From: "Cédric Le Goater" <clg@redhat.com>
To: "Philippe Mathieu-Daudé" <philmd@linaro.org>, qemu-devel@nongnu.org
Cc: Alex Williamson <alex.williamson@redhat.com>,
Richard Henderson <richard.henderson@linaro.org>,
Paolo Bonzini <pbonzini@redhat.com>,
Eduardo Habkost <eduardo@habkost.net>,
Marcel Apfelbaum <marcel.apfelbaum@gmail.com>,
Yanan Wang <wangyanan55@huawei.com>,
Zhao Liu <zhao1.liu@intel.com>
Subject: Re: [PATCH v2 7/9] cpu: Introduce cpu_get_phys_bits()
Date: Thu, 6 Mar 2025 15:41:02 +0100 [thread overview]
Message-ID: <b73d5989-f5a8-4370-be6f-6f6bb609a38e@redhat.com> (raw)
In-Reply-To: <4dc45084-fb9d-4fed-88b0-18b0034c8c74@linaro.org>
On 3/6/25 11:37, Philippe Mathieu-Daudé wrote:
> On 30/1/25 14:43, Cédric Le Goater wrote:
>> The Intel CPU has a complex history regarding setting of the physical
>> address space width on KVM. A 'phys_bits' field and a "phys-bits"
>> property were added by commit af45907a1328 ("target-i386: Allow
>> physical address bits to be set") to tune this value.
>>
>> In certain circumstances, it is interesting to know this value to
>> check that all the conditions are met for optimal operation. For
>> instance, when the system has a 39-bit IOMMU address space width and a
>> larger CPU physical address space, we expect issues when mapping the
>> MMIO regions of passthrough devices and it would good to report to the
>> user. These hybrid HW configs can be found on some consumer grade
>> processors or when using a vIOMMU device with default settings.
>>
>> For this purpose, add an helper routine and a CPUClass callback to
>> return the physical address space width of a CPU.
>>
>> Cc: Richard Henderson <richard.henderson@linaro.org>
>> Cc: Paolo Bonzini <pbonzini@redhat.com>
>> Cc: Eduardo Habkost <eduardo@habkost.net>
>> Cc: Marcel Apfelbaum <marcel.apfelbaum@gmail.com>
>> Cc: "Philippe Mathieu-Daudé" <philmd@linaro.org>
>> Cc: Yanan Wang <wangyanan55@huawei.com>
>> Cc: Zhao Liu <zhao1.liu@intel.com>
>> Signed-off-by: Cédric Le Goater <clg@redhat.com>
>> ---
>> include/hw/core/cpu.h | 9 +++++++++
>> include/hw/core/sysemu-cpu-ops.h | 6 ++++++
>> cpu-target.c | 5 +++++
>> hw/core/cpu-system.c | 11 +++++++++++
>> target/i386/cpu.c | 6 ++++++
>> 5 files changed, 37 insertions(+)
>>
>> diff --git a/include/hw/core/cpu.h b/include/hw/core/cpu.h
>> index fb397cdfc53d12d40d3e4e7f86251fc31c48b9f6..1b3eead102ce62fcee55ab0ed5e0dff327fa2fc5 100644
>> --- a/include/hw/core/cpu.h
>> +++ b/include/hw/core/cpu.h
>> @@ -748,6 +748,14 @@ int cpu_asidx_from_attrs(CPUState *cpu, MemTxAttrs attrs);
>> */
>> bool cpu_virtio_is_big_endian(CPUState *cpu);
>> +/**
>> + * cpu_get_phys_bits:
>> + * @cpu: CPU
>> + *
>> + * Return the physical address space width of the CPU @cpu.
>> + */
>> +uint32_t cpu_get_phys_bits(const CPUState *cpu);
>> +
>> #endif /* CONFIG_USER_ONLY */
>> /**
>> @@ -1168,6 +1176,7 @@ void cpu_exec_unrealizefn(CPUState *cpu);
>> void cpu_exec_reset_hold(CPUState *cpu);
>> const char *target_name(void);
>> +uint32_t target_phys_bits(void);
>> #ifdef COMPILING_PER_TARGET
>> diff --git a/include/hw/core/sysemu-cpu-ops.h b/include/hw/core/sysemu-cpu-ops.h
>> index 0df5b058f50073e47d2a6b8286be5204776520d2..210b3ed57985525795b81559e41e0085969210d5 100644
>> --- a/include/hw/core/sysemu-cpu-ops.h
>> +++ b/include/hw/core/sysemu-cpu-ops.h
>> @@ -81,6 +81,12 @@ typedef struct SysemuCPUOps {
>> */
>> bool (*virtio_is_big_endian)(CPUState *cpu);
>> + /**
>> + * @get_phys_bits: Callback to return the physical address space
>> + * width of a CPU.
>> + */
>> + uint32_t (*get_phys_bits)(const CPUState *cpu);
>> +
>> /**
>> * @legacy_vmsd: Legacy state for migration.
>> * Do not use in new targets, use #DeviceClass::vmsd instead.
>> diff --git a/cpu-target.c b/cpu-target.c
>> index 667688332c929aa53782c94343def34571272d5f..88158272c06cc42424d435b9701e33735f080239 100644
>> --- a/cpu-target.c
>> +++ b/cpu-target.c
>> @@ -472,3 +472,8 @@ const char *target_name(void)
>> {
>> return TARGET_NAME;
>> }
>> +
>> +uint32_t target_phys_bits(void)
>> +{
>> + return TARGET_PHYS_ADDR_SPACE_BITS;
>> +}
>> diff --git a/hw/core/cpu-system.c b/hw/core/cpu-system.c
>> index 6aae28a349a7a377d010ff9dcab5ebc29e1126ca..05067d84f4258facf4252216f17729e390d38eae 100644
>> --- a/hw/core/cpu-system.c
>> +++ b/hw/core/cpu-system.c
>> @@ -60,6 +60,17 @@ hwaddr cpu_get_phys_page_attrs_debug(CPUState *cpu, vaddr addr,
>> return cc->sysemu_ops->get_phys_page_debug(cpu, addr);
>> }
>> +uint32_t cpu_get_phys_bits(const CPUState *cpu)
>> +{
>> + CPUClass *cc = CPU_GET_CLASS(cpu);
>> +
>> + if (cc->sysemu_ops->get_phys_bits) {
>> + return cc->sysemu_ops->get_phys_bits(cpu);
>> + }
>> +
>> + return target_phys_bits();
>
> I might have suggested to return TARGET_PHYS_ADDR_SPACE_BITS
> by default in v1, I'm not sure about it anymore. Maybe default
> to 0 and have each target register its helper if necessary?
sure. No problem. I can change that.
Thanks,
C.
>
>> +}
>> +
>> hwaddr cpu_get_phys_page_debug(CPUState *cpu, vaddr addr)
>> {
>> MemTxAttrs attrs = {};
>> diff --git a/target/i386/cpu.c b/target/i386/cpu.c
>> index b5dd60d2812e0c3d13c1743fd485a9068ab29c4f..01cf9a44963710a415c755c17582730f75233143 100644
>> --- a/target/i386/cpu.c
>> +++ b/target/i386/cpu.c
>> @@ -8393,6 +8393,11 @@ static bool x86_cpu_get_paging_enabled(const CPUState *cs)
>> return cpu->env.cr[0] & CR0_PG_MASK;
>> }
>> +
>> +static uint32_t x86_cpu_get_phys_bits(const CPUState *cs)
>> +{
>> + return X86_CPU(cs)->phys_bits;
>> +}
>> #endif /* !CONFIG_USER_ONLY */
>> static void x86_cpu_set_pc(CPUState *cs, vaddr value)
>> @@ -8701,6 +8706,7 @@ static const struct SysemuCPUOps i386_sysemu_ops = {
>> .get_memory_mapping = x86_cpu_get_memory_mapping,
>> .get_paging_enabled = x86_cpu_get_paging_enabled,
>> .get_phys_page_attrs_debug = x86_cpu_get_phys_page_attrs_debug,
>> + .get_phys_bits = x86_cpu_get_phys_bits,
>> .asidx_from_attrs = x86_asidx_from_attrs,
>> .get_crash_info = x86_cpu_get_crash_info,
>> .write_elf32_note = x86_cpu_write_elf32_note,
>
next prev parent reply other threads:[~2025-03-06 14:41 UTC|newest]
Thread overview: 35+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-01-30 13:43 [PATCH v2 0/9]vfio: Improve error reporting when MMIO region mapping fails Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 1/9] util/error: Introduce warn_report_once_err() Cédric Le Goater
2025-01-30 14:25 ` Markus Armbruster
2025-01-30 16:03 ` Cédric Le Goater
2025-01-30 16:28 ` Cédric Le Goater
2025-01-30 17:55 ` Alex Williamson
2025-01-30 21:26 ` Cédric Le Goater
2025-01-31 8:30 ` Markus Armbruster
2025-01-30 13:43 ` [PATCH v2 2/9] vfio/pci: Replace "iommu_device" by "vIOMMU" Cédric Le Goater
2025-02-10 14:28 ` Philippe Mathieu-Daudé
2025-01-30 13:43 ` [PATCH v2 3/9] vfio: Rephrase comment in vfio_listener_region_add() error path Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 4/9] vfio: Introduce vfio_get_vfio_device() Cédric Le Goater
2025-02-10 14:32 ` Philippe Mathieu-Daudé
2025-02-10 16:19 ` Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 5/9] vfio: Improve error reporting when MMIO region mapping fails Cédric Le Goater
2025-02-10 14:36 ` Philippe Mathieu-Daudé
2025-02-10 16:17 ` Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 6/9] vfio: Remove reports of DMA mapping errors in backends Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 7/9] cpu: Introduce cpu_get_phys_bits() Cédric Le Goater
2025-02-10 14:40 ` Philippe Mathieu-Daudé
2025-03-06 10:37 ` Philippe Mathieu-Daudé
2025-03-06 14:41 ` Cédric Le Goater [this message]
2025-01-30 13:43 ` [PATCH v2 8/9] vfio: Check compatibility of CPU and IOMMU address space width Cédric Le Goater
2025-01-30 18:58 ` Alex Williamson
2025-01-31 12:42 ` Cédric Le Goater
2025-01-31 13:23 ` Gerd Hoffmann
2025-01-31 17:03 ` Cédric Le Goater
2025-02-06 7:54 ` Gerd Hoffmann
2025-02-06 17:05 ` Cédric Le Goater
2025-01-31 22:18 ` Alex Williamson
2025-02-06 8:22 ` Gerd Hoffmann
2025-03-06 10:33 ` Philippe Mathieu-Daudé
2025-09-05 13:04 ` Daniel Kral
2025-09-08 8:35 ` Cédric Le Goater
2025-01-30 13:43 ` [PATCH v2 9/9] vfio: Remove superfluous error report in vfio_listener_region_add() Cédric Le Goater
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=b73d5989-f5a8-4370-be6f-6f6bb609a38e@redhat.com \
--to=clg@redhat.com \
--cc=alex.williamson@redhat.com \
--cc=eduardo@habkost.net \
--cc=marcel.apfelbaum@gmail.com \
--cc=pbonzini@redhat.com \
--cc=philmd@linaro.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
--cc=wangyanan55@huawei.com \
--cc=zhao1.liu@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).