Skip to contentSkip to conversation

coresight: tmc: allocating memory when needed

DISCUSSION Mathieu Poirier 1 participant 1 message Apr 4, 2016 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 20160402053209.GA15353@mwanda

#1 lore raw
Quoted 11 lines · Dan Carpenter

Hello Mathieu Poirier,

The patch 06ec4671e5d1: "coresight: tmc: allocating memory when needed" from Dec 9, 2015, leads to the following static checker warning:

drivers/hwtracing/coresight/coresight-tmc-etf.c:125 tmc_enable_etf_sink_sysfs()
warn: passing devm_ allocated variable to kfree. 'buf'

drivers/hwtracing/coresight/coresight-tmc-etf.c
109 static int tmc_enable_etf_sink_sysfs(struct coresight_device *csdev, u32 mode)

Quoted 7 lines
   110  {
   111          u32 val;
   112          bool allocated = false;
   113          char *buf = NULL;
                      ^^^^^^^^^^

This bogus initialization disables GCC's checks for uninitialized memory. Don't disable static analysis for no reason.

Yes, this one doesn't do anything.

Quoted 14 lines
   114          unsigned long flags;
   115          struct tmc_drvdata *drvdata = dev_get_drvdata(csdev->dev.parent);
   116
   117          /* Allocating the memory here while outside of the spinlock */
   118          buf = devm_kzalloc(drvdata->dev, drvdata->size, GFP_KERNEL);
   119          if (!buf)
   120                  return -ENOMEM;
   121
   122          spin_lock_irqsave(&drvdata->spinlock, flags);
   123          if (drvdata->reading) {
   124                  spin_unlock_irqrestore(&drvdata->spinlock, flags);
   125                  kfree(buf);
                        ^^^^^^^^^^
Quoted 5 lines

This is a double free because the buffer is allocated with devm_kzalloc(). There are lots more double frees. We double free in tmc_read_unprepare_etf() as well. It's hard to find all the bugs without knowing the code more than I do. Please review and test unloading.

This is obviously the wrong kind of memory allocator and the same will be true on the ETF side since they were tailored to be as similar as possible. I'll revise this.

Thanks,
Mathieu

Quoted 7 lines
   126                  return -EBUSY;
   127          }
   128

regards,
dan carpenter