[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <de6d9fa7-abc1-47a1-9935-b544a7ecd537@amd.com>
Date: Fri, 30 Aug 2024 01:18:29 +0700
From: "Suthikulpanit, Suravee" <suravee.suthikulpanit@....com>
To: Uros Bizjak <ubizjak@...il.com>, linux-kernel@...r.kernel.org,
iommu@...ts.linux.dev
Cc: joro@...tes.org, robin.murphy@....com, vasant.hegde@....com,
jgg@...dia.com, kevin.tian@...el.com, jon.grimm@....com,
santosh.shukla@....com, pandoh@...gle.com, kumaranand@...gle.com
Subject: Re: [PATCH] iommu/amd: Modify set_dte_entry() to use 128-bit cmpxchg
operation
On 8/20/2024 1:15 AM, Uros Bizjak wrote:
>
>
> On 19. 08. 24 18:18, Suravee Suthikulpanit wrote:
>> The current implementation does not follow the 128-bit write
>> requirement to update DTE as specified in the AMD I/O Virtualization
>> Techonology (IOMMU) Specification.
>>
>> In addition, the function is used to program several DTE fields
>> (e.g. stage1 table, stage2 table, domain id, and etc.), which is
>> difficult to keep track with current implementation.
>>
>> Therefore, introduce new a new dte256_t data type and a helper function
>> update_dte_256(), which uses two try_cmpxchg128 operations to update
>> 256-bit DTE.
>>
>> Also, separate logic for setting up the GCR3 Table Root Pointer, GIOV,
>> GV,
>> GLX, and GuestPagingMode into another helper function
>> set_dte_gcr3_table().
>>
>> Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@....com>
>
> [...]
>
>> +static void update_dte256(struct amd_iommu *iommu, u16 devid, struct
>> dte256 *new)
>> +{
>> + struct dev_table_entry *dev_table = get_dev_table(iommu);
>> + struct dte256 *ptr = (struct dte256 *)&dev_table[devid];
>> + struct dte256 old = {
>> + .qw_lo.data = ptr->qw_lo.data,
>> + .qw_hi.data = ptr->qw_hi.data,
>> + };
>> +
>> + /* Update qw_lo */
>> + if (!try_cmpxchg128(&ptr->qw_lo.data, &old.qw_lo.data,
>> new->qw_lo.data))
>> + goto err_out;
>> +
>> + /* Update qw_hi */
>> + if (!try_cmpxchg128(&ptr->qw_hi.data, &old.qw_hi.data,
>> new->qw_hi.data)) {
>> + /* Restore qw_lo */
>> + try_cmpxchg128(&ptr->qw_lo.data, &new->qw_lo.data,
>> old.qw_lo.data);
>
> You should use plain cmpxchg128() when the result is unused.
>
> Uros.
I will update in v2.
Thanks,
Suravee
Powered by blists - more mailing lists