 1 file changed, 2 insertions(+)
xen/arch/arm/guest_walk.c +2 -0
index 43d3215304..0feb7b76ec 100644
@@ -154,8 +154,10 @@ static bool guest_walk_sd(const struct vcpu *v,
             mask = (1ULL << L1DESC_SUPERSECTION_SHIFT) - 1;
             *ipa = gva & mask;
             *ipa |= (paddr_t)(pte.supersec.base) << L1DESC_SUPERSECTION_SHIFT;
+#ifndef CONFIG_ARM_PA_32
             *ipa |= (paddr_t)(pte.supersec.extbase1) << L1DESC_SUPERSECTION_EXT_BASE1_SHIFT;
             *ipa |= (paddr_t)(pte.supersec.extbase2) << L1DESC_SUPERSECTION_EXT_BASE2_SHIFT;
+#endif /* CONFIG_ARM_PA_32 */
         }
 
         /* Set permissions so that the caller can check the flags by herself. */
Signature · 1 line
2.17.1
#41 lore raw

Re: [XEN v2 08/11] xen/arm: guest_walk: LPAE specific bits should be enclosed within "ifndef CONFIG_ARM_PA_32"

Quoted 5 lines · Ayan Kumar Halder

In the subsequent patch, we introduce "CONFIG_ARM_PA_32" to support 32 bit physical addresses. Thus, the code specific to "Large Page Address Extension" (ie LPAE) should be enclosed within "ifndef CONFIG_ARM_PA_32".

Quoted 8 lines

Refer xen/arch/arm/include/asm/short-desc.h, "short_desc_l1_supersec_t"
unsigned int extbase1:4; /* Extended base address, PA[35:32] */
unsigned int extbase2:4; /* Extended base address, PA[39:36] */

Thus, extbase1 and extbase2 are not valid when 32 bit physical addresses are supported.

This patch should be merged with patch 9: we shouldn't start to use a new kconfig symbol before it is defined.

Quoted 26 lines

---

Changes from -
v1 - 1. Extracted from "[XEN v1 8/9] xen/arm: Other adaptations required to support 32bit paddr".

 xen/arch/arm/guest_walk.c | 2 ++
 1 file changed, 2 insertions(+)
xen/arch/arm/guest_walk.c +2 -0
index 43d3215304..0feb7b76ec 100644
@@ -154,8 +154,10 @@ static bool guest_walk_sd(const struct vcpu *v,
             mask = (1ULL << L1DESC_SUPERSECTION_SHIFT) - 1;
             *ipa = gva & mask;
             *ipa |= (paddr_t)(pte.supersec.base) << L1DESC_SUPERSECTION_SHIFT;
+#ifndef CONFIG_ARM_PA_32
             *ipa |= (paddr_t)(pte.supersec.extbase1) << L1DESC_SUPERSECTION_EXT_BASE1_SHIFT;
             *ipa |= (paddr_t)(pte.supersec.extbase2) << L1DESC_SUPERSECTION_EXT_BASE2_SHIFT;
+#endif /* CONFIG_ARM_PA_32 */
         }
 
         /* Set permissions so that the caller can check the flags by herself. */
Signature · 1 line
2.17.1
#42 lore raw

Re: [XEN v2 08/11] xen/arm: guest_walk: LPAE specific bits should be enclosed within "ifndef CONFIG_ARM_PA_32"

Hi Stefano,

Quoted 10 lines · Stefano Stabellini
Quoted 9 lines · Ayan Kumar Halder

In the subsequent patch, we introduce "CONFIG_ARM_PA_32" to support 32 bit physical addresses. Thus, the code specific to "Large Page Address Extension" (ie LPAE) should be enclosed within "ifndef CONFIG_ARM_PA_32".

Refer xen/arch/arm/include/asm/short-desc.h, "short_desc_l1_supersec_t"
unsigned int extbase1:4; /* Extended base address, PA[35:32] */
unsigned int extbase2:4; /* Extended base address, PA[39:36] */

Quoted 7 lines
Quoted 4 lines

Thus, extbase1 and extbase2 are not valid when 32 bit physical addresses are supported.

This patch should be merged with patch 9: we shouldn't start to use a new kconfig symbol before it is defined.

I have asked the adaptation to be in a separate patch to avoid having a melting-pot patch.

So if you want the new Kconfig to be defined first, then we should reshuffle it.

BTW, the reshuffle will likely be necessary if you want to check for truncation in earlier patch.

Cheers,

Signature · 1 line
Julien Grall
#43 lore raw

PATCH v2 9/11 xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

In reply to Ayan Kumar Halder

We have introduced a new config option to support 32 bit physical address. By default, it is disabled.
ARM_PA_32 cannot be enabled on ARM_64 as the memory management unit works on 48bit physical addresses.
On ARM_32, it can be used on systems where large page address extension is not supported.

---
Changes from -

v1 - 1. No changes.

 xen/arch/arm/Kconfig                 | 9 +++++++++
 xen/arch/arm/include/asm/page-bits.h | 2 ++
 xen/arch/arm/include/asm/types.h     | 7 +++++++
 3 files changed, 18 insertions(+)
xen/arch/arm/Kconfig +9 -0
index 239d3aed3c..aeb0f7252e 100644
@@ -39,6 +39,15 @@ config ACPI
 config ARM_EFI
 	bool
 
+config ARM_PA_32
+	bool "32 bit Physical Address"
+	depends on ARM_32
+	default n
+	---help---
+
+	  Support 32 bit physical addresses.
+	  If unsure, say N
+
 config GICV3
 	bool "GICv3 driver"
 	depends on !NEW_VGIC
xen/arch/arm/include/asm/page-bits.h +2 -0
index 5d6477e599..8f4dcebcfd 100644
@@ -5,6 +5,8 @@
 
 #ifdef CONFIG_ARM_64
 #define PADDR_BITS              48
+#elif CONFIG_ARM_PA_32
+#define PADDR_BITS              32
 #else
 #define PADDR_BITS              40
 #endif
xen/arch/arm/include/asm/types.h +7 -0
index 083acbd151..f9595b9098 100644
@@ -37,9 +37,16 @@ typedef signed long long s64;
 typedef unsigned long long u64;
 typedef u32 vaddr_t;
 #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else
 typedef u64 paddr_t;
 #define INVALID_PADDR (~0ULL)
 #define PRIpaddr "016llx"
+#endif
 typedef u32 register_t;
 #define PRIregister "08x"
 #elif defined (CONFIG_ARM_64)
Signature · 1 line
2.17.1
#44 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 45

Quoted 12 lines · Ayan Kumar Halder
xen/arch/arm/include/asm/types.h +6 -0
@@ -37,9 +37,16 @@ typedef signed long long s64;
 typedef unsigned long long u64;
 typedef u32 vaddr_t;
 #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else

With our plan to consolidate basic type definitions into xen/types.h the use of ARM_PA_32 is problematic: Preferably we'd have an arch- independent Kconfig setting, much like Linux'es PHYS_ADDR_T_64BIT (I don't think we should re-use the name directly, as we have no phys_addr_t type), which each arch selects (or not) accordingly. Perhaps architectures already selecting 64BIT wouldn't need to do this explicitly, and instead 64BIT could select the new setting (or the new setting could default to Y when 64BIT=y).

Jan

#45 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

Hi Jan,

Quoted 13 lines · Jan Beulich

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 45

Quoted 12 lines · Ayan Kumar Halder
xen/arch/arm/include/asm/types.h +6 -0
@@ -37,9 +37,16 @@ typedef signed long long s64;
  typedef unsigned long long u64;
  typedef u32 vaddr_t;
  #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else
Quoted 8 lines

With our plan to consolidate basic type definitions into xen/types.h the use of ARM_PA_32 is problematic: Preferably we'd have an arch- independent Kconfig setting, much like Linux'es PHYS_ADDR_T_64BIT (I don't think we should re-use the name directly, as we have no phys_addr_t type), which each arch selects (or not) accordingly. Perhaps architectures already selecting 64BIT wouldn't need to do this explicitly, and instead 64BIT could select the new setting (or the new setting could default to Y when 64BIT=y).

Looking at some of the instances where 64BIT is currently used (xen/drivers/passthrough/arm/smmu.c), I understand that it is used to define the width of **virtual** address.

Thus, it would not conflict with ARM_PA_32 (which is for physical address).

Later on if we wish to introduce something similar to Linux's PHYS_ADDR_T_64BIT (which is arch agnostic), then ARM_PA_32 can be selected by "!PHYS_ADDR_T_64BIT" && "CONFIG_ARM_32". (We may decide to drop ARM_PA_32 config option, but we can still reuse ARM_PA_32 specific changes).

Also, then we will need to support 32 bit physical address (ie !PHYS_ADDR_T_64BIT) with 32 bit virtual address (ie !64BIT) and 64 bit virtual address (ie 64BIT).

Let's confine the current changes to Arm architecture only (as I don't have knowledge about x86 or RISCV). When similar changes will be done for other architectures, then we can think about introducing an architecture agnostic option.

- Ayan

Quoted 2 lines

Jan

#46 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

Quoted 37 lines · Ayan Kumar Halder

Hi Jan,

Quoted 13 lines · Jan Beulich

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 45

Quoted 12 lines · Ayan Kumar Halder
xen/arch/arm/include/asm/types.h +6 -0
@@ -37,9 +37,16 @@ typedef signed long long s64;
  typedef unsigned long long u64;
  typedef u32 vaddr_t;
  #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else
Quoted 8 lines

With our plan to consolidate basic type definitions into xen/types.h the use of ARM_PA_32 is problematic: Preferably we'd have an arch- independent Kconfig setting, much like Linux'es PHYS_ADDR_T_64BIT (I don't think we should re-use the name directly, as we have no phys_addr_t type), which each arch selects (or not) accordingly. Perhaps architectures already selecting 64BIT wouldn't need to do this explicitly, and instead 64BIT could select the new setting (or the new setting could default to Y when 64BIT=y).

Looking at some of the instances where 64BIT is currently used (xen/drivers/passthrough/arm/smmu.c), I understand that it is used to define the width of **virtual** address.

Thus, it would not conflict with ARM_PA_32 (which is for physical address).

Later on if we wish to introduce something similar to Linux's PHYS_ADDR_T_64BIT (which is arch agnostic), then ARM_PA_32 can be selected by "!PHYS_ADDR_T_64BIT" && "CONFIG_ARM_32". (We may decide to drop ARM_PA_32 config option, but we can still reuse ARM_PA_32 specific changes).

Quoted 8 lines

Also, then we will need to support 32 bit physical address (ie !PHYS_ADDR_T_64BIT) with 32 bit virtual address (ie !64BIT) and 64 bit virtual address (ie 64BIT).

Let's confine the current changes to Arm architecture only (as I don't have knowledge about x86 or RISCV). When similar changes will be done for other architectures, then we can think about introducing an architecture agnostic option.

I disagree, not the least with the present goal of reworking xen/types.h vs asm/types.h. By having an arch-independent Kconfig symbol, paddr_t could also be manufactured uniformly in xen/types.h.

Jan

#47 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

Quoted 42 lines · Jan Beulich
Quoted 41 lines · Ayan Kumar Halder

Hi Jan,

Quoted 13 lines · Jan Beulich

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 45

Quoted 12 lines · Ayan Kumar Halder
xen/arch/arm/include/asm/types.h +6 -0
@@ -37,9 +37,16 @@ typedef signed long long s64;
  typedef unsigned long long u64;
  typedef u32 vaddr_t;
  #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else
Quoted 8 lines

With our plan to consolidate basic type definitions into xen/types.h the use of ARM_PA_32 is problematic: Preferably we'd have an arch- independent Kconfig setting, much like Linux'es PHYS_ADDR_T_64BIT (I don't think we should re-use the name directly, as we have no phys_addr_t type), which each arch selects (or not) accordingly. Perhaps architectures already selecting 64BIT wouldn't need to do this explicitly, and instead 64BIT could select the new setting (or the new setting could default to Y when 64BIT=y).

Looking at some of the instances where 64BIT is currently used (xen/drivers/passthrough/arm/smmu.c), I understand that it is used to define the width of **virtual** address.

Thus, it would not conflict with ARM_PA_32 (which is for physical address).

Later on if we wish to introduce something similar to Linux's PHYS_ADDR_T_64BIT (which is arch agnostic), then ARM_PA_32 can be selected by "!PHYS_ADDR_T_64BIT" && "CONFIG_ARM_32". (We may decide to drop ARM_PA_32 config option, but we can still reuse ARM_PA_32 specific changes).

Also, then we will need to support 32 bit physical address (ie !PHYS_ADDR_T_64BIT) with 32 bit virtual address (ie !64BIT) and 64 bit virtual address (ie 64BIT).

Quoted 8 lines
Quoted 4 lines

Let's confine the current changes to Arm architecture only (as I don't have knowledge about x86 or RISCV). When similar changes will be done for other architectures, then we can think about introducing an architecture agnostic option.

I disagree, not the least with the present goal of reworking xen/types.h vs asm/types.h. By having an arch-independent Kconfig symbol, paddr_t could also be manufactured uniformly in xen/types.h.

Jan is only asking to introduce the new Kconfig symbol as an arch-independent symbol. In other words, rename ARM_PA_32 to PADDR_32 (or something like that) and move the symbol to xen/arch/Kconfig.

I think that's reasonable. And PADDR_32 could be forced to always be unselected on x86: I don't think Jan is asking you to rework the whole of xen/arch/x86 and xen/commons to build on x86 with paddr_t set to uint32_t.

#48 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

In reply to Ayan Kumar Halder

Hi Ayan,

Quoted 4 lines · Ayan Kumar Halder

We have introduced a new config option to support 32 bit physical address. By default, it is disabled.
ARM_PA_32 cannot be enabled on ARM_64 as the memory management unit works on 48bit physical addresses.

I don't understand the "cannot" here. It is possible to have a 64-bit HW that support only 32-bit physical address.

After your series, I also don't see any restriction in Xen to enable ARM_PA_32.

Whether we want to do it is a different discussion. I don't have any strong opinion. But the wording should be clarified.

Quoted 19 lines

On ARM_32, it can be used on systems where large page address extension is not supported.

---
Changes from -

v1 - 1. No changes.

  xen/arch/arm/Kconfig                 | 9 +++++++++
  xen/arch/arm/include/asm/page-bits.h | 2 ++
  xen/arch/arm/include/asm/types.h     | 7 +++++++
  3 files changed, 18 insertions(+)
xen/arch/arm/Kconfig +0 -0
index 239d3aed3c..aeb0f7252e 100644

xen/arch/arm/Kconfig config ACPI · line 48

Quoted 10 lines
xen/arch/arm/Kconfig +7 -0
@@ -39,6 +39,15 @@ config ACPI
  config ARM_EFI
  	bool
 
+config ARM_PA_32
+	bool "32 bit Physical Address"
+	depends on ARM_32
+	default n
+	---help---
+
+	  Support 32 bit physical addresses.

The description is a bit misleading. If you select N, then you can still still boot on HW supporting only 32-bit physical address.

It is only not clear from the description why a user may want to select it.

From an external interface PoV, I think it would be better if we let the user decide how much physical address bits they want Xen to support.

In the Kconfig, this would translate as a "choice". For Arm64, there will only be one (48 bits) where-as Arm32 there would be two (32, 40).

For an internal interface PoV, this could still translate to select ARM_PA_32 (or whichever name we decide) to indicate the type of paddr_t.

Quoted 29 lines
xen/arch/arm/Kconfig +2 -0
+	  If unsure, say N
+
  config GICV3
  	bool "GICv3 driver"
  	depends on !NEW_VGIC
xen/arch/arm/include/asm/page-bits.h +2 -0
index 5d6477e599..8f4dcebcfd 100644
@@ -5,6 +5,8 @@
 
  #ifdef CONFIG_ARM_64
  #define PADDR_BITS              48
+#elif CONFIG_ARM_PA_32
+#define PADDR_BITS              32
  #else
  #define PADDR_BITS              40
  #endif
xen/arch/arm/include/asm/types.h +3 -0
index 083acbd151..f9595b9098 100644
@@ -37,9 +37,16 @@ typedef signed long long s64;
  typedef unsigned long long u64;
  typedef u32 vaddr_t;
  #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 52

Quoted 10 lines
xen/arch/arm/include/asm/types.h +4 -0
+#define PADDR_SHIFT BITS_PER_LONG
+#define PRIpaddr PRIx32
+#else
  typedef u64 paddr_t;
  #define INVALID_PADDR (~0ULL)
  #define PRIpaddr "016llx"
+#endif
  typedef u32 register_t;
  #define PRIregister "08x"
  #elif defined (CONFIG_ARM_64)

Cheers,

Signature · 1 line
Julien Grall
#49 lore raw

Re: [XEN v2 09/11] xen/arm: Introduce ARM_PA_32 to support 32 bit physical address

In reply to Ayan Kumar Halder

Hi,

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 43

Quoted 12 lines · Ayan Kumar Halder
xen/arch/arm/include/asm/types.h +4 -0
index 083acbd151..f9595b9098 100644
@@ -37,9 +37,16 @@ typedef signed long long s64;
  typedef unsigned long long u64;
  typedef u32 vaddr_t;
  #define PRIvaddr PRIx32
+#if defined(CONFIG_ARM_PA_32)
+typedef u32 paddr_t;
+#define INVALID_PADDR (~0UL)
+#define PADDR_SHIFT BITS_PER_LONG

You define PADDR_SHIFT but don't seem to use it. Is it intended?

xen/arch/arm/include/asm/types.h typedef signed long long s64; · line 52

Quoted 9 lines
xen/arch/arm/include/asm/types.h +3 -0
+#define PRIpaddr PRIx32
+#else
  typedef u64 paddr_t;
  #define INVALID_PADDR (~0ULL)
  #define PRIpaddr "016llx"
+#endif
  typedef u32 register_t;
  #define PRIregister "08x"
  #elif defined (CONFIG_ARM_64)

Cheers,

Signature · 1 line
Julien Grall
#50 lore raw

PATCH v2 10/11 xen/arm: Restrict zeroeth_table_offset for ARM_64

In reply to Ayan Kumar Halder

zeroeth_table_offset is not accessed by ARM_32.
Also, when 32 bit physical addresses are used (ie ARM_PA_32=y), this causes an overflow.

---
Changes from -

v1 - Removed the duplicate declaration for DECLARE_OFFSETS.

 xen/arch/arm/include/asm/lpae.h | 4 ++++
 xen/arch/arm/mm.c               | 7 +------
 2 files changed, 5 insertions(+), 6 deletions(-)
xen/arch/arm/include/asm/lpae.h +4 -0
index 3fdd5d0de2..2744e0eebf 100644
@@ -259,7 +259,11 @@ lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr);
 #define first_table_offset(va)  TABLE_OFFSET(first_linear_offset(va))
 #define second_table_offset(va) TABLE_OFFSET(second_linear_offset(va))
 #define third_table_offset(va)  TABLE_OFFSET(third_linear_offset(va))
+#ifdef CONFIG_ARM_64
 #define zeroeth_table_offset(va)  TABLE_OFFSET(zeroeth_linear_offset(va))
+#else
+#define zeroeth_table_offset(va)  0
+#endif
 
 /*
  * Macros to define page-tables:
xen/arch/arm/mm.c +1 -6
index fab54618ab..95784e0c59 100644
@@ -207,12 +207,7 @@ void dump_pt_walk(paddr_t ttbr, paddr_t addr,
 {
     static const char *level_strs[4] = { "0TH", "1ST", "2ND", "3RD" };
     const mfn_t root_mfn = maddr_to_mfn(ttbr);
-    const unsigned int offsets[4] = {
-        zeroeth_table_offset(addr),
-        first_table_offset(addr),
-        second_table_offset(addr),
-        third_table_offset(addr)
-    };
+    DECLARE_OFFSETS(offsets, addr);
     lpae_t pte, *mapping;
     unsigned int level, root_table;
 
Signature · 1 line
2.17.1
#51 lore raw

Re: [XEN v2 10/11] xen/arm: Restrict zeroeth_table_offset for ARM_64

Quoted 19 lines · Ayan Kumar Halder

zeroeth_table_offset is not accessed by ARM_32.
Also, when 32 bit physical addresses are used (ie ARM_PA_32=y), this causes an overflow.

---
Changes from -

v1 - Removed the duplicate declaration for DECLARE_OFFSETS.

 xen/arch/arm/include/asm/lpae.h | 4 ++++
 xen/arch/arm/mm.c               | 7 +------
 2 files changed, 5 insertions(+), 6 deletions(-)
xen/arch/arm/include/asm/lpae.h +0 -0
index 3fdd5d0de2..2744e0eebf 100644

xen/arch/arm/include/asm/lpae.h lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr); · line 262

Quoted 4 lines
xen/arch/arm/include/asm/lpae.h +1 -0
@@ -259,7 +259,11 @@ lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr);
 #define first_table_offset(va)  TABLE_OFFSET(first_linear_offset(va))
 #define second_table_offset(va) TABLE_OFFSET(second_linear_offset(va))
 #define third_table_offset(va)  TABLE_OFFSET(third_linear_offset(va))
+#ifdef CONFIG_ARM_64

Julien was asking for a selectable Kconfig option that would allow us to have 32-bit paddr_t even on ARM_64. If we do that, assuming we are on aarch64, and we set VTCR_T0SZ to 0x20, hence we get 32-bit IPA, are we going to have a 3-level or a 4-level p2m pagetable?

In any case I think this should be:
#ifndef CONFIG_PADDR_32

And if it doesn't work today on aarch64 due to pagetable levels or other reasons, than I would make CONFIG_PADDR_32 not (yet) selectable on ARM_64 (until it is fixed).

xen/arch/arm/include/asm/lpae.h lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr); · line 265

Quoted 3 lines
xen/arch/arm/include/asm/lpae.h +2 -0
 #define zeroeth_table_offset(va)  TABLE_OFFSET(zeroeth_linear_offset(va))
+#else
+#define zeroeth_table_offset(va)  0

Rather than 0 it might be better to have 32, hence zeroing the input address

Quoted 25 lines
xen/arch/arm/include/asm/lpae.h +1 -0
+#endif
 
 /*
  * Macros to define page-tables:
xen/arch/arm/mm.c +1 -6
index fab54618ab..95784e0c59 100644
@@ -207,12 +207,7 @@ void dump_pt_walk(paddr_t ttbr, paddr_t addr,
 {
     static const char *level_strs[4] = { "0TH", "1ST", "2ND", "3RD" };
     const mfn_t root_mfn = maddr_to_mfn(ttbr);
-    const unsigned int offsets[4] = {
-        zeroeth_table_offset(addr),
-        first_table_offset(addr),
-        second_table_offset(addr),
-        third_table_offset(addr)
-    };
+    DECLARE_OFFSETS(offsets, addr);
     lpae_t pte, *mapping;
     unsigned int level, root_table;
 
Signature · 1 line
2.17.1
#52 lore raw

Re: [XEN v2 10/11] xen/arm: Restrict zeroeth_table_offset for ARM_64

Hi,

Title: For me "restrict" means that the code macro cannot be used if !ARM_64. But this is not the case here.

Quoted 2 lines · Stefano Stabellini
Quoted 1 line · Ayan Kumar Halder

zeroeth_table_offset is not accessed by ARM_32.

I don't quite understand this sentence. The helper is used by 32-bit arm. The output may not be used thought.

I would suggest to say that there no zeroeth level on Arm 32-bit. But...

Quoted 2 lines
Quoted 2 lines

Also, when 32 bit physical addresses are used (ie ARM_PA_32=y), this causes an overflow.

... this is the most important part.

Quoted 11 lines
Quoted 11 lines

---
Changes from -

v1 - Removed the duplicate declaration for DECLARE_OFFSETS.

  xen/arch/arm/include/asm/lpae.h | 4 ++++
  xen/arch/arm/mm.c               | 7 +------
  2 files changed, 5 insertions(+), 6 deletions(-)
Quoted 14 lines

xen/arch/arm/include/asm/lpae.h lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr); · line 262

Quoted 9 lines
xen/arch/arm/include/asm/lpae.h +1 -0
index 3fdd5d0de2..2744e0eebf 100644
@@ -259,7 +259,11 @@ lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr);
  #define first_table_offset(va)  TABLE_OFFSET(first_linear_offset(va))
  #define second_table_offset(va) TABLE_OFFSET(second_linear_offset(va))
  #define third_table_offset(va)  TABLE_OFFSET(third_linear_offset(va))
+#ifdef CONFIG_ARM_64

Julien was asking for a selectable Kconfig option that would allow us to have 32-bit paddr_t even on ARM_64. If we do that, assuming we are on aarch64, and we set VTCR_T0SZ to 0x20, hence we get 32-bit IPA, are we going to have a 3-level or a 4-level p2m pagetable?

It will start at level 1. So 3-level page-table.

Quoted 3 lines

In any case I think this should be:
#ifndef CONFIG_PADDR_32

+1

Quoted 4 lines

And if it doesn't work today on aarch64 due to pagetable levels or other reasons, than I would make CONFIG_PADDR_32 not (yet) selectable on ARM_64 (until it is fixed).

+1

Quoted 6 lines

xen/arch/arm/include/asm/lpae.h lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr); · line 265

Quoted 3 lines
xen/arch/arm/include/asm/lpae.h +2 -0
  #define zeroeth_table_offset(va)  TABLE_OFFSET(zeroeth_linear_offset(va))
+#else
+#define zeroeth_table_offset(va)  0

Rather than 0 it might be better to have 32, hence zeroing the input address

I don't understand why you suggest 32. The macro is meant to return the index in the 0th table. So return 0 is correct here.

Quoted 13 lines
Quoted 13 lines
xen/arch/arm/include/asm/lpae.h +1 -0
+#endif
 
  /*
   * Macros to define page-tables:
xen/arch/arm/mm.c +0 -0
index fab54618ab..95784e0c59 100644
@@ -207,12 +207,7 @@ void dump_pt_walk(paddr_t ttbr, paddr_t addr,
  {
      static const char *level_strs[4] = { "0TH", "1ST", "2ND", "3RD" };
Quoted 8 lines

xen/arch/arm/mm.c void dump_pt_walk(paddr_t ttbr, paddr_t addr, · line 210

Quoted 8 lines
xen/arch/arm/mm.c +1 -6
      const mfn_t root_mfn = maddr_to_mfn(ttbr);
-    const unsigned int offsets[4] = {
-        zeroeth_table_offset(addr),
-        first_table_offset(addr),
-        second_table_offset(addr),
-        third_table_offset(addr)
-    };
+    DECLARE_OFFSETS(offsets, addr);

This wants to be explained in the commit message.

Quoted 6 lines
Quoted 6 lines
xen/arch/arm/mm.c +0 -0
      lpae_t pte, *mapping;
      unsigned int level, root_table;
 
Signature · 1 line
2.17.1

Cheers,

Signature · 1 line
Julien Grall
#53 lore raw

Re: [XEN v2 10/11] xen/arm: Restrict zeroeth_table_offset for ARM_64

Quoted 9 lines · Julien Grall
Quoted 7 lines

xen/arch/arm/include/asm/lpae.h lpae_t mfn_to_xen_entry(mfn_t mfn, unsigned int attr); · line 265

Quoted 4 lines

#define zeroeth_table_offset(va)
TABLE_OFFSET(zeroeth_linear_offset(va))

xen/arch/arm/include/asm/lpae.h +2 -0
+#else
+#define zeroeth_table_offset(va)  0

Rather than 0 it might be better to have 32, hence zeroing the input address

I don't understand why you suggest 32. The macro is meant to return the index in the 0th table. So return 0 is correct here.

This suggestion was a mistake, 0 is fine.

#54 lore raw

PATCH v2 11/11 xen/arm: p2m: Enable support for 32bit IPA

In reply to Ayan Kumar Halder

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

 xen/arch/arm/p2m.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +7 -3
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
     register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
 #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
         panic("P2M: Not able to support %u-bit IPA at the moment\n",
               p2m_ipa_bits);
 
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
     val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif
     val |= VTCR_SL0(0x1); /* P2M starts at first level */
 #else /* CONFIG_ARM_64 */
     static const struct {
Signature · 1 line
2.17.1
Acked-by on v2 11/11
#55 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Quoted 5 lines · Ayan Kumar Halder

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

Quoted 36 lines

---
Changes from -

v1 - New patch.

 xen/arch/arm/p2m.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +7 -3
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
     register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
 #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
         panic("P2M: Not able to support %u-bit IPA at the moment\n",
               p2m_ipa_bits);
 
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
     val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif
     val |= VTCR_SL0(0x1); /* P2M starts at first level */
 #else /* CONFIG_ARM_64 */
     static const struct {
Signature · 1 line
2.17.1
#56 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

In reply to Ayan Kumar Halder

Hi Ayan,

Quoted 25 lines · Ayan Kumar Halder

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---
  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +1 -1
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
      register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
  #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
          panic("P2M: Not able to support %u-bit IPA at the moment\n",
                p2m_ipa_bits);

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2279

Quoted 10 lines
xen/arch/arm/p2m.c +6 -2
 
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Cheers,

Signature · 1 line
Julien Grall
#57 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Quoted 1 line · Julien Grall

Hi Ayan,

Hi Julien,

Quoted 29 lines · Ayan Kumar Halder
Quoted 27 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +2 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
Quoted 8 lines
Quoted 6 lines
xen/arch/arm/p2m.c +4 -0
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

- Ayan

Quoted 3 lines

Cheers,

#58 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Hi,

Quoted 42 lines · Ayan Kumar Halder
Quoted 1 line · Julien Grall

Hi Ayan,

Hi Julien,

Quoted 29 lines · Ayan Kumar Halder
Quoted 27 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +2 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
Quoted 8 lines
Quoted 6 lines
xen/arch/arm/p2m.c +4 -0
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Quoted 1 line

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

It is using both the ID_AA64MMFR0_EL1 but also p2m_ipa_bits to decide the size.

Quoted 2 lines

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

I looked at the Arm Arm and indeed it doesn't look like there are equivalent for ID_AA64MMFR0_EL1.PARange.

However, my point was less about reading the system register but more about the fact we could have the code a bit more generic and avoid the assumption that PADDR_BITS is only modified when CONFIG_ARM_PA_32 is set.

Cheers,

Signature · 1 line
Julien Grall
#59 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Quoted 1 line · Julien Grall

Hi,

Hi Julien,

Quoted 44 lines · Ayan Kumar Halder
Quoted 42 lines · Julien Grall
Quoted 1 line

Hi Ayan,

Hi Julien,

Quoted 37 lines · Ayan Kumar Halder
Quoted 31 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +6 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
Quoted 2 lines

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Quoted 3 lines
Quoted 1 line

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

It is using both the ID_AA64MMFR0_EL1 but also p2m_ipa_bits to decide the size.

Ack.

Quoted 7 lines
Quoted 3 lines

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

I looked at the Arm Arm and indeed it doesn't look like there are equivalent for ID_AA64MMFR0_EL1.PARange.

However, my point was less about reading the system register but more about the fact we could have the code a bit more generic and avoid the assumption that PADDR_BITS is only modified when CONFIG_ARM_PA_32 is set.

I had a rework at the patch. Please let me know if the following looks better.

xen/arch/arm/p2m.c +0 -0
index 948f199d84..bc3bdf5f3e 100644
@@ -2266,14 +2266,35 @@ void __init setup_virt_paging(void)

     register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;

 #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    static const struct {
+        unsigned int pabits; /* Physical Address Size */ +        unsigned int t0sz;   /* Desired T0SZ */
+        unsigned int sl0;    /* Desired SL0 */
+    } pa_range_info[] __initconst = {
+        [0] = { 32,     32,     1 },
+        [1] = { 40,     24,     1 },
+    };
+    int i = 0;
+
+    if ( p2m_ipa_bits < PADDR_BITS )
+        panic("P2M: Not able to support %u-bit IPA at the moment\n", +              p2m_ipa_bits);
+
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+
+    for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
+        if ( p2m_ipa_bits == pa_range_info[i].pabits )
+            break;
+
+    if ( i == ARRAY_SIZE(pa_range_info) )
         panic("P2M: Not able to support %u-bit IPA at the moment\n", p2m_ipa_bits);

-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;

-    val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
-    val |= VTCR_SL0(0x1); /* P2M starts at first level */
+    val |= VTCR_T0SZ(pa_range_info[i].t0sz);
+    val |= VTCR_SL0(pa_range_info[i].sl0);

+

- Ayan

Quoted 3 lines

Cheers,

#60 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Hi,

Quoted 67 lines · Ayan Kumar Halder
Quoted 1 line · Julien Grall

Hi,

Hi Julien,

Quoted 44 lines · Ayan Kumar Halder
Quoted 42 lines · Julien Grall
Quoted 1 line

Hi Ayan,

Hi Julien,

Quoted 37 lines · Ayan Kumar Halder
Quoted 31 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +6 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
Quoted 2 lines

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Quoted 3 lines
Quoted 1 line

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

It is using both the ID_AA64MMFR0_EL1 but also p2m_ipa_bits to decide the size.

Ack.

Quoted 7 lines
Quoted 3 lines

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

I looked at the Arm Arm and indeed it doesn't look like there are equivalent for ID_AA64MMFR0_EL1.PARange.

However, my point was less about reading the system register but more about the fact we could have the code a bit more generic and avoid the assumption that PADDR_BITS is only modified when CONFIG_ARM_PA_32 is set.

I had a rework at the patch. Please let me know if the following looks better.

xen/arch/arm/p2m.c +0 -0
index 948f199d84..bc3bdf5f3e 100644

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2273

Quoted 9 lines
xen/arch/arm/p2m.c +4 -1
@@ -2266,14 +2266,35 @@ void __init setup_virt_paging(void)
      register_t val = 
 VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
  #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    static const struct {
+        unsigned int pabits; /* Physical Address Size */
+        unsigned int t0sz;   /* Desired T0SZ */
+        unsigned int sl0;    /* Desired SL0 */

Hmmm... Why don't you include the root order? In fact...

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2276

Quoted 3 lines
xen/arch/arm/p2m.c +3 -0
+    } pa_range_info[] __initconst = {
+        [0] = { 32,     32,     1 },
+        [1] = { 40,     24,     1 },

... looking at the ARMv7 specification (see B3-1345 in ARM DDI 0406C.d), the root page-table is only concatenated for 40-bits.

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2282

Quoted 6 lines
xen/arch/arm/p2m.c +6 -0
+    };
+    int i = 0;
+
+    if ( p2m_ipa_bits < PADDR_BITS )
+        panic("P2M: Not able to support %u-bit IPA at the moment\n",
+              p2m_ipa_bits);

This check seems unnecessary now.

Quoted 9 lines
xen/arch/arm/p2m.c +9 -0
+
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+
+    for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
+        if ( p2m_ipa_bits == pa_range_info[i].pabits )
+            break;
+
+    if ( i == ARRAY_SIZE(pa_range_info) )

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2297

Quoted 10 lines
xen/arch/arm/p2m.c +3 -4
          panic("P2M: Not able to support %u-bit IPA at the moment\n",
                p2m_ipa_bits);
 
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
-    val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
-    val |= VTCR_SL0(0x1); /* P2M starts at first level */
+    val |= VTCR_T0SZ(pa_range_info[i].t0sz);
+    val |= VTCR_SL0(pa_range_info[i].sl0);
+

I think you can share a lot of code between arm64 and arm32. A diff below (not tested):

xen/arch/arm/include/asm/p2m.h +0 -6
index 91df922e1c9f..05f26780a29d 100644
@@ -14,16 +14,10 @@
 /* Holds the bit size of IPAs in p2m tables.  */
 extern unsigned int p2m_ipa_bits;
 
-#ifdef CONFIG_ARM_64
 extern unsigned int p2m_root_order;
 extern unsigned int p2m_root_level;
 #define P2M_ROOT_ORDER    p2m_root_order
 #define P2M_ROOT_LEVEL p2m_root_level
-#else
-/* First level P2M is always 2 consecutive pages */
-#define P2M_ROOT_ORDER    1
-#define P2M_ROOT_LEVEL 1
-#endif
 
 struct domain;
 
xen/arch/arm/p2m.c +0 -0
index 948f199d848f..6fda0b92230a 100644
@@ -2264,17 +2264,6 @@ void __init setup_virt_paging(void)
 {
     /* Setup Stage 2 address translation */
     register_t val = 

VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA; -
-#ifdef CONFIG_ARM_32
- if ( p2m_ipa_bits < 40 )
- panic("P2M: Not able to support %u-bit IPA at the moment\n",
- p2m_ipa_bits);
-

-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
-    val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
-    val |= VTCR_SL0(0x1); /* P2M starts at first level */

-#else /* CONFIG_ARM_64 */
static const struct {
unsigned int pabits; /* Physical Address Size */
unsigned int t0sz; /* Desired T0SZ, minimum in comment */
@@ -2286,29 +2275,26 @@ void __init setup_virt_paging(void)

         [0] = { 32,      32/*32*/,  0,          1 },
         [1] = { 36,      28/*28*/,  0,          1 },
         [2] = { 40,      24/*24*/,  1,          1 },
+#ifdef CONFIG_ARM_64
         [3] = { 42,      22/*22*/,  3,          1 },
         [4] = { 44,      20/*20*/,  0,          2 },
         [5] = { 48,      16/*16*/,  0,          2 },
         [6] = { 52,      12/*12*/,  4,          2 },
         [7] = { 0 }  /* Invalid */
+#endif
     };

     unsigned int i;
     unsigned int pa_range = 0x10; /* Larger than any possible value */

+#ifdef CONFIG_ARM_64
/*
* Restrict "p2m_ipa_bits" if needed. As P2M table is always
configured
* with IPA bits == PA bits, compare against "pabits". */
if ( pa_range_info[system_cpuinfo.mm64.pa_range].pabits <
p2m_ipa_bits )

         p2m_ipa_bits = pa_range_info[system_cpuinfo.mm64.pa_range].pabits;
-
-    /*
-     * cpu info sanitization made sure we support 16bits VMID only if all
-     * cores are supporting it.
-     */

- if ( system_cpuinfo.mm64.vmid_bits == MM64_VMID_16_BITS_SUPPORT )
- max_vmid = MAX_VMID_16_BIT;
+#endif

     /* Choose suitable "pa_range" according to the resulted 
"p2m_ipa_bits". */
     for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
@@ -2324,12 +2310,22 @@ void __init setup_virt_paging(void)
     if ( pa_range >= ARRAY_SIZE(pa_range_info) || 

!pa_range_info[pa_range].pabits )
panic("Unknown encoding of ID_AA64MMFR0_EL1.PARange %x\n",
pa_range);

-    val |= VTCR_PS(pa_range);
-    val |= VTCR_TG0_4K;
+#ifdef CONFIG_ARM_64
+    /*
+     * cpu info sanitization made sure we support 16bits VMID only if all
+     * cores are supporting it.
+     */

+ if ( system_cpuinfo.mm64.vmid_bits == MM64_VMID_16_BITS_SUPPORT )
+ max_vmid = MAX_VMID_16_BIT;

     /* Set the VS bit only if 16 bit VMID is supported. */
     if ( MAX_VMID == MAX_VMID_16_BIT )
         val |= VTCR_VS;
+#endif
+
+    val |= VTCR_PS(pa_range);
+    val |= VTCR_TG0_4K;
+
     val |= VTCR_SL0(pa_range_info[pa_range].sl0);
     val |= VTCR_T0SZ(pa_range_info[pa_range].t0sz);

@@ -2341,7 +2337,7 @@ void __init setup_virt_paging(void)

            p2m_ipa_bits,
            pa_range_info[pa_range].pabits,
            ( MAX_VMID == MAX_VMID_16_BIT ) ? 16 : 8);
-#endif
+
     printk("P2M: %d levels with order-%d root, VTCR 0x%"PRIregister"\n",
            4 - P2M_ROOT_LEVEL, P2M_ROOT_ORDER, val);

Cheers,

Signature · 1 line
Julien Grall
#61 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Quoted 1 line · Julien Grall

Hi,

Hi Julien,

I need some inputs.

Quoted 80 lines · Ayan Kumar Halder
Quoted 67 lines · Julien Grall
Quoted 1 line

Hi,

Hi Julien,

Quoted 44 lines · Ayan Kumar Halder
Quoted 42 lines · Julien Grall
Quoted 1 line

Hi Ayan,

Hi Julien,

Quoted 37 lines · Ayan Kumar Halder
Quoted 31 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +6 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
Quoted 2 lines

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Quoted 3 lines
Quoted 1 line

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

It is using both the ID_AA64MMFR0_EL1 but also p2m_ipa_bits to decide the size.

Ack.

Quoted 7 lines
Quoted 3 lines

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

I looked at the Arm Arm and indeed it doesn't look like there are equivalent for ID_AA64MMFR0_EL1.PARange.

However, my point was less about reading the system register but more about the fact we could have the code a bit more generic and avoid the assumption that PADDR_BITS is only modified when CONFIG_ARM_PA_32 is set.

I had a rework at the patch. Please let me know if the following looks better.

xen/arch/arm/p2m.c +0 -0
index 948f199d84..bc3bdf5f3e 100644

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2272

Quoted 8 lines
xen/arch/arm/p2m.c +4 -1
@@ -2266,14 +2266,35 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
   #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    static const struct {
+        unsigned int pabits; /* Physical Address Size */
+        unsigned int t0sz;   /* Desired T0SZ */
+        unsigned int sl0;    /* Desired SL0 */

Hmmm... Why don't you include the root order? In fact...

Quoted 5 lines

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2275

Quoted 3 lines
xen/arch/arm/p2m.c +3 -0
+    } pa_range_info[] __initconst = {
+        [0] = { 32,     32,     1 },
+        [1] = { 40,     24,     1 },

... looking at the ARMv7 specification (see B3-1345 in ARM DDI 0406C.d), the root page-table is only concatenated for 40-bits.

Sorry, I could not follow how you inferred this. Can you paste the relevant snippet ?

Quoted 9 lines

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2281

Quoted 7 lines
xen/arch/arm/p2m.c +6 -0
+    };
+    int i = 0;
+
+    if ( p2m_ipa_bits < PADDR_BITS )
+        panic("P2M: Not able to support %u-bit IPA at the moment\n",
+              p2m_ipa_bits);

This check seems unnecessary now.

We still need this check as arm_smmu_device_cfg_probe() invokes p2m_restrict_ipa_bits(size).

And referARM IHI 0062D.cID070116 (SMMU arch version 2.0 Specification), the IPA address size can be

0.

0b0000 32-bit.

1.

0b0001 36-bit.

10.

0b0010 40-bit.

11.

0b0011 42-bit.

100.

0b0100 44-bit.

101.

0b0101 48-bit.

So if p2m_ipa_bits = 36 bits and PADDR_BITS = 40, then we want to panic.

- Ayan

Quoted 138 lines
Quoted 10 lines
xen/arch/arm/p2m.c +9 -0
+
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+
+    for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )
+        if ( p2m_ipa_bits == pa_range_info[i].pabits )
+            break;
+
+    if ( i == ARRAY_SIZE(pa_range_info) )

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2296

Quoted 10 lines
xen/arch/arm/p2m.c +3 -4
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
 
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
-    val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
-    val |= VTCR_SL0(0x1); /* P2M starts at first level */
+    val |= VTCR_T0SZ(pa_range_info[i].t0sz);
+    val |= VTCR_SL0(pa_range_info[i].sl0);
+

I think you can share a lot of code between arm64 and arm32. A diff below (not tested):

xen/arch/arm/include/asm/p2m.h +0 -6
index 91df922e1c9f..05f26780a29d 100644
@@ -14,16 +14,10 @@
  /* Holds the bit size of IPAs in p2m tables.  */
  extern unsigned int p2m_ipa_bits;
 
-#ifdef CONFIG_ARM_64
  extern unsigned int p2m_root_order;
  extern unsigned int p2m_root_level;
  #define P2M_ROOT_ORDER    p2m_root_order
  #define P2M_ROOT_LEVEL p2m_root_level
-#else
-/* First level P2M is always 2 consecutive pages */
-#define P2M_ROOT_ORDER    1
-#define P2M_ROOT_LEVEL 1
-#endif
 
  struct domain;
 
xen/arch/arm/p2m.c +0 -11
index 948f199d848f..6fda0b92230a 100644
@@ -2264,17 +2264,6 @@ void __init setup_virt_paging(void)
  {
      /* Setup Stage 2 address translation */
      register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
-
-#ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
-        panic("P2M: Not able to support %u-bit IPA at the moment\n",
-              p2m_ipa_bits);
-
-    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
-    val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
-    val |= VTCR_SL0(0x1); /* P2M starts at first level */
-#else /* CONFIG_ARM_64 */
      static const struct {

         unsigned int pabits; /* Physical Address Size */ unsigned int t0sz;   /* Desired T0SZ, minimum in comment */

xen/arch/arm/p2m.c +1 -0
@@ -2286,29 +2275,26 @@ void __init setup_virt_paging(void)
          [0] = { 32,      32/*32*/,  0,          1 },
          [1] = { 36,      28/*28*/,  0,          1 },
          [2] = { 40,      24/*24*/,  1,          1 },
+#ifdef CONFIG_ARM_64
          [3] = { 42,      22/*22*/,  3,          1 },
          [4] = { 44,      20/*20*/,  0,          2 },
          [5] = { 48,      16/*16*/,  0,          2 },
          [6] = { 52,      12/*12*/,  4,          2 },

         [7] = { 0 }  /* Invalid */
+#endif
     };

     unsigned int i;
     unsigned int pa_range = 0x10; /* Larger than any possible value */

+#ifdef CONFIG_ARM_64
     /*
      * Restrict "p2m_ipa_bits" if needed. As P2M table is always configured
      * with IPA bits == PA bits, compare against "pabits".
      */
     if ( pa_range_info[system_cpuinfo.mm64.pa_range].pabits < p2m_ipa_bits ) p2m_ipa_bits = pa_range_info[system_cpuinfo.mm64.pa_range].pabits; -
-    /*
-     * cpu info sanitization made sure we support 16bits VMID only if all
-     * cores are supporting it.
-     */
-    if ( system_cpuinfo.mm64.vmid_bits == MM64_VMID_16_BITS_SUPPORT ) -        max_vmid = MAX_VMID_16_BIT;
+#endif

     /* Choose suitable "pa_range" according to the resulted "p2m_ipa_bits". */ for ( i = 0; i < ARRAY_SIZE(pa_range_info); i++ )

xen/arch/arm/p2m.c +3 -2
@@ -2324,12 +2310,22 @@ void __init setup_virt_paging(void)
      if ( pa_range >= ARRAY_SIZE(pa_range_info) || !pa_range_info[pa_range].pabits )
          panic("Unknown encoding of ID_AA64MMFR0_EL1.PARange %x\n", pa_range);
 
-    val |= VTCR_PS(pa_range);
-    val |= VTCR_TG0_4K;
+#ifdef CONFIG_ARM_64
+    /*
+     * cpu info sanitization made sure we support 16bits VMID only if 

all
+     * cores are supporting it.
+     */
+    if ( system_cpuinfo.mm64.vmid_bits == MM64_VMID_16_BITS_SUPPORT ) +        max_vmid = MAX_VMID_16_BIT;

     /* Set the VS bit only if 16 bit VMID is supported. */ if ( MAX_VMID == MAX_VMID_16_BIT )
         val |= VTCR_VS;

xen/arch/arm/p2m.c +5 -0
+#endif
+
+    val |= VTCR_PS(pa_range);
+    val |= VTCR_TG0_4K;
+

     val |= VTCR_SL0(pa_range_info[pa_range].sl0); val |= VTCR_T0SZ(pa_range_info[pa_range].t0sz);

xen/arch/arm/p2m.c +1 -1
@@ -2341,7 +2337,7 @@ void __init setup_virt_paging(void)
             p2m_ipa_bits,
             pa_range_info[pa_range].pabits,
             ( MAX_VMID == MAX_VMID_16_BIT ) ? 16 : 8);
-#endif
+
      printk("P2M: %d levels with order-%d root, VTCR 0x%"PRIregister"\n",
             4 - P2M_ROOT_LEVEL, P2M_ROOT_ORDER, val);
 

Cheers,

#62 lore raw

Re: [XEN v2 11/11] xen/arm: p2m: Enable support for 32bit IPA

Quoted 82 lines · Ayan Kumar Halder
Quoted 80 lines · Julien Grall

Hi,

Quoted 67 lines · Ayan Kumar Halder
Quoted 1 line · Julien Grall

Hi,

Hi Julien,

Quoted 44 lines · Ayan Kumar Halder
Quoted 42 lines · Julien Grall
Quoted 1 line

Hi Ayan,

Hi Julien,

Quoted 37 lines · Ayan Kumar Halder
Quoted 31 lines

VTCR.T0SZ should be set as 0x20 to support 32bit IPA.
Refer ARM DDI 0487I.a ID081822, G8-9824, G8.2.171, VTCR, "Virtualization Translation Control Register" for the bit descriptions.

---
Changes from -

v1 - New patch.

  xen/arch/arm/p2m.c | 10 +++++++---

  1 file changed, 7 insertions(+), 3 deletions(-)
xen/arch/arm/p2m.c +6 -2
index 948f199d84..cfdea55e71 100644
@@ -2266,13 +2266,17 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
     #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    if ( p2m_ipa_bits < PADDR_BITS )
           panic("P2M: Not able to support %u-bit IPA at the moment\n",
                 p2m_ipa_bits);
   -    printk("P2M: 40-bit IPA\n");
-    p2m_ipa_bits = 40;
+    printk("P2M: %u-bit IPA\n",PADDR_BITS);
+    p2m_ipa_bits = PADDR_BITS;
+#ifdef CONFIG_ARM_PA_32
+    val |= VTCR_T0SZ(0x20); /* 32 bit IPA */
+#else
Quoted 2 lines

      val |= VTCR_T0SZ(0x18); /* 40 bit IPA */
+#endif

I am wondering whether this is right time to switch to an array like the arm64 code? This would allow to use 32-bit IPA also when Xen support 64-bit physical address.

Quoted 3 lines
Quoted 1 line

In AArch64, we use ID_AA64MMFR0_EL1.PARange to determine the physical address range supported at runtime. This is then used as an index into pa_range_info[] to determine t0sz, root_order, etc.

It is using both the ID_AA64MMFR0_EL1 but also p2m_ipa_bits to decide the size.

Ack.

Quoted 7 lines
Quoted 3 lines

However, for AArch32 I do not see an equivalent register (similar to ID_AA64MMFR0_EL1) or any register to determine the physical address range. Thus, I will prefer to keep the code as it is unless you suggest any alternative.

I looked at the Arm Arm and indeed it doesn't look like there are equivalent for ID_AA64MMFR0_EL1.PARange.

However, my point was less about reading the system register but more about the fact we could have the code a bit more generic and avoid the assumption that PADDR_BITS is only modified when CONFIG_ARM_PA_32 is set.

I had a rework at the patch. Please let me know if the following looks better.

xen/arch/arm/p2m.c +0 -0
index 948f199d84..bc3bdf5f3e 100644

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2272

Quoted 8 lines
xen/arch/arm/p2m.c +4 -1
@@ -2266,14 +2266,35 @@ void __init setup_virt_paging(void)
       register_t val = VTCR_RES1|VTCR_SH0_IS|VTCR_ORGN0_WBWA|VTCR_IRGN0_WBWA;
 
   #ifdef CONFIG_ARM_32
-    if ( p2m_ipa_bits < 40 )
+    static const struct {
+        unsigned int pabits; /* Physical Address Size */
+        unsigned int t0sz;   /* Desired T0SZ */
+        unsigned int sl0;    /* Desired SL0 */

Hmmm... Why don't you include the root order? In fact...

Quoted 6 lines
Quoted 5 lines

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2275

Quoted 3 lines
xen/arch/arm/p2m.c +3 -0
+    } pa_range_info[] __initconst = {
+        [0] = { 32,     32,     1 },
+        [1] = { 40,     24,     1 },

... looking at the ARMv7 specification (see B3-1345 in ARM DDI 0406C.d), the root page-table is only concatenated for 40-bits.

Sorry, I could not follow how you inferred this. Can you paste the relevant snippet ?

Use of concatenated second-level translation tables
A stage 2 translation with an input address range of 31-34 bits can start the translation either:
• With a first-level lookup, accessing a first-level translation table with 2-16 entries.
• With a second-level lookup, accessing a set of concatenated second-level translation tables.

Quoted 32 lines
Quoted 9 lines

xen/arch/arm/p2m.c void __init setup_virt_paging(void) · line 2281

Quoted 7 lines
xen/arch/arm/p2m.c +6 -0
+    };
+    int i = 0;
+
+    if ( p2m_ipa_bits < PADDR_BITS )
+        panic("P2M: Not able to support %u-bit IPA at the moment\n",
+              p2m_ipa_bits);

This check seems unnecessary now.

We still need this check as arm_smmu_device_cfg_probe() invokes p2m_restrict_ipa_bits(size).

And referARM IHI 0062D.cID070116 (SMMU arch version 2.0 Specification), the IPA address size can be

0.

   0b0000 32-bit.

1.

   0b0001 36-bit.

10.

   0b0010 40-bit.

11.

   0b0011 42-bit.

100.

Quoted 7 lines

   0b0100 44-bit.

101.

   0b0101 48-bit.

So if p2m_ipa_bits = 36 bits and PADDR_BITS = 40, then we want to panic.

We can have the same situation on Arm64 (PADDR_BITS = 48), yet we don't panic(). So I don't quite understand why we would need to differ...

Or are you saying that for 64-bit we also need such check? If so, then I am still not sure why because p2m_ipa_bits represent the guest physical address space and PADDR_BITS the Xen physical address space. It is valid to have both of them differing.

Cheers,

Signature · 1 line
Julien Grall
#63 lore raw
In reply to Ayan Kumar Halder

Hi Ayan,

Quoted 11 lines · Ayan Kumar Halder

Hi All,

Please have a look at https://lists.xenproject.org/archives/html/xen-devel/2022-11/msg01465.html
for the context.

The benefits of using 32 bit physical addresses are as follows :-

1. It helps to use Xen on platforms (for eg R52) which supports 32 bit physical addresses and has no support for large page address extension.

Looking at your entire series, you keep using "large page address extension". LPAE is ARMv7A thing and it is defined as "large *physical* address extension". So it would be good to stick to the proper naming.

~Michal