[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <e7358027-6a8b-85b0-3ccd-cb09b3a9ce7d@amd.com>
Date: Fri, 21 Mar 2025 12:52:35 -0700
From: Lizhi Hou <lizhi.hou@....com>
To: Jeff Hugo <jeff.hugo@....qualcomm.com>, <ogabbay@...nel.org>,
<jacek.lawrynowicz@...ux.intel.com>, <mario.limonciello@....com>,
<dri-devel@...ts.freedesktop.org>
CC: <linux-kernel@...r.kernel.org>, <min.ma@....com>, <max.zhen@....com>,
<sonal.santan@....com>, <king.tam@....com>
Subject: Re: [PATCH V1] accel/amdxdna: Add BO import and export
On 3/21/25 08:15, Jeff Hugo wrote:
> On 3/6/2025 11:03 AM, Lizhi Hou wrote:
>> +struct drm_gem_object *
>> +amdxdna_gem_prime_import(struct drm_device *dev, struct dma_buf
>> *dma_buf)
>> +{
>> + struct dma_buf_attachment *attach;
>> + struct drm_gem_object *gobj;
>> + struct sg_table *sgt;
>> + int ret;
>> +
>> + attach = dma_buf_attach(dma_buf, dev->dev);
>> + if (IS_ERR(attach))
>> + return ERR_CAST(attach);
>> +
>> + get_dma_buf(dma_buf);
>> +
>> + sgt = dma_buf_map_attachment_unlocked(attach, DMA_BIDIRECTIONAL);
>> + if (IS_ERR(sgt)) {
>> + ret = PTR_ERR(sgt);
>> + goto fail_detach;
>> + }
>> +
>> + gobj = drm_gem_shmem_prime_import_sg_table(dev, attach, sgt);
>> + if (IS_ERR(gobj)) {
>> + ret = PTR_ERR(gobj);
>> + goto fail_unmap;
>> + }
>> +
>> + gobj->import_attach = attach;
>> + gobj->resv = dma_buf->resv;
>> +
>> + return gobj;
>> +
>> +fail_unmap:
>> + dma_buf_unmap_attachment_unlocked(attach, sgt, DMA_BIDIRECTIONAL);
>> +fail_detach:
>> + dma_buf_detach(dma_buf, attach);
>> + dma_buf_put(dma_buf);
>
> You attach() and then get(), so normal "reverse order" cleanup would
> be put(), then detach(). That is not what you do here. Should this be
> reordered, or should you get() then attach() first?
I referred drm_gem_prime_import_dev(). And I agree with you. It looks
better to get() before attach(). I will respin V2 which will also
contain another small update for this patch.
Thanks,
Lizhi
>
>
Powered by blists - more mailing lists