On Mon, Jul 15, 2024 at 5:33 PM Clément Léger <cleger@xxxxxxxxxxxx> wrote:
>
> MASK_VAL() was added a way to handle bit_offset and bit_width for
> registers located in system memory address space. However, while suited
> for reading, it does not work for writing and result in corrupted
> registers when writing values with bit_offset > 0. Moreover, when a
> register is collocated with another one at the same address but with a
> different mask, the current code results in the other registers being
> overwritten with 0s. The write procedure for SYSTEM_MEMORY registers
> should actually read the value, mask it, update it and write it with the
> updated value. Moreover, since registers can be located in the same
> word, we must take care of locking the access before doing it. We should
> potentially use a global lock since we don't know in if register
> addresses aren't shared with another _CPC package but better not
> encourage vendors to do so. Assume that registers can use the same word
> inside a _CPC package and thus, use a per _CPC package lock.
>
> Fixes: 2f4a4d63a193 ("ACPI: CPPC: Use access_width over bit_width for system memory accesses")
> Signed-off-by: Clément Léger <cleger@xxxxxxxxxxxx>
>
> ---
> drivers/acpi/cppc_acpi.c | 44 ++++++++++++++++++++++++++++++++++++----
> include/acpi/cppc_acpi.h | 2 ++
> 2 files changed, 42 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
> index 1d857978f5f4..2e99cf1842ee 100644
> --- a/drivers/acpi/cppc_acpi.c
> +++ b/drivers/acpi/cppc_acpi.c
> @@ -170,8 +170,11 @@ show_cppc_data(cppc_get_perf_ctrs, cppc_perf_fb_ctrs, wraparound_time);
> #define GET_BIT_WIDTH(reg) ((reg)->access_width ? (8 << ((reg)->access_width - 1)) : (reg)->bit_width)
>
> /* Shift and apply the mask for CPC reads/writes */
> -#define MASK_VAL(reg, val) (((val) >> (reg)->bit_offset) & \
> +#define MASK_VAL_READ(reg, val) (((val) >> (reg)->bit_offset) & \
> GENMASK(((reg)->bit_width) - 1, 0))
> +#define MASK_VAL_WRITE(reg, prev_val, val) \
> + ((((val) & GENMASK(((reg)->bit_width) - 1, 0)) << (reg)->bit_offset) | \
> + ((prev_val) & ~(GENMASK(((reg)->bit_width) - 1, 0) << (reg)->bit_offset))) \
>
> static ssize_t show_feedback_ctrs(struct kobject *kobj,
> struct kobj_attribute *attr, char *buf)
> @@ -857,6 +860,7 @@ int acpi_cppc_processor_probe(struct acpi_processor *pr)
>
> /* Store CPU Logical ID */
> cpc_ptr->cpu_id = pr->id;
> + spin_lock_init(&cpc_ptr->rmw_lock);
>
> /* Parse PSD data for this CPU */
> ret = acpi_get_psd(cpc_ptr, handle);
> @@ -1062,7 +1066,7 @@ static int cpc_read(int cpu, struct cpc_register_resource *reg_res, u64 *val)
> }
>
> if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
> - *val = MASK_VAL(reg, *val);
> + *val = MASK_VAL_READ(reg, *val);
>
> return 0;
> }
> @@ -1071,9 +1075,11 @@ static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
> {
> int ret_val = 0;
> int size;
> + u64 prev_val;
> void __iomem *vaddr = NULL;
> int pcc_ss_id = per_cpu(cpu_pcc_subspace_idx, cpu);
> struct cpc_reg *reg = ®_res->cpc_entry.reg;
> + struct cpc_desc *cpc_desc;
>
> size = GET_BIT_WIDTH(reg);
>
> @@ -1106,8 +1112,34 @@ static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
> return acpi_os_write_memory((acpi_physical_address)reg->address,
> val, size);
>
> - if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
> - val = MASK_VAL(reg, val);
> + if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY) {
> + cpc_desc = per_cpu(cpc_desc_ptr, cpu);
> + if (!cpc_desc) {
> + pr_debug("No CPC descriptor for CPU:%d\n", cpu);
> + return -ENODEV;
> + }
> +
> + spin_lock(&cpc_desc->rmw_lock);
> + switch (size) {
> + case 8:
> + prev_val = readb_relaxed(vaddr);
> + break;
> + case 16:
> + prev_val = readw_relaxed(vaddr);
> + break;
> + case 32:
> + prev_val = readl_relaxed(vaddr);
> + break;
> + case 64:
> + prev_val = readq_relaxed(vaddr);
> + break;
> + default:
> + ret_val = -EFAULT;
> + goto out_unlock;
I would do
spin_unlock(&cpc_desc->rmw_lock);
return -EFAUL;
here to avoid the check below which is redundant in this path and the
label would not be necessary then.
LGTM otherwise.
> + };
> + val = MASK_VAL_WRITE(reg, prev_val, val);
> + val |= prev_val;
> + }
>
> switch (size) {
> case 8:
> @@ -1134,6 +1166,10 @@ static int cpc_write(int cpu, struct cpc_register_resource *reg_res, u64 val)
> break;
> }
>
> +out_unlock:
> + if (reg->space_id == ACPI_ADR_SPACE_SYSTEM_MEMORY)
> + spin_unlock(&cpc_desc->rmw_lock);
> +
> return ret_val;
> }
>
> diff --git a/include/acpi/cppc_acpi.h b/include/acpi/cppc_acpi.h
> index 930b6afba6f4..e1720d930666 100644
> --- a/include/acpi/cppc_acpi.h
> +++ b/include/acpi/cppc_acpi.h
> @@ -64,6 +64,8 @@ struct cpc_desc {
> int cpu_id;
> int write_cmd_status;
> int write_cmd_id;
> + /* Lock used for RMW operations in cpc_write() */
> + spinlock_t rmw_lock;
> struct cpc_register_resource cpc_regs[MAX_CPC_REG_ENT];
> struct acpi_psd_package domain_info;
> struct kobject kobj;
> --
> 2.45.2
>
