Skip to contentSkip to conversation

drm/vmwgfx: Use dma-coherent memory for high-bandwidth port messaging

PATCH In review Thomas Hellström (VMware) 2 participants 6 messages Nov 13, 2019 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/vmwgfx: Use dma-coherent memory for high-bandwidth port messaging

From: Thomas Hellstrom <[email protected]>

With AMD-SEV high-bandwidth port messaging runs into trouble since the message content is encrypted using the vm-specific key, and the hypervisor is unable to read it.

So use unencrypted dma-coherent bounce buffers for temporary message storage space. Allocating that memory is expensive so a future optimization might include a static unencrypted memory area for messages.

---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.h |  4 +--
 drivers/gpu/drm/vmwgfx/vmwgfx_msg.c | 45 +++++++++++++++++------------
 3 files changed, 31 insertions(+), 22 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +2 -2
index 81a95651643f..fc0283659c41 100644
@@ -908,13 +908,13 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 
 	snprintf(host_log, sizeof(host_log), "vmwgfx: %s-%s",
 		VMWGFX_REPO, VMWGFX_GIT_VERSION);
-	vmw_host_log(host_log);
+	vmw_host_log(dev->dev, host_log);
 
 	memset(host_log, 0, sizeof(host_log));
 	snprintf(host_log, sizeof(host_log), "vmwgfx: Module Version: %d.%d.%d",
 		VMWGFX_DRIVER_MAJOR, VMWGFX_DRIVER_MINOR,
 		VMWGFX_DRIVER_PATCHLEVEL);
-	vmw_host_log(host_log);
+	vmw_host_log(dev->dev, host_log);
 
 	if (dev_priv->enable_fb) {
 		vmw_fifo_resource_inc(dev_priv);
drivers/gpu/drm/vmwgfx/vmwgfx_drv.h +2 -2
index b18842f73081..a77bf72cb9ac 100644
@@ -1389,9 +1389,9 @@ int vmw_bo_cpu_blit(struct ttm_buffer_object *dst,
 		    struct vmw_diff_cpy *diff);
 
 /* Host messaging -vmwgfx_msg.c: */
-int vmw_host_get_guestinfo(const char *guest_info_param,
+int vmw_host_get_guestinfo(struct device *dev, const char *guest_info_param,
 			   char *buffer, size_t *length);
-int vmw_host_log(const char *log);
+int vmw_host_log(struct device *dev, const char *log);
 
 /* VMW logging */
 
drivers/gpu/drm/vmwgfx/vmwgfx_msg.c +27 -18

152 lines not on this page · raw message

Signature · 6 lines
2.21.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 1/1 drm/vmwgfx: Use coherent memory if there are dma mapping size restrictions

From: Thomas Hellstrom <[email protected]>

We're gradually moving towards using DMA coherent memory in most situations, although TTM interactions with the DMA layers is still a work-in-progress. Meanwhile, use coherent memory when there are size restrictions meaning that there is a chance that streaming dma mapping of large buffer objects may fail.
Also move DMA mask settings to the vmw_dma_select_mode function, since it's important that we set the correct DMA masks before calling the dma_max_mapping_size() function.

---
 drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 31 +++++++----------------------
 1 file changed, 7 insertions(+), 24 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +7 -24
index fc0283659c41..1e1de83908fe 100644
@@ -569,7 +569,10 @@ static int vmw_dma_select_mode(struct vmw_private *dev_priv)
 		[vmw_dma_map_populate] = "Caching DMA mappings.",
 		[vmw_dma_map_bind] = "Giving up DMA mappings early."};
 
-	if (vmw_force_coherent)
+	(void) dma_set_mask_and_coherent(dev_priv->dev->dev, DMA_BIT_MASK(64));
+
+	if (vmw_force_coherent ||
+	    dma_max_mapping_size(dev_priv->dev->dev) != SIZE_MAX)
 		dev_priv->map_mode = vmw_dma_alloc_coherent;
 	else if (vmw_restrict_iommu)
 		dev_priv->map_mode = vmw_dma_map_bind;
@@ -582,30 +585,15 @@ static int vmw_dma_select_mode(struct vmw_private *dev_priv)
 		return -EINVAL;
 
 	DRM_INFO("DMA map mode: %s\n", names[dev_priv->map_mode]);
-	return 0;
-}
 
-/**
- * vmw_dma_masks - set required page- and dma masks
- *
- * @dev: Pointer to struct drm-device
- *
- * With 32-bit we can only handle 32 bit PFNs. Optionally set that
- * restriction also for 64-bit systems.
- */
-static int vmw_dma_masks(struct vmw_private *dev_priv)
-{
-	struct drm_device *dev = dev_priv->dev;
-	int ret = 0;
-
-	ret = dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64));
 	if (dev_priv->map_mode != vmw_dma_phys &&
 	    (sizeof(unsigned long) == 4 || vmw_restrict_dma_mask)) {
 		DRM_INFO("Restricting DMA addresses to 44 bits.\n");
-		return dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(44));
+		return dma_set_mask_and_coherent(dev_priv->dev->dev,
+						 DMA_BIT_MASK(44));
 	}
 
-	return ret;
+	return 0;
 }
 
 static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
@@ -674,7 +662,6 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 		dev_priv->capabilities2 = vmw_read(dev_priv, SVGA_REG_CAP2);
 	}
 
-
 	ret = vmw_dma_select_mode(dev_priv);
 	if (unlikely(ret != 0)) {
 		DRM_INFO("Restricting capabilities due to IOMMU setup.\n");
@@ -746,10 +733,6 @@ static int vmw_driver_load(struct drm_device *dev, unsigned long chipset)
 	if (dev_priv->capabilities & SVGA_CAP_CAP2_REGISTER)
 		vmw_print_capabilities2(dev_priv->capabilities2);
 
-	ret = vmw_dma_masks(dev_priv);
-	if (unlikely(ret != 0))
-		goto out_err0;
-
 	dma_set_max_seg_size(dev->dev, min_t(unsigned int, U32_MAX & PAGE_MASK,
 					     SCATTERLIST_MAX_SEGMENT));
 
Signature · 6 lines
2.21.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

A message that is not in the archive (sent off-list or not yet received): [email protected]

#3 lore raw

Re: [PATCH] drm/vmwgfx: Use coherent memory if there are dma mapping size restrictions

Hi,

Quoted 27 lines · Christoph Hellwig
Quoted 26 lines · Thomas Hellström (VMware)

From: Thomas Hellstrom <[email protected]>

We're gradually moving towards using DMA coherent memory in most situations, although TTM interactions with the DMA layers is still a work-in-progress. Meanwhile, use coherent memory when there are size restrictions meaning that there is a chance that streaming dma mapping of large buffer objects may fail.
Also move DMA mask settings to the vmw_dma_select_mode function, since it's important that we set the correct DMA masks before calling the dma_max_mapping_size() function.

---
  drivers/gpu/drm/vmwgfx/vmwgfx_drv.c | 31 +++++++----------------------
  1 file changed, 7 insertions(+), 24 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +0 -0
index fc0283659c41..1e1de83908fe 100644
@@ -569,7 +569,10 @@ static int vmw_dma_select_mode(struct vmw_private *dev_priv)
  		[vmw_dma_map_populate] = "Caching DMA mappings.",
  		[vmw_dma_map_bind] = "Giving up DMA mappings early."};
 
Quoted 5 lines

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c static int vmw_dma_select_mode(struct vmw_private *dev_priv) · line 572

Quoted 2 lines
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +1 -1
-	if (vmw_force_coherent)
+	(void) dma_set_mask_and_coherent(dev_priv->dev->dev, DMA_BIT_MASK(64));

Please don't use void cast on returns. Also this genercally can't fail, so if it fails anyway it is fatal, and you should actually return an error.

OK. Will fix.

Quoted 6 lines

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c static int vmw_dma_select_mode(struct vmw_private *dev_priv) · line 575

Quoted 3 lines
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +2 -0
+	if (vmw_force_coherent ||
+	    dma_max_mapping_size(dev_priv->dev->dev) != SIZE_MAX)

I don't think this is the right check to see if swiotlb would be used. You probably want to call dma_addressing_limited(). Please also add a comment explaining the call here.

If I understand things correctly, dma_addressing_limited() would always return false on vmwgfx, at least if the "restrict to 44 bit" option is removed. The "odd" modes we want to catch is someone setting swiotlb=force, or someone enabling SEV. In both cases, vmwgfx would break down due to limited DMA space even if streaming DMA was fixed with appropriate sync calls.

The same holds for vmw_pvscsi (which we discussed before) where the large queue depth will typically fill the SWIOTLB causing excessive non-fatal error logging. dma_max_mapping_size() currently catch these modes, but best I guess would be dma_swiotlb_forced() function that could be used in cases like this?

Quoted 6 lines

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c static int vmw_dma_select_mode(struct vmw_private *dev_priv) · line 589

Quoted 3 lines
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +0 -0
  	if (dev_priv->map_mode != vmw_dma_phys &&

AFAIK vmw_dma_phys can't even be set in current mainline and is dead code. Can you check if I'm missing something? Certainly all three branches above don't set it.

I'll remove that dead code for v2.

Quoted 12 lines

drivers/gpu/drm/vmwgfx/vmwgfx_drv.c static int vmw_dma_select_mode(struct vmw_private *dev_priv) · line 593

Quoted 6 lines
drivers/gpu/drm/vmwgfx/vmwgfx_drv.c +2 -1
  	    (sizeof(unsigned long) == 4 || vmw_restrict_dma_mask)) {
  		DRM_INFO("Restricting DMA addresses to 44 bits.\n");
-		return dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(44));
+		return dma_set_mask_and_coherent(dev_priv->dev->dev,
+						 DMA_BIT_MASK(44));

Can you keep setting the DMA mask together?

E.g. make the whole thing something like:

static int vmw_dma_select_mode(struct vmw_private *dev_priv)

Quoted 10 lines
{
	if (dma_addressing_limited(dev_priv->dev->dev) || vmw_force_coherent) {
		/*
		 * XXX: what about AMD iommu?  virtio-iommu?  Also
		 * swiotlb should be always available where we can
		 * address more than 32-bit of memory..
		 */
		if (!IS_ENABLED(CONFIG_SWIOTLB) &&
		    !IS_ENABLED(CONFIG_INTEL_IOMMU))
			return -EINVAL;

The above check is only to see if coherent memory functionality is really compiled in into TTM. We have a patchset that got lost in the last merge window to fix this properly and avoid confusion about this. I'll rebase on that.

Quoted 7 lines
		dev_priv->map_mode = vmw_dma_alloc_coherent;
	} else if (vmw_restrict_iommu) {
		dev_priv->map_mode = vmw_dma_map_bind;
	} else {
		dev_priv->map_mode = vmw_dma_map_populate;
	}
Quoted 8 lines
	/*
	 * On 32-bit systems we can only handle 32 bit PFNs. Optionally set
	 * that restriction also for 64-bit systems.
	 */
	return dma_set_mask_and_coherent(dev->dev,
			(IS_ENABLED(CONFIG_64BIT) && !vmw_restrict_dma_mask) ?
			64 : 44);
}

Thanks,

Thomas

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#4 lore raw

PATCH 1/1 drm/vmwgfx: remove set but not used variable 'srf'

In reply to Thomas Hellström (VMware)

From: YueHaibing <[email protected]>

drivers/gpu/drm/vmwgfx/vmwgfx_surface.c:339:22:
warning: variable srf set but not used [-Wunused-but-set-variable]

'srf' is never used, so can be removed.

---
 drivers/gpu/drm/vmwgfx/vmwgfx_surface.c | 2 --
 1 file changed, 2 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_surface.c +0 -2
index 29d8794f0421..de0530b4dc1b 100644
@@ -336,7 +336,6 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res)
 {
 
 	struct vmw_private *dev_priv = res->dev_priv;
-	struct vmw_surface *srf;
 	void *cmd;
 
 	if (res->func->destroy == vmw_gb_surface_destroy) {
@@ -360,7 +359,6 @@ static void vmw_hw_surface_destroy(struct vmw_resource *res)
 		 */
 
 		mutex_lock(&dev_priv->cmdbuf_mutex);
-		srf = vmw_res_to_srf(res);
 		dev_priv->used_memory_size -= res->backup_size;
 		mutex_unlock(&dev_priv->cmdbuf_mutex);
 	}
Signature · 6 lines
2.21.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#5 lore raw
In reply to Thomas Hellström (VMware)
Quoted 6 lines · Thomas Hellström (VMware)

From: Thomas Hellstrom <[email protected]>

With AMD-SEV high-bandwidth port messaging runs into trouble since the message content is encrypted using the vm-specific key, and the hypervisor is unable to read it.

Quoted 6 lines

So use unencrypted dma-coherent bounce buffers for temporary message storage space. Allocating that memory is expensive so a future optimization might include a static unencrypted memory area for messages.

Hi Thomas,

The 0day team has been doing clang builds for us and sending the results to our mailing list for triage; this patch causes the following warning. Seems legitimate, mind taking a look at it and resolving it how you see fit?

Cheers,
Nathan

Quoted 127 lines · kbuild test robot

CC: [email protected]
In-Reply-To: <[email protected]>
References: <[email protected]>
TO: "Thomas Hellström (VMware)" <[email protected]>
CC: [email protected], Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>, Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>
CC: Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>

Hi "Thomas,

I love your patch! Perhaps something to improve:

[auto build test WARNING on linus/master]
[also build test WARNING on v5.4-rc7 next-20191113]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Thomas-Hellstr-m-VMware/drm-vmwgfx-Use-dma-coherent-memory-for-high-bandwidth-port-messaging/20191114-020818
base:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 0e3f1ad80fc8cb0c517fd9a9afb22752b741fa76
config: x86_64-rhel-7.6 (attached as .config)
compiler: clang version 10.0.0 (git://gitmirror/llvm_project 335ac2eb662ce5f1888e2a50310b01fba2d40d68)
reproduce:
        # save the attached .config to linux build tree
        make ARCH=x86_64 

If you fix the issue, kindly add following tag

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:441:6: warning: variable 'reply_handle' is used uninitialized whenever '||' condition is true [-Wsometimes-uninitialized]

           if (vmw_send_msg(&channel, msg) ||
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~
   drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:467:47: note: uninitialized use occurs here
           dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
                                                        ^~~~~~~~~~~~
   drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:441:6: note: remove the '||' if its condition is always false
           if (vmw_send_msg(&channel, msg) ||
               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:421:37: note: initialize the variable 'reply_handle' to silence this warning
           dma_addr_t req_handle, reply_handle;
                                              ^
                                               = 0

1 warning generated.

vim +441 drivers/gpu/drm/vmwgfx/vmwgfx_msg.c

89da76fde68de1 Sinclair Yeh      2016-04-27  400  
89da76fde68de1 Sinclair Yeh      2016-04-27  401  
89da76fde68de1 Sinclair Yeh      2016-04-27  402  /**
89da76fde68de1 Sinclair Yeh      2016-04-27  403   * vmw_host_get_guestinfo: Gets a GuestInfo parameter
89da76fde68de1 Sinclair Yeh      2016-04-27  404   *
89da76fde68de1 Sinclair Yeh      2016-04-27  405   * Gets the value of a  GuestInfo.* parameter.  The value returned will be in
89da76fde68de1 Sinclair Yeh      2016-04-27  406   * a string, and it is up to the caller to post-process.
89da76fde68de1 Sinclair Yeh      2016-04-27  407   *
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  408   * @dev: Pointer to struct device used for coherent memory allocation
89da76fde68de1 Sinclair Yeh      2016-04-27  409   * @guest_info_param:  Parameter to get, e.g. GuestInfo.svga.gl3
89da76fde68de1 Sinclair Yeh      2016-04-27  410   * @buffer: if NULL, *reply_len will contain reply size.
89da76fde68de1 Sinclair Yeh      2016-04-27  411   * @length: size of the reply_buf.  Set to size of reply upon return
89da76fde68de1 Sinclair Yeh      2016-04-27  412   *
89da76fde68de1 Sinclair Yeh      2016-04-27  413   * Returns: 0 on success
89da76fde68de1 Sinclair Yeh      2016-04-27  414   */
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  415  int vmw_host_get_guestinfo(struct device *dev, const char *guest_info_param,
89da76fde68de1 Sinclair Yeh      2016-04-27  416  			   char *buffer, size_t *length)
89da76fde68de1 Sinclair Yeh      2016-04-27  417  {
89da76fde68de1 Sinclair Yeh      2016-04-27  418  	struct rpc_channel channel;
89da76fde68de1 Sinclair Yeh      2016-04-27  419  	char *msg, *reply = NULL;
6073a09210e06f Himanshu Jha      2018-03-22  420  	size_t reply_len = 0;
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  421  	dma_addr_t req_handle, reply_handle;
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  422  	int req_len = strlen("info-get ") + strlen(guest_info_param) + 1;
89da76fde68de1 Sinclair Yeh      2016-04-27  423  
89da76fde68de1 Sinclair Yeh      2016-04-27  424  	if (!vmw_msg_enabled)
89da76fde68de1 Sinclair Yeh      2016-04-27  425  		return -ENODEV;
89da76fde68de1 Sinclair Yeh      2016-04-27  426  
89da76fde68de1 Sinclair Yeh      2016-04-27  427  	if (!guest_info_param || !length)
89da76fde68de1 Sinclair Yeh      2016-04-27  428  		return -EINVAL;
89da76fde68de1 Sinclair Yeh      2016-04-27  429  
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  430  	msg = dma_alloc_coherent(dev, req_len, &req_handle, GFP_KERNEL);
1a4adb05632e90 Ravikant B Sharma 2016-11-08  431  	if (!msg) {
3fbeccf8ceb165 Thomas Hellstrom  2018-06-20  432  		DRM_ERROR("Cannot allocate memory to get guest info \"%s\".",
3fbeccf8ceb165 Thomas Hellstrom  2018-06-20  433  			  guest_info_param);
89da76fde68de1 Sinclair Yeh      2016-04-27  434  		return -ENOMEM;
89da76fde68de1 Sinclair Yeh      2016-04-27  435  	}
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  436  	snprintf(msg, req_len, "info-get %s", guest_info_param);
89da76fde68de1 Sinclair Yeh      2016-04-27  437  
f37230c0ad4810 Thomas Hellstrom  2018-05-23  438  	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM))
f37230c0ad4810 Thomas Hellstrom  2018-05-23  439  		goto out_open;
89da76fde68de1 Sinclair Yeh      2016-04-27  440  
f37230c0ad4810 Thomas Hellstrom  2018-05-23 @441  	if (vmw_send_msg(&channel, msg) ||
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  442  	    vmw_recv_msg(dev, &channel, (void *) &reply, &reply_len,
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  443  			 &reply_handle))
f37230c0ad4810 Thomas Hellstrom  2018-05-23  444  		goto out_msg;
89da76fde68de1 Sinclair Yeh      2016-04-27  445  
f37230c0ad4810 Thomas Hellstrom  2018-05-23  446  	vmw_close_channel(&channel);
89da76fde68de1 Sinclair Yeh      2016-04-27  447  	if (buffer && reply && reply_len > 0) {
89da76fde68de1 Sinclair Yeh      2016-04-27  448  		/* Remove reply code, which are the first 2 characters of
89da76fde68de1 Sinclair Yeh      2016-04-27  449  		 * the reply
89da76fde68de1 Sinclair Yeh      2016-04-27  450  		 */
89da76fde68de1 Sinclair Yeh      2016-04-27  451  		reply_len = max(reply_len - 2, (size_t) 0);
89da76fde68de1 Sinclair Yeh      2016-04-27  452  		reply_len = min(reply_len, *length);
89da76fde68de1 Sinclair Yeh      2016-04-27  453  
89da76fde68de1 Sinclair Yeh      2016-04-27  454  		if (reply_len > 0)
89da76fde68de1 Sinclair Yeh      2016-04-27  455  			memcpy(buffer, reply + 2, reply_len);
89da76fde68de1 Sinclair Yeh      2016-04-27  456  	}
89da76fde68de1 Sinclair Yeh      2016-04-27  457  
89da76fde68de1 Sinclair Yeh      2016-04-27  458  	*length = reply_len;
89da76fde68de1 Sinclair Yeh      2016-04-27  459  
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  460  	dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  461  	dma_free_coherent(dev, req_len, msg, req_handle);
89da76fde68de1 Sinclair Yeh      2016-04-27  462  
f37230c0ad4810 Thomas Hellstrom  2018-05-23  463  	return 0;
f37230c0ad4810 Thomas Hellstrom  2018-05-23  464  
f37230c0ad4810 Thomas Hellstrom  2018-05-23  465  out_msg:
f37230c0ad4810 Thomas Hellstrom  2018-05-23  466  	vmw_close_channel(&channel);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  467  	dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
f37230c0ad4810 Thomas Hellstrom  2018-05-23  468  out_open:
f37230c0ad4810 Thomas Hellstrom  2018-05-23  469  	*length = 0;
f37230c0ad4810 Thomas Hellstrom  2018-05-23  470  
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  471  	DRM_ERROR("Failed to get guest info \"%s\".", guest_info_param);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  472  	dma_free_coherent(dev, req_len, msg, req_handle);
f37230c0ad4810 Thomas Hellstrom  2018-05-23  473  	return -EINVAL;
89da76fde68de1 Sinclair Yeh      2016-04-27  474  }
89da76fde68de1 Sinclair Yeh      2016-04-27  475  

:::::: The code at line 441 was first introduced by commit
:::::: f37230c0ad481091bc136788ff8b37dc86300c6d drm/vmwgfx: Fix host logging / guestinfo reading error paths

Quoted 6 lines

:::::: TO: Thomas Hellstrom <[email protected]> :::::: CC: Thomas Hellstrom <[email protected]>

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/hyperkitty/list/[email protected] Intel Corporation

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#6 lore raw

Hi, Nathan,

Quoted 15 lines · Nathan Chancellor
Quoted 6 lines · Thomas Hellström (VMware)

From: Thomas Hellstrom <[email protected]>

With AMD-SEV high-bandwidth port messaging runs into trouble since the message content is encrypted using the vm-specific key, and the hypervisor is unable to read it.

Quoted 6 lines

So use unencrypted dma-coherent bounce buffers for temporary message storage space. Allocating that memory is expensive so a future optimization might include a static unencrypted memory area for messages.

Hi Thomas,

Quoted 7 lines

The 0day team has been doing clang builds for us and sending the results to our mailing list for triage; this patch causes the following warning. Seems legitimate, mind taking a look at it and resolving it how you see fit?

Cheers,
Nathan

This should be harmless as dma_free_coherent() with reply == NULL is a nop, but anyway I'll respin to silence the warning.

Thanks,

Thomas

Quoted 128 lines · kbuild test robot
Quoted 127 lines

CC: [email protected]
In-Reply-To: <[email protected]>
References: <[email protected]>
TO: "Thomas Hellström (VMware)" <[email protected]>
CC: [email protected], Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>, Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>
CC: Thomas Hellstrom <[email protected]>, Brian Paul <[email protected]>

Hi "Thomas,

I love your patch! Perhaps something to improve:

[auto build test WARNING on linus/master]
[also build test WARNING on v5.4-rc7 next-20191113]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Thomas-Hellstr-m-VMware/drm-vmwgfx-Use-dma-coherent-memory-for-high-bandwidth-port-messaging/20191114-020818
base:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 0e3f1ad80fc8cb0c517fd9a9afb22752b741fa76
config: x86_64-rhel-7.6 (attached as .config)
compiler: clang version 10.0.0 (git://gitmirror/llvm_project 335ac2eb662ce5f1888e2a50310b01fba2d40d68)
reproduce:
         # save the attached .config to linux build tree
         make ARCH=x86_64

If you fix the issue, kindly add following tag

All warnings (new ones prefixed by >>):

Quoted 1 line
Quoted 1 line

drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:441:6: warning: variable 'reply_handle' is used uninitialized whenever '||' condition is true [-Wsometimes-uninitialized]

            if (vmw_send_msg(&channel, msg) ||
                ^~~~~~~~~~~~~~~~~~~~~~~~~~~
    drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:467:47: note: uninitialized use occurs here
            dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
                                                         ^~~~~~~~~~~~
    drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:441:6: note: remove the '||' if its condition is always false
            if (vmw_send_msg(&channel, msg) ||
                ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    drivers/gpu/drm/vmwgfx/vmwgfx_msg.c:421:37: note: initialize the variable 'reply_handle' to silence this warning
            dma_addr_t req_handle, reply_handle;
                                               ^
                                                = 0
    1 warning generated.

vim +441 drivers/gpu/drm/vmwgfx/vmwgfx_msg.c

89da76fde68de1 Sinclair Yeh      2016-04-27  400
89da76fde68de1 Sinclair Yeh      2016-04-27  401
89da76fde68de1 Sinclair Yeh      2016-04-27  402  /**
89da76fde68de1 Sinclair Yeh      2016-04-27  403   * vmw_host_get_guestinfo: Gets a GuestInfo parameter
89da76fde68de1 Sinclair Yeh      2016-04-27  404   *
89da76fde68de1 Sinclair Yeh      2016-04-27  405   * Gets the value of a  GuestInfo.* parameter.  The value returned will be in
89da76fde68de1 Sinclair Yeh      2016-04-27  406   * a string, and it is up to the caller to post-process.
89da76fde68de1 Sinclair Yeh      2016-04-27  407   *
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  408   * @dev: Pointer to struct device used for coherent memory allocation
89da76fde68de1 Sinclair Yeh      2016-04-27  409   * @guest_info_param:  Parameter to get, e.g. GuestInfo.svga.gl3
89da76fde68de1 Sinclair Yeh      2016-04-27  410   * @buffer: if NULL, *reply_len will contain reply size.
89da76fde68de1 Sinclair Yeh      2016-04-27  411   * @length: size of the reply_buf.  Set to size of reply upon return
89da76fde68de1 Sinclair Yeh      2016-04-27  412   *
89da76fde68de1 Sinclair Yeh      2016-04-27  413   * Returns: 0 on success
89da76fde68de1 Sinclair Yeh      2016-04-27  414   */
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  415  int vmw_host_get_guestinfo(struct device *dev, const char *guest_info_param,
89da76fde68de1 Sinclair Yeh      2016-04-27  416  			   char *buffer, size_t *length)
89da76fde68de1 Sinclair Yeh      2016-04-27  417  {
89da76fde68de1 Sinclair Yeh      2016-04-27  418  	struct rpc_channel channel;
89da76fde68de1 Sinclair Yeh      2016-04-27  419  	char *msg, *reply = NULL;
6073a09210e06f Himanshu Jha      2018-03-22  420  	size_t reply_len = 0;
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  421  	dma_addr_t req_handle, reply_handle;
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  422  	int req_len = strlen("info-get ") + strlen(guest_info_param) + 1;
89da76fde68de1 Sinclair Yeh      2016-04-27  423
89da76fde68de1 Sinclair Yeh      2016-04-27  424  	if (!vmw_msg_enabled)
89da76fde68de1 Sinclair Yeh      2016-04-27  425  		return -ENODEV;
89da76fde68de1 Sinclair Yeh      2016-04-27  426
89da76fde68de1 Sinclair Yeh      2016-04-27  427  	if (!guest_info_param || !length)
89da76fde68de1 Sinclair Yeh      2016-04-27  428  		return -EINVAL;
89da76fde68de1 Sinclair Yeh      2016-04-27  429
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  430  	msg = dma_alloc_coherent(dev, req_len, &req_handle, GFP_KERNEL);
1a4adb05632e90 Ravikant B Sharma 2016-11-08  431  	if (!msg) {
3fbeccf8ceb165 Thomas Hellstrom  2018-06-20  432  		DRM_ERROR("Cannot allocate memory to get guest info \"%s\".",
3fbeccf8ceb165 Thomas Hellstrom  2018-06-20  433  			  guest_info_param);
89da76fde68de1 Sinclair Yeh      2016-04-27  434  		return -ENOMEM;
89da76fde68de1 Sinclair Yeh      2016-04-27  435  	}
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  436  	snprintf(msg, req_len, "info-get %s", guest_info_param);
89da76fde68de1 Sinclair Yeh      2016-04-27  437
f37230c0ad4810 Thomas Hellstrom  2018-05-23  438  	if (vmw_open_channel(&channel, RPCI_PROTOCOL_NUM))
f37230c0ad4810 Thomas Hellstrom  2018-05-23  439  		goto out_open;
89da76fde68de1 Sinclair Yeh      2016-04-27  440
f37230c0ad4810 Thomas Hellstrom  2018-05-23 @441  	if (vmw_send_msg(&channel, msg) ||
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  442  	    vmw_recv_msg(dev, &channel, (void *) &reply, &reply_len,
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  443  			 &reply_handle))
f37230c0ad4810 Thomas Hellstrom  2018-05-23  444  		goto out_msg;
89da76fde68de1 Sinclair Yeh      2016-04-27  445
f37230c0ad4810 Thomas Hellstrom  2018-05-23  446  	vmw_close_channel(&channel);
89da76fde68de1 Sinclair Yeh      2016-04-27  447  	if (buffer && reply && reply_len > 0) {
89da76fde68de1 Sinclair Yeh      2016-04-27  448  		/* Remove reply code, which are the first 2 characters of
89da76fde68de1 Sinclair Yeh      2016-04-27  449  		 * the reply
89da76fde68de1 Sinclair Yeh      2016-04-27  450  		 */
89da76fde68de1 Sinclair Yeh      2016-04-27  451  		reply_len = max(reply_len - 2, (size_t) 0);
89da76fde68de1 Sinclair Yeh      2016-04-27  452  		reply_len = min(reply_len, *length);
89da76fde68de1 Sinclair Yeh      2016-04-27  453
89da76fde68de1 Sinclair Yeh      2016-04-27  454  		if (reply_len > 0)
89da76fde68de1 Sinclair Yeh      2016-04-27  455  			memcpy(buffer, reply + 2, reply_len);
89da76fde68de1 Sinclair Yeh      2016-04-27  456  	}
89da76fde68de1 Sinclair Yeh      2016-04-27  457
89da76fde68de1 Sinclair Yeh      2016-04-27  458  	*length = reply_len;
89da76fde68de1 Sinclair Yeh      2016-04-27  459
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  460  	dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  461  	dma_free_coherent(dev, req_len, msg, req_handle);
89da76fde68de1 Sinclair Yeh      2016-04-27  462
f37230c0ad4810 Thomas Hellstrom  2018-05-23  463  	return 0;
f37230c0ad4810 Thomas Hellstrom  2018-05-23  464
f37230c0ad4810 Thomas Hellstrom  2018-05-23  465  out_msg:
f37230c0ad4810 Thomas Hellstrom  2018-05-23  466  	vmw_close_channel(&channel);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  467  	dma_free_coherent(dev, reply_len + 1, reply, reply_handle);
f37230c0ad4810 Thomas Hellstrom  2018-05-23  468  out_open:
f37230c0ad4810 Thomas Hellstrom  2018-05-23  469  	*length = 0;
f37230c0ad4810 Thomas Hellstrom  2018-05-23  470
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  471  	DRM_ERROR("Failed to get guest info \"%s\".", guest_info_param);
6bdb21230a2a01 Thomas Hellstrom  2019-11-13  472  	dma_free_coherent(dev, req_len, msg, req_handle);
f37230c0ad4810 Thomas Hellstrom  2018-05-23  473  	return -EINVAL;
89da76fde68de1 Sinclair Yeh      2016-04-27  474  }
89da76fde68de1 Sinclair Yeh      2016-04-27  475

:::::: The code at line 441 was first introduced by commit
:::::: f37230c0ad481091bc136788ff8b37dc86300c6d drm/vmwgfx: Fix host logging / guestinfo reading error paths

Quoted 6 lines
Quoted 6 lines

:::::: TO: Thomas Hellstrom <[email protected]> :::::: CC: Thomas Hellstrom <[email protected]>

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/hyperkitty/list/[email protected] Intel Corporation

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel